MODALITY IN GRAMMAR AND DISCOURSE

TYPOLOGICAL STUDIES IN LANGUAGE (TSL)

A companion series to the journal "STUDIES IN LANGUAGE"

Honorary Editor: Joseph H. Greenberg

General Editor: T. Givón
Associate General Editor: Michael Noonan
Assistant Editors: Spike Gildea, Suzanne Kemmer

Editorial Board:
Wallace Chafe (Santa Barbara) Ronald Langacker (San Diego)
Bernard Comrie (Los Angeles) Charles Li (Santa Barbara)
R.M.W. Dixon (Canberra) Andrew Pawley (Canberra)
Matthew Dryer (Buffalo) Doris Payne (Oregon)
John Haiman (St Paul) Frans Plank (Konstanz)
Kenneth Hale (Cambridge, Mass.) Jerrold Sadock (Chicago)
Bemd Heine (Köln) Dan Slobin (Berkeley)
Paul Hopper (Pittsburgh) Sandra Thompson (Santa Barbara)
Andrej Kibrik (Moscow)

Volumes in this series will be functionally and typologically oriented, covering specific
topics in language by collecting together data from a wide variety of languages and language
typologies. The orientation of the volumes will be substantive rather than formal, with the
aim of investigating universals of human language via as broadly defined a data base as
possible, leaning toward cross-linguistic, diachronic, developmental and live-discourse data.
The series is, in spirit as well as in fact, a continuation of the tradition initiated by C. Li (Word
Order and Word Order Change, Subject and Topic, Mechanisms for Syntactic Change) and
continued by T. Givón (Discourse and Syntax) and P. Hopper (Tense-Aspect: Between
Semantics and Pragmatics).

Volume 32

Joan Bybee and Suzanne Fleischman (eds)

Modality in Grammar and Discourse

 MODALITY
IN GRAMMAR
AND DISCOURSE
Edited by

JOAN BYBEE
University of New Mexico
SUZANNE FLEISCHMAN
University of California, Berkeley

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
 The paper used in this publication meets the minimum requirements of
American National Standard for Information Sciences — Permanence of
Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Modality in grammar and discourse / edited by Joan Bybee, Suzanne Fleischman.
p. cm. — (Typological studies in language, ISSN 0167-7373; v. 32)
Papers from a symposium on Mood and Modality held at the University of New Mexico in
1992.
Includes bibliographical references and index.
1. Modality (Linguistics)--Congresses. 2. Grammar, Comparative and general—Mood-
Congresses. I. Bybee, Joan L. II. Fleischman, Suzarne. III. Series.
P299.M6M64 1995
415--dc20 95-17034
ISBN 90 272 2925 2 (hb.) / 90 272 2926 0 (pb.) (European; alk. paper) CIP
ISBN 1-55619-639-3 (hb.) / 1-55619-640-7 (pb.) (U.S.; alk. paper)
© Copyright 1995 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands
John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
To the memory of Dwight Bolinger
 Contents

Introduction 1

I. Agent-Oriented and Epistemic Modality 15

Agent-Oriented vs. Epistemic Modality: Some Observations on
German Modals 17
Bernd Heine
The Expression of Root and Epistemic Possibility in English 55
Jennifer Coates
Contextuel Conditions for the Interpretation of poder and deber in
Spanish 67
Carmen Silva-Corvalán
The Obligation Modality in Western Nilotic Languages 107
Edith Bavin
The Gestural Expression of Modality in ASL 135
Sherman Wilcox and Phyllis Wilcox

II The Interactional Basis of Modality 163

The Development of Epistemic Sentence-ending Modal Forms and
Functions in Korean Children 165
Soonja Choi
The Interactional Basis of the Mandarin Modal néng 'can' 205
Jiansheng Guo
The Discourse and Interactive Functions of Obligation Expressions 239
John Myhill and Laura A. Smith
Apprehensional Epistemics 293
Frantisek Lichtenberk
Moods and MetaMessages: Alienation as a Mood 329
John Raiman
Viii Contents

III Irrealis Modality and Subjunctive 347

The Realis-Irrealis Distinction in Caddo, the Northern Iroquoian
Languages, and English 349
Wallace Chafe
On the Relativity of Irreality 367
Marianne Mithun
The Grammaticalization of Irrealis in Tok Pisin 389
Suzanne Romaine
The Evaluative Function of the Spanish Subjunctive 429
Patricia V Lunn

IV Modality and Other Categories of Grammar 451

Negation and the Modals of Possibility and Necessity 453
Frank Palmer
A Functional Theory of Complementizers 473
Zygmunt Frajzyngier
The Semantic Development of Past Tense Modals in English 503
John Bybee
Imperfective and Irrealis 519
Suzanne Fleischman

Subject index 553

Language index 565
Author index 569
 Modality in Grammar and Discourse
An Introductory Essay*

Joan Bybee
University of New Mexico
Suzanne Fleischman
University of California, Berkeley

This volume is the result of a symposium on Mood and Modality held at the
University of New Mexico in 1992, the goal of which was to bring together
linguists whose research has targeted this area of grammar but whose ap­
proaches to it reflect differing perspectives on functional linguistics. The
symposium was planned so as to include a diversity of languages, of foci
(synchronic and diachronic), and of theoretical orientations, especially with
regard to the interaction of morphosyntactic, semantic and discourse-prag­
matic factors.
Given the complexity of this linguistic domain, the many and diverse
ways it comes to be expressed in different languages, and the tendency of
researchers to work narrowly within the confines of their own theoretical and
methodological frameworks, there was reason to fear that symposium partici­
pants might not find adequate common ground nor a sufficiently common
metalanguage to be able to communicate with one another. However, just as a
similar symposium on tense and aspect a decade earlier (Hopper 1982)
confirmed the status of those categories as valid cross-language categories of
grammar—what we refer to, following Bybee & Dahl (1989), as 'gram
types'—, so too the current symposium succeeded in demonstrating that
despite differences in terminology, language areas, and theoretical perspec­
tives, we were in effect all examining similar phenomena and could partici­
pate in meaningful dialogue about our data and analyses. Thus while the
papers in this volume present a wide range of topics and perspectives, they
2 Modality in Grammar and Discourse

nonetheless converge around a number of key issues, and in the aggregate

seem to have succeeded in moving us toward a better understanding of the
functions of modality and its forms of expression in natural language.
In the course of the symposium several essential issues came up repeat­
edly regarding both the categories in question and the levels of linguistic
analysis at which they operate. One of the most basic of these issues concerns
the relationship between 'mood' and 'modality'.

Mood and modality

As used here, mood refers to a formally grammaticalized category of the verb

which has a modal function. Moods are expressed inflectionally, generally in
distinct sets of verbal paradigms, e.g. indicative, subjunctive, optative, im­
perative, conditional, etc., which vary from one language to another in respect
to number as well as to the semantic distinctions they mark. Modality, on the
other hand, is the semantic domain pertaining to elements of meaning that
languages express. It covers a broad range of semantic nuances—jussive,
desiderative, intentive, hypothetical, potential, obligative, dubitative, horta­
tory, exclamative, etc.—whose common denominator is the addition of a
supplement or overlay of meaning to the most neutral semantic value of the
proposition of an utterance, namely factual and declarative.1
In the terms of the framework set forth in Bybee and Dahl (1989),
modality is a semantic domain, while moods, as formal categories of gram­
mar, can be either cross-language gram types (e.g. conditional or subjunctive)
or language-specific categories (e.g. the Delayed Imperative in Buriat, an
Altaic language; note that we distinguish these two levels by use of an initial
upper case letter to signal language- or family-specific categories, retaining
lower case for cross-language gram types.)
Modality is expressed in language in a variety of ways: morphological,
lexical, syntactic, or via intonation. These are not mutually exclusive. Thus in
the Spanish sentence dudo que haya ganado el premio T doubt (that) he won
the prize', the 'dubitative' modality is conveyed redundantly by both the
lexical meaning of the main verb and the subjunctive mood of the subordi­
nate-clause verb. In this volume we will be concerned primarily with gram­
matical (morphological and syntactic) expressions of modality, including
forms that may be currently undergoing grammaticalization.
An Introductory Essay 3

Function and its relation to linguistic form

A second issue we wish to clarify at the outset of our discussion concerns the
relationship between domains of modality and their expression in natural
languages. In some approaches to modality, function is studied for its own
sake. In this volume we take the position that analysis of function should
explain distribution of form. That is, the modal categories we operate with do
not lead an autonomous existence in some abstract logical or semantic space;
rather, they correspond to—indeed are determined by—(a) the formal distinc­
tions made in particular languages, (b) documented pathways of language
change, and (c) prominent cross-language patterns of form-function correla­
tion.
In the area of modality, however, cross-language comparison has been a
difficult task, for several reasons. First, because the semantic/functional do­
main of modality is so broad; second, because modality, as we have discov­
ered, lends itself best to investigation in social, interactive contexts
(elaborated below); third, and conceivably most important, because of the
extent to which languages differ in their mapping of the relevant semantic
content onto linguistic form. A case in point that came up repeatedly in the
course of our discussions concerns the category irrealis and the nature of the
realis/irrealis distinction. The languages that came under our scrutiny differed
in terms of what they classify as realis and what they classify as irrealis,
prompting us to question whether the distinction is cross-linguistically valid
at all, and if so, whether it corresponds to a gram-type distinction, such as
perfective/imperfective, or whether it more closely resembles a supercategory
such as mood.
As noted above, it also became clear over the course of our discussions at
the symposium that many of the functions of modality are inextricably em­
bedded in contexts of social interaction and, consequently, cannot be de­
scribed adequately apart from their contextual moorings in interactive
discourse. Several papers in this volume explore particular modalities specifi­
cally as they occur in contexts of face-to-face communication. But even in
those that do not make explicit reference to contexts of social interaction, this
factor is nonetheless in evidence.
4 Modality in Grammar and Discourse

Modal categories and their associated nomenclature

Our linguistic understanding of modality has its roots in modal logic (a

branch of philosophy of language) and in particular in the distinction between
'deontic' and 'epistemic' modality. Modal logic has to do with the notions of
possibility and necessity, and its categories epistemic and deontic concern
themselves with these notions in two different domains. Epistemic modality
has to do with the possibility or necessity of the truth of propositions, and is
thus involved with knowledge and belief (Lyons 1977:793). Deontic modal­
ity, on the other hand, is concerned with the necessity or possibility of acts
performed by morally responsible agents (Lyons 1977: 823), and is thus
associated with the social functions of permission and obligation.
The epistemic notion is of considerable use to linguists, given that many
languages have grammatical markers which function explicitly to express an
evaluation of the truth of a proposition. Accordingly, most linguists under­
stand epistemic modality as expressing the degree of a speaker's commitment
to the truth of the proposition contained in an utterance. Thus, one way
epistemic possibility is expressed in English is by may and might (we may/
might lose the election), while epistemic necessity is expressed by must (they
must have won the election). However, as applied to natural language, there is
no reason to restrict the epistemic notion just to necessity and possibility, as is
traditional in philosophy of language. For one thing, commitment to the truth
of a proposition is often a matter of degree. For another, epistemic modality
can be seen as overlapping with, or even encompassing, another grammatical
category, namely evidentiality 2 For our purpose, the former expansion of the
epistemic notion will be adhered to (i.e. construal along a continuum) but not
the latter. The symposium from which this volume emerges deliberately
excluded the study of evidentials, which were the subject of an earlier meeting
in this same series (see Chafe and Nichols 1986).
Deontic modality has also proven to be a useful concept for linguists;
however, its translation into linguistic categories has not been as smooth as in
the case of epistemic modality (cf. Bybee 1985, Bybee, Pagliuca and Perkins
1991), As understood in philosophy of language, deontic modality focuses on
the notions of obligation and permission. It is found in directives that grant
permission (you may go now) or impose obligations (eat your vegetables!), as
well as in statements that report deontic conditions (Yeltsin should slow down
reforms in Russia; graduate students can check out books for the whole semester).
An Introductory Essay 5

A problem with the deontic notion for linguists, however, is the imperfect
nature of its fit with the corresponding linguistic categories that we encounter
in the world's languages as well as in language change: i.e., it is at once too
broad and too narrow. For one thing, unlike 'agent-oriented' modality—a
supercategory label that will be used by most papers in this volume in
preference to 'deontic' 3 —deontic modality fails to distinguish subcategories
that are expressed inflectionally (i.e. as grammaticalized moods), such as
imperative, from lexical ox periphrastic (i.e. auxiliary) expressions of obliga­
tion or permission. Furthermore, deontic modality as traditionally understood
excludes certain semantically related notions such as ability (physical and
mental) and desire that have linguistic expression similar to that of permission
and obligation. While one argument for the category 'deontic' might be the
well-documented pathway of change whereby deontic modals over time come
to acquire epistemic functions, in actual fact this change affects a broader
range of meanings than the term 'deontic' indicates.
The traditional division of modality into epistemic and deontic reveals
some interesting cases of polysemy in which the same form can be used for
both types of modality. Thus English may can express either deontic permis­
sion (you may come in now) or epistemic possibility (this may be your lucky
day!), while must can express deontic obligation (you must be here by seven)
as well as inferred probability (that must be the mailman at the door). A
diachronic view of this polysemy yields the observation, documented in many
languages, that so-called deontic meanings typically evolve into epistemic
meanings. 4
With regard to the difficulties attaching to 'deontic' as a supercategory
label, even this change from deontic to epistemic meaning in fact affects a
broader range of categories. While it is true that obligation markers may come
to be used for epistemic functions such as probability or inference, in the case
of permission markers it is not 'permission' per se that licenses a meaning of
epistemic possibility (e.g. in the case of may). Virtually all permission mark­
ers can be traced back to expressions of ability, permission being just one
sense of a more generalized 'root-possibility' meaning that arises from ability
(Bybee, Perkins and Pagliuca 1994). Root possibility predicates general ena­
bling conditions (e.g. it can take three hours to get there). These include
permission, which is a social enabling condition. Bybee (1988) has shown
that it is the root possibility sense that gives rise to epistemic possibility.
A second instance of categories other than narrowly-defined deontic
6 Modality in Grammar and Discourse

categories developing epistemic meanings is that of verbs indicating desire

(and obligation) evolving into futures. This development parallels the deontic
to epistemic shift in that a change occurs from a modal expression predicating
conditions on an agent—an 'agent-oriented' modality—to a modal expression
that has an entire proposition in its scope and communicates the speaker's
stance with regard to the truth of that proposition—a 'speaker-oriented'
modality.
For the reasons suggested above, Bybee (1985) proposed a change in the
categorial nomenclature of modals as follows: Agent-oriented modality
encompasses all modal meanings that predicate conditions on an agent with
regard to the completion of an action referred to by the main predicate, e.g.
obligation, desire, ability, permission and root possibility. Epistemic modal­
ity retains its traditional definition: epistemics are clausal-scope indicators of
a speaker's commitment to the truth of a proposition. Markers of directives,
such as imperatives, optatives or permissives, which represent speech acts
through which a speaker attempts to move an addressee to action, are called
speaker-oriented.
It will be observed that the distinction between agent-oriented and
speaker-oriented modalities cross-cuts the traditional category of deontic
modality. Agent-oriented modals include deontic statements (statements that
describe obligations and permission), while speaker-oriented modals include
speech-act types such as imperatives that impose conditions of obligation. A
prime motive for replacing the deontic category by an agent-oriented category
is that the latter better reflects general morphosyntactic trends in expression
type: i.e., there is a strong, quasi-universal tendency for agent-oriented modal­
ity to be expressed by verbs, auxiliaries or non-bound particles, whereas the
remaining two types (speaker-oriented and epistemic) are often expressed
inflectionally (Bybee 1985). A second universal pattern is diachronic, namely
the tendency referred to above whereby the agent-oriented modalities develop
predictably into the other two types (Bybee, Perkins and Pagliuca 1994).

Agent-oriented and epistemic modality

As noted above, many modal forms exhibit a systematic polysemy between

agent-oriented and epistemic meanings. The theoretical issue raised by such
polysemy is whether a form should be regarded as having a single underlying
basic meaning (a so-called invariant meaning) that combines with contextual
An Introductory Essay 7

factors to yield differing interpretations for individual tokens, or whether,

alternatively, it has a complex set of semantic properties that by definition
incorporate meanings derived from context. A more specific issue that this
polysemy raises is that of targeting the particular contextual factors that favor
one interpretation over the other. In this regard, Bernd Heine's paper focuses
on agent-oriented modals that can also express epistemic meanings, using
data from German. His analysis isolates contextual factors as well as concep­
tual properties resident in the meanings of the modals that operate together to
enable addressees to correctly choose an agent-oriented or an epistemic
reading. Jennifer Coates' contribution builds on Heine's analysis and dis­
cusses the particular case of root possibility (in English). Applying the
conceptual properties Heine proposes to contrast agent-oriented and epis­
temic interpretations of modals, Coates shows that this distinction is weaker
with respect to the modal domain of possibility than with respect to other
areas of modal meaning.
Analyzing the Spanish modals poder 'can' and deber 'ought to,' Car­
men Silva-Corvalán also argues for the importance of context in the interpre­
tation of modal meanings. However, she rejects the idea that these modals are
inherently polysemous, positing instead an invariant meaning for each one:
'does not preclude X' for poder and 'favors, requires or entails X' for deber.
These invariant meanings are more generalized than the agent-oriented mean­
ings and require supplementation from context to produce their actual inter­
pretations in particular utterances. Many of the contextual factors
Silva-Corvalán points to resemble those proposed by Heine.
Edith Bavin takes a diachronic and comparative approach to some of
these same issues, examining the development of obligation markers in West­
ern Nilotic languages. While obligation markers in these languages have
developed from sources different from those found in European languages
(i.e. from impersonal constructions rather than from agent-oriented modal
verbs), the same sorts of contextual factors influence addressees' interpreta­
tions of them as agent-oriented or epistemic.
Sherman and Phyllis Wilcox present one of the first analyses of modal­
ity in American Sign Language, showing once again the familiar pathway of
change whereby agent-oriented modals eventually acquire epistemic mean­
ings. The Wilcoxes also isolate a set of linguistic parameters that distinguish
the meanings of ASL modals, and describe further a set of gestural parameters
that iconically mirror the semantic properties of the modal markers.
8 Modality in Grammar and Discourse

The interactional basis of modality

In recent years, an increased understanding of many grammatical categories

has come about through examination of these categories in the actual contexts
in which they are used—what is referred to as 'discourse' or 'situation'
context. Whereas for the analysis of tense, aspect, transitivity, ergativity,
reference, etc., linguists have looked for the most part at narrative discourse,
and with good result, modality cannot be studied solely with respect to
narrative, since many modal functions surface only in face-to-face interactive
discourse. That is, they typically depend not just on a monologic speaker (the
narrator in narrative discourse), but on a dialogic (explicitly or by implication)
speaker-addressee interaction. This is the case in the imposing of obligations
(through statements or through directives), in the giving and receiving of
permission, and in exchanges of information, with appropriate expressions of
commitment to the truth of that information. In fact, modals can be viewed as
strategic linguistic tools for the construction of social reality, as demonstrated
by Julie Gerhardt (1985,1990) in detailed studies of the use of modals in child
language.
Two papers in this second section of the volume highlight the social
functions of modality by analyzing chronologically the development of the
functions to which children put modal forms. Soonja Choi's paper on sen­
tence-ending particles in Korean shows how children learn to integrate infor­
mation they have to offer into the aggregate of information shared by speaker
and addressee. Because of the strong interactive function of the Korean
particles and the way they figure in the construction of appropriate discourse,
these particles are acquired earlier by Korean children than are the more
purely epistemic markers of languages such as English. Jiansheng Guo's
study of the acquisition of Mandarin neng (roughly 'can') emphasizes that
children's command of a modality marker depends on their association of that
marker with particular contexts of interaction. The meanings of modals in
general, Guo argues, are rooted in the social, interactional functions of lan­
guage; and in the case of neng, which he interprets as functioning in various
ways as a challenge to the addressee, the information-exchanging function is
clearly subordinate to the interactive function.
The interactional functions of modals can also be observed in adult
language. Using texts that represent interactive discourse in written form (i.e.,
conversations in written texts of several different genres), John Myhill and
An Introductory Essay 9

Laura Ann Smith undertake to make cross-linguistic comparisons of the use

of obligation markers in languages of different types (English, Mandarin,
Biblical Hebrew and Hopi). An important finding of their study is that
obligation expressions can carry a range of subtle nuances that render them
appropriate for carrying out speech-acts other than simply imposing obliga­
tions or making reference to obligations already in force. These include
evaluating the effects of actions, explaining actions that might be construed as
impolite, expressing lack of sympathy, and persuading one's conversational
partner to do something.
Also included in this section is Frantisek Lichtenberk's comparative
study of the functions of the 'apprehensional epistemic' modality found in
various Austronesian languages. While the methodology of his study differs
from that of the three preceding papers in this section in not being based on a
data corpus of explicitly interactive discourse, the forms Lichtenberk investi­
gates—whose functions include issuing a warning to the addressee ('watch
out! you may get sick') as well as conveying the speaker's apprehension about
something that might happen to the addressee—are clearly suited to
negotiatory discourse. Moreover, the changes they undergo in certain of the
languages investigated clearly demonstrate their dependence on speaker-
addressee interaction.
We also include in this section a highly original—and unorthodox (in the
context of traditional analyses of modality)—paper by John Haiman that
helps delimit the range of grammaticalizable modalities by targeting a set of
speaker attitudes (the 'sarcastive', the 'guiltive', the 'mass-productive') that
as far as we know have never become grammaticalized as moods. Although
sarcasm functions like a mood, insofar as it expresses a speaker's attitude
toward the proposition of an utterance produced in a dialogic exchange, its
formal markers (segmental or suprasegmental) never seem to make it into the
service sector of natural languages, i.e. they never become grammatical.
Haiman speculates on why this is the case.

Irrealis modality and subjunctive

A term widely used in discussions of modality, especially with respect to

Native American and Indo-Pacific languages, is 'irrealis', often contrasted
with 'realis.' Irrealis refers to a very broad conceptual category that covers a
wide range of non-assertive modal meanings and receives formal expression
 10 Modality in Grammar and Discourse

in certain languages. Unfortunately, it is difficult to determine in many

instances whether the modal meaning of an utterance is contributed specifi­
cally by the so-called Irrealis marker or by some other element (lexical or
morphosyntactic) of the discourse context, in which case the Irrealis marker is
functionally redundant. This ambiguity calls into question the labeling of
certain language-specific forms as Irrealis markers. A similar problem arises
in the description of Subjunctives in European languages; like Irrealis mark­
ers, Subjunctives occur in a wide range of non-assertive contexts, and the
status of their semantic content is very much in question.
A second and not insignificant problem with irrealis as a cross-language
gram-type is the degree to which languages vary in their assignment of
notional categories to the grammatical category Irrealis. Each language that
operates with this category seems to make its own determination as to which
notional categories will be considered irrealis. Thus, for both irrealis and
subjunctive, it is difficult to circumscribe a focal meaning for the gram-type.
Several approaches to the latter problem are demonstrated in the papers of our
third section.
For Wallace Chafe, the realis/irrealis distinction is based on a funda­
mental assumption on the part of language users—an assumption which
presumably operates in the same way for users of any given language—that
some of their ideas belong to the domain of objective reality while others have
their source in the imagination. This basic cognitive principle of judged
reality vs. unreality is expressed formally by the Realis/irrealis distinction in
Caddo and the Northern Iroquoian languages, albeit in slightly differing
contexts and through the use of different formal devices. Chafe sees another
instantiation of this distinction in English speakers' judgments about the
referentiality or non-referentiality of indefinite arguments.
Marianne Mithun also treats Irrealis categories in Native American
languages in a paper that foregrounds the theoretical problem of the cross-
language variability of irrealis as a gram-type. Appealing to the same underly­
ing distinction as that described by Chafe, Mithun argues that this common
cognitive distinction gets applied in different ways in different languages, a
state of affairs which accounts for the fact that Irrealis functions vary so
widely across languages. She insists, however, that this cross-linguistic varia­
tion is not random, and explains the different language-specific construals of
irrealis described in her paper as the respective outcomes of differing dia-
chronic developments.
An Introductory Essay 11

The paper by Suzanne Romaine takes a diachronic look at the Tok Pisin
particle bai, which now functions primarily as a future marker but which also
has a variety of modal functions (as do most future markers). Drawing on an
extensive corpus of data (synchronic and diachronic, spoken as well as
written), Romaine traces the stages through which the clause-initial time
adverb baimbai 'by and by' grammaticalizes into the reduced pre-verbal
marker bai and in the process acquires a set of future and 'irrealis' functions.
Her study also points out the striking conformity of these developments to
cross-linguistically established trends.
Irrealis modality is also a primary focus of Suzanne Fleischman's paper,
summarized in the section below on the interaction of modality with other
categories of grammar.
As noted above, there are certain similarities between Irrealis categories
and the Subjunctives of European languages. Patricia Lunn's paper is repre­
sentative of recent work on the Spanish Subjunctive in its appeal to pragmatic
considerations—contextual and interactional factors—to account for the use
of Subjunctive vs. Indicative forms in several varieties of discourse. Lunn
shows that Subjunctive coding is not limited to unreal and non-assertive
propositions, as suggested in traditional accounts of the Spanish Subjunctive;
this mood can also be used to signal background information in literary texts,
and in journalistic discourse to mark particular information as 'common
knowledge'.

Modality and other categories of grammar

In the final section of this volume we group together four papers that in
different ways explore the interaction between modality and other domains of
grammar, specifically: negation, complementizers, past tense, and imperfec-
tive aspect. In these papers we see once again the crucial role that interactive
contexts play in shaping the meanings that result from the combinations of
grammatical categories.
Frank Palmer's contribution investigates the systematic irregularity
(this is not an oxymoron) that we find across languages in the behavior of
modals and in the meanings that emerge when modals appear under the scope
of negation. A particularly widespread irregularity involves the strategies
languages use to express the semantic notions of 'necessary-not' and 'not-
necessary'. This may involve use of a different verb altogether from the one
 12 Modality in Grammar and Discourse

normally used in affirmative contexts, as in English mustn't vs. needn't, or a

'displacement' of the negative marker for 'necessary-not' to a syntactic
position that corresponds formally to 'not-necessary', as in French and Italian
(Fr. il ne faut pas partir, It. non deve venire).
Zygmunt Frajzyngier's paper advances the claim that in various unre­
lated languages modality appears to be expressed by complementizers, which
function to code the modality of embedded clauses. More specifically, he
links the presence of one or more complementizers (treated as a parametric
variation within the GB framework), as well as their syntactic position in
embedded clauses, to the presence or absence (synchronic or diachronic) of
certain modalities in the sentence as a whole. Of particular interest in this
paper are cases of clauses with multiple complementizers expressing different
types of modality, e.g. agent-oriented and epistemic. The paper also has
implications for formal syntactic theories that posit COMP as a component of
the sentence.
When agent-oriented modals combine with past tense the resulting unit
often undergoes a meaning change, losing the past-tense component of its
meaning and coming to signal a weakened version of its original modal
meaning in the present tense. Joan Bybee's paper explores the reasons for this
development by studying the uses of would and should in texts from Middle
English and Early Modern English. Bybee argues that these modals lose their
past sense because of the implication that a modal condition in past tense
continues into present time, pointing to the conclusion that modal meaning is
heavily influenced by the interactive contexts in which it is used.
The last paper in the volume, by Suzanne Fleischman, surveys a wide
range of evidence pointing to an overlap (synchronic and diachronic) between
the aspectual gram-type imperfective and irrealis modality, and poses the
question of why, in so many unrelated languages, verb forms marked for
imperfective aspect come to acquire meanings subsumable under the broad
modal heading of irrealis. Since none of the languages Fleischman refers to
has a formally grammaticalized Irrealis category, and in light of the notorious
elusiveness of irrealis pointed out above, her definition of irrealis is worth
noting here. She characterizes 'irrealis' as prototype category, at the semantic
level, expressing a spectrum of meanings that signal a speaker's lack of belief
in or lack of commitment to any of the following: the reality or referentiality
of a situation; the possibility that an agent's wishes, hopes or intentions will
effectively be realized; the authenticity of an utterance or a chunk of dis-
An Introductory Essay 13

course; or the normalcy of a discourse or of a communicative situation. These

parameters are illustrated with data from various languages and discourse
genres.
So, here goes, readers. We hope you enjoy the papers. And don't be put off by
the realization that mood is a grammatical category with an attitude! We'll
get it straightened out one of these days.

NOTES
* We are grateful to Greg Thomson for assisting the editors and authors in manuscript
preparation and copy editing. The indexes were prepared by Lisa Dasinger and Jacki
Trademan.
1 In this volume we avoid the term 'mode' because of the problematic ambiguities it
presents, being used with widely different meanings in the grammars of different
languages. In many European languages it translates what is here referred to as 'mood',
while in the grammars of certain non-European languages it is used to label categories
whose meanings fall under the headings of tense and aspect. We are aware, of course,
that languages often bundle tense, aspect, and mood information into portmanteau
morphology, thereby making it difficult to decide how to label such categories.
2 The term 'evidential' was first introduced by Jakobson (1957) as a tentative label for a
verbal category that indicates the source of the information on which a speaker's
statement is based. As currently understood, evidentiality covers a range of distinctions
involved in the identification of the source of one's knowledge. Various languages have
grammaticalized evidential markers indicating whether or not the speaker vouches
personally for the information contained in a statement. (See Chafe and Nichols 1986,
Willett 1988).
3 In place of the traditional distinction of linguistically-relevant modals into 'epistemic'
and 'deontic', Bybee (1985) recategorizes the modals into 'agent-oriented,' 'speaker-
oriented', and 'epistemic'. These categories will be defined and elaborated on below.
Other categories distinguished in modal logic, e.g. 'dynamic' and 'alethic' modalities
(cf. Lyons 1977:791, Palmer 1986:102-103), will not be discussed here, being less
germane to the analysis of modality in natural language.
4 The earlier 'deontic' meanings may or may not be preserved. English must, for example,
retains its obligative meaning, whereas might has lost its earlier abilitative meaning.

REFERENCES
Bybee, Joan L. 1985. Morphology: A Study of the Relation between Meaning and Form
(=Typological Studies in Language, 9) Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan L. 1988. "Semantic Substance vs. Contrast in the Development of Grammati­
cal Meaning." Proceedings of the Berkeley Linguistic Society. 14.247-264.
Bybee, Joan & Östen Dahl. 1989. "The Creation of Tense and Aspect Systems in the
Languages of the World". Studies in Language 13.51-103
14 Modality in Grammar and Discourse

Bybee, Joan L., William Pagliuca, & Revere D. Perkins. 1991. "Back to the Future".
Approaches to Grammaticalization. ed. by E. C. Traugott &. B. Heine, vol. 2, 17-58.
Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan L., Revere D. Perkins & William Pagliuca. 1994. The Evolution of Grammar:
Tense, Aspect and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Chafe, Wallace L., & Johanna Nichols. 1986. Evidentiality: The Coding of Epistemology
in Language. Norwood, N.J.: Ablex.
Gerhardt, Julie. 1985. "On the Use of will and gonna: Toward a Description of Activity
Types for Child Language". Discourse Processes 8.143-75.
Gerhardt, Julie. 1990. "The Relation of Language to Context in Children's Speech: The
Role of hafta Statements in Structuring 3-Year-Old's Discourse". IPrA Papers in
Pragmatics A. 1-57.
Hopper, Paul, ed. 1982. Tense-Aspect: Between Semantics and Pragmatics. Amsterdam:
John Benjamins.
Jakobson, Roman. 1957. Shifters, Verbal Categories, and the Russian Verb. Cambridge,
Mass: Russian Language Project, Department of Slavic Languages and Literatures,
Harvard University. (Repr. The Selected Writings of Roman Jakobson. Vol 2.130-147
[1971] The Hague: Mouton.)
Lyons, John. 1977. Semantics , vol. 2. Cambridge: Cambridge University Press.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Willett, T. 1988. "A Cross-Linguistic Survey of the Grammaticization of Evidentiality".
Studies in Language. 12.51-97.
 I
Agent-Oriented and Epistemic Modality
 Agent-Oriented vs. Epistemic Modality
Some Observations on German Modals1

Bernd Heine
University of Cologne

1. Introduction

Recent research on the domain of modality has focussed in particular on the

nature of one basic distinction, that between what has variously been referred
to as deontic, root, objective, pragmatic or agent-oriented modality on the one
hand and subjective, hypothetical, or epistemic modality on the other.2 A
number of generalizations have been proposed about the nature of agent-
oriented as opposed to epistemic modality, such as the following:
a. In many languages, agent-oriented and epistemic meanings are
expressed by means of one and the same expression.
b. Wherever there is adequate historical evidence available on the
evolution of a language, agent-oriented meanings have been shown
to be older than epistemic ones (Shepherd 1982, Bybee and
Pagliuca 1985, Traugott 1989).
c. Epistemic meanings are said to be more strongly subjective than
agent-oriented ones, and the transition from agent-oriented to epis­
temic uses of a given linguistic expression has been described as
involving subjectification (cf. Langacker 1985), whereby mean­
ings tend to become increasingly based in the speaker's subjective
belief state/attitude toward the proposition (Traugott 1989:35).
d. In the ontogenesis of language, agent-oriented modality develops
prior to epistemic modality (but see Soonja Choi, this volume for
different observations). For example, for both English and German
 18 Bernd Heine

children, modal verbs at first serve agent-oriented rather than epis-

temic functions (Stephany 1989:4).
e. Similarly, in second language acquisition, agent-oriented uses of
modal elements tend to appear earlier than epistemic ones (Ramat
1992).
There are a number of questions that the study of these two kinds of
modality raises, such as the following:
(i) What factors can be held responsible for the fact that linguistic
expressions for agent-oriented modality are extended to also ex­
press epistemic modality?
(ii) If a linguistic item is used for the expression of both agent-oriented
and epistemic modality, are we dealing with an instance of ho-
monymy or of polysemy, that is, does that item simultaneously
belong to two different categories or only to one? The former view
is favored by Coates (1983), Palmer (1986), as well as a number of
other authors, while the latter is argued for by Sweetser (1982),
Traugott (1989) and others.
(iii) Is the transition from one kind of modality to another discrete/
discontinuous or gradual/continuous?
(iv) How can the shift from agent-oriented to epistemic modality be
described in a model based on grammaticalization theory?
While these questions have been addressed in a good deal of research
over the past decade, none has been answered satisfactorily, nor will the
present paper attempt to provide conclusive answers. Most discussions relat­
ing to these questions have centered around modality in English. This paper
will focus on a language different from but closely related to English, and its
aim is to provide some fresh data that might be relevant to answering the
above questions.

2. German modals

2.1 German and English modals compared

A number of authors have pointed out that German modals differ from their
English counterparts essentially in being more verb-like (Abraham
Agent-Oriented vs. Epistemic Modality 19

1989:357); some (cf. Jenkins 1972:9-12; Steele et al. 1981; Abraham 1992)
go so far as to argue that a categorial boundary separates German and English
modals, the latter belonging to a category called AUX and the former to that
of Verb. German modals such as können 'can' or müssen 'must', etc. differ
from their English counterparts, e.g., in the following ways:

a. They exhibit inflections (e.g., ich kann T can', du kannst 'you

can'). 3
b. They may appear in non-finite constructions (e.g., um kommen zu
können 'in order to be able to come').
c. They may iterate (e.g., er will arbeiten können 'he claims that he
can work').
d. They are not restricted to any fixed order with respect to each other
or to the perfect auxiliary haben 'have' (e.g., er hat arbeiten
gekonnt 'he has been able to work' vs. er kann gearbeitet haben 'he
may have worked').
e. They may also occur in certain constructions without any other
verb in the same clause (e.g., er kann Handstand 'he can do a
handstand').
f. "Auxiliary reduction" does not apply to them.4
There are, however, a number of properties that German modals share
with their English counterparts, e.g. (cf. Helbig and Buscha 1988:107ff.; 123ff.):
(i) They express a restricted range of grammatical functions.
(ii) They form a closed set of entities.
(iii) They normally do not passivize.
(iv) They do not form imperatives.
(v) They typically require main verbs, rather than noun phrases or
adverbial phrases, as complements.
(vi) In their presence, main verbs are used in a non-finite (infinitival)
form.
(vii) They lack a present tense 3 sg. inflection (-s in English and -t in
German).
(viii) They do not take the infinitive marker (zu) to introduce the main
verb.
(ix) Finally, and most importantly in this context, German and English
modals are both used to express agent-oriented as well as epistemic
meanings.
20 Bernd Heine

What this suggests is that we are dealing with a difference in degree

rather than in kind: while German modals are more verb-like than their
English cognates, the two are similar enough to be comparable. For more
details, see Heine (1993).

2.2. A survey

In the present section, eight German modals are looked at with a view to
determining how and to what extent they are associated with agent-oriented/
deontic and epistemic modality, respectively, or, to use the terminology of
German grammarians, with "objective" and "subjective" modality. The mo­
dals in question are:
müssen 'must'
sollen 'should'
können 'can'
wollen 'want, will'
mögen 'like, may'
dürfen 'be allowed to, may'
möchten 'want to, would like to'
werden 'will'
One may wonder what justification there is to include werden, essentially
a future tense marker, within the paradigm of modal auxiliaries. While I do
not wish to argue that werden is a modal, as has been done elsewhere (see
especially Vater 1970), the main reason for including it here is that like "true
modals" it exhibits a distinction between non-epistemic and epistemic uses.
Note furthermore that können, like English can, has two non-epistemic
senses: root possibility and permission (cf. Bybee, Perkins, and Pagliuca
1994). These two senses will not be distinguished in the quantitative analysis
that follows; our concern will be essentially with the distinction between non-
epistemic and epistemic modality, rather than between agent-oriented and
epistemic modality. These distinctions will be clarified below.
In the association of the eight modals with these two categories of
modality in given utterances, the following three situations will be distin­
guished, where "A" stands for agent-oriented and "E" for epistemic modality:
Agent-Oriented vs. Epistemic Modality 21

A, E: The relevant utterance is automatically associated with either an A

or an E interpretation, with no further contextual clues required,
i.e., either A or E constitutes the "basic meaning" or focal sense
(see Heine, Claudi and Hünnemeyer 1991) of the utterance con­
cerned.
A l , El : The relevant utterance does not have a focal sense of A or E, but is
associated with specific linguistic or extra-linguistic contexts that
are suggestive of either an A or an E interpretation. We may say
that in such cases A or E constitutes a non-focal sense.
A2, E2: The relevant utterance is not automatically associated with either A
or E. There are, however, highly specific contexts where an A or an
E interpretation is conceivable. In such cases, A or E will be said to
constitute a marginal sense.
This threefold classification is somewhat crude; a more fine-grained
typology would be desirable. For the present analysis, however, this classifi­
cation is sufficient to study the problems at issue here. I illustrate the three
situations with two examples.5
(1) Er muß mindestens 1,80 m sein.
he must at.least 1.80 metres be
'He must be at least six foot.'
a. [E: "On the basis of the evidence available I am led to conclude
that..."]
b. [A: "They are looking for a new goal-keeper (he has to be at
least six foot)"]
The most likely reading of muß in (1) is the epistemic one (cf. (la), hence
(1) is said to have a focal epistemic sense (E). (1) has, however, a non-focal
agent-oriented sense (Al) in addition, which is foregrounded, e.g., in a
context like (lb).
An example of a marginal sense is provided by the utterance in (2) below.
While (2) has a focal A sense ("He is obliged to come"), contexts can be
conceived of that would invite an E2 interpretation. A context like (2a), for
example, calls up the marginal epistemic sense, conveying the meaning:
"I have every reason to conclude that he is coming".
22 Bernd Heine

(2) Er muß kommen.

he must come
'He has to cóme.'
a. ["I can already hear his voice (so he must be coming)"]

In what follows, quantitative data are presented that bear on the question
of whether and to what extent certain standardized utterances are associated
with the two kinds of modality. To this end, index values are calculated on the
basis of the three-fold classification proposed above: a value of 1.00 is given
to an utterance whenever a focal sense is involved, 0.66 in the case of a non-
focal sense, and 0.33 in the case of a marginal sense. I illustrate this procedure
with reference to examples (1) and (2) above: the utterance in (1) (Er muß
mindestens 1.80 m sein) would receive an index value of 1.00 for its focal
epistemic sense and of 0.66 for its non-focal agent-oriented sense. Utterance
(2), on the other hand, would have an index value of 1.00 for its focal agent-
oriented sense and of 0.33 for its marginal epistemic sense. Values are
calculated for declarative (e.g., Er muß kommen 'He must come'), interroga­
tive (Muß er kommen? 'Does he need to come?'), and negative utterances (Er
muß nicht kommen 'He need not come'), for main verb perfect forms (Er muß
gekommen sein 'He must have come'), auxiliary past tense forms (Er mußte
kommen 'He had to come'), as well as for first- (Ich muß kommen 'I must
come'), second- (Du mußt kommen 'You must come'), and third-person
subject utterances (Er muß kommen 'He must come'). The sum totals are
divided by the total number of utterances considered, thereby establishing
average index values for agent-oriented and epistemic modalities. The maxi­
mum value attainable is 1.00, the minimum value 0 (zero). For example, a
modal will receive the maximum value of agent-oriented modality if in all
utterances considered it has a focal agent-oriented sense, and the minimum
value if it has neither focal, non-focal or marginal agent-oriented senses; see
APPENDIX for details. 6
A number of calculations have been carried out on the basis of selected
parameters; the results obtained are summarized in Tables 1 through 4. In
Table 1 the eight modals are compared in terms of the degree to which they
correlate with each of the two kinds of modality. As the index values suggest,
the modals can be arranged along a scale for each of the two modalities. At
Agent-Oriented vs. Epistemic Modality 23

Table 1. Index of modality of eight German modals (Affirmative/negative, declarative/

interrogative, present/past tense uses; see APPENDIX)
Index value of modality
Agent-oriented Epistemic
Mochten 0.92 0
Dürfen 0.74 0
Wollen 0.75 0.15
Müssen 0.51 0.37
Mögen 0.47 0.37
Werden 0.59 0.46
Können 0.46 0.50
Sollen 0.44 0.50

one end of the scale are möchten and dürfen, which have the highest values for
agent-oriented modality and exhibit no epistemic uses, at least not in the
corpus considered. [FN 7] At the other end we find können and sollen, which
have the lowest values for agent-oriented and the highest values for epistemic
modality.
The extent to which these modals are associated with agent-oriented and
epistemic modality appears to correlate with their relative degree of grammat-
icalization in general and with prototypical 'verbiness' in particular: thus, the
highest A values and the lowest E values are found with the items most
prototypically verb-like. Wollen 'will', for instance, is more verbal (has more
verbal properties) than modals like müssen 'must', können 'can', or sollen
'should', in that it can still take subordinate clauses introduced by daß 'that'
as complements and, accordingly, does not require subject identity between
the auxiliary and the main verb, as in (3). Also it can be passivized, as in (4)
(Vater 1970).
(3) Hans will, daß Anna zu Hause bleibt.
Hans wants that Anna at home stays
'Hans wants Anna to stay home.'
(4) Von Hans wird gewollt, daß Anna zu Hause bleibt.
by Hans is wanted that Anna at home stays
'Hans is requested to leave Anna at home.'
24 Bernd Heine

Table 2. Index of modality of four German expressions (on the basis of eight German
modals; see Appendix)
Index value of modality
Agent-oriented Epistemic
a. Sie kommt. 'She is coming.' 0.87 0.02
b. Sie weiß es. 'She knows it.' 0.77 0.32
c. Sie hat viel Geld. 'She has a lot of money.' 0.67 0.27
d. Sie kennt ihn. 'She knows him.' 0.37 0.59

Table 3. Index of modality of eight German modals according to various grammatical

distinctions (see Appendix)
Type of proposition Index value of modality
Agent-oriented Epistemic
a. Perfect
(e.g., Er hat kommen müssen. 'He had to come.') 0.75 0
b. Interrogative
(e.g., Muß er kommen? 'Does he need to come?') 0.72 0.11
c. Negative
(e.g., Er muß nicht kommen. 'He need not come.') 0.74 0.26
d. Past
(e.g., Er mußte kommen. 'He had to come.') 0.62 0.20
e. Present
(e.g., Er muß kommen. 'He must come.') 0.60 0.43
f. Main verb perfect
(e.g., Er muß gekommen sein. 'He must have come.') 0.19 0.55
g. Main verb progressive
(e.g., Er muß am Kommen sein. 'He must be coming.') 0.02 0.64

Table 4. Index of modality of eight German modals according to grammatical person of

the sentence subject (Affirmative/negative, declarative/interrogative, present/
past uses; see Appendix)
Index value of modality
Agent-oriented Epistemic
1st person subject 0.70 0.18
2nd person subject 0.69 0.25
3rd person subject 0.57 0.35
Agent-Oriented vs. Epistemic Modality 25

The behavior of the modals also differs greatly in accordance with the
type of main verb figuring in the utterances. Thus, verbs of action and telic
verbs associate predominantly with agent-oriented modality; cf. utterance (a)
in Table 2, involving the motion verb kommen 'come'. The situation is more
complicated in the case of stative verbs: while some, like wissen 'know
(information)' and haben 'have', gravitate toward agent-oriented uses,
kennen 'know, be acquainted with', has a majority of epistemic uses.
The modal behavior of the eight items considered also varies in accord­
ance with the type of proposition and the tense and aspect in which the modals
occur, as can be seen in Table 3. According to this table, in Perfect construc­
tions the modals are confined to agent-oriented uses (a). Even with verbs such
as kennen, which generally associate strongly with epistemic modality, an
epistemic reading is unlikely once the modal auxiliary has a Perfect form.
Furthermore, in Interrogative (b) and Negative (c) propositions and in the
Past tense (d) the modals correlate primarily with agent-oriented rather than
with epistemic modality. Epistemic uses prevail, however, if the main verb is
in the Perfect (f), and Progressive constructions8 are almost entirely associ­
ated with epistemic modality (g).
Finally, the interpretation of an utterance is also affected by the gram­
matical person of the subject. As Table 4 suggests, an utterance is most likely
to have an agent-oriented interpretation when the subject referent is first-
person and least likely when it is third-person; conversely, epistemic modality
correlates most strongly with third-person and least strongly with first-person
subjects.

3. Context

3.1 The linguistic context

The quantitative data on German modals presented in the preceding section

allow us to isolate several linguistic factors that appear to be relevant for
characterizing the uses of the two kinds of modality. These data suggest that
an utterance is very likely to have a focal agent-oriented sense if:
a. any of the modals möchten, dürfen or wollen is involved;
b. the main verb is an action or a terminative verb;
c. the modal is in the perfect or past tense.
26 Bernd Heine

Furthermore, a modal is more likely to have agent-oriented than epistemic

modality as its focal sense if:
d. it occurs in interrogative rather than in declarative utterances;
e. it occurs in negative rather than in affirmative utterances;
f. the subject is first or second (rather than a third) person.
Thus, the more of the linguistic properties (a) through (f) an utterance
has, the more likely it is to express agent-oriented rather than epistemic
modality. Conversely, a focal epistemic sense is more likely if the main verb
is in a progressive or perfect form or is stative.
What these observations suggest is that the agent-oriented vs. epistemic
distinction works in roughly the same way with German as with English
modals. For example, according to the findings on English modals presented
by Coates, the following associations can be observed: (a) All agent-oriented
modals can occur in interrogative constructions where epistemic modals are
not normally found (1983:244-245). (b) Epistemic but not agent-oriented uses
are associated with the progressive aspect (1983:246). (c) The core meanings
of agent-oriented modals are crucially associated with agentivity, that is, the
action referred to in the main predication is carried out by someone (or
something) using their own energy (Coates 1983:245). (d) Epistemic meaning
is typically associated with existential subjects and stative verbs (1983:245;
see also Bybee, Perkins, and Pagliuca 1994). We will return to these cross-
linguistic observations below.

3.2 Contextual frames

While the above linguistic characterization allows us to predict within limits

whether a given utterance will receive an agent-oriented or an epistemic
interpretation, it does not account for a considerable number of actual uses of
German modals. Many times, when queried whether a given utterance was
suggestive of agent-oriented or epistemic modality, an informant's reaction
was to search for a relevant context based on past experience. If the informant
could remember a context that matched the utterance in question, then he or
she would answer the question with reference to that experience. If not, then
he or she would be likely to declare the utterance unacceptable. Thus, rather
than semantic, syntactic or other rule-governed behavior, it was the presence
or absence of a cluster of related experiences, let us call it the contextual
Agent-Oriented vs. Epistemic Modality 27

frame, that seemed to be crucial in deciding between the two modalities.

Consider the utterance in (5):
(5) Das Bier sollte kalt sein.
the beer should cold be
'The beer should be cold.'

In accordance with the observations made in 3.1, (5) is strongly suggestive of

a focal epistemic sense since there is an inanimate subject and a stative main
verb. Nevertheless, the majority of our informants, all native speakers of High
German, described (5) as having a focal agent-oriented sense. What appears to
account for the agent-oriented interpretation of (5) is its association with a
contextual frame like a party situation where someone is offered a warm beer.
One possible reaction would be to answer with (5), and in such a case, (5) can
only receive an agent-oriented interpretation. In accordance with the terminol­
ogy introduced in Section 2, the contextual information just presented may be
said to constitute the focal contextual frame for (5). A non-focal frame would
be provided, e.g., by (5a) which triggers an epistemic interpretation of (5).
5a ["I put the drinks into the fridge more than two hours ago
(so the beer should be cold by now)"]

Coates (1983:233) notes that the interpretation of English modals as

agent-oriented (that is, "root" in her terminology) depends in most cases on
the presence of agentivity. This also applies to a large extent to German
modals, though agentivity need not be, and frequently is not, explicitly
expressed; what matters is the contextual frame associated with the use of the
modal. The utterance in (5) has no formally expressed agent; its association
with focal agent-oriented modality derives from contextually recoverable
agentivity, i.e., the speaker wishes to urge someone to put the beer into the
fridge, or to replace the bottle of warm beer by a bottle of cold beer. What this
suggests is that any attempt at characterizing the relationship between the two
types of modality strictly in terms of the linguistic categories that typically
correlate with each one is likely to achieve little more than probabilistic
approximations; it is unlikely to explain the relationship.
The notion of a contextual frame is similar to what others have called an
'inferential schema', or simply a 'frame', i.e. a body of knowledge evoked by
the language user in order to provide an inferential basis for the understanding
28 Bernd Heine

of an utterance (cf. Levinson 1983:281). It is beyond the scope of the present

paper to define the parameters that are relevant for classifying contextual
frames as either focal or non-focal; presumably they have to do with factors
such as the following:
a. Contextual clues: Focal frames require a minimum of contextual
clues, they are triggered spontaneously without involving major
mental effort or imagination. Non-focal frames, on the other hand,
require a certain amount of mental energy.
b. Knowledge of the world: An example like (6) has a focal epistemic
sense, since we cannot manipulate time in such a way as to license
an agent-oriented interpretation.
(6) Es muß drei Uhr sein.
it must three o'clock be
'It must be three o'clock.'
c. Social norms: We believe in a positive world and expect other
people to behave in a socially acceptable way. Therefore, while
(7a) may receive either an agent-oriented or an epistemic interpre­
tation, (7b) receives only an epistemic sense, since we do not
expect someone to want someone else to behave in a socially
disapproved way.
(7) a Er muß tapfer sein.
he must brave be
(i) 'He has to be brave.'
(ii) 'He must be brave.'
b Er muß feige sein.
he must cowardly be
'He must be a coward.'
Conversely, utterances having human agents as subjects and action verbs
as predicates, that is, structures typically associated with agent-oriented mo­
dality, trigger a focal epistemic meaning if they are associated with contextual
frames that discourage an agent-oriented interpretation, as in (8).
(8) Er muß seine Frau schlagen.
he must his wife beat
'He is said to beat his wife.'
Agent-Oriented vs. Epistemic Modality 29

These are but a few of the factors that may be relevant to the evocation of
contextual frames, possible additional factors being frequency and/or inten­
sity of experience, perceptual salience, propensity for memory storage, or the
stereotypes an utterance tends to evoke.

3.3 Conceptual properties

In addition to contextual frames, the choice between the two kinds of modality
depends crucially on certain conceptual properties. Agent-oriented uses of
German modals are associated with the following properties:
a. There is some force (F) that is characterized by an "element of
will" (Jespersen 1924:320-1), i.e., that has an interest in an event
either occurring or not occurring.9
b. The event is to be performed typically by a controlling agent (C).
c. The event is dynamic (D), i.e., it involves the manipulation of a
situation and is conceived of typically as leading to a change of
state.
d. The event has not yet taken place at reference time, i.e., its occur­
rence, if it does in fact take place (see (e) below), will be later than
the reference time (L).
e. The event is non-factual (Palmer 1986:96), though there is a certain
degree of probability that it will occur (P).
Take utterance (9), for example, where F is the speaker (9a) or some
other force the speaker has in mind (9b), and C is the sentence subject er. The
event is dynamic (D), as can be concluded from the use of the action verb
kommen, and is supposed to take place later than at reference time (L).
Finally, the modal muß expresses a high degree of probability (= strong
obligation) that the event will in fact take place (P).
(9) Er muß kommen.
he must come
'He has to come.'
a. [A: "I insist that he comes"]
b. [A: "His boss insists that he comes"]
Differences between the various modals exist with regard to the degree of
probability that the event will take place: the probability is low in the case of
30 Bernd Heine

können and dürfen (permission) but high in the case of müssen (strong
obligation), with the remaining modals being intermediate. But the modals
differ also in other respects. There are essentially three kinds of relevant
participants: the speaker (S), the modal force (F), and the agent (C), the last of
these being typically, though not necessarily, coded as the sentence subject.
While F is different from C in the case of modals such as müssen 'must',
sollen 'shall, should' or können 'can' (= "subject-external modality"), as can
be seen, e.g., in (9), it is identical with C in the case of the modals wollen,
mögen, and möchten (= "subject-internal modality"), as in (10), where the
subject ich 'I' is both the modal force F and the agent C, and in this instance
also S.

(10) Ich möchte kommen.

I want, to come
'I would like to come.'
F can refer to either the speaker or to some other human being, or even to
a non-human entity. Thus, in (11), F may be either a human authority, e.g., the
speaker, or a non-human (e.g., religious, institutional, or moral) power.
(11) Du darfst nicht stehlen.
you may not steal
'Thou shalt not steal.'
There is no necessary relationship between F and C on the one hand, and
certain kinds of nominal referents on the other. For example, in the agent-
oriented reading of (5), repeated below for convenience, it remains unclear, in
the absence of additional contextual clues, whether F refers to the speaker or
to some other force, or whether C refers to the hearer, to another agent, or to
no explicit agent at all.
(5) Das Bier sollte kalt sein.
the beer should cold be
'The beer should be cold.'
While there are no compelling clues as to the referential identity of F in
(5), nevertheless, some kind of F is implied, as can be derived, e.g., from the
fact that a possible response to (5) would be to inquire about the referent of F
by asking: Wér möchte, daß das Bier kalt ist? 'Who wants the beer to be cold?'
Agent-Oriented vs. Epistemic Modality 31

While prototypical instances of agent-oriented modality are character­

ized by the presence of the properties listed above, prototypical instances of
epistemic modality lack all properties except P. Thus, in the epistemic reading
of (5), there is neither F nor C, no D is involved, and the later-than-reference-
time constraint (L) does not apply (see (14) below).
Since P is shared by the two kinds of modality, it will not be considered in the
remainder of this paper.

3.4 On the transition from agent-oriented to epistemic meanings

While we can distinguish the two kinds of modality on the basis of the
conceptual properties C, F, D and L, these properties are not of equal impor­
tance. Consider, for example, utterance (12b). The use of muß does not entail
L; that is, there is no later-than-reference-time constraint involved: the action
of schlafen can be assumed to take place over an interval that includes
reference time, yet (12b) has a focal agent-oriented sense.
(12) a. Warum ist Klaus nicht hier?
why is Klaus not here
'Why is Klaus not here?'
b. Er muß schlafen.
he must sleep
'He has to sleep.'
It might seem as if F is also irrelevant in (12b) since no explicit force is
mentioned. Yet one could inquire about a possible F by asking: Wér ist dafür
verantwortlich, daß er schlafen muß? 'Who is responsible for the fact that he
has to sleep?' Thus, F may be said to be potentially present in (12b), albeit
backgrounded.
In examples like (12b), where L does not apply, the difference between
agent-oriented and epistemic modality becomes minimal: While (12b) has a
focal sense of agent-oriented modality, it may also be understood to convey an
epistemic meaning, i.e. 'he must be sleeping.' In such cases we are dealing
with what Coates refers to as merger: the agent-oriented and the epistemic
senses of (12b) are in a both/and relationship, they are mutually compatible;
in order to understand (12b) it is therefore not necessary to decide which of the
two senses is intended since they are not mutually exclusive (see Section 4).
32 Bernd Heine

Furthermore, there are contexts in the use of müssen 'must' where both C
and L are absent and D is largely irrelevant, as in (13). Verbs that behave in a
similar way to frieren 'freeze, be cold' are leiden 'suffer', weinen 'cry', or
gähnen 'yawn'.
(13) Er muß frieren.
he must freeze
'He is freezing/feeling cold.'
In examples like (13), F is some unidentified force, and we are dealing
with an instance of focal agent-oriented and non-focal epistemic modality.
Once F is eliminated, however, an agent-oriented interpretation is ruled out:
no F is implied in (14), which therefore expresses exclusively epistemic
modality.
(14) Ihn muß frieren.
him must freeze
'He must be freezing/cold.'
What the observations made in this section suggest is that of the four
properties considered, only F is obligatorily connected with agent-oriented
modality: modal utterances that lack F lack the "element of will" to which
Jespersen (1924:320-21) referred. In such cases we are dealing with an
epistemic rather than an agent-oriented sense.
Typically, the lack of F will leave all other components of the modal
concept unaffected. Thus, P remains the same whether or not F is present: the
degree of probability that the situation described in the utterance will obtain is
high in the case of müssen 'must' and low in the case of können 'can',
irrespective of F. Take, for example, the modal sollen 'shall, should': in the
agent-oriented interpretation of (15), F may refer either to the speaker (15a)
or to another person (15b), while in the epistemic interpretation of (15), the
speaker is excluded as source of the modal force (15c); the modal source is
associated with someone other than the speaker, as suggested by the fact that
(15d) would not be an appropriate paraphrase of (15).

(15) Sie soll heute zu Hause sein.

she should today at home be
'She should be at home today.'
Agent-Oriented vs. Epistemic Modality 33

a. [A: "I want her to be at home today"]

b. [A: "Someone wants her to be at home today"]
c. [E: "Someone claims that she is at home today"]
d. [ *"I claim that she is at home today"]
In many of its agent-oriented uses, sollen excludes the speaker as a
source of F, as can be seen in (16), where the subject is a first-person referent.
(16) Ich soll kommen.
I should come
'I should come.'
[A: "Someone wants me to come"; *"I want to come"]

3.5 Discussion

The distinction between the two categories of modality has been associated
with a number of factors. First, as pointed out above, one major distinguishing
property is that agent-oriented senses are likely to refer to situations that are
supposed to obtain later than at reference time (= property L). Thus, an
utterance like (17) (= (5)) has future reference when used in its agent-oriented
sense (17a) but present reference when used epistemically (17b).
(17) Das Bier sollte kalt sein.
the beer should cold be
T h e beer should be cold.'
a [A: "I want the beer to be cold (so you'd better put it into the
fridge again")]
b [E: "I have reason to assume that the beer (standing in front of
me) is cold"]
Similar observations have been made for English modals. Coates
(1983:235), for example, notes that, with the exception of root can, English
modals in their agent-oriented uses always have future reference (see also
Palmer 1986:97; Bybee, Perkins, and Pagliuca 1994):
[...] one of the conditions of using Root MUST, SHOULD or OUGHT (=
'Obligation') or Root MAY or CAN (= 'Permission') is that the speaker
believes that the action referred to in the main predication has not already
been achieved. In other words, commands, recommendations and permis­
sion-granting utterances all refer to an action which will be carried out at a
time subsequent to the utterance (Coates 1983:233).
34 Bernd Heine

As noted in 3.4, the association of agent-oriented modality with deictic time is

not as rigid in the case of German modals as Coates suggests it is for English
modals. Note further that there are contexts where the later-than-reference-
time constraint does not apply, e.g., in the case of modals used in the past
tense, as in (18), where exclusively agent-oriented modality is involved.
(18) a Er mußte es tun.
he had.to it do
'He had to do it.'
b Er wollte zu Hause sein.
he wanted at home be
'He had intended to be at home.'
What this suggests is that the later-than-reference-time constraint is a
concomitant, rather than a defining, property of agent-oriented modality.
Furthermore, Bybee, Perkins and Pagliuca (1994) observe that the two
modalities expressed by English must correlate with mutually exclusive lin­
guistic environments: in the Past tense, in the Present tense with a stative verb,
and in the Progressive, must can only have an epistemic ("inferred certainty")
reading, while in the non-past with dynamic verbs, must is only agent-oriented
and refers to the future. Apart from the strong correlation between progressive
aspect and epistemic modality observed in Section 2 (Table 4), the German
cognate müssen does not exhibit a similarly rigid distinction: for example,
dynamic verbs may have a focal epistemic meaning and stative verbs a focal
agent-oriented meaning in the present tense, as the following examples show,
respectively:
(19) Er muß kommen; ich höre schon seine Stimme.
he must come I hear already his voice
'He must be coming; I hear his voice already.'
(20) Er muß heute nachmittag unbedingt zu Hause sein.
he must today afternoon by.all.means at home be
'He definitely has to be at home this afternoon.'
Even in English, the distinction is not as clear-cut as Bybee, Perkins and
Pagliuca (1994) suggest. Consider an utterance like (21), where must may
have either an agent-oriented (21a) or an epistemic interpretation (21b)
(Perlmutter 1970:115; Calbert 1973:7); see also (26) below.
Agent-Oriented vs. Epistemic Modality 35

(21) Clyde must work hard.

a [A: "Clyde is obliged to work hard"]
b [E: "It must be the case that Clyde works hard"]
Second, with reference to property C, Coates (1983:245) observes that
the core meanings of agent-oriented modals in English are crucially associ­
ated with agentivity. This applies to German modals as well, as we saw in
Section 2, where action verbs, which typically require an agent, were shown
to trigger an agent-oriented interpretation of modals in the majority of exam­
ples. At the same time, however, agentivity is not decisive for defining agent-
oriented modality in German, as (17) illustrates. The same may be said for
property D, also demonstrated by (17), which contains neither a dynamic
predicate nor an agent-like participant. This raises the following question: If
agent-oriented modality can be defined with reference to one property only,
namely F, what accounts for its strong association with properties C, D, and L?
There appears to be a straightforward answer: since the presence of F
implies that an event may, should or must happen, this entails that the event
will occur later than now, and more often than not an event that happens will
involve agents and actions, less typically also inanimate forces. Thus, the
presence of F suggests that we are dealing with a kind of utterance that tends
to be associated with agents and dynamic processes typically leading to
changes of state. If, however, a modal utterance lacks F, i.e., if it expresses
epistemic modality, it is likely to be associated with states and time-stable
situations. It is not surprising therefore that epistemic modality is closely
associated with utterances involving existential subjects and Stative verbs, or
verbs in the progressive and perfect aspects.
That existential subjects, impersonal subjects and dummy subjects are
strongly associated with epistemic, rather than with agent-oriented modality,
has been argued repeatedly in the literature on English modality (Coates 1983;
Bybee, Perkins, and Pagliuca forthc). Bybee, Perkins, and Pagliuca (1994) in
fact observe that the presence of a dummy subject precludes an agent-oriented
reading in cases like (22):
(22) It should take me about four hours to get there.
While we have not been able to make corresponding observations about
German, there are contextual frames where the presence of a dummy subject
does favor an epistemic reading. Thus, while (23) is associated equally with
36 Bernd Heine

an agent-oriented and an epistemic interpretation, (24) may be said to have a

focal epistemic reading (24b), but only a non-focal agent-oriented reading
(24a).
(23) Der Brief muß in zwei Tagen da sein.
the letter must in two days there be
'The letter must be there in two days.'
(24) Es muß zwei Tage dauern, bis der Brief da ist.
it must two days last until the letter there is
'It must take two days until the letter is there.'
a. [A: "... I'll be in trouble if the letter arrives earlier"]
b. [E: "... since it is a long way to get there"]
Yet even in the presence of existential/dummy subjects and durative
verbs like dauern 'last', an agent-oriented sense is possible (24a). What might
account for the stronger epistemic force in the case of contextual frames like
(24) is the fact that the absence of a human or animate subject is also likely to
suggest absence of a manipulating force F and, hence, to favor an epistemic
interpretation. Note, however, that German has a paradigm instance of an
impersonal marker, which is man 'one', and this marker is almost exclusively
associated with agent-oriented modality. Thus, even in utterances with a main
verb kennen 'know' as their predicate, which strongly correlates with epis­
temic modality (as we saw in Section 2.2, Table 2), there is a focal agent-
oriented sense if the sentence subject is man, as in (25).

(25) Man muß sie kennen.

one must her know
'You have to know her.'
There appears to be an obvious reason for this: man only occurs as a
subject, refers exclusively to human referents, is associated with human
activities, and tends to imply some manipulating force. It is therefore not
surprising that it promotes agent-oriented, rather than epistemic, interpreta­
tions.
Agent-Oriented vs. Epistemic Modality 37

4. On models

4.1 Metaphor model vs. context model

A number of studies of grammaticalization have claimed that the develop­

ment of grammatical categories involves a metaphorical process (see espe­
cially Claudi and Heine 1986; Sweetser 1988; Heine, Claudi and Hiinnemeyer
1991; Stolz 1991; see also Bybee and Pagliuca 1985:73). For example, there
are good reasons to argue that the shift of the English be going to-construction
from a concrete/lexical meaning to the abstract/grammatical meaning of
future tense has a metaphorical base. First, this shift involves a transfer from
the "real-world", the world of referential entities and kinetic activities, to the
world of discourse, to entities that exist in acts of speech or, e.g. with
reference to the development of modal auxiliaries, a transfer from the socio-
physical world to the world of reason and belief (Sweetser 1984:24). Thus, we
are dealing with a transfer from one domain of human conceptualization to
another. Second, the metaphoric nature of this transfer can also be derived
from the fact that our be going to-example meets the criteria commonly
applied to define metaphor, e.g., that the statement, if taken literally, is false.
While in the sentence Sally is going to town the phrase is going to has its
literal meaning of motion in space, it has a "transferred sense" in sentences
like Sally is going to wake up in a minute, where the literal meaning is ruled
out.9
Essentially the same kind of metaphorical transfer can be observed in
other instances of grammaticalization: they involve a transfer from one cogni­
tive domain to another, with the effect that, from a certain stage onwards, an
expression undergoing this shift becomes false or meaningless if taken liter­
ally, that is, if interpreted as designating its original meaning. A number of
such source domains and their corresponding target domains are discussed in
Heine, Claudi and Hiinnemeyer (1991). It goes without saying that we are
dealing here with a specific kind of metaphorical transfer that has been
described as "emerging metaphor" (Heine, Claudi and Hiinnemeyer 1991:60-62).
There is, however, at least one problem with this metaphorical model.
This problem concerns the continuous nature of grammaticalization. Meta­
phor is commonly assumed to involve a discrete "jump" from one domain to
another - a process that is hard to reconcile with the gradient nature of
38 Bernd Heine

grammaticalization chains (Heine 1992). In the case of be going to, the shift
from physical motion to grammatical function proceeded in a chain-like
manner. This is suggested on the one hand by historical evidence (cf. Pérez
1990) and on the other hand, by the present use patterns of the construction.
That conceptual shift in the process of grammaticalization is chain-like
and continuous is due to the particular circumstances giving rise to new
grammatical meanings: such meanings are derived from existing meanings
via context extension (Aijmer 1985), and through the conventionalization of
invited inferences or conversational implicatures (Traugott and König 1991).
This reasoning has been advanced with reference to the transition from agent-
oriented to epistemic meanings. Thus, Traugott (1989:50-51) argues that from
'permission' one can implicate 'expectation': thus, if I say You may go, I may,
in the right circumstances, implicate that I want you to go, from which you
may infer that you have some obligation to go. With reference to the modality
of obligation, Traugott and König observe:
[...] must in the epistemic sense of 'I conclude that' derived from the
obligative sense of 'ought to' by strengthening of conversational inferences
and subjectification. If I say She must be married in the obligation sense, I
invite the inference that she will indeed get married. This inference is of
course epistemic, pertaining to a state of affairs that is anticipated to be true
at some later time (Traugott and König 1991:209).

This overall process, of which pragmatic strengthening appears to be a part, is

called 'context-induced reinterpretation' by Heine, Claudi and Hünnemeyer
(1991, Chapter 3), because it is specific contexts (or "circumstances") that
invite new inferences and ultimately lead to the emergence of (new) gram­
matical meanings.
To summarize, in order to account for the conceptual shift from concrete/
lexical to abstract/grammatical meaning we have two main models at our
disposal: one that rests on a metaphorical interpretation of the process and
may therefore be called the metaphor model, and another relying on context-
induced reinterpretation, which I call the context model. While in the past,
scholars have argued in favor of either one or the other, more recent investiga­
tions suggest that both models are required to understand the process con­
cerned.
There are, however, two contrasting positions with regard to the role
these models should play in a theory of grammaticalization. On the one hand,
there is the position of Traugott (1989), Traugott and König (1991), and of
Agent-Oriented vs. Epistemic Modality 39

Bybee, Perkins and Pagliuca (1994.), according to which different kinds of

grammaticalization have to be distinguished; and while some kinds are han­
dled best in terms of the metaphor model, others have to be accounted for in
terms of the context model. Among these authors again, there are differing
views as to where exactly the models apply. While all agree that the metaphor
model typically applies in the case of concrete/lexical structures and the
context model in the case of more strongly grammaticalized structures, disa­
greement exists as to where to draw the boundary between the two. For
Traugott (1989) and Traugott and König (1991), the transfer from concrete
source propositions to auxiliary functions of tense and aspect is metaphori­
cally structured11 while the transfer from agent-oriented (deontic) to epistemic
meanings is not:
But the shift to epistemics of conclusion, belief, knowledge, hear-say,
hypothetical conditionality, and so forth has little of the analogical mapping
from one conceptual domain onto another that is characteristic of metaphor
(Traugott 1989:50).

Sweetser (1982) on the other hand suggests that the conceptual shift from the
agent-oriented to the epistemic domain is a clear instance of a metaphorical
process. Bybee, Pagliuca and Perkins (forthc.) again argue that the English
modal auxiliaries must and should both have experienced an extension from
agent-oriented to epistemic modality, but whereas the extension of must is
suggestive of metaphorical transfer, that of should might be due to the
conventionalization of implicature and, hence, would have to be dealt with in
terms of the context model.
The different behavior of must and should had already been pointed out
by Coates (1983:14ff.). In her work on English modal auxiliaries, she ob­
served that in the transition from one kind of modality to another there are
three types of indeterminacy, which she refers to, respectively, as gradience,
ambiguity, and merger. Gradience concerns the nature of the continuum of
meaning, e.g., in the transition from the core of ABILITY to the periphery of
POSSIBILITY of English can. Ambiguity may be described in terms of the
following properties (cf. Coates 1983:15-16):

a. A given expression has two senses and it is not possible to decide

from the immediate context which of these is intended.
b. The two senses are in an either/or kind of indeterminacy since
either of them may be chosen.
40 Bernd Heine

c. These senses belong to different categories, viz. the categories of

agent-oriented (= root) and of epistemic modality, respectively.
d. The categories concerned are discrete.
The only example of ambiguity in Coates' work on English modal auxiliaries
involves must, which is ambiguous between an agent-oriented (26a) and an
epistemic sense (26b).
(26) He must understand that we mean business.
a. [A: "It is essential that he understand that we mean business"]
b. [E: "Surely he understands that we mean business"]
Merger shares with ambiguity properties (a) and (d). Instead of properties
(b) and (c), however, it has (b') and (c'), respectively:
b'. The two senses are in a both/and relationship, i.e., they are mutu­
ally compatible. In order to understand the relevant expression, it is
not necessary to decide which of the two senses is intended.
c'. In certain contexts, the two senses are "neutralized", i.e., they are
not mutually exclusive.
As examples of merger, Coates (1983:17) cites English should and ought to,
as in (27):
(27) a. A: Newcastle Brown is a jolly good beer.
B: Is it?
C: Well it ought to be at that price.
b. [A: Obligation of the maker to provide a good beer]
c. [E: Logical assumption - "the beer costs a lot, therefore it is
good"]
According to Heine, Claudi and Hünnemeyer (1991) on the other hand,
for all instances of grammaticalization analyzed so far, both the metaphor and
the context models are required, each relating to a different level: the context
model to the micro-level and the metaphor model to the macro-level of
conceptual shift. The former perspectivizes the continuous nature of concep­
tual shift, that is, the fact that this shift proceeds in a series of overlapping,
contextually defined extensions, while the macro-level perspectivizes the fact
that, given enough micro-level extensions, conceptual shift will cross bounda­
ries between cognitive domains, like that between the domain of concrete,
Agent-Oriented vs. Epistemic Modality 41

"real-world" phenomena and that of abstract grammatical functions. In the

case of the be goingto-examplealluded to above, the context model takes care
of the many micro-level extensions to be observed both in the historical and
the present uses of this construction, including the numerous situations of
ambiguity between physical motion and intention, or between intention and
prediction, or even among all three senses. The metaphor model, on the other
hand, is concerned with and accounts for the macro-level shift from a concrete
source proposition to an abstract grammatical function.12
The data considered in the previous sections suggest that the distinction
between agent-oriented and epistemic modality is of a discrete nature in the
case of German modals, being based on presence vs. absence of the modal
force F. It was noted, however, that three additional properties tend to be
associated with this distinction, namely presence vs. absence of a controlling
agent (C), of a dynamic event (D), and of the later-than-reference-time
constraint (L). The more of these properties are present, the more pronounced
the semantic contrast between the two interpretations is likely to be and the
more likely it is that we are dealing with instances of ambiguity; conversely,
the more of these properties that are absent, the more we approach what
Coates refers to as merger. Take, for example, utterance (13), repeated here as
(28):
(28) Er muß frieren.
he must freeze
'He is freezing/feeling cold.'
(They let him freeze.')
Muß in (28) is likely to involve a modal force F, hence, its focal sense is an
agent-oriented one, although apart from F, none of the properties normally
accompanying agent-oriented uses, i.e. C, D or L, is clearly present. There is,
however, an almost equally pronounced epistemic sense ["He must be freez­
ing"], and the two senses can be said to be suggestive of merger, in that they
satisfy criteria (b') and (c') mentioned above: given the right context, there is
not much of a difference whether the hearer interprets (28) in its agent-
oriented or in its epistemic sense. (28a) below would be an example of such a
context: the contextual information Er hat keinen Mantel an ("He isn't
wearing a coat") may be interpreted alternatively as the force that is responsi­
ble for the content expressed by (28) or else as a presupposition leading to the
logical assumption underlying the epistemic interpretation of (28).
42 Bernd Heine

(28) a. ["Er hat keinen Mantel an;..."]

'He isn't wearing a coat;...'

It is in utterances like (28), used in contexts like (28a), that an inferential

mechanism could have been at work (Bybee, Pagliuca and Perkins 1994)
whereby an agent-oriented sense of obligation gives rise to an epistemic sense
of inferred certainty.
What this would seem to suggest is that the transition from the agent-
oriented to the epistemic sense of (28) involves both discontinuity and conti­
nuity; discontinuity because even in a context like (28a), the two modalities
can clearly be distinguished, and continuity because, given the right context, it
does not matter which of the two senses is intended by the speaker and
selected by the hearer: the information value remains essentially the same. A
similar situation is found in the case of modals other than müssen, which
means that, for all modals considered, the same descriptive framework can be
employed. 13
To summarize, the fact that there is only one criterial property that
distinguishes the two kinds of modality in German modals would seem to
suggest, first, that an element of discreteness separates the two modalities.
Second, it suggests that the transition from agent-oriented to epistemic modal­
ity may be regarded as an instance of semantic bleaching in that it involves the
"bleaching out" of property F, i.e., of Jespersen's "element of will"
(1924:320-1). At the same time, however, there are also gains: with the
elimination of F, the modals acquire new senses that are characterized by what
Traugott (1989:49) calls the "strengthening of focus on knowledge, belief,
and the speaker's attitude toward the proposition".

Third, such observations suggest that there are reasons to regard the transition
from agent-oriented to epistemic concepts as being metaphorically structured,
as has been argued for independently by Sweetser (1982; 1988), Bybee and
Pagliuca (1985), and Heine, Claudi and Hünnemeyer (1991): it involves a
transfer from one domain of human experience to another, from the
sociophysical world to the epistemic world (Sweetser 1982:492-5), the former
being a dynamic world of willful human beings who act and are capable of
imposing their will on other agents, the latter essentially a static world, one
that may but need not be associated with human participants (Heine, Claudi
and Hiinnemeyer 1991:176-178).
Agent-Oriented vs. Epistemic Modality 43

4.2 Another example

In a paper on "Temporal distance: a basic linguistic metaphor", Fleischman

(1989) observes that distance in the past frequently serves to move a real event
from actuality to a hypothetical world, and that expressions for temporal
distance are often pressed into service to express, inter alia, social distance
and politeness, as in (29):
(29) / thought/was thinking about asking you to dinner.
In this way, Past tense forms referring to a present situation are "felt to be less
assertive, more deferential, whence more polite" than corresponding Present-
tense forms:
The risk of a possible negative reply is reduced since no invitation has
actually been issued. Use of the PAST tense in speech acts of this type
serves to cancel - or at least attenuate - their illocutionary force [...]
(Fleischman 1989:9).

The result is that temporal distance serves as a metaphorical vehicle for social/
interpersonal distance as well as for a number of other notions relating to
evidentiality, speaker subjectivity, etc.
A different explanation is volunteered by Bybee (This volume) with
regard to English modal verbs used in the Past tense. She discusses some of
the same evidence adduced by Fleischman but describes the phenomenon
without reference to metaphor; rather, her parameters are the particular se­
mantics characterizing modals on the one hand, and the use of past tense on
the other:
[...] the use of wanted to in present time in Modern English is possible
because it implies that certain conditions on carrying out the wanted predi­
cate may not be met. The so-called polite or remote uses of Past tense as in
(21) / wanted to ask you a question.
arise in the same way. That is, (21) implies that there might be conditions
that are unmet. Among these implied conditions is the question of whether
the addressee wants to be asked a question — thus the deferential use of the
Past. (Bybee, this volume).

According to Bybee, a series of factors work together to produce the

deferential or polite senses apparent in the two English examples presented
above:
44 Bernd Heine

a. First is the fact that modal verbs are stative verbs. When modals are
used in the Past what is asserted is that a state existed before the
moment of speech, but not that the state necessarily still exists in
the present.
b. Since the state may still exist in the present, a modal in the Past
tense may be used in contexts where it has present relevance: it
may receive the interpretation "that the modality is still in effect,
and [that] the predicate action will be carried out if the right
conditions are met" (Bybee, this volume).
c. The combination of the modal sense and the Past sense is said to
produce a hypothetical reading; compared to Present tense modals,
modals in the Past tense may then express hypothetical situations
in conditional apodosis, as can be seen in (30a), as opposed to
(30b):
(30) a. If I saw Judy, I would tell her the news.
b. If I see Judy, I will tell her the news.
d. In the context of a dialogic discourse (typically involving first and
second person referents), the hypothetical sense may then be ex­
ploited to express a deferential or polite sense, as in (29) above,
e.g., to tone down or weaken the illocutionary force of an otherwise
assertive speech act, as described in the above quotation by
Fleischman.
As is to be expected in such processes of grammaticalization, with the
development of new senses, old uses may gradually lose in significance. For
example, the more the hypothetical sense gains ground, the more the past
sense disappears. Stage (a) is no longer relevant for English Past tense modals
such as would and could', for should, Bybee (This volume) adds that "one
could argue that there are no past uses at all".
I do not wish to delve into the question as to whether the development
sketched in (a) through (d) is appropriate only to languages with a fuzzy/
overlapping zone between the morphological categories of past tense and
subjunctive, or to languages lacking a subjunctive category altogether. In
German, for example, Bybee's scenario does not seem to apply in cases where
there is a well marked formal distinction between Past tense and Subjunctive:
Agent-Oriented vs. Epistemic Modality 45

the role of a hypothetical or deferential sense is invariably assumed by the

Subjunctive rather than the Past tense, as can be seen in (31).
(31) Ich kann Ihnen helfen.
'I can help you.' PRESENT

Ich könnte Ihnen helfen. SUBJUNCTIVE

(Similar meaning, but more hypothetical, or polite, deferential)
Ich konnte Ihnen helfen.
'I was able to help you.' PAST
(Past significance only)

It would seem, however, that this question does not touch the main point made
by Bybee. What is obvious, is that Bybee's account offers a legitimate
alternative to that of Fleischman, one that rests on a series of inferences that
can be described in terms of context-induced reinterpretation (Heine, Claudi
and Hiinnemeyer 1991, Ch. 3).
Bybee (this volume) argues that Fleischman's metaphorical explanation
of a transfer from temporal distance to social distance in certain politeness
forms does not account for the fact that it is always modal verbs that are
involved in this transfer. Yet Bybee's account does not explain how it is that
modals which express a situation involving unmet conditions are also used for
marking a certain kind of social relationship (politeness). What I wish to argue
here again (cf. Heine, Claudi and Hiinnemeyer 1991), is that an explanation
for a conceptual shift in the process of grammaticalization must have two
components: one relates to the macro-effects of the process which are sugges­
tive of a shift from one domain of human conceptualization to another, in this
case from the domain of time to that of social relations; the other relates to the
micro-effects which are suggestive of a gradient sequence of contextual and
inferential extensions leading to the emergence of chain-like use patterns of
the linguistic items concerned.
While Fleischman's analysis highlights the macro-level of the process,
Bybee's approach perspectivizes the micro-level. The way in which the two
models are interrelated is sketched graphically in Figure 1 (cf. Heine, Claudi
and Hiinnemeyer 1991, Ch. 4). Thus, rather than contradicting or excluding
one another, the two models complement each other and are both required for
a better understanding of grammaticalization.
46 Bernd Heine

Domain
TIME SOCIAL RELATIONS

Context

Figure 1. A metaphor-and-context model for some of the senses of English would and
should.

5. Conclusions

In previous research it has been established that epistemic uses of modals tend
to develop out of agent-oriented uses; that is, the latter are older than the
former (see especially Shepherd 1982; Bybee and Pagliuca 1985; Traugott
1989; Bybee, Pagliuca and Perkins 1994.). From this it follows that contexts
associated with agent-oriented modality reflect more conservative or less
grammaticalized use patterns than contexts associated with epistemic modal­
ity. While there may be exceptions, this conclusion seems to hold true for
most of the data considered here; it is also confirmed by diachronic analysis.14
With reference to German this would mean in particular that the most con­
servative behavior is encountered if
a. modals such as möchten 'would like to', dürfen 'be allowed to,
may', or wollen 'want', rather than modals like sollen 'should' or
können 'can' are involved,
b. the main verb is an action verb like kommen 'come',
c. the modal is used in the perfect or the past tense,
d. the modal occurs in interrogative rather than in declarative utter­
ances,
e. the modal occurs in negative rather than in affirmative utterances, or
f. if the subject is a first or second person, rather than a third person
referent.
Agent-Oriented vs. Epistemic Modality 47

Some related observations have also been made for English modals. Traugott
(1989:52), for example, notes that older meanings tend to be maintained
longer in negative environments, and draws attention to the relative paucity of
epistemic (as opposed to agent-oriented) uses of must not and mustn't in
British English and to the maintenance of the volitional sense of will in We
won't go.
One main claim made in this paper is that it is not possible to describe the
transition between agent-oriented and epistemic modality in German modals
exclusively in terms of linguistic categorization. What has to be taken into
consideration in addition are the contextual frames with which particular uses
of a modal are associated. Utterances having human agents as their subject
and dynamic main verbs may still have a focal epistemic sense if associated
with a contextual frame that discourages an agent-oriented interpretation.
More important, however, is the presence vs. absence of the modal force F in
determining which variety of modality a given modal utterance receives.
Finally, as we saw in Section 4, the transition from one kind of modality
to another has elements of discontinuity as well as continuity, and can thus be
accounted for by appeal to both the metaphor model and the context model of
grammaticalization. In this respect, the problem looked at in this paper does
not differ significantly from other instances of grammaticalization studied so far.

NOTES

1 I wish to express my gratitude to Werner Abraham, Joan Bybee, Eithne Carlin, Wallace
Chafe, Ulrike Claudi, Zygmunt Frajzyngier, Tom Givón, Christa Kilian-Hatz, Christa
König, Andreas "Donald" Lessau, Frank Lichtenberk, Franz Potyka, Heinz Roberg,
Dan Slobin, Thomas Stolz, Eve Sweetser, as well as a number of other participants of the
Albuquerque symposium on Mood and Modality for valuable discussions on this paper,
most of all to Suzanne Fleischman. I am also indebted to the Deutsche Forschungs­
gemeinschaft (German Research Society) for its financial support.
2 In the following, y will use the terms "agent-oriented" and "deontic" as defined by
Bybee, Pagliuca and Perkins (1991, 1994), with slight modifications to be specified
below. Note that the distinctions referred to above are not only terminological; for
Coates (1983), for example, "root modality" embraces "deontic" and "dynamic catego­
ries", and "epistemic modality" also includes "alethic" modality.
3 See Steele et al. (1981:260-264) and Heine (1993) for additional exemplification.
4. Note, however, that, compared to full-fledged verbs, they have a reduced paradigm of
personal inflections, lacking e.g. the present tense 3sg. suffix -t; see below.
 48 Bernd Heine

5 Square brackets "[]" are used in this paper for contextual information relevant for the
semantic interpretation of an utterance.
6 While it is hoped that the observations made in the course of the survey will reveal some
salient semantic characteristics of German modals, the survey is biased in several ways.
For example, the three informants, all students of the University of Cologne having some
acquaintance with linguistics, had to take the three-fold classification proposed here for
granted in their responses.
7. The situation would be different if subjunctive uses of modals were included.
8 I am referring to the non-standard progressive of the form Er ist am Essen 'He is eating'
which, although not acceptable in High German, is nevertheless found in some form or
other in many modern German dialects.
9 Cf. the notion of preference as defined by Givón (1990:529).
10. For further evidence on the metaphorical nature of conceptual shift in grammaticaliza-
tion, see Heine, Claudi and Hünnemeyer 1991.
11 The examples provided by Traugott (1989:207) involve such developments as that from
GO to future, from COME to perfect, or from BE AT/IN to progressive. The process
concerned is described by Traugott as one involving "spatio-temporal metaphors".
12 In addition to these two models, a third model has to be distinguished, namely the
bleaching or containment model (Tom Givón, p.c.; see Heine, Claudi and Hiinnemeyer
1991:108ff.). We will not further deal with this model here, essentially since it is
confined to semantic substance while our primary concern is with the cognitive and
pragmatic strategies underlying the process of grammaticalization (but see below).
13 As noted above, this does not necessarily apply to werden, whose status as a modal is controversial.
14 The present paper is based exclusively on synchronic findings.

REFERENCES

Abraham, Werner 1989. "Futur-Typologie in den germanischen Sprachen." Abraham and

Janssen 1989. Pp. 345-369.
Abraham, Werner 1992. The Aspectual Source of the Epistemic-Root Distinction of
Modal Verbs. Paper presented at the Symposium on Mood and Modality,
Albuquerque, University of New Mexico, May 8-10, 1992.
Abraham, Werner and Theo Janssen 1989. Tempus - Aspekt - Modus: Die lexikalischen
und grammatischen Formen in den germanischen Sprachen. Tübingen: Max
Niemeyer Verlag.
Ahlqvist, Anders ed. 1982. Paper from the Fifth International Conference on Historical
Linguistics. Amsterdam: John Benjamins.
Aijmer, Karin 1985. The semantic development of will. Fisiak 1985. Pp. 11-21.
Bybee, Joan L. and William Pagliuca 1985. Cross Linguistic Comparison and the Devel­
opment of Grammatical Meaning. Fisiak, J. 1985. Pp. 59-83.
Bybee, Joan L., William Pagliuca, and Revere D. Perkins 1991. Back to the Future.
Traugott, Elizabeth C. and Bernd Heine 1991.2. Pp. 17-58.
Agent-Oriented vs. Epistemic Modality 49

Bybee, Joan L., Revere D. Perkins, and William Pagliuca forthc. The Grammaticization of
Tense, Aspect, and Modality in the Languages of the World. Chicago, London:
University of Chicago Press.
Bybee, Joan L. This volume. The Semantic Development of Past Tense Modals in
English. Paper prepared for the Symposium on Mood and Modality, University of
New Mexico, Albuquerque, 8 - 1 0 May, 1992.
Calbert, Joseph P. 1971. "Modality and Case Grammar." Working Papers in Linguistics
(Ohio State University), August 1971:85-132.
Claudi, Ulrike and Bernd Heine 1986. "On the Metaphorical Base of Grammar." Studies
in Language 10,2:297-335
Coates, Jennifer 1983. The Semantics of the Modal Auxiliaries. London/Canberra: Croom
Helm.
Fisiak, Jacek, ed. 1985. Historical Semantics, Historical Word Formation. Berlin:
Mouton de Gruyter.
Fleischman, Suzanne 1989. "Temporal Distance: A Basic Linguistic Metaphor." Studies
in Language 13,1:1-50.
Givón, T. 1990. Syntax: A Functional-Typological Introduction. Volume II. Amsterdam/
Philadelphia: John Benjamins.
Haiman, John 1985. Iconicity in Syntax. (Typological Studies in Language, 6.) Amster­
dam/Philadelphia: John Benjamins.
Heine, Bernd 1992. "Grammaticalization Chains." Studies in Language 16,2:335-68.
Heine, Bernd 1993. Auxiliaries. New York: Oxford University Press.
Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer 1991. Grammaticalization: A
Conceptual Framework. Chicago: University of Chicago Press.
Helbig, Gerhard and Joachim Buscha 1988. Deutsche Grammatik. Ein Handbuch für den
Ausländerunterricht. Eleventh edition. Leipzig: VEB Enzyklopädie.
Jacobs, R. and P. Rosenbaum, eds 1970. Readings in English Transformational Gram­
mar. Waltham, Mass.: Ginn and Co.
Jenkins, L. 1972. Modality in English Syntax. Indiana University Linguistics Club:
Mimeograph.
Jespersen, Otto 1924. The philosophy of grammar. London: Allen and Unwin.
Lakoff, George & Mark Johnson 1980. Metaphors We Live by. Chicago/London: Univer­
sity of Chicago Press.
Langacker, Ronald W. 1985. "Observations and Speculations on subjectivity." In Haiman
1985. Pp. 109-150.
Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Pérez, Aveline 1990. "Time in Motion: Grammaticalisation of the Be Going to Construc­
tion in English." La Trobe University Working Papers in Linguistics 3:49-64.
Perlmutter, D. M. 1970. "The Two Verbs 'Begin'." In Jacobs, R. and P. Rosenbaum 1970.
Pp. 107-119.
Ramat, Anna Giacalone 1991. "Grammaticalization Processes in the Area of Temporal
and Modal Relations." To appear Studies in Second Language Acquisition 14 (1992).
Shepherd, Susan C. 1982. "From Deontic to Epistemic: An Analysis of Modals in the
History of English, Creoles, and Language Acquisition." In: Ahlqvist, Anders 1982.
Pp. 316-323.
 50 Bernd Heine

Steele, Susan M., Adrian Akmajian, Richard Demers, Eloise Jelinek, Chisato Kitagawa,
Richard Oehrle, and Thomas Wasow 1981. An Encyclopedia of AUX: A study in
Cross-Linguistic Equivalence. (Linguistic Inquiry Monographs, 5.) Cambridge,
Mass./London: MIT Press.
Stephany, Ursula 1989. "Modality in First Language Acquisition: The State of the Art."
To appear in Proceedings of the Symposium on Modality in Language Acquisition,
Berlin, Free University, 1-3 May, 1989.
Stolz, Thomas 1991. Von der Grammatikalisierbarkeit des Körpers. I: Vorbereitung.
Typescript, University of Bochum.
Sweetser, Eve Eliot 1982. Root and Epistemic Modals: Causality in Two Worlds."
Berkeley Linguistics Society 8: 484-507.
Sweetser, Eve Eliot 1984. Semantic Structure and Semantic Change: A Cognitive Linguis­
tic Study of Modality, Perception, Speech Acts, and Logical Relations. Ph. D. disserta­
tion, University of California at Berkeley.
Sweetser, Eve Eliot 1988. Grammaticalization and Semantic Bleaching. Berkeley Lin­
guistics Society 14:389-405.
Traugott, Elizabeth C. 1989. "On the Rise of Epistemic Meanings in English: An Example
of Subjectification in Semantic Change." Language 65,1:31-55.
Traugott, Elizabeth C. & Bernd Heine, eds 1991.1, 1991.2. Approaches to Grammatical­
ization. 2 volumes. Amsterdam/Philadelphia: John Benjamins.
Traugott, Elizabeth C. and Ekkehard König 1991. "The Semantics-Pragmatics of Gram­
maticalization Revisited." Traugott, E. C. and B. Heine 1991.1. Pp. 189-218.
Vater, Heinz 1970. On the Generation of Modal Verbs. Paper read at the Ethno-Linguistic
Seminar, Bloomington, Indiana.

ABBREVIATIONS

A/E = focal agent-oriented/epistemic sense

Al/El = non-focal agent-oriented/epistemic sense
A2/E2 = marginal agent-oriented/epistemic sense

APPENDIX

Agent-oriented vs. epistemic modality of eight German modals

The data presented below have been elicited from three native speakers of High German.
While a number of discrepancies were noted among these informants, the overall ratios
turned out to be essentially the same.
Agent-Oriented vs. Epistemic Modality 51

A. Linguistic data

1. Present
Er muß kommen A E2 Er muß Geld haben E Er muß sie kennen E
Er soll kommen A E Er soll G. haben E Er soll sie kennen E
Er kann kommen A E2 Er kann G. haben A E Er kann sie kennen E
Er will kommen A Er will G. haben A E Er will sie kennen E
Er mag kommen A E2 Er mag G. haben A2 E Er mag sie kennen E
Er darf kommen A Er darf G. haben A Er darf s. kennen
Er möchte kommen A Er möchte G haben A Er möchte s. kennen A
Er wird kommen A E2 Er wird G. haben A2 E Er wird sie kennen A2 E

2. Past
Er mußte kommen,. A Er mußte Geld haben. Er mußte sie kernten.
sollte A E sollte E sollte
konnte A konnte A konnte
wollte A wollte A wollte E2
mochte A mochte A2 mochte
durfte A durfte A durfte A

3. Perfect
hat kommen Er hat Geld haben Er hat ihn kennen
müssen. A müssen. A2 müssen. A2
sollen. A sollen. A2 sollen. A2
können. A können. können. A2
wollen. A wollen. A wollen. A2

4. Negation
Er muß nicht kommen. A Er muß kein Geld haben. Er muß sie nicht kennen. A2 E
soll A soll A soll Al E
kann A kann Al E kann E
will A will A will A E
mag A mag A2 mag A El
darf A darf A darf A
möchte A möchte A möchte A
wird A wird Al E wird E
52 Bernd Heine

5. Question
Muß er kommen? A Muß er Geld haben ? A2 Muß er sie kennen? A E
Soll A Soll A1 Soll
Kann A Kann A El Kann E
Will A Will A Will El
Mag A Mag Al Mag
Darf A Darf A Darf A
Möchte A Möchte A Möchte A
Wird A Wird Al Wird E2

6. Main verb in perfect

Er muß gekommen sein. E Er muß Geld gehabt haben. E Er muß sie gekannt haben. E
soll E soll E soll E
kann E kann El kann E
will E will E will E
mag E mag E mag E
darf darf darf A2
möchte möchte E2 möchte A2
wird E wird E wird E

7. Second person subject

Du mußt kommen. A Du mußt Geld haben. E Du mußt sie kennen. E
sollst A sollst E sollst E
kannst A kannst A E2 kannst E2
willst A willst A E2 willst E2
magst magst E2 magst E
darfst A darfst A darfst A2
möchtest möchtest A möchtest A2
wirst A1 wirst Al E wirst E

8. First person subject

Ich muß kommen. A Ich muß Geld haben. A Ich muß sie kennen. E
soll A soll E2 soll E
kann A kann A1 kann El
will A will A will A2
mag A mag E2 mag E
darf A darf A2 darf Al
möchte A möchte A möchte A
werde A werde A werde E
Agent-Oriented vs. Epistemic Modality 53

0 = minimal value)
Index of modality
(1.00- maximal,
B Quantitative data (Based on A above)

modality
Value of

Maximum number of points

Number of instances
i

J
J

Agent-oriented
Agent-oriented

Epistemic

Epistemic
Kind of proposition

1 Present (3rd person subject) 41 43 24 72 0.57 0.60

2 Past 33 10 18 54 0.61 0.19
3 Perfect 21 0 12 36 0.58 0
4 Negation 56 23 24 72 0.78 0.32
5 Question 52 11 24 72 0.72 0.15
6 Main verb in perf. 2 54 24 72 0.03 0.75
7 2nd p. subject 33 26 24 72 0.46 0.36
8 1st p. subject 39 16 24 72 0.54 0.22

Kind of expression

Er kommt. 137 28 58 174 0.79 0.16

Er hat Geld. 94 61 58 174 0.54 0.35
Er kennt sie. 46 94 58 174 0.26 0.54

Kind of modal

Dürfen 47 0 7 63 0.75 0
Möchten 41 1 6 54 0.76 0.18
Wollen 48 23 8 72 0.67 0.32
Können 38 30 8 72 0.53 0.42
Müssen 31 31 8 72 0.43 0.43
Mögen 23 25 7 63 0.37 0.40
Werden 25 32 6 54 0.46 0.59
Sollen 30 40 8 72 0.42 0.56
 The Expression of Root and Epistemic
Possibility in English1

Jennifer Coates
Roehampton Institute

1. The root/epistemic distinction

The distinction between root (or agent-oriented or deontic) modality and

epistemic modality has proved enormously useful to those attempting to
describe the modal systems obtaining in the world's languages. The analysis
of English is no exception (see Coates 1983; Haegeman 1983; Palmer 1990).
Moreover, there is general agreement on the definition of these terms. Epis­
temic modality is concerned with the speaker's assumptions or assessment of
possibilities, and in most cases it indicates the speaker's confidence or lack of
confidence in the truth of the proposition expressed. Root modality encom­
passes meanings such as permission and obligation, and also possibility and
necessity. This means that the root/epistemic distinction cuts across the neces­
sity/obligation and possibility/permission distinctions (see diagram below).
As in many languages, in English the same linguistic forms express both root
and epistemic meanings. So, for example, may can express both root and
epistemic possibility; must can express both root and epistemic necessity.

Root Epistemic

CAN permission possibility possibility MAY

MAY

MUST obligation necessity necessity MUST

HAVE TO ← → HAVE TO

Figure 1. Meaning and the root/epistemic distinction

56 Jennifer Coates

As far as the expression of necessity is concerned, this polysemy is unprob-

lematic: the root/epistemic distinction remains distinct, as the following ex­
amples illustrate. Must, in (la), expresses the root meaning of obligation, in
(lb) it expresses the (weaker) root meaning of necessity, and in (lc) it
expresses epistemic necessity.

(1) a. You must finish this before dinner ROOT

b. All students must obtain the consent of the Dean ROOT
c. I must have a temperature EPISTEMIC

It is my impression, however, that there is some confusion about the root/

epistemic distinction when it is applied to possibility. In this paper, I want to
re-examine this area of modality, and will argue that: (i) the semantic contrast
between root and epistemic possibility is considerably weaker than in other
root/epistemic pairs; (ii) where the same linguistic form expresses both root
and epistemic possibility, instances of merger (see Coates 1983:17) are
common (may); (iii) where a linguistic form expresses predominantly root
possibility, epistemic readings are likely to develop (can).

2. Root and epistemic possibility

In order to discuss the root/epistemic contrast in relation to the expression of

possibility in English, I shall make use of the set of properties developed by
Heine (this volume) in his analysis of the German modals. Heine argues that
the following properties are criterial:
a. There is some force F that has an interest in an event either occurring or
not occurring (Heine comments that F "is characterised by some 'ele­
ment of will', to use the wording of Jespersen (1924:320-1)". (Heine this
volume)
b. That event is to be performed by some agent A.
c. The event is dynamic (D).
d. The event has not yet taken place at reference time and, if it does take
place, that will be at a time later than reference time (L).
e. The event is non-factual, but there is a certain degree of probability that it
will occur (P).
The Expression of Root and Epistemic Possibility in English 57

In an utterance such as she must go to bed now, F may refer to the speaker
(T insist that she goes to bed now') or to an absent parent ('her mother insists
that she goes to bed at this time') or to any other source of power the speaker
may have in mind. The pronoun she is the agent (A), the event — go — is
dynamic (D), the event has not yet taken place at the moment the utterance is
produced (L), and there is a high probability (P) that the event referred to will
take place.
As Heine (this volume) says: "While prototypical instances of agent-
oriented modality are characterised by the presence of the properties [i.e. F,
A, D, L and P], prototypical instances of epistemic modality lack all proper­
ties except P". This means that the fewer of these properties there are in any
given instance, the weaker will be the semantic contrast between the two
interpretations of these sentences. If we look at the examples given in (1)
above, we can justify the claim that the semantic contrast between root and
epistemic meaning is strong here by applying Heine's criteria. In (la) and (lb)
all of the properties are present, while only P is present in (lc). In other words,
examples of root must in English would normally be classified as prototypical
examples of agent-oriented modality, while examples of epistemic must
would be classified as prototypically epistemic.
Let's turn to the contrast between root and epistemic possibility. Examples
(2) and (3) below are typical instances of the expression of root possibility:2
(2) well I think there is a place where I can get a cheap kettle
(S. 1.4.62)
(3) I am afraid this is the bank's final word. I tell you so that you may
make arrangements elsewhere. (W.7.9.37)
Can (see example 2) is the normal exponent of root possibility in English,
while may, as illustrated in example (3), is the exponent of root possibility in
more formal contexts (in this case, a letter from a Bank Manager). Properties
A, D, and L are present in both (2) and (3), but F (force) is absent.
In everyday discourse, can expressing root possibility is most commonly
found in examples like (4) and (5) below, general statements of possibilities
with impersonal subjects.
(4) certain things can be sex-linked to the Y chromosome (S.5b.2.54)
(5) first thing in the morning they come, you can hear the whistle
(S.1.14A.43)
 58 Jennifer Coates

These two examples display none of Heine's properties apart from P. Even
more perplexing are archetypal examples of can such as (6) below, which
Palmer (1990:152-4) describes as 'existential':
(6) Lions can be dangerous
Palmer justifies his use of the term 'existential' by claiming that paraphrases
involving 'some' {some lions are dangerous) or 'sometimes' {lions are some­
times dangerous) are more appropriate than paraphrases using 'possible for'.
What is intriguing about examples like (6) is that, besides lacking F, A, D, and
L (the properties associated with non-epistemic modality), they also lack P. In
other words, some examples of can lack all the properties which Heine claims
are normally associated with modal meaning.
Although we have seen that examples of root possibility vary in the
number of properties associated with them (examples (2) and (3) are associ­
ated with properties A, D, L and P, examples (4) and (5) are associated only
with P, while example (6) is associated with none of the properties), they all
share one characteristic: absence of F. In his analysis of German modals,
Heine argues that only F is obligatorily connected with root modality: "once
the use of a modal is characterised by a lack of F then it lacks that 'element of
will' that Jespersen (1924:320-1) referred to, and we are dealing with an
epistemic rather than an agent-oriented sense" (Heine, this volume). On the
basis of this argument, (2), (3), (4), (5) and (6) all fail as instances of root
modality.
Let's look now at the expression of epistemic possibility in English.
Epistemic possibility has many exponents, notably maybe, perhaps, I think,
possibly, probably and the modal auxiliaries may, might and could. Examples
are given in (7), (8), (9) and (10) below.

(7) that may be yellow fever, I'm not sure (S.4.2.65)

(8) I may be a few minutes late, but don't know (S.7.3E.6)
(9) / think it's unlikely actually but he might do it today (S.8.1A.18)
(10) The only snag is that it has been raining ... and I could get held up
for anything up to a week. (W.7.2.29)
These examples are not associated with F. However, examples (8), (9) and
(10) are associated with L (since they refer to events that are to occur later
than reference time), and example (9) is also associated with A and D. Thus,
only (7) is prototypically epistemic in Heine's sense.
The Expression of Root and Epistemic Possibility in English 59

Example Properties Type of modality

F A D L P

la + + + + + root obligation
1b + + + + root necessity
lc + epistemic necessity
2/3 - + + + + root possibility
4/5 +
6 - - - - - 'existential'
7
8
9
-
+ +
+
+ + epistemic possibility

Figure 2.
+
Matrix showing presence or absence of Heine's properties

A summary of the properties associated with examples given so far is

presented in the figure above. This shows that, using Heine's criteria alone,
we are unable to distinguish clearly between instances of root and epistemic
possibility in English. This figure includes examples (la), (lb) and (lc) to
show how clearly Heine's properties distinguish between root and epistemic
necessity. But in the case of examples of root and epistemic possibility in
English, this table would force us to conclude that there is often no difference
(cf. examples (2) and (9) which share four properties; examples (4) and (7)
which both have one property, P). While the use of these criteria has helped to
demonstrate that the distinction between root and epistemic possibility is
weak, it is not the case that examples are indistinguishable. So what distin­
guishes root from epistemic possibility?
The crucial distinction between forms expressing root possibility in
English and forms expressing epistemic possibility in English is that the latter
involve Subjectivity (which I shall refer to as S). Forms involving S can be
defined as 'devices whereby the speaker, in making an utterance, simultane­
ously comments upon that utterance and expresses his attitude to what he is
saying' (Lyons 1977:739). As examples (7), (8), (9) and (10) above illustrate,
S is an integral component of the expression of epistemic possibility. The
speakers in (7), (8), and (9) and the letter-writer in (10) are not only making
statements but are simultaneously expressing their lack of confidence in the
propositions expressed in these utterances. In every case here, speaker's
uncertainty is encoded in an accompanying phrase which reinforces the
 60 Jennifer Coates

modal: I'm not sure in (7), don't know in (8), I think it's unlikely in (9), and the
only snag is in (10). (We can compare earlier examples of root possibility: I can
get a cheap kettle (2) and you can hear the whistle (5). Both are statements of
fact; subjectivity is not involved. The difference between I can get a kettle and I
may get a kettle is that in the latter the speaker's uncertainty is encoded too.)
We therefore need to add S to the matrix to show that this acts as the
criterial property where the expression of possibility is concerned, (see Figure 3)
If we ignore P, which is common to both root and epistemic modality
(except in unusual cases like (6)), we can see that the reason for the weakness
of the root/epistemic distinction in the expression of possibility is the absence,
in utterances involving root possibility, of properties normally associated with
root meaning. Exponents of root necessity (e.g. lb) differ from exponents of
epistemic necessity (e.g. lc) on five measures, whereas exponents of root
possibility (e.g. 2/3) may differ from exponents of epistemic possibility (e.g.
9) by as little as one property (presence or absence of S).
One of the reasons that the weakness of this distinction has been unprob-
lematic for speakers of English is that root and epistemic possibility are, by
and large, expressed by different linguistic forms. May is the only modal form
which regularly expresses both root and epistemic possibility, and when
expressing root possibility it is restricted to the most formal contexts (as in (3)
above). The extent of overlap between the two forms is small, as the following
statistics show (these record my analysis of a representative sample of 200
cases of can and 200 cases of may, all examples of spoken British English
taken from the Survey of English Usage):

Example Properties Type of modality

F A D L P S

la + + + + + - root obligation
lb + + + + + - root necessity
2/3 — + + + +
root possibility
4/5 + :}
6 - - - - - - 'existential'

lc - - - - + + epistemic necessity
7 - - - - +
8
9
-
-
-
+
-
+
+
+
+
+ + epistemic possibility

+
Figure 3. Matrix to show the distinction between root and epistemic meanings in English

+
 The Expression of Root and Epistemic Possibility in English 61

Table 1. The use of can and may in contemporary spoken English

can: Root Possibility 129 may. Epistemic Possibility 147

Ability 41 Permission 32
Permission 10 Root Possibility 7
Undecidable 20 Valediction 1
TOTAL 200 Undecidable 13
TOTAL 200

As these figures demonstrate, each linguistic form is overwhelmingly associ­

ated with one particular meaning {can with root possibility, may with epis­
temic possibility). The category 'undecidable' refers to those examples which
I could not assign unproblematically to one meaning rather than another.
Often, this was because there was not enough contextual evidence to permit a
definite reading; in the case of may, however, some examples exhibit
'merger', that is, the phenomenon whereby utterances containing may com­
bine elements of both root and epistemic meaning (see 3.1 below).

3. The linguistic consequences of a weak root/epistemic contrast

In the previous section, I have demonstrated the weakness of the root/epis­

temic distinction in the expression of possibility. As I have argued, the
blurring of the root/epistemic boundary in this semantic area is not often
problematic in English, given the use of different linguistic forms. However,
there are certain interesting linguistic developments in this area which are a
direct result, I would argue, of the weakness of the distinction. One involves
the growing instances of merger with may. The other is the development of
epistemic readings for can in contemporary spoken American English.

3.1 May and merger

Merger, as I have explained elsewhere (see Coates 1983; Leech and Coates
1980), refers to instances where two meanings co-exist in a both/and relation­
ship. In other words, two readings are available for a given utterance, but
instead of having to choose one meaning and discard the other (as with
ambiguous examples), the hearer is able to process both meanings. Merger
occurs quite frequently in more formal texts (example 11) and is becoming
endemic in academic writing (as example (12) illustrates):
62 Jennifer Coates

(11) or the pollen may be taken from the stamens of one rose and
transferred to the stigma of another (W. 10.3.27)
(12) ... the process of simplification ... through which even forms and
distinctions present in all the contributory dialects may be lost
(Trudgill 1986:126)
In both these examples, the only property clearly present is P. Properties
associated with root meaning (F, A, and D) are absent, while S, normally
criterial for epistemic meaning, is not typically associated with this kind of
formal style with its passives and inanimate subjects. So instances such as
these lack clear markers of either root or epistemic meaning. The two mean­
ings merge, and the reader is not required to choose one or the other:3

(11') Root: 'it's possible for the pollen to be taken ...'

Epistemic: 'it's possible that the pollen will be taken ...'
(12') Root: 'it's possible for forms and distinctions ... to be lost'
Epistemic: 'it's possible that forms and distinctions ... will be lost'
It is significant that such examples of merger come typically from the
(formal) written domain. In spoken language, exponents of root and epistemic
possibility are normally distinguished prosodically: for example, may and can
do not receive stress when expressing root possibility, but may, when it
expresses epistemic possibility, is normally stressed. Epistemic may is also
typically associated with fall-rise intonation (see example (13) below):
(13) I may be wrong (S1.2.38)
While writers can exploit the both/and relationship of root and epistemic
meanings when talking about possibilities, speakers are constrained by pro­
sodic factors to choose one or the other. If a speaker uttered example (11) with
a fall-rise nucleus on may, for example, the only interpretation available to
hearers would be epistemic.

3.2 The development of epistemic can

Given the historical pattern of epistemic meanings developing from root

meanings (Bybee & Pagliucca 1985; Traugott 1989), it would not be surpris­
ing if can were to develop an epistemic reading. Moreover, the homogeneous
picture of can given so far in this paper is not the whole truth: can does have
some well-established specialised epistemic uses.
The Expression of Root and Epistemic Possibility in English 63

First, it provides the missing negative form in the epistemic must para­
digm (see Palmer, this volume). The invariant form can't (not cannot or can
not) expresses 'it's necessarily the case that ... not ...' (nec ~ p) or 'it's not
possibly the case that' (~ poss p) in examples such as (14). (Note the stress on
can't and the fall-rise intonation contour.)
(14) [speaker describes friends arriving early]
I almost phoned them up and said come a bit later — and then I
thought oh they 've probably left by now — so I didn 't and — twelve
thirty, now that... can't be them, and it was (S.2.7.6)
Second, can is used in interrogative constructions to express epistemic
possibility. Example (15) is a cliché of pop music and Hollywood-style films,
but it makes the point clear:
(15) Can it be true?
['Is it possible that this is true? that she loves me?']
Such examples have a clear relationship with the use of can't discussed above,
in that (15) could be glossed it can't be true! (it must be false). Example (16)
comes from a radio discussion of Government policy on pensions:
(16) Can that be sensible ?
['Is it possibly the case that that is sensible?']
The speaker was clearly trying to make the point That can't be sensible!
In British English, these are still the only contexts in which a form of can
is used with epistemic meaning (though could is making headway as an
alternative to might in the expression of tentativeness — see example 10). But
in American English, can is starting to appear in other contexts. The following
example occurred during the Symposium on Mood and Modality (held at the
University of New Mexico in May, 1992) as a participant finished her presen­
tation:
(17) we hope this coding system can be useful [to other linguists work­
ing in the field]
This utterance meant something like 'we hope there's a chance that this
system will be useful'. For British speakers, this utterance is not possible: a
British speaker would have to say 'we hope this coding system will be useful',
thereby losing the subjective force. As the person who uttered (17) com-
64 Jennifer Coates

mented to me afterwards, can in this utterance is 'a sort of hedge'. American

speakers seem to have no problems with utterances of this kind.
So if can is developing an epistemic meaning in the United States but not
in Britain, what is different in the linguistic environment in the United States?
I would like to suggest three factors:
a. can is less commonly used to express permission in American English;
may is the normal exponent of permission (Coates & Leech 1980);
b. the 'bleaching' of root can is further advanced in American English, with
the majority of examples not associated with F, A, D or L;
c. may is the chief exponent of epistemic possibility in British English, but
is less common in American English, where may has connotations of
formality.
More generally, there are many reasons why we would expect can to
develop epistemic possibility readings. First, all the other modal auxiliaries in
English express both root and epistemic meaning. Second, it seems to be the
case that epistemic meanings derive from earlier, non-epistemic meanings
(Traugott 1989:52). Third, the evidence from child acquisition research is
suggestive: children develop deontic meaning much earlier than epistemic
meaning (Stephany 1986); Guo (this volume) claims that the Mandarin form
neng {can) is starting to be used to express epistemic possibility in children's
speech. Fourth, the occurrence of merger with examples of may (illustrated in
Section 3.1) illustrates the fuzziness of the root/epistemic boundary in the
expression of possibility. Fifth, as this paper has attempted to demonstrate,
root and epistemic possibility are only weakly distinguished.
Given these circumstances, and a historic pattern in the development of
can from ability meanings to permission meanings to root possibility mean­
ings (Bybee 1988), I would predict that initially examples of epistemic can
will co-occur with syntactic features such as inanimate subject and stative
verb, and in contexts where accompanying words support an epistemic read­
ing. Example (17) is of this type, with an inanimate subject this coding system,
and the phrase I hope introducing subjectivity to the utterance. However, it is
difficult to imagine can becoming a serious contender in the expression of
epistemic possibility unless it can develop a stressed alternative to the usual
[kən]; until that happens, utterances like I can come will be processed as root.
The Expression of Root and Epistemic Possibility in English 65

4. Conclusions

In this paper, I have argued that the distinction between root and epistemic
meanings is much weaker in the case of possibility than in other areas of
modal meaning. I have demonstrated that the weakness of this distinction
arises from the nature of root possibility, typical examples of which are not
associated with Heine's properties F, A, D and L (normally criterial for root
meaning). As a direct consequence of the weakness of the root/epistemic
distinction, instances of merger are common (in the case of may), and epis­
temic readings are beginning to occur in declaratives with can. Speakers will
exploit the potentialities of the English modal system to say the things they
need to say. Whether this means that can, like the other English modal
auxiliaries, will develop the full range of epistemic meanings remains to be
seen.

NOTES

1 I would like to record my gratitude to Bas Aarts, Joan Bybee, Suzanne Fleischmann,
Talmy Givón, Bernd Heine and Charles Meyer for their comments on earlier drafts of
this paper.
2 Examples from this point onwards will be taken from the Survey of English Usage,
University College, London (prosodic information omitted).
3 Can is also beginning to be involved in merger, in contexts where the speaker/writer
wishes to hedge what they are saying. A nice example occurs on Inland Revenue (UK)
Tax Forms: False statements can result in prosecution.

REFERENCES

Bybee, Joan. 1988. "Semantic Substance Versus Contrast in the Development of Gram­
matical Meaning". Proceedings of the Fourteenth Berkeley Linguistic Society 247-279.
Bybee, Joan & William Pagliuca. 1985. "Cross-Linguistic Comparisons and the Develop­
ment of Grammatical Meaning". Historical Semantics and Historical Word Formation
ed. by J. Fisiak, 59-84. The Hague: Mouton.
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries, London: Croom Helm.
Coates, Jennifer & Geoffrey Leech. 1980. "The Meanings of the Modals in Modern
British and American English". York Papers in Linguistics 8.23-34.
Guo, Jiansheng. This volume. "The Interactional Stucturing of Meaning: Children's Use
and Development of the Mandarin Modal neng 'can'".
66 Jennifer Coates

Haegeman, Liliane. 1983. The Semantics of Will in Present-day British English. Brussels:
Paleis der Academiën.
Heine, Bernd. This volume. "Agent-Oriented vs. Epistemic Modality — Some Observa­
tions on German Modals".
Jespersen, Otto. 1924. The Philosophy of Grammar. London: Allen & Unwin.
Leech, Geoffrey & Jennifer Coates. 1980. "Semantic Indeterminacy and the Modals".
Studies in English Linguistics ed. by Sidney Greenbaum, Geoffrey Leech & Jan
Svartvik, 79-90. London: Longman.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Palmer, Frank. 1990. Modality and the English Modals. (2nd edition). London: Longman.
Stephany, Ursula. 1986. "Modality". Language Acquisition ed. by Paul Fletcher &
Michael Garman. (2nd edition) 375-400. Cambridge: Cambridge University Press.
Traugott, Elizabeth Close. 1989. "On the Rise of Epistemic Meanings in English: An
Example of Subjectification in Semantic Change". Language 65.31-55.
Trudgill, Peter. 1986. Dialects in Context. Oxford: Blackwell.
Contextual Conditions for the Interpretation
of 'poder' and 'deber' in Spanish*

Carmen Silva-Corvalán
University of Southern California

1. Introduction

This study deals with some aspects of the grammar of poder 'can, may' and
deber 'must' in Spanish. Poder and deber are considered to be modal verbs or
modal auxiliaries both on syntactic and semantic grounds (see, among others,
Marcos Marín 1975:211; Narbona 1989; Rivero 1977).1 Many grammars of
Spanish (e.g. Gili Gaya 1976; Hernández 1986; Real Academia Española
1973) include querer 'want', saber 'know', osar 'dare', soler 'be used to', etc.
as 'modal verbs' as well. Syntactically, however, poder, deber, osar and soler
are different in that they share the requirement to occur exclusively in con­
struction with an Infinitive with an equivalent subject, a constraint which does
not apply to querer and saber (see Narbona 1989, and Rivero 1977 for further
discussion of the syntax and semantics of these and other periphrastic verbal
constructions). Semantically, poder and deber stand apart as the only modals
that can make a statement, in at least one interpretation, about the possibility
of p, as the paraphrases of examples (1 a-b), given in (2 a-b), show.

(1) a. Jon puede/debe tocar el clarinete.

b. Jon quiere/sabe/osa/suele tocar el clarinete.
a. 'Jon may/must play the clarinet'
b. 'Jon wants to/knows how to/dares to/is in the habit of playing
the clarinet'
(2) a. It is possible/very likely that Jon plays the clarinet.
b. It is the case that Jon wants/knows how to/dares to play/often
plays the clarinet.
68 Carmen Silva-Corvalán

There are, then, syntactic and semantic arguments for treating poder and
deber as a class in Spanish. In contrast to modal verbs in English, whose
syntactic and semantic characteristics have been studied in depth (see Coates
1983, Perkins 1982 & 1983, Palmer 1977 & 1979, Halliday 1970, and numer­
ous references in these studies and in the present volume), Spanish modals
have received little attention, perhaps due to the fact that their syntactic
characteristics have not appeared to be as distinct as those of the English
modals when compared with other verbs which may occur in auxiliary posi­
tion in verbal periphrases.
Note, however, that poder and deber pose a number of special syntactic
and semantic questions of general and specific interest: What syntactic con­
straints apply to these modal verbs? What is the most adequate way to
describe their semantics? Can a monosemantic approach account for the
various uses oí poder and deber? What is the relation between the messages
conveyed by poder and by deber (as in examples 3-4)? What role do the
linguistic and extra-linguistic contexts play in the interpretation of the mean­
ing of these forms? Indeed, as can be seen in the suggested readings of the
following examples, poder and deber can convey different messages, which
include such notions as ability, permission, possibility, and necessity.

(3) Juan puede venir.

'John can/may come.'
Possible paraphrases:
a. John is able to come.
b. John is allowed to come.
c. It is possible for John to come/that John will come.
(4) Juan debe venir.
'John must come.'
Possible paraphrases'.
a. John has the obligation to come.
b. It is very likely/I can confidently infer that John will come.
In addition, poder and deber cannot be interpreted to convey possibility
in their non-finite forms, as the unacceptabihty of the paraphrases in (5 b) and
(6 b) indicate.
 Interpretation of 'poder' and 'deber' 69

(5) a. Juan va a poder cruzar el río.

'John's going to be able/be allowed to cross the river.'
Unacceptable reading'.
b. *Va a ser posible que Juan cruce el río.
'It may be the case that John will cross the river'
(6) a. Juan va a deber cruzar el río.
'John's going to have to cross the river.'
Unacceptable reading:
b. *Es muy posible que Juan vaya a cruzar el río
'It's very likely that John's going to cross the river'
By contrast, when combined with conditional verb morphology, poder is
interpreted to convey possibility (ex. 7), while deber can only convey (weak)
obligation, as in example (8).
(7) Juan podría hacerlo.
'John could do it'
Possible reading'.
It may be possible for John to do it.
(8) Juan debería hacerlo.
'John should do it'
Unacceptable reading:
*'It's very likely that John would do it'
In this essay I examine the factors that determine the different interpreta­
tions of poder and deber in different contexts, and argue for the adequacy of a
monosemantic approach in the analysis of the semantics of these verbs. I
propose to show that the 'modal verbs' each have an invariant meaning (IM),
but are interpreted to convey different contextual meanings (CM) as a conse­
quence of their interaction with other elements in the context (e.g. animacy of
NPs, aspect, tense).
I approach the problem mainly through an examination of examples of
poder and deber in actual language use.2 My analysis suggests that a monose­
mantic approach is more appropriate for Spanish on at least two grounds: (1)
The different meanings proposed (ability, possibility, permission, politeness)
are shown to be a function of the interaction between the modal and other
70 Carmen Silva-Corvalán

elements (linguistic and extra-linguistic) in the discourse. (2) A monoseman-

tic perspective, which admits the possibility of synonymy in context, can
account for the choice of one "synonymous expression" over another on the
basis of their different basic meanings. The invariant meanings proposed for
poder and for deber justify the choice of one modal over the other in a
particular utterance where both may be interpreted to convey, for instance,
epistemic possibility (more strictly 'epistemic necessity' in the case of deber,
but I will not make a difference between possibility and necessity here). While
the analysis of the data allows me to argue against a polysemantic approach to
the meaning of modals, it also leads me to propose that their contexts of
occurrence may be polysemantic, i.e., at times more than one interpretation of
the meaning of a given modalized utterance is possible. Thus, the inclusion of
a specific context (defined by a combination of features, e.g. animacy and
agentivity of entities) in one or another set of contexts associated with the
inference of different messages (e.g. possibility, permission) is in some cases
fuzzy.

2. Theoretical frameworks

Linguistic studies of modality in natural languages differ at least in three

respects: (1) the definition and types of modality identified; (2) the establish­
ment of what constitutes 'true' modality; and (3) the theoretical approach used
in accounting for the semantics of modality.
Thus, Palmer (1979), for instance, uses modality to refer to the meanings
of the English modal verbs, classified on the basis of their formal features. By
contrast, Halliday (1970) considers modalities to be only those meanings
related to the assessment of probability. A modality is expressed by a form
which represents "the speaker's assessment of the probability of what he is
saying, or the extent to which he regards it as self-evident" (p. 328). There­
fore, though all modal verbs in English may express modalities, in Halliday's
view they are not modality when they express permission, volition, ability,
obligation, etc.3
Bybee (1985:169) explicitly establishes a more sensible and appropriate
difference between mood, a verbal category which has also been considered to
represent the speaker's assessment of the degree of factuality of the proposi­
tion, and modality in the following terms: modality is "a conceptual domain
[including root and epistemic meanings] which may take various types of
Interpretation of 'poder' and 'deber' 71

linguistic expression," while mood refers to a subdivision of this conceptual

domain and is realized as "a marker on the verb that signals how the speaker
chooses to put the proposition into the discourse context." (165) That is,
mood, a grammatical category signalled by verbal affixes, is one among
several means of expressing modalities.
Different theoretical models have been adopted in studies of modality in
English (Coates 1983; Haegeman 1983; Kratzer 1977; Perkins 1983; Sweetser
1982). Haegeman, Kratzer and Perkins propose a basic or core-meaning
approach to account for the meaning of modals,4 arguing that this core
meaning is present in all their uses and is compatible in different contexts with
the various notions which have been associated with the modals (e.g. neces­
sity, permission). Out of context, however, i.e. in their lexical entry form, only
the basic meaning is present.
Coates rejects both a purely monosemantic and a purely polysemantic
approach: the former, she insists, fails to deal with the problem of indetermi­
nacy in real language, while the latter has to assume discrete categories of
meaning in a semantic field characterized precisely by indeterminacy. In­
stead, she seeks to reconcile the two approaches observing, on the one hand,
that the root/epistemic distinction is discrete,5 but on the other, that examples
assigned to the two categories are indeterminate inasmuch as they cover a
range of possible meanings. Quite naturally, then, Coates directs her attention
to 'fuzzy set' theory (Zadeh 1965, 1972), a model which allows her to account
for clearly classifiable discrete cases as belonging to the core of the set, as
well as for various degrees of indeterminacy or gradience as belonging to the
skirt or the periphery.
In contrast, Sweetser views the 'possibility' use of modals as a meta­
phorical extension of root meaning. Adopting the 'force dynamics' semantics
of Talmy (1982, cited in Sweetser 1982), she argues that the reasoning
processes underlying epistemic meanings are "subject to compulsions, obliga­
tions, and other modalities, just as our real-world actions are subject to
modalities of the same sort" (484), i.e. to those modalities (ability, necessity,
permission) considered to convey root or deontic meanings. Accordingly, an
understanding of the epistemic world of possibilities and logical conclusions
in terms of the sociophysical world explains quite naturally the connection
between, for instance, epistemic may as the absence of a barrier to the
speaker's conclusions about the truth of the proposition, and permission
-granting (i.e. root) may as the absence of a potential barrier for the truth of the
proposition in the sociophysical world.
72 Carmen Silva-Corvalán

Sweetser further agrees with the observation made by other linguists (e.g.
Coates, Kratzer, Perkins) that pragmatic factors will determine which world
the modal is taken as operating in. Thus, the identification of a real-world
cause will determine a root reading, while the identification of a body of
premises will determine an epistemic one.6 In either case, the contextual
interpretation of such modals as can and may (as well as of verbs like let and
allow) appears to be that of "taking away a potential barrier" (485). Sweetser
acknowledges that there may be ambiguity between real-world force and
epistemic force, which in turn implies contextual ambiguity, but she leaves
open the question of the basic meaning or basic contribution that different
modals make to the interpretation of an utterance.
Like Coates, I feel that "the imprecision of our knowledge of the world
might be inherent" (p.ll), and that fuzzy set theory may be of use in the
analysis of modality (cf. Sweetser on contextual ambiguity). Contrary to
Coates (and perhaps also to Sweetser), however, I argue that modals have
basic, core meanings along the lines of the definitions proposed by Perkins
(1983), while fuzziness applies rather to the contexts, linguistic and extra
-linguistic, in which modals are used. Fuzziness or graded membership of
contexts allows for various interpretations of modalized propositions, and the
precise point at which one interpretation is no longer possible is usually
difficult to establish. Contextually inferred messages (e.g. ability, possibility),
in principle infinite, have been mistakenly considered to be the, or part of the,
meaning of the modal.7
The question of the basic meanings conveyed by linguistic expressions,
as opposed to the meanings, messages or pragmatic implicatures which these
forms appear to have in specific contexts of use, is a recurrent theme in
semantic-pragmatic studies. With particular respect to closed grammatical
systems (e.g. pronouns, verbal affixes), there seems to be general agreement
that a distinction between basic, invariant, or systemic meaning, and second­
ary, implicated, or non-systemic meanings or functions must be made (Bello
1977; Bull 1971; García 1975; King 1992; Silva-Corvalán 1991). By contrast,
there is little agreement as to what exactly constitutes the semantic substance
of the linguistic units under analysis.
With respect to poder and deber, I propose that these modals may be
characterized as having invariant meanings which account for the use of the
modal instead of a non-modalized verb, or for the choice of one modal over
the other in particular discourse contexts. Poder and deber interact with other
Interpretation of 'poder' and 'deber' 73

linguistic and extra-linguistic elements that contribute to producing the multi­

ple sentential and/or discourse meanings that are compatible with the seman­
tics of the modals.
I am suggesting, then, the existence of three meaning components or types
of meaning in language: (1) de-contextualized, systemic invariant meaning;
(2) contextualized meaning; and (3) prototypical discourse meaning.8 These
meaning components are not specific to modals, but characterize all linguistic
elements; contextualized and prototypical meanings are related to decontextu-
alized invariant meanings in systematic, principled ways which need to be
empirically ascertained.
Invariant meaning (IM) refers to the meaning which underlies, or is
present in, all uses of a modal; contextualized meaning (CM) is the message
which the modal conveys, or the analyst infers that the modal conveys in a
specific context. Contextualized meanings derive from the interaction of the
modal verb with morphosyntactic, semantic, prosodie, and pragmatic factors
which constitute its context of use. Thus, every linguistic element contributes
a specific meaning to the total, but the CM goes beyond the sum of these
meanings because it incorporates pragmatic factors that may not be context
-independent as the IMs of grammatical forms (as opposed to lexical forms)
appear to be. Prototypical discourse meaning (PM) refers to the most frequent
message (contextualized meaning) that the modal conveys in a corpus of
language data.
A PM tends to correspond to the meaning that most language users (and
sometimes even linguists) assign to a form. This confusion arises from the fact
that the inferred message is so frequent that speakers assume it is part of the
meaning of the form (cf. Faltz 1989, cited in Bybee, Perkins & Pagliuca 1994:
Ch. 6).
IMs and CMs correspond in part to what Bosch (1985) calls 'context
independent lexical meaning' and 'context dependent contextual notions',
respectively. Contextual notions are in principle infinite and unique for each
new context, an observation which leads Bosch to propose that "we should
give up the notion of the meaning or interpretation of a sentence or utterance
as an identifiable unit or thing altogether" (257). By contrast, it seems to me
that there is reason to propose an intermediate level of contextual meanings
which are associated with or derived from classes or types of contexts, rather
along the lines of what García (1975:276-7) discusses in terms of "the chief
uses of the meaning" of a form or "great classes of messages", all of which are
74 Carmen Silva-Corvalán

characterized by the same common reason for using one form rather than
another. Thus, though there is in principle no end to the messages that may be
postulated, I would like to propose that CMs may cluster into "great classes of
messages" (the most frequent ones becoming the PMs) that underlie groups of
utterances where the modal is inferred to convey the same or similar meanings.
The term 'modality', as the expression of speaker attitude towards the
contents of a proposition, is a semantic notion covering a range of meanings:
certainty, probability, possibility, belief, obligation, necessity, permission,
volition, intention, doubt, prediction, and denial. The speaker's attitude may
be conveyed through verb morphology (mood), lexically, syntactically, pro-
sodically, or by a combination of these linguistic devices. Here, Ï focus on
lexical and affixal manifestations of epistemic and root modalities. For exam­
ple, the proposition Juan viene mañana 'John is coming tomorrow' can be
modalized to express speaker uncertainty or prediction about the probability
of actualization of the event encoded in the proposition (henceforth p) by
using the modal verbs (henceforth m) poder or deber. Thus, Juan puede/debe
venir mañana, may be paraphrased, for instance, as 'Speaker believes that it is
possible for Juan to come/that it is very likely that Juan will come tomorrow'.
Marked with Future morphology, on the other hand, these modals may not
convey possibility, but rather dynamic or deontic modalities: Juan podrá/
deberá venir mañana 'Juan will be able to/will have to come tomorrow'.
I distinguish between epistemic (EP) and root possibility (RP) contextual
meanings (cf. Coates 1983). Epistemic and root possibility are concerned with
the speaker's assessment of or assumptions about the likelihood that the
content of a proposition is or may become true. EP and RP differ in that the
former involves the speaker in logical inference and has p and m in its scope,
while the latter has only p in its scope and is agent/event oriented, i.e., the
subject of the modalized infinitive is normally a willful agent and/or the
infinitive is a dynamic verb.9 This difference has consequences for negation,
such that negation affects p in the case of epistemic sense, but m and p in the
case of root sense. When an affirmative modal (example 9) interpreted to
convey RP is negated, the negative form is interpreted as a negative fact
(example 10).10 To preserve the possibility interpretation, therefore, root
modality must be negated on p (example 11), while epistemic modality may
be negated on m, as example (12) shows.
Interpretation of 'poder' and 'deber' 75

(9) Ahí pueden fumar. Root possibility

'They may smoke there'
Paraphrase'.
It's possible for them to smoke there.
(10) Ahí no pueden fumar. Permission negated.
'They can't smoke there.'
Paraphrase'.
It is the case that they are not allowed to smoke there.
(11) Ahí pueden no fumar. Epistemic possibility
'They may not smoke there.'
Paraphrase:
It is possible that they do not smoke there.
(12) Juan no debe estar en casa. Epistemic possibility11
'John must not be home.'
Paraphrase'.
It is very likely/I confidently infer that John is not home.
I further use the term root modality to encompass both the deontic and
dynamic categories of modal logic, namely obligation, necessity, permission
(see examples 13-14), and ability (example 15), noting that though ability is
not usually considered to be truly a modality, it is relatable to deontic modal­
ity, and at times may be indistinguishable from RP, as Bybee, Perkins &
Pagliuca (1994) note to be the case in a large number of languages.
(13) Debes comer para sobrevivir. Obligation
'You must eat to survive.'
(14) [Father to young son]
Puedes hablar cuando yo me calle. Permission
'You can speak when I stop talking.'
(15) Juan puede andar en bicicleta; aprendió de pequeño. Ability
'John can ride a bicycle; he learnt when he was a child.'
The invariant meanings (IM) for poder and deber that I propose here are
approximately those that Perkins (1982) postulates for the English modals can
and must.12 The definition of modals is formulated with variables ranging over
different systems of laws and principles and different sets of circumstances or
76 Carmen Silva-Corvalán

contexts of use. It incorporates four elements: 1) the invariant or basic

meaning of the modal, which relates the variables C and X (defined below); 2)
K, representing one of various systems of laws or principles (e.g. social,
magic, natural, rational laws, which embody our knowledge of the world,
including those principles which regulate linguistic interaction) according to
which the modalized proposition can be interpreted; (3) X, the third element,
represents the event, state of affairs, etc. referred to in the proposition; and (4)
C, which stands for a set of circumstances, presupposed or explicitly identi­
fied, under which K is relevant.
The circumstances include the structure of p, which may consist of an
n-place predicate (y,z), where y = subject, z = predicate. This interpretation of
C differs from that of Perkins, for whom C is a variable which may represent
such notions as 'personal interactions', 'deontic source', 'subject-oriented'.
The problem is that it is not clear how Perkins's definition accounts for the
role of linguistic characteristics of p, including mood for instance, in the
interpretation of the meaning of modalized p.
The various elements that bear upon the interpretation of a modal are
summarized in the following formula: K (C..[IM]..X), which I illustrate
below. The ensuing sections study usesofpoder and deber and focus on C and
X in order to examine the interaction between these variables and the specific
semantic contribution proposed for the modals.

3. Poder

I assume that the formula proposed by Perkins (1982) for can captures the
meaning of poder. K (C does not preclude X). This formula incorporates both
the IM and the elements that contribute to the CM. Thus, poder presents the
IM 'does not preclude' and it relates actualization of X to a set of circum­
stances C, such that with reference to a set of principles K, the circumstances
C do not preclude X. The values of K and C are recognized (and specified by
the analyst) in context; these variables (which include morphosyntactic, se­
mantic and prosodic features) plus the modal chosen determine whether X is
interpreted under a dynamic, deontic or epistemic modality (Heine (this
volume) proposes similar variables in his stimulating account of German
modals).
Interestingly, the specific semantic contribution proposed for poder,
'does not preclude', appears to agree with the meaning postulated for can by
Interpretation of 'poder' and 'deber' 11

Bybee, Perkins & Pagliuca (1994: Ch. 6), i.e., 'enabling conditions exist', as
well as with the meaning suggested by Sweetser (1982) in terms of the
presence or absence of 'barriers'. The question might arise why I agree with
Perkins in assigning a negative meaning to poder. This is because it seems to
me that a negative meaning captures more appropriately than a positive
meaning (such as 'enables' or 'makes possible') the sense of 'difficulty
overcome' which appears to be incorporated in most of the CMs of poder.
This is taken up again later in this essay.
Example 16 illustrates how the definition of poder would lead to an
interpretation of ability assuming certain values for K and C. Different values
for K and C would obviously lead to different interpretations.

(16) Mi hija puede contar hasta diez.

'My daughter can count up to ten'
IM: K (C does not preclude X)
(i) K: natural/biological laws (e.g. learning capacity)
(ii) C: (a) speaker's previous experience of p (empirical circumstan­
ces, i.e. speaker's evidence that p is true)
(b) y = agent ('my daughter'); z = event ('count up to 10');
time orientation: generic (i.e. valid for an unspecified past,
present and future time)
(iii) X: my daughter counts up to 10

CM: given humans' learning capacity and the circumstances in C, ad­

dressee interprets that speaker intends to convey the message that y (his/her
daughter) has the ability for z ('count up to ten').
Given an affirmative sentence, the elements in C that appear to bear upon
the CM include (cf. Heine, this volume):
(i) the presence or absence of a deontic source,
(ii) evidentiality or previous experience of p (cf. Givón 1982, Wright
1990),
(iii) degree of agentivity of the subject,
(iv) aspect of the situation (stative, dynamic, etc.),
(v) morphological aspect and mood,
such that various combinations of these elements account for the meanings
traditionally associated with poder, 'ability, permission, possibility', i.e. these
so-called meanings are indeed contextual inferences.
78 Carmen Silva-Corvalán

Pottier (1976) discusses the same three meanings for pouvoir 'can' in
terms of 'pouvoir physique/possibilité interne', 'pouvoir par autorisation/
possibilité externe', and 'probabilité/possibilité neutre' (pp.39-41). In addi­
tion, Palmer (1977:2) observes that epistemic modality should be seen as a
more basic concept "since what is (epistemically) possible includes what can
(dynamically) be done". This led me to think, for a while, that the IM of poder
should explicitly include 'possibility'. It became obvious to me later, how­
ever, that 'possibility' is a CM which may be derived from 'does not pre­
clude', since not precluding X in fact implies (though not equivalently)
making X possible.

3.1 Poder in Present tense form.

The data examined indicate that examples of poder in Present tense affirma­
tive form group into five sets of CMs (or "great classes of messages"),
discussed below, with the following readings: permission, ability, mitigation,
root possibility, and epistemic possibility. It is in principle possible, however,
that other sets of CMs might be identified in a much larger corpus of data.
It must be emphasized that many cases are not easy to fit neatly into one
of these five CMs. The indeterminacy of many cases may be viewed as
stemming from possible indeterminacies of K and C in natural language use,
e.g. the strength of the deontic source, weak indication of evidentiality.
Analogous observations made by Coates in regard to the meaning of English
modals led her to develop a prototype model of analysis. As in her English
data, my data contain a number of examples with contexts which appear to
represent the core set of features associated with a certain CM, but many
examples fall somewhere in the skirt or towards the periphery of the set.
Let us look now at examples of poder in contexts whose features have led
me to propose five different modalities (sets of CMs).

3.1.1 Permission.
Whenever a strong deontic source, i.e. some person or institution that clearly
creates obligation or permission (Lyons 1977:843), is identified either explic­
itly or implicitly, the CM inferred is permission, as in example (14). The most
clearly identifiable context for permission, then, includes a strong deontic
source, an agentive subject, and future time orientation. The transcribed
Interpretation of 'poder' and 'deber' 79

conversations examined do not offer these clearly identifiable contexts. This

is most likely due to the nature of the data, which do not contain pragmatic
situations that would elicit the granting of permission. Example (17) is repre­
sentative: it seems to assume a deontic source, but its non-specific agent and
the generic rather than future time orientation make our interpretation move in
the direction of root possibility.
(17) En España a los dieciocho años ya puedes votar, (M18,p34) 13
'In Spain you can vote when you're eighteen.'
(i) K: social laws (governments have authority over citizens)
(ii) C: (a) deontic source implied (legal code of laws)
(b) y = agent; z = event; time orientation: generic
(iii) X: y,z
CM: y is allowed to z
(Possible CM: it is possible for y to z)
Both deontic source and set of laws under which a specific example is to
be interpreted may lead to indeterminacy. On the one hand, written codes of
laws (e.g. traffic regulations, state laws) constitute strong deontic sources and
contribute to a more clearly identifiable CM of permission. On the other hand,
ethical principles may be somewhat weaker sources of obligation, and a
parent to a child, or a teacher to a student, may or may not constitute a strong
deontic source. Weaker deontic sources (e.g. people at the same authority
level, abstract entities, such natural phenomena as river currents) determine
contextual meanings which move away from permission towards interpreta­
tions of ability or possibility, as in example (18).

(18) Creo que ésta es una democracia y que uno puede hacer lo que
quiera. (M25,p.6)
T think that this is a democracy and one can do what one wants.'
In example (18) the deontic source is an abstract entity, 'democracy' (as
opposed to Franco's dictatorship), and X is not a specific event. In addition,
time orientation is generic. Given these values for C and X, example (18) may
lend itself to an interpretation of root possibility as well as permission.
That these inferred meanings are contextually determined is sup­
ported by the fact that a change in the environment of the modal has conse­
quences on its interpretation, as shown in (19).
80 Carmen Silva-Corvalán

(19) [Father responding to son's request for permission to go out]

Confío en ti, puedes hacer lo que quieras.
T trust you, you can/may do what you want.'
A specific animate deontic source, a specific agent, and the pragmatics of
the interaction (request for permission) make the permission CM of poder
much less peripheral.

3.1.2 Ability.
The IM 'does not preclude' is interpreted as 'ability' when the following
contextual factors obtain: (a) K: natural or rational principles that permit us to
infer either innate or learned capacities to do z. (b) C: (i) the speaker has
evidence that X has been previously actualized, or that at least z has been, and
the circumstances are such that z may be valid also for a 'new' y; (ii) y is
animate, agentive, and specific; z is an event; core time orientation is generic
or present; if y is 'new', it is future.
Examples (20) and (21) illustrate contextual conditions for the CM 'ability':
(20) Ella os puede decir todo, todo referente a la droga. (M1 l,p39)
'She can tell you everything, everything about drugs.'
(21) El árabe y el ruso, pues, no los ha llegado a hablar, pero vamos,
para conversación puede defenderse. (M26)
'Arabic and Russian, well, she hasn't managed to speak them, but,
in conversation she can get by.'
Note that saber 'to know' may also convey 'ability for z', as in Yo sé
nadar 'I know how to swim'. Saber, however, simply asserts knowledge or
ability for; it does not implicate precluding circumstances as poder does.
Therefore, saber is not compatible with circumstances that imply that X is
somewhat surprising or that it involves effort. Thus, when y performs an event
that involves effort, or the overcoming of a given barrier, or a certain degree
of difficulty, p may be modalized with poder (ex. 22) but not with saber
(example. 23):
(22) Juan puede nadar mariposa por horas.
'John can swim the butterfly (stroke) for hours.'
(23) *Juan sabe nadar mariposa por horas.
*'John knows how to swim the butterfly (stroke) for hours.'
Interpretation of 'poder' and 'deber' 81

Further evidence for the implication of 'overcoming difficulty' is provided by

a comparison of example (21) with its non-modalized version in (24).
(24) El árabe y el ruso, pues, no los ha llegado a hablar, pero vamos,
para conversación se defiende.
'Arabic and Russian, well, she hasn't managed to speak them,
but,in conversation she gets by.'
Without poder, example (24) says that the subject referent (the speaker's
daughter) does well in conversation, a statement which is too assertive pre­
cisely because it does not convey the idea that this person encounters a certain
amount of difficulty in speaking Arabic and Russian. Example (25) is also
illustrative.
(25) No, no claro, vamos, de hecho esas voces que se graban, tú con tu
oído no las escuchas...solamente las puedes escuchar cuando
rebobinas la cinta y escuchas lo que se grabó. (M5,585)
'No, no, well, in fact some of those voices which you record you
can't hear with your naked ear...you can only hear them when you
rewind the tape and listen to what was recorded.'
From example (25) one infers that a certain amount of difficulty has to be
overcome for the modalized event to take place. In fact, the speaker explicitly
states that only after rewinding the tape can certain voices be heard. The
implication of the existence of precluding cirsumstances derived from the IM
of poder weakens the strength of the assertion.

3.1.3 Mitigation.

By weakening the assertiveness of utterances, the use oí poder may be said to

convey a CM of mitigation, which may at times be interpreted as politeness.
This is clearly seen in utterances containing a verbum dicendi, especially with
a first person subject. The fixed conversational expression puedo/podemos
decir 'I/we might say', shown in example (26), constitutes a typical example
of mitigated assertiveness.
(26) Pero en la vida social, podemos decir, no nota diferencias con, con
las gentes que están en cualquier barrio madrileño. (M26,212)
'But in social life, we might say, you don't notice differences
among people who live in any barrio in Madrid.'
82 Carmen Silva-Corvalán

3.1.4 Root possibility.

The clearest context for the interpretation of root possibility includes (a) a
non-agentive, non-volitional, non-specific subject (y); (b) a stative verb (i.e.,
z is not an event); (c) no evidence of previous actualization of p; (d) non-past
time orientation. Thus, the only CM that may be inferred for examples (27) &
(28) is RP.
(27) Ya puede tener todos los millones que quiera [la persona sigue
siendo buena, si es buena por naturaleza]. (M9,p.7)
'(S)he may have all the millions (s)he wants [the person continues
to be good, if (s)he's good by nature].'
(28) Una música diferente, a, digamos, a la música que tú
(non-specific) puedes escuchar de un piano, de un violin, etc.
(M5,663)
'Music which is different, shall we say, from the music you
(non-specific) might hear from a piano, a violin, etc.'
Many examples, however, with animate subjects and more or less dy­
namic verbs may be interpreted to convey both ability and RP (ex. 29), or to
be closer to an ability reading (ex. 30a) or to an RP interpretation (ex. 31).
(29) Argentina puede salir muy bien del paso, en cinco o seis años.
(M9,A464)
'Argentina can/may very well solve its problems, in five or six
years'
(30) No se dan cuenta que salen estos militares y creo xx, entonces
como son unos bestias, pues pueden xx unas matanzas (a) y pueden
hacer cosas mhm tremendas. (M10,pl3)
'They don't realize that these military men go out into the streets
and I think xx, then since they are like animals, well they can xx
killings (a) and they can do mhmm terrible things.'
(31) El colegio está cerquísima. Puede ir andando. (M1 l,p9)
'The school is very near. You may/can walk there'

3.1.5 Epistemic possibility.

EP involves the speaker's logical inference and lack of confidence about the
possibility of actualization of p. There are a number of clear examples of EP
interpretation in the data examined. In these clear cases, C includes one or
Interpretation of 'poder' and 'deber' 83

more of the following elements: (a) inanimate, non-specific subjects, or

obligatory subjectless sentences; (b) a stative verb (e.g. ser 'to be', estar 'to
be', querer 'to want', saber 'to know'); (c) negative consequences of z for y.
Further, when a negated Infinitive is modalized by affirmative poder, an
epistemic sense must be inferred (as in example 11). Examples (32-33)
illustrate.

(32) ...que si hoy se cierra el plazo, dentro de tres meses ya puede ser el
examen. (M5,135)
'...if the deadline is today, then the exam may be within three
months.'
CM: Based on previous experience, given that today is the deadline to
register for the exam, I infer that it is possible that the exam may be
given in three months.
(33) Puede ser ésa una diferencia respecto a otros regímenes tota­
litarios. (M5,250)
'That may be a difference with respect to other totalitarian re­
gimes.'
In the case of epistemic possibility, K includes rational laws (people can
make predictions, deduce, infer, etc.). As for C, lack of evidentiality and
stative aspect are indeed crucial. If evidentiality exists, an example may then
be interpreted as conveying RP, as in (34) and (35):
(34) No es un clima muy estándar. Efectivamente, te puede llover y al
día siguiente hacer un sol espléndido. (M6,p2)
'It's not an even climate. In fact, it can rain one day and the
following day the sun may be shining brightly'
CM'. I have evidence that the climate is not even. I may infer with
confidence that it is possible for it to rain one day and the following
day be sunny.
(35) Luisa puede ser la Decana de Filosofía, porque ha renunciado a su
otro cargo.
'Louise can be the Dean of Philosophy because she has resigned
her other job.'
CM: Now that she has resigned her other job, I infer that it is possible for
Louise to be the Dean of Philosophy.
84 Carmen Silva- Corvalán

Example (35) appears to be indeterminate between a RP and an ability

CM. This is to be expected when the context does not offer core contextual
conditions because it seems rather clear that CMs are all related through the
assumption of ability for z existing: if y is assumed to have ability for z, then
z is possible in a given world and y may be allowed, advised, etc. to z.
Poder has become lexicalized in the expressions puede que 'maybe' and
puede ser que 'it may be that', which can only convey EP, as illustrated in
examples (36) & (37).
(36) Pero ya eso la policía no te explica nada. Si alguna vez coincides
con algún amigo, amigo te hablo que lo has visto dos o tres veces -
entonces puede que a lo mejor te cuente algo. (M24,pl0)
'But the police don't explain anything to you. If you happen to run
into a friend, friend I mean someone you've seen two or three times
- then he may perhaps tell you something.'
(37) Pues aquí ese coche puede ser que sobre el setenta y dos al setenta
y cuatro como mucho dejaron de fabricarlo. (M9,A353)
'Well here, that car, maybe they stopped making it around seventy
two or seventy four at the latest.'
Support for the lexicalized status of puede que and puede ser que is
offered by the fact that they may only have third person singular morphology,
and that they are subjectless. With respect to negation, only puede ser que
may be negated and, as it is epistemic possibility, negation affects only p.
Examples (38-39) are illustrative.
(38) a. *No puede que Juan lo engañe.
b. Puede que Juan no lo engañe.
'It is possible that John is not cheating him'
(39) a. No puede ser que Juan lo engañe.
b. Puede ser que Juan no lo engañe.
a. 'It can't be possible that John is cheating him'
b. 'It is possible that John is not cheating him'

3.2 Conclusion

The data examined indicate that the most frequent CM of poder when used in
Present tense form is root possibility (58 of 110 cases studied). Ability, EP,
and mitigation account for 9%, 8%, and 4%, respectively. Furthermore, 9
Interpretation of 'poder' and 'deber' 85

examples are indeterminate between ability and RP, and 20 examples are
indeterminate between permission and RP.
Two further points must be emphasized: (1) Some of the 58 cases of RP
are in the periphery of this CM, close to readings of permission or ability; and
(2) The nature of the data (conversational discourse) must have consequences
on the frequency of occurrence of the CMs identified, such that other dis­
course genres would most likely yield different percentages of distribution of
CMs.
It is clear that we are dealing with intersecting contextual sets. On the
other hand, it is not clear to me that a strict prototype approach is the most
appropriate one to account for the facts. Indeed, a prototype model does not
acknowledge the relevance or validity of necessary or sufficient conditions in
the interpretation of CMs. Nevertheless, there do appear to be certain neces­
sary contextual features in every CM, and there are cases when the necessary
conditions are also sufficient to identify a CM as, for instance, 'permission': a
specific strong deontic source, a specific agent, direct speech, a dynamic
situation, positive consequences of p (ex. 40); or 'epistemic possibility':
non-agentive, non-intentional 'doer', stative situation, negative consequences
of p (ex. 41).

(40) [Teacher to child]

Puedes salir de la sala ahora.
'You may leave the room now'
(41) El mundo puede estar al borde de una crisis nuclear.
'The world may be about to suffer a nuclear disaster'
What happens to be the case is that these sets of contextual conditions are
frequently not a question of plus or minus in conversation, but of more or less.
It seems that the contextual features which are necessary conditions in the
permission and ability sets —deontic source and evidentiality— are precisely
the fuzzy ones. The fuzziness of contexts plus the logical relations between
ability, necessity, mitigation, permission, RP and EP often lead to intersecting
CMs.

4. Deber

Like must in English, deber can be used in either an epistemic or a deontic

sense, as shown in examples (42-43). An example of the type of (42) involves
86 Carmen Silva-Corvalán

'epistemic necessity', while an example of the type of (41) (with poder)

involves 'epistemic possibility' (cf. Lyons 1977: 791). However, given that
necessity and possibility are related notions, I have chosen to use the term
epistemic possibility in the semantic analysis of deber as well.
(42) [Loud car engine noises]
Ese debe ser Fernando. Epistemic possibility
'That must be Fernando.'
Interpretation'.
On the basis of frequent previous associations between loud engine
noises and Fernando's car, speaker confidently infers that the
person in the loud car is Fernando.
(43) Debes estudiar para aprobar el examen. Deontic obligation/necessity
'You must study to pass the exam.'
Interpretation'.
Common knowledge indicates that it is required/necessary to study
in order to pass an exam.
The IM of deber has some points of overlap with that of poder, but while
poder 'does not preclude X', deber 'favors X', 'requires X', or 'entails X'.
That is, while poder communicates the speaker's lack of strong belief in the
possibility of actualization of p, deber communicates confidence in the reali­
zation of p; it has connotations of 'very likely', 'necessary', and 'appropriate'.
The contribution of deber to the meaning of modalized p appears to be, then,
to require X as essential in the light of a set of circumstances C, interpreted
under a system of laws K, as represented in the formula: K (C requires X).
This definition is slightly different from the one proposed by Perkins
(1982:255) for the core meaning of must: K (C entails X). I have substituted
the term 'requires' for 'entails' in order to avoid misunderstanding due to a
reading of 'entails' in its strict logical sense. I use 'requires' in the sense of
'demanding as necessary or essential'. Thus, deber presents the IM 'requires'
and it relates actualization of X to a set of circumstances C, such that with
reference to a set of principles K, the circumstances C require X. As in the
case of poder, the values of K and C are recognized in context; these variables
(which include morphosyntactic, semantic and prosodic features) plus the
modal chosen determine whether X is interpreted under a dynamic, deontic or
epistemic modality.
Example (44) illustrates how the definition of deber would lead to an
interpretation of epistemic possibility assuming certain values for K and C.
Interpretation of 'poder' and 'deber' 87

(44) Tener un coche allí [en USA] debe ser algo alucinante. (M25,p5)
'To have a car there [in the USA] must be out of this world.'
IM: K (C requires X)
(i) K: rational laws (capacity to infer from available evidence)
(ii) C: (a) absence of deontic source
(b) preceding discourse gives some indication that p is true
(c) y = inanimate subject; z = stative; time orientation: generic
(iii) X: having a car in the USA is out of this world
CM: given humans' rational capacity to infer the likelihood of p
from available evidence, I interpret that the speaker intends to
convey the message that y ('to have a car in the USA') is very
probably z (' be out of this world'). (Epistemic possibility
sense).
By postulating for deber a core meaning of 'requirement of X', it is
possible to account for such CMs as obligation, advice, and probability,
depending on whether a number of contextual circumstances prompt a read­
ing of more or less forcible requirement. These circumstances include: (1)
Animacy and agentivity of y; (2) Nature of the deontic source, if any; (3)
Speaker approving or disapproving attitude towards p; (4) Adverse or positive
consequences of p.

4.1 Deber in Present tense form

Deber occurs infrequently in the data examined. Epistemic possibility seems

to be more frequently expressed by the lexicalized expression puede que,
while the auxiliary tener que 'to have to' is more commonly used to convey
deontic necessity. Only 26 cases of Present tense affirmative deber (de), plus
6 negative examples are attested in the data. The CMs of the 26 positive
examples range from 'high likelihood' (24 examples) to 'obligation/neces­
sity' (2 examples). The negative examples range from 'obligation/necessity'
to 'advice.'

4.1.1 Epistemic possibility to necessity

Spanish grammars note a difference between deber 'must' (obligation) and
deber de 'must' (possibility), but at the same time acknowledge that this
lexical opposition is being lost. The data examined support this observation
(cf. Sirbu 1988). The opposition deber - deber de is disappearing in the
88 Carmen Silva-Corvalán

Present tense, but appears to be more stable in other tenses (see Section 5 on
tense and modality). Deber de occurs in only 4 of the 24 examples of
epistemic possibility (see example 45). Furthermore, two examples with
deber de in the Imperfect Indicative do not convey possibility but rather
advice or obligation, as shown in (46).
(45) A: Es que se para [el coche] y luego-
B: xxx (incomprehensible)
A: Sí. Es que debe de estar estropeado el freno. (M24,p27)
A: 'It's that it stalls [the car] and then-'
B: xxx (unintelligible)
A: 'Yes. It's that the brake must be broken.'
(46) Yo exactamente no lo sé. Tendría que ser [el referendum] a prime­
ros del año que viene o algo así. Debían de solucionarlo antes de
terminar la jefatura de ellos [los socialistas] .(M24,p29)
'I don't know exactly. It [the referendum] would have to be at the
beginning of next year or something like that. They [the socialists]
should resolve this question before the end of their term in office.'
Deber is more strictly modal in its semantics than poder in that it is in all
contexts non-factual, i.e., propositions modalized by deber cannot be inter­
preted as 'actualized' ,15 This semantic difference accounts for the fact that one
of deber's CMs is EP, but not RP, which is related to ability and evidentiality.
In its EP contexts, deber may be preceded by negation but is not affected by it.
Examples (47) a-c are illustrative (see note 11).
(47) a. El alto debe ser Pepe.
b. El alto no debe ser Pepe.
c. *El alto debe no ser Pepe.
a. T h e tall one must be Pepe.'
b. 'The tall one can't be Pepe.'
c. *'The tall one must not be Pepe.'
The core contextual features associated with epistemic possibility are as
follows:
(i) K: rational laws
(ii) C: (a) evidence that p may be true
(b) absence of deontic source
Interpretation of 'poder' and 'deber' 89

(c) speaker attitude towards p is neutral

(d) adverse consequences of p
(e) y = inanimate; z = stative; time orientation: non-past
A human subject, a non-stative situation, speaker disapproval of not-p,
and adverse consequences of not-p determine readings which move away
from assessment of epistemic possibility towards the necessity end of the CM
continuum, examples (48a) and (48b) illustrate such a case.
(48) A: a. Porque se estropean las construcciones de agua [en La
Granja]. Por lo visto ya están muy pasadas de agua y pues
se deben estropear. Entonces las cuidan.
B: Las están arreglando.
A: b. Exactamente, las deben cuidar. (M24,p34)
A: a. 'Because the fountains break [at La Granja]. It's obvious
that they're too old (oversaturated) and so, of course they
{must) break. So they take care of them.'
B: 'They're fixing them.'
A: b. 'Exactly, they have to (must) take care of them.'
In (48a), a combination of core factors: inanimate subject, speaker neu­
tral attitude (i.e. neither approval nor disapproval) towards p, adverse conse­
quences of p, and evidence favoring the truth of p, determine a CM of
epistemic possibility. In (48b), on the other hand, C (human subject, non
-stative situation, adverse consequences of not-p) allows an interpretation of
necessity.

4.1.2 Mitigation
Changes in the contextual circumstances of deber in (48b): animate and
agentive subject, a possible though weak deontic source, strong evidence that
p is true, adverse consequences of not-p, bring about a CM which is indeter­
minate between high likelihood, categorical necessity or requirement, and
mitigation.
Note that example (48) illustrates intriguing uses of deber in utterances
which repeat the content of an immediately preceding non-modalized p. This
appears to indicate that in certain cases deber is used to convey a degree of
speaker uncertainty in the truth of p as a pragmatic strategy for mitigating the
degree of assertiveness of the utterance. The communicative purposes which
motivate the use of this strategy need to be further investigated. The choice of
90 Carmen Silva-Corvalán

deber over poder, which may also be interpreted to convey 'mitigation',

appears to be related to the higher degree of certainty implied by the IM of
deber.

4.1.3 Obligation to advice

The six negative constructions with deber in the data all express weak
obligation or advice:
(49) Nadie debe pensar por ti, ni nadie debe decidir por ti nunca, en
ninguna circunstancia. (M18,p27)
'No one should think for you, and no one should decide for you
ever, under no circumstances.'
I will not say much here about the expression of deontic obligation. I
would like to point out, however, that haber que 'to be necessary' and tener
que 'to have to' typically express necessity and obligation in affirmative
constructions in Present tense, as in (50) (cf. Sirbu 1988).
(50) Por ejemplo, hay que dar tres pagas al año, que dice el gobierno,
pues tres pagas al año se tienen que dar. Que hay que dar un mes
de permiso, pues un mes de permiso se tiene que dar. (M9,pl7)
'For instance, it is necessary to give three extra paychecks every
year, the government says, well then three extra paychecks have to
be given. And it's necessary to give one month of leave, well then
one month of leave has to be given.'
What is relevant, however, is that tener que also occurs, albeit quite
infrequently, in possibility contexts, as seen in (51):
(51) Y me pasó. Me pegó una pasada un 600, que yo digo, 'Pero la
madre del cordero, ese tío tiene que llevar un turbo ahí detrás. '
(M9,p3)
'And he passed me. A 600 passed me so fast, that I go, "Good
heavens, that guy must have a turbo behind".'
There are several indications that the use of tener que in possibility
contexts may be a recent development in Spanish: (a) The fact that neither
tener que nor the main verb Infinitive can be negated (exx. 52a-b); (b)
Infrequent occurrence of tener que in contexts of possibility; and (c) This
'meaning' of possibility is not acknowledged in current grammars and dic­
tionaries.16
Interpretation of 'poder' and 'deber7 91

(52) a. *Ese tío va tan lento que no tiene que sentirse muy bien.
b. *Ese tío va tan lento que tiene que no sentirse muy bien.
a. 'That guy's going so slowly that he doesn't have to be feeling
well'
b. 'That guy's going so slowly that he has to not be feeling well'

4.2 Summary

There appear to be a number of contextual sets of circumstances associated

with different CMs of deber. It seems possible to identify core contextual
circumstances invariably associated with epistemic possibility. By contrast,
fuzzy contexts make obligation, advice, and mitigation not always easy to
differentiate. As in the case oí poder, agentivity, evidentiality, and the possi­
ble consequences of p play an important part in the establishment of CMs.
Furthermore, ex. 46 suggests how verb morphology interacts with lexical
modalities. The next section explores this question further.

5. Modality and tense mood aspect

This section discusses the co-occurrence of poder and deber with verbal
markers of tense, mood and aspect other than 'Present tense'. Proposing IMs
for these verbal affixes is beyond the scope of this study. Therefore, my
analysis will consider only what appears to be their contextual contribution.

5.1 Non-finite forms and Imperative

I have observed in the introductory section that EP and RP are not possible
CMs when the modals are in Infinitive form (ex. 53). This restriction extends
to the Present Participle (ex. 54). Furthermore, neither modal is interpretable
in the Imperative (ex. 55). Thus, one might say that, like English modals,
poder and deber are 'defective' in this respect: they lack CMs of possibility in
the Infinitive and the Present Participle, and lack the Imperative altogether. In
the Infinitive (53) and the Present Participle (54) only root modality CMs are
inferred (e.g. obligation, ability). With regard to the Past Participle, the
behavior of poder and deber differs: while deber appears to retain many of its
possible CMs, including EP (56), poder in the Past Participle may convey
ability (57), but cannot be interpreted to convey RP (58) nor permission.
92 Carmen Silva-Corvalán

(53) Juan quiere deber/poder hacerlo.

'John wants to have the obligation/be able to do it.'
(54) Juan está debiendo/pudiendo hacerlo hace tiempo.
'John is having to/being able to do it for a long time now.'
(55) *Debe/puede hacerlo!.
*'Have to (must)/be able to (can) do it!'
(56) El árbol ha debido caerse con la lluvia.
lit: the tree has had to fall because of the rain
'The tree must have fallen because of the rain.'
(57) Juan ha podido irse.
'John has been able to leave.'
(58) *El árbol ha podido caerse con la lluvia.
* T h e tree has had the ability to fall because of the rain.'
Not a possible CM: The tree must have fallen because of the rain.
Verb morphology has been shown to be modal in that it can contribute to
the proposition the meaning of 'more or less assertiveness', where assertive­
ness is defined as speaker belief or confidence in the truth of the proposition.
Degrees of assertiveness are pragmatically inferred to convey degrees of
hypotheticality. These two notions correlate with verb morphology roughly as
in the scale below (cf. Fleischman 1989; Klein-Andreu 1986; Silva-Corvalán
1989).
Greatest assertiveness
factual Preterite Indicative
Present indicative
least hypothetical Future indicative
Imperfect indicative
Present subjunctive
most hypothetical Conditional Indicative
Imperfect subjunctive
Least assertiveness
In addition, verb morphology expresses tense and aspect oppositions,
roughly as follows: past time perspective (Preterite, Imperfect), omni-temporal
or atemporal time perspective (Present), future time perspective (Future, Condi­
tional, and perhaps also the Subjunctive forms). In regard to morphological
aspect, it may be argued that Preterite and Future are perfective,17 Imperfect
and Conditional Indicative, and Subjunctive forms are Imperfective, and Present
Indicative seems to be unmarked for aspect (cf. Silva-Corvalán 1991).
Interpretation of 'poder' and 'deber' 93

Let us now examine examples of poder and deber in non-Present form.

5.2 Imperfect and Preterite Indicative

31 examples of poder in Imperfect Indicative were examined: 6 are non-past

time oriented and exhibit desiderative (ex. 59) and RP CMs (cf. Bybee, this
volume); the rest have past time orientation and convey CMs of ability,
permission, RP (ex. 60), and obligation or advisability (this last one only in
one negative example, 61).

(59) Nos podíamos parecer a Argentina y, nosotros, a Estados Unidos

en eso, en eso, ¡en el dinero! (M9,A446)
'I wish we were (lit: we could be) like Argentine and, we, like the
United States, in that, in that, with respect to money!'
(60) Y estaba - todo la ca-, las calles estaban heladas. Y teníamos que
dar saltos, porque nos escurríamos y nos podíamos matar, ¿ com­
prendes? (M10,p6)
'And it was - the whole st- the streets were icy. And we had to jump
around, because it was slippery and we could kill ourselves, you
see?'
(61) Y cuando llegué a Madrid, ya definitivamente pues me planteé el
problema que yo no podía estar mirando al techo. Tenía que hacer
algo, ¿no? (M10,p3)
'And when I returned to Madrid, I definitely faced the question that
I couldn't just stare at the ceiling. I had to do something, right?'
Example (59) is interesting in that it has a CM not identified for poder in
Present tense, namely desiderative, in contexts with future time orientation.
This CM is inferred when the speaker's attitude is clearly approving of p. If
speaker attitude is neutral, however, the CM of podía with future time
orientation is simply possibility, as in (62).
(62) Si estás aburrido podías ir al cine.
'If you're bored you could go to the movies.'
Although it appears as if in Spanish root possibility in the past is marked
by the Imperfect morpheme (-bal-ía), this is not so. Past time orientation is
inferred from other signals in the context, as illustrated by the minimal pair
sentences in (63), where (63 a) has a future time orientation (signalled by el
próximo año 'next year'), while (63 b) may only be past (signalled by antes
'before, in the past').
94 Carmen Silva-Corvalán

(63) a. Si te apoyaba la ITT, podías ser Presidente el próximo año.

'If ITT supported you, you could be President next year.'
b. Si antes te apoyaba la ITT, podías ser Presidente, pero ahora
ya ese apoyo no sirve.
'If in the past ITT supported you, you could be President, but
now their support is useless.'
Halliday (1970:337) notes that "modality itself is not subject to variation
in tense". Contrary to Halliday, Coates (1983:107) discusses, for instance,
could as expressing, among other meanings, root possibility in the past, i.e., as
'past' of can. Example (64) is from Coates (ex. ii, p. 108).
(64) Past of can = Root Possibility
With all but one of the cookers the grid of the grill pan could be at
one of two possible distances from the heat. The exception was the
Cannon, which had four available positions.
'Exceptuando una de las cocinas, la rejilla del grill podía estar en
una de dos posibles distancias del calor. La excepción era la
Cannon, que tenía cuatro posiciones posibles.'
It seems to me that it is more appropriate to state that could is compatible
with past time contexts, just as podía is in Spanish, since past time in (64) is
indeed indicated by the verb forms was and had in the second sentence.
In Spanish, however, poder may occur with Preterite morphology and
unambiguously refer to past root possibility in certain contexts which do not
contain any other time reference (cf. Bybee's (this volume) interesting study
of the development of the hypothetical senses of past modals in English).
Compare examples (65) & (66) (fuller contexts are given in examples 69-70).
(65) Pudo marcharse.
could-3psg leave
'He was able to leave'
(66) Pudo pasar algo terrible.
might happen something terrible
'Something terrible might've happened.'
Example (65) conveys factuality of the event. By contrast, contextual
features, namely inanimate subject, stative verb, adverse consequences of p,
and speaker disapproval of p, determine a root possibility interpretation of
Interpretation of 'poder' and 'deber' 95

example (66), which is counter-factual; the Preterite suffix indicates that this
is root possibility in the past.
I have proposed that the IM of poder implies that a certain amount of
effort must be expended in 'not precluding X'. The relation between this IM
and the CM of (66) is, then, easy to establish: speakers do not assume that
entities may spend effort in the actualization of an event whose consequences
are adverse, so the only possible reading for (66) is of non-actualization or
non-factuality. Thus, this CM does not appear to be fuzzy, but rather a
category with clear boundaries defined by three features: adverse conse­
quences of p, speaker's disapproval of p, Preterite morphology. This analysis
is confirmed by the acceptability of (67) and (68 a), and the unacceptability
(under 'normal circumstances') of (68 b).

(67) Con esfuerzo, pudo marcharse finalmente.

'After trying hard, he was finally able to leave.'
(68) a. Estoy contenta, porque pudo pasar algo terrible, pero no pasó.
'I'm happy because something terrible might've happened, but
it didn't.'
b. * Estoy contenta porque pudo pasar algo terrible.
*'I'm happy because something terrible could happen'
I have excerpted the modalized propositions in (65-66) from actual
occurrences in the corpus (exx. 69 & 70) in order to contrast them in minimal
contexts, thus highlighting the contribution of adverse vs. neutral conse­
quences of p.
(69) Tuvo esa suerte y se pudo marchar. Le dieron un lectorado. (M10,p3)
'He was lucky and managed to leave. They gave him a lectureship.'
(70) Fue muy grave aquello. Pudo pasar algo terrible, porque si los
militares se echan a la calle - con seguridad ocurría una masacre.
Yo no sé qué hubiera ocurrido. (M10,pl3)
'That was really a critical situation. Something terrible could've
happened, because if the military get onto the streets - surely there
would've been a massacre. I don't know what would've hap­
pened.'
Adverse consequences of p is a crucial feature for CM root possibility
with Preterite poder. Despite the animacy of the subject, while example (71)
allows the inference that the situation did occur, (72) can only convey possi-
 96 Carmen Silva-Corvalán

bility (unless, of course, the speaker had wanted to become extremely poor,
but I would not consider this to be a 'normal social circumstance').
(71) Pudimos quedarnos en Inglaterra.
'We could stay in England.'
(72) Pudimos quedarnos en la miseria. [La situación estaba muy mala.]
could-we stay-us in the poverty
'We could've gone broke.' [The situation was bad.]
In its actual, more extended context of occurrence, example (71) in fact
conveys root possibility: staying in England was possible but not actualized.
However, this has to be made explicit by the speaker (pero él no quiso 'but he
didn't want to'), as shown in the actual utterance (73).
(73) Pero pudimos quedarnos en Inglaterra, porque le ofrecieron una
permanencia, pero él no quiso. (M10,p4)
'But we could have stayed (lit: could stay) in England, because
they offered him a permanent position, but he didn't want to.'
Otherwise, possibility with Preterite poder may be expressed if the modal
is combined with the Perfect Infinitive, as in English, such that example (74)
does not then require any further context to be interpreted as a possibility
which existed in the past, i.e. as counterfactual in the present.18
(74) Pudimos habernos quedado.
'We could have stayed.'
There are no examples with Preterite poder plus Perfect Infinitive in the
data from Madrid. I note, however, that in my variety of Chilean Spanish, this
tense combination is required in examples (66), (70), (72), & (73). My
intuitions in this respect are confirmed by the responses of four native speak­
ers from Chile and Argentina to an elicitation test given to determine the
meanings associated with poder and deber in various tense forms. This may
reflect a difference between Madrid and Latin American Spanish, but conver­
sational data from this latter variety must be examined before any further
proposals can be made in this respect.
The Perfect Infinitive does occur, however, with Present, Imperfect, and
Conditional forms of poder. All examples have past time orientation and
stress the non-factuality or hypotheticality of p. This higher degree of hypotheti-
cality becomes more evident through a comparison of examples (60) and (75).
Interpretation of 'poder' and 'deber' 97

(75) Tenía yo esto aquí que se podía haber utilizado.

T had this here that could've been used.'
Imperfective aspect in example (60) allows the interpretation that 'killing
ourselves' is omnitemporally possible if the contextual circumstances re­
ferred to in the example obtain. By contrast, Perfective aspect of modalized p
in (75) conveys that the possibility of p existed but is no longer available.
Examples with deber in Preterite and Imperfect are scarce (19 cases). It is
marked for tense only when combined with ser 'to be' in contexts which
explicitly indicate uncertainty about p (ex. 76).
(76) Debió ser eso de las 10 u 11 de la noche [cuando empezaron los
dolores del parto]. Y, no sé si nació a las 8 de la mañana.
(M24,pl8)
'It must've been at around ten or eleven at night [when labor pains
started]. And, I think she was born at eight in the morning.'
All other cases of Preterite and Imperfect deber convey obligation or
advice, even when combined with Perfect Infinitive, as shown in (77). Other­
wise, they are lexically differentiated (i.e. by using deber de) in contexts of
possibility, as illustrated by (78) and (79).19
(77) Por eso a mí me parece que debiste haber grabado a personas muy
jóvenes -porque tienen el lenguaje cheli.(M10,pl5)
'That's why it seems to me that you should've recorded very young
people - because they speak this cheli language.'
(78) Y mi madre debió de nacer en el año once o doce, porque tiene 75
o 74 años. (M10,p2)
'And my mother must have been born in 1911 or 1912, because
she's 75 or 74 years old.'
(79) Y la cogieron en el aeropuerto justo al llegar a Las Palmas.
Entonces allí debía, debía de haberlo llevado [la droga]. No sé
cómo funciona esto. (M24,p9)
'And they caught her at the airport just when she was arriving at
Las Palmas. So she must have taken it [the drug] there. I don't
know how that works.'
98 Carmen Silva-Corvalán

In regard to the Preterite/Imperfect opposition, I can only suggest at this

point that they reflect two different degrees of speaker assertiveness: stronger
and weaker, respectively. Thus, mutatis mutandis, in the case of deber,
Preterite morphology turns the IM requirement into obligation, while Imper­
fect morphology makes it less assertive and allows a peripheral reading of
weak obligation.
In my variety of Spanish, where opposition with deber de has disap­
peared, the expression of epistemic possibility in the past retains deber in the
Present and marks tense with the Perfect Infinitive. Thus, examples (78) &
(79) are rendered as in (80) & (81), respectively.

(80) Debe haber sido eso de las 10 u 11 de la noche ...

must-Pres. have been at around ...
(81) Y mi madre debe haber nacido en el...
must-Pres. have been born in ...

5.3 Conditional and Future

There are no occurrences of deber with future morphology in the data, and
only one of poder, with CM permission:
(82) Entonces esa persona que quiera abortar - podrá abortar siempre
que esto lo diga en las primeras semanas de su embarazo. (M10,p7)
'So a person who may want to have an abortion - will be able to
have one provided she requests it during the first weeks of preg­
nancy.'

In example (82), permission is associated with future time orientation

and identification of a deontic source (the law). If time orientation is present,
Future morphology independently conveys possibility, i.e. when combined
with non-modal verbs (see note 17), and it does so with poder as well.
Consider examples (83) a-b.
(83) a. Si tú lo dices, será (be-Fut.) cierto.
'If you say so, it must be true.'
b. Si tú lo dices, podrá ser cierto.
poder-Fut. be true
'If you say so, it may be true.'
Interpretation of 'poder' and 'deber' 99

The translations provided for examples (83) a-b are intended to capture
the fact that poder, because it incorporates the notion of effort, makes p more
hypothetical than when it is modalized simply by Future morphology.
By contrast, deber in the Future does not appear to be compatible with an
epistemic modality context, as I show in (84). Only root modality CMs may
be inferred for deber with Future morphology (85).
(84) a. ?Si tú lo dices, deberá ser cierto.
b. Si tú lo dices, será cierto.
'If you say so, it must-fut be true.'
(85) La inauguración deberá ser mañana.
'The inauguration will have to be tomorrow'
There are 6 examples with poder in Conditional form; they all occur in
root possibility contexts. Indeed, since Conditional morphology conveys very
weak assertiveness, it is not surprising that in every case it communicates
probability of p, regardless of the context. That is, non-modal verbs modal­
ized by Conditional morphology are interpreted to refer to hypothetical events
as well. Example (86) illustrates.
(86) Podrían hacerme una virguería - se suele decir una virguería, o
sea una cosa bonita -, arreglarme los dientes, quedarían perfectos.
(Mll,p27)
'They could do me a 'virguería' - they usually say 'virguería', you
know, something pretty -, fix my teeth, they would look perfect.'
On the other hand, deber in the Conditional may receive a deontic or an
epistemic interpretation depending on the same contextual features estab­
lished for Present deber. The contribution of Conditional morphology is that
of adding weak speaker assertiveness which, in a deontic context, creates a
CM of mitigated obligation or advice (ex. 87), and in an epistemic context, a
CM of weaker confidence in the truth of p (ex. 88).20
(87) El gobierno debería dar una explicación. (M5,p21)
'The government should give an explanation.'
(88) No vamos a saber exactamente las consecuencias, ¿no? [de entrar
al Mercado Común Europeo] En principio, debería de ser bueno.
(M18,pl5)
'We won't know exactly what the consequences will be, right? [of
entering the Common Market] In theory, it should be good.'
 100 Carmen Silva-Corvalán

5. Conclusions

It seems to me that an intriguing conclusion of this study is that, contrary to

what is usually believed about the determining and specifying effect of
context on the meaning of linguistic elements, a number of examples of poder
and deber have proved to have indeterminate CMs precisely because their
context of use makes them so. Assuming, then, basic IMs for these modals,
one may conclude that indeterminacy is a consequence of the fuzziness of
contexts (including such extra-linguistic factors as systems of beliefs, as­
sumed shared knowledge, interpersonal relations, etc.)-I have focused on this
phenomenon from the point of view of the hearer/analyst. However, there is
no a priori reason to deny that this would be valid for speakers as well, i.e.,
speakers themselves may not intend to communicate one message to the
exclusion of all other possible ones compatible with a given context of
occurrence. A number of examples in the data could be not only cases in
which there are no striking differences between two interpretations of an
utterance, but ones in which a speaker may either intend two interpretations to
be drawn, or not be aware of or concerned about indeterminacies or ambigui­
ties.
I have argued that contextual indeterminacy rather than indeterminate
modal meaning accounts for the fuzzy membership of modalized propositions
in one or another 'meaning category.' I would like to propose that this is valid
for all Romance languages and also for English. With respect to English, for
instance, Coates (p. 31) states that must has two main meanings (both consti­
tuting fuzzy sets): root (obligation/necessity) and epistemic (logical necessity/
confident inference), which she illustrates with the following examples:
(89) 'You must play this ten times over,' Miss Jarrova would say, point­
ing with relentless fingers to a jumble of crotchets and quavers.
(90) That place must make quite a profit for it was packed out and has
been all week.
My proposal of monosemanticity and contextual fuzziness predicts that
changing the contextual environment of must in (89) and (90) may cause
indeterminate interpretations of the CM of must. This is precisely what
examples (89 a) and (90 a) show:
(89) a. 'You must play this often,' Miss Jarrova would say when she
heard her play it so beautifully.
Interpretation of 'poder' and 'deber' 101

(90) a. That place must make a good profit. They'll close it down if it
doesn't.
Contextual modification of (89) and (90) leads to indeterminacy between
root and epistemic interpretations of (89a) and (90a), i.e. to fuzzy CM of must,
Admittedly, Coates's analysis may in essence be the same as mine. Yet, it
seems to me more appropriate to acknowledge explicitly that certain forms at
least do not have multiple meanings. Rather, they have basic meanings which
make them compatible with a number of contexts; this form-context interac­
tion may bring about different interpretations of the meaning of the form,
which I have referred to as CMs.
In regard to the questions posed at the start, I hope to have shown that a
monosemantic approach which goes beyond a level of IM and studies linguis­
tic forms in specific contexts of use adequately captures the relation between
IM and CMs of modals, and indeed reflects this interrelatedness more appro­
priately than a polysemantic analysis, which does not necessarily imply
affinity between various meanings. Given my view of language as a system of
communication and a particular instance of human behavior, I have sought
answers to linguistic questions within and beyond the specific linguistic
system under examination. Thus, crucial external factors which have rel­
evance to the interpretation of modalized utterances have been identified:
identification and strength of deontic sources, evidentiality, negative conse­
quences of propositions (which are likely to be culture specific), and speaker
attitude towards the content of propositions. This appears to be conclusive
evidence in support of a non-autonomous semantic approach to the study of
meaning.

NOTES

* I would like to thank Joan Bybee, Jenny Coates and Erica García for their extensive
comments and insightful criticism of an earlier version of this paper. My thanks as well
to Franco D'Introno, Wolfgang Klein, and Michael Perkins, who have contributed
helpful observations. Improvements in this present version also owe much to Erica, and,
likewise, to Flora Klein-Andreu, Roger Wright and especially to the editors of this
volume. I am grateful to all of them for their careful reading of my work and for their
penetrating comments. Errors and misinterpretations remain my sole responsibility.
1 Grammars note a difference between deber 'must' (obligation) and deber de 'must'
(possibility), but at the same time acknowledge that this opposition is being lost as the
form without the preposition de has extended to contexts of epistemic modality. I refer to
this change later.
 102 Carmen Silva-Corvalán

I do not examine here the transitive verb deber 'to owe' (Te debo $1 'I owe you
$1'), because at this synchronic stage of the language it clearly is a different lexical item
from modal deber.
2 The corpus of spoken data consists of approximately six hours of recorded conversations
with seven speakers from Madrid. The data were collected by me and a group of
graduate students participating in a seminar which I conducted in Madrid in the summer
of 1985.1 would like to thank my students for sharing their data with me. Included in this
study are three adult women and four adult men. They are all middle class 'madrileños',
except one man (M9), who works as a handyman in one of the Colegios Mayores of the
University.
3 Halliday refers to these as modulations and notes that though modalities and modula­
tions are not identical, they are semantically related, such that their partial reduction to a
single network "expresses the closeness of fit of the two systems" (347)] i.e. when they
do not convey the meaning of probability.
4 Haegeman deals only with will, and Kratzer with can and must, while Perkins examines
a wide range of modal expressions.
5 Coates includes the deontic and dynamic categories of modal logic in a root category,
and epistemic and alethic modalities in an epistemic one.
6 This analysis is similar to Lyons's (1977:843) proposal that the identification of a
deontic source (an entity with authority) is a prerequisite to the interpretation of a modal
as conveying obligation or necessity.
7 In his analysis of poder 'can, may', for instance, Narbona (1989:88) appears to identify
two verbs poder, a modal that does not occur with a compound Infinitive (no example is
offered), and a non-modal with the meaning of possibility that does admit this construc­
tion (Puede habérselo prestado '(He) may have lent it to him). It seems to me that
Narbona's proposition that there is a dichotomous modal/non-modal opposition identi­
fied by the possibility of co-occurrence with a compound Infinitive is not correct;
suggesting the existence of two verbs poder would force him to create at least two lexical
entries for this verb, a solution that he criticizes elsewhere in his work. Pottier (1976:39)
adopts a similar position when he argues that pouvoir 'can, may' and devoir 'must'
clearly represent a case of multiple polysemy. Likewise, Huot (1974, cited in Thibault
1991) states that there are two different auxiliary verbs devoir in French.
8 There is at least one other type of meaning which is also conveyed linguistically, namely
social meaning. I am not concerned with social meaning here.
9 'Agent-oriented' is here used in a similar, though not identical, way in which Bybee &
Pagliuca (1985) have used the term. Indeed, they propose to refer to the notions of
ability, obligation, desire and intention as 'agent-oriented modalities' because they
"predicate conditions of either an internal or external nature on a willful agent" (p. 63).
10 Miller and Kwilosz (1981) investigate the interaction of modality and negation in
English through a study of speakers' judgements of the meaning of sentences. Their
experiments show that "There is apparently a tendency for verbs of necessity to take de
re negation [i.e. only p is negated], and for verbs of possibility to take de dicto negation
[i.e. m and p are negated], but there are exceptions in both cases." This tendency also
seems to apply to poder and deber, but I do not examine this question in depth here.
Interpretation of 'poder' and 'deber' 103

11 For the EP reading to obtain, the Infinitive must receive stronger stress than the modal.
12 Perkins's definition is in turn modelled on Miller's (1978) discussion of Wertheimer's
(1972) characterization of the meaning of OUGHT.
13 Information given in parentheses identifies the tape and either page of the transcription
or tape counter number where example is located. Examples without this information are
made up by the author.
14 By 'new' y I mean a subject referent for whom the speaker infers that z is valid, as in M.
va a poder caminar después de la operación 'M. will be able to walk after the operation',
on the basis of the speaker's previous evidence that z has been valid for a different
subject referent under similar circumstances.
15 Compare poder and deber in the Preterite: Pudo cruzar el río 'He was able to [and did]
cross the river' (situation occurred in the past); Debió cruzar el río 'He must have
crossed the river' (the situation most likely occurred in the past).
16 Moliner (1981), for instance, notes that tener que may express necessity, obligation, and
strong resolution, but does not include 'possibility.'
17 The perfectivity of the Future (-rá) in Spanish is debatable, but there is evidence that
supports this analysis whenever this form fulfills its 'tense' function (Silva-Corvalán
1991). By contrast, in contexts where the Future is used to express hypotheticality in the
present, its aspect seems to be Imperfective. Note, however, that the expression of
hypotheticality is subject to some intriguing restrictions with dynamic verbs, as illus­
trated in (i) and (ii):
(i) A: ¿Por qué no vino Pepe?
'Why didn't Pepe come?'
B: Estará enfermo.
'He may be/is probably ill'
(ii) A: ¿Por qué no está Pepe aquí?
'Why isn't Pepe here?'
B: *Trabajará hoy.
*'He'll work today.'
Estará trabajando hoy.
'He may be working today.'
18 The Perfect Infinitive by itself expresses conditionality (counterfactual protasis) in
certain special constructions of the type of (i) B:
(i) A: Me habría gustado ir contigo a Chile.
B: Haberlo dicho antes, te habría invitado.
have-it said before,
'Had you said it before, I would've invited you.'
19 This fact poses a challenging question in regard to the process of loss of deber de from
the system. The observation that past morphology appears to favor its retention, espe­
cially with non-stative verbs, needs to be examined statistically in a larger corpus of
data. See, also, examples 46, 87, and 88, which give evidence of insecurity in the use of
deber de.
20 Note that in example 88 the speaker uses deber de, but in Chilean Spanish deber alone
would also be interpreted to convey EP.
 104 Carmen Silva-Corvalán

REFERENCES

Bello, Andrés. 1977. Gramática de la lengua castellana (With notes from Rufino J.
Cuervo). 10th edition. Buenos Aires: Sopena.
Bosch, Peter. 1985. Lexical Meaning Contextualized. Meaning and the Lexicon.
ed. by G.Hoppenbrouwers, P. Seuren & A. Weijters, 251-258. Dordrecht: Foris.
Bull, William E. 1971. Time, Tense, and the Verb. 4th printing. Berkeley: University of
California Press.
Bybee, Joan. 1985. Morphology. Amsterdam: John Benjamins.
Bybee, Joan. (This volume). "The Semantic Development of Past Tense Modals in
English."
Bybee, Joan & William Pagliuca. 1985. "Cross-Linguistic Comparison and the Develop­
ment of Grammatical Meaning." Historical Semantics. Historical Word Formation,
ed. by Jacek Fisiak, 59-83. Berlin: Mouton.
Bybee, Joan, Revere Perkins & William Pagliuca. 1994. The Evolution of Grammar:
Tense, Aspect and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries. London: Croom Helm.
Fleischman, Suzanne. 1989. "Temporal Distance: A Basic Linguistic Metaphor." Studies
in Language 13.1-50.
García, Erica C. 1975. The Role of Theory in Linguistic Analysis: The Spanish Pronoun
System. Amsterdam: North- Holland.
Gili Gaya, Samuel. 1976, Curso superior de sintaxis española. Barcelona: Biblograf.
Givón, Talmy. 1982. "Evidentiality and Epistemic Space." Studies in Language 6.23-49
Haegeman, Liliane M.V. 1983. The Semantics of Will in Present-day British English:
A Unified Account. Brussels: Verhandeling Letteren, 45.103.
Halliday, M.A.K. 1970. "Functional Diversity in Language as Seen from a Consideration
of Modality and Mood in English. Foundations of Language 6.322-361.
Heine, Berndt. (This volume). "Agent-Oriented vs. Epistemic Modality: Some Observa­
tions on German Modals."
Hernández, César. 1986. Gramática funcional del español. Madrid: Gredos
King, Larry D. 1992. The Semantic Structure of Spanish. Amsterdam: John Benjamins.
Klein-Andreu, Flora. 1986. "Speaker-Based and Reference-Based Factors in Language:
Non-past Conditional Sentences in Spanish." Studies in Romance Linguistics, ed. by
Osvaldo Jaeggli, & Carmen Silva-Corvalán, 99-119. Amsterdam: Foris.
Kratzer, Angelika. 1977. "What 'Must' and 'Can' Must and Can Mean. Linguistics and
Philosophy 1.337-355.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Marcos Marín, Francisco. 1975. Aproximación a la gramática española. 3rd edition.
Madrid: Cincel.
Miller, George A. 1978. "Semantic Relations Among Words." Linguistic Theory and
Psychological Reality, ed. by M. Halle, J. Bresnan, & G.A. Miller, 60-118. Cam­
bridge, Mass.: MIT Press.
Miller, George A., & Donna M. Kwilosz. 1981. "Interactions of Modality and Negation in
English." Elements of Discourse Understanding, ed. by Aravind Joshi, Bonnie
Webber, & Ivan Sag, 201-216. Cambridge: Cambridge University Press.
Interpretation of 'poder' and 'deber' 105

Moliner, María. 1981. Diccionario de uso del español. Madrid: Gredos.

Narbona, Antonio. 1989. Sintaxis española: nuevos y viejos enfoques. Barcelona: Ariel.
Palmer, F.R. 1977. "Modals and Actuality." Journal of Linguistics 13.1-23.
Palmer, F.R. 1979. Modality and the English Modals. London: Longman.
Perkins, Michael R. 1982. The Core Meanings of the English Modals. Journal of Linguis­
tics 18.245-273.
Perkins, Michael R. 1983. Modal Expressions in English. Norwood, N.J.: Ablex.
Pottier, Bernard. 1976. Sur la Formulation des Modalités en Linguistique. Modalités.
Langages, special issue No 43.39-46.
Real Academia Española. 1973. Esbozo de una nueva gramática de la lengua española.
Madrid: Espasa-Calpe.
Rivero, María Luisa. 1977. Estudios de gramática generativa del español. Madrid:
Cátedra.
Silva-Corvalán, Carmen. 1985. "Modality and Semantic Change." Historical Semantics.
Historical Word Formation, ed. by Jacek Fisiak, 547-572. Berlin: Mouton.
Silva-Corvalán, Carmen. 1989. "The Pragmastylistics of Hypothetical Discourse." The
Pragmatics of Style, ed. by Leo Hickey, 87-105. London & New York: Routledge.
Silva-Corvalán, Carmen. 1991. "Invariant Meanings and Context-Bound Functions of
Tense in Spanish." The Function of Tense in Texts, ed. by J. Gvozdanovic & T.
Janssen, in collaboration with Östen Dahl, 255-270. Amsterdam: North- Holland.
Sirbu Dumitrescu, Domnita. 1988. "Contribución al estudio de los verbos modales en
español." Híspanla 71.139-147.
Sweetser, Eve. 1982. "Root and Epistemic Modals: Causality in two Worlds." Proceedings
of the Eighth Annual Meeting of the Berkeley Linguistics Society, ed. by Macaulay,
Monica, Orin Gensler, et al., 484-507. Berkeley: Berkeley Linguistics Society.
Thibault, Pierrette. 1991. "Semantic Overlaps of French Modal Expressions." Language
Variation and Change 3.191- 222.
Wertheimer, R. 1972. The Significance of Sense: Meaning, Modality and Morality. Ithaca:
Cornell University Press.
Wright, Roger. 1990. "Fact and Fiction in Language and Literature." Liverpool Papers in
Language and Discourse 3.1-37.
Zadeh, L.A. 1965. "Fuzzy sets." Information and Control 8.338-353.
Zadeh, L.A. 1972. "A Fuzzy-Set-Theoretic Interpretation of Linguistic Hedges." Journal
of Cybernetics 2.4-34.
 The Obligation Modality in Western
Nilotic Languages

Edith L. Bavin
La Trobe University

1. Introduction

Bybee (1985) includes obligation and necessity in the category of agent-

oriented modality which reports the existence of internal and external condi­
tions on an agent with respect to the completion of the action expressed in the
main predicate (Bybee, Perkins & Pagliuca 1994:195). Strong obligation
('must' ) and weak obligation ('should') involve external pressures; in con­
trast, necessity usually involves some physical conditions (Bybee et al. 1994).
This paper examines the forms, uses and diachronic development of the
markers of obligation in Lango and Acholi, two Western Nilotic languages. In
addition, some data from three other languages in this group are presented for
comparison purposes. 1 The source for the gram which typically marks obliga­
tion and necessity in Lango and Acholi is an impersonal form of the verb 'to
be fitting, to be suitable'. Lexical items with these meanings have been
documented as the sources of obligation grams in other languages in other
parts of the world, as reported in Bybee et al. (1994). What is of particular
interest are the semantic and syntactic changes required for a third person
form of the verb to become a marker of agent-oriented modality, which
attributes modality to the subject of the clause. These changes involve extend­
ing the syntactic context in which the verb is used.
Grams which mark modalities often have different interpretations de­
pending on the context. For instance, the examples to be discussed below
illustrate that the obligation markers in Lango and Acholi may be interpreted
 108 Edith L. Bavin

with a strong or weak obligation sense, or as marking the intention of the

speaker, or with an epistemic function. One way to handle such synchronic
variability is to view each interpretation as a different meaning, that is to
allow polysemy since polysemy can be viewed as comprising a set of overlap­
ping senses, as in Coates (1983). However, the view taken in this paper is that
interpretations are derived from context. When there is synchronic variation
in the interpretation of a form, this can be indicative of gradual change in
progress. Over time, implications made from the context may become part of
the meaning of the form; thus, diachronic extensions can be viewed as
grammaticalizations of what were once synchronic extensions of basic mean­
ings, as argued by Fleischman (1989).
There is other support for explaining synchronic variation from a dia­
chronic perspective. For example, Bybee (1985: 191-196) argues that the
category of mood is best viewed as a set of diachronically related functions,
and Traugott (1989:49) discusses diachronic changes from nonepistemic to
epistemic meanings for modal markers, attributing the changes to an increase
in subjectivity: as there is an increase in the coding of speaker informativeness
about his or her attitude, this is reflected in changes from nonepistemic to
epistemic meanings.
The forms most usually used to mark obligation in five Western Nilotic
languages are discussed in the next section. The morpho-syntactic properties
of the obligation constructions in Lango and Acholi and the lexical sources of
these obligation particles are discussed in Section 3. Different contextual
interpretations for the particles are illustrated in Section 4, while the sources
for other grammatical markers in the languages are discusssed in Section 5.
Following this, the use of the verb meaning 'to want' as another marker of
agent-oriented modalities is illustrated in Section 6.

2. The Western Nilotic languages

2.1 Introduction

There are five languages in the Western Nilotic group: Acholi, Lango,
Dhopadhola, Alur and DhoLuo. With the exception of DhoLuo, the languages
are spoken in Uganda, with Acholi extending into the Sudan. DhoLuo is
spoken in Western Kenya, so it is separated physically from the other four
The Obligation in Western Nilotic 109

languages. The languages share a high percentage of vocabulary items. Based

on a survey of vocabulary items in Ugandan languages, Ladefoged, Glick &
Criper (1968) report the number of vocabulary items shared between Acholi
and Lango is the same as that between Acholi and Alur (89%). The number of
vocabulary items shared between Acholi and Dhopadhola is reported to be
76%. However between Lango and Alur it is 84%, between Lango and
Dhopadhola 73%, and between Alur and Dhopadhola, 73%.
All five languages are often claimed to be mutually intelligible, and are
referred to collectively as Luo. However, there are significant differences in
the grammatical categories. (See Bavin 1982, 1990 for discussion on some of
the grammatical categories in Lango and Acholi). In addition, there are
phonological differences. For example, Dhopadhola, Alur and Dholuo have a
series of dental stops and a number of fricatives, but Lango and Acholi do not.

2.2 Obligation markers

2.2.1 The forms

Lango and Acholi are similar in the way the modality of obligation is en­
coded. In Lango (example 1) the particle myero is used and in Acholi (exam­
ple 2), the form is omyero or myero. Some speakers use one form, other
speakers use the other and another group of speakers use both forms. In Alur
(example 3) mako is the obligation marker, and in Dholuo (example 4)
obligation is expressed with nyaka. In contrast, Dhopadhola (example 5),
makes use of a verbal affix ripo which appears between the Subject Prefix and
the verb stem. Thus the marking of obligation is one area of difference among
the languages.

(1) a. Myero a-ngol pii

must l:sG-cross water
T must cross the water.'
b. Dako-ni myero nen-a oyotoyot
woman-DET must see-l:SG quickly
This woman should see me immediately.'
(2) a. Omyero i-tim
must 2.SG do
'You must do it.'
 110 Edith L. Bavin

b. Omyero doki a-wil ngom

must again l:sG-buy land
'I must buy more land.'
(3) a. Omako a-cith-i kawoni
must l:sG-go-suBJ now
'I must leave now.'
b. Ka i-mito cam omako i-bed ku cente
if 2:SG-want food must 2:sG-stay with money
'If you want food, you must have money.'
(4) a. Nyaka i-chiem mos
must 2:SG-eat slowly
'You must eat more slowly.'
b. Nyaka wa-kony-e
Must l:PL-help-REF
'We must help each other.'
(5) a. A-ripo-woth
l:SG-must-go
'I must go.'
b. I-ripo-woth
2:SG-must-go
'You must go.'
The verb in all five languages carries a Subject Prefix to agree with the person
and number of the subject, as illustrated in the sentences above and in (6) from
Lango. The Subject Prefix is generally a short form of the free pronoun.
Dhopadhola is distinct from the other four languages in using go as the Third
Person Singular pronoun and jo as the Third Plural form; en 'Singular' and gin
'Plural' are used in the other languages.
The Subject Prefix appears on the lexical verb following myero in Lango,
(o)myero in Acholi, omako in Alur and nyaka in Dholuo. In sentences with a
pronominal subject, it is optional to include the free pronoun, but the Subject
Prefix is required, as illustrated in (6). The particles themselves do not change
in form when the subject changes. Note in (6b) that both the auxiliary tye 'to
be (locative)' and the main verb carry a Subject Prefix, but the Obligation
marker myero does not.
The Obligation in Western Nilotic 111

(6) Lango
a. (An) a-ngweco
I l:SG-run
'I ran.'
b. A-tye a-ngweco
l:SG-be l:SG-run
'I am running.'
c. Myero onwongo dong a-ngwec oko
must PI then l:SG-run already
T should have run.'

2.2.2 Other properties

Other syntactic properties associated with the obligation construction are the
position of the subject noun or pronoun, the position of the Negative, the form
of the lexical verb following the obligation particle, and changes in interpreta­
tion when the particle follows an aspect marker. These properties are discuss-
sed below.
In all the languages except Dhopadhola, in which the obligation gram is
affixed to the verb, an overt subject can appear either before or after the
obligation particle. For example, in (7) from Alur, the subject noun could be
placed following omako. The subject pronoun in (8), from Lango, could also
appear in sentence initial position, but most speakers seem to place a subject
pronoun, if used, following myero, (recall that subject pronouns are op­
tional). 2 Other grammatical morphemes can seperate the obligation marker
from the lexical verb, as indicated in (6c) in which onwongo 'Past Imperfec-
tive' and dong 'then' seperate the particle from the lexical verb. The Negative
marker pe may appear between the particle and the lexical verb in Lango and
Acholi, as in (9c) from Acholi, but it generally appears at the beginning of a
subordinate clause, as in (9a) and (9b), without a difference in interpretation.

(7) Alur
Awiya omako o-dik-i i-paco tin
children must 3:SG-return-SUBJ LOC-home today
'The children must go home today.'
 112 Edith L. Bavin

(8) Lango
Onwongo gin myero o-bol odilo
PI they must 3:SG-throw ball
They would have thrown the ball.'
(9) a. Lango
Te bedo kobo atin gwok-ere ca ni pe
NARR stay talk child dog-POSS DET COMP NEG
dong myero gwee
then must bark
'He then told his puppy that he mustn't bark.'
b. Lango
Onwongo o-punyo-wa ni pe myero o-mede
PI 3:PL-teach-1:PL COMP not must 1:PL-continue
kede kodi luwenyi
with way fighting
'We were being taught that we shouldn't fight.'
c. Acholi
Myero pe o-bol odilo
Must NEG 3:SG-throw ball
'He shouldn't have thrown the ball.'
Another property of the obligation construction concerns the shape of the
lexical verb following the particle. In all five languages, the Subjunctive form
of the verb is used. Most indicative verbs in the Western Nilotic languages are
disyllabic and end with a final vowel o (e.g. from Acholi and Lango: neno
'see', miyo 'give', ryemo 'chase', keto 'put, bino 'come', camo 'eat', mito
'want', ringo 'run'). There is also a small set of mono-syllabic verbs in each
language (e.g. from Acholi: coo 'wake up', aa start from', and oo 'arrive').
For the disyllabic verbs, the Subjunctive form that follows (o)myero is formed
by omitting the final vowel (e.g. nen from neno 'see'), and for some verbs a
final vowel i replaces the o. Driberg (1923) discussing Lango, and Crazzolara
(1955) discussing Acholi, specify that the Subjunctive verbs which add a final
i in place of the o are phonologically determined; however this is no longer
true as speakers now use the suffix i only on motion verbs, or not at all. 3 For
The Obligation in Western Nilotic 113

example, Acholi speakers vary as to whether they use i on the verb ngwec
'run' and dok 'return', but they do not use a final i on non-motion verbs.
The Subjunctive form also follows wek 'let, allow', and gwok 'prohibi­
tion'. Crazzolara (1955:17) describes its function as hortative, and lists the
following meanings for the form: wish, invitation, exhortation, command,
prohibition. However, there are other uses. For example, in Lango and Acholi,
nen is used with the meaning 'to be seen'. The form is also used for the
imperative; in this function, also, there is variability in the use of the suffix i
(e.g. ngweci or ngwec 'run!'), although more Lango speakers than Acholi tend
to use the suffix.
Note that in (10) the Acholi speaker uses a final vowel i on the verb
following myero; he also produced (11), in which bed does not have a suffix.
A few verbs in each language are idiosyncratic since they always appear in the
Reflexive form, no matter what the context (e.g. medde 'continue, which is
illustrated in (12)).
(10) Myero a-ngwec-i
must l:sG-run-suBJ
T must run.'
(11) Myero a-bed piny
must l:SG-sit down
'I must sit down.'
(12) Myero a-med-de i-kwan iyonge mwaka aryo
should l:SG-continue-REF with-study after year two
'I must continue studying in two years/I intend to continue study­
ing in two years.'
In examples (l)-(5), the lexical verbs following the Obligation markers
are ngol 'cross' and nen 'see' from Lango, tim 'do' and wil 'buy' from Acholi,
cithi 'go' and bed 'stay' from Alur, chiem 'eat' and kony 'help' from Dholuo,
and woth 'go' from Dhopadhola. These are all Subjunctive forms of the
respective verbs, and all have a Subject Prefix. In contrast, verb complements
of equi-verbs do not appear in the Subjunctive form, and it is the matrix verb
that carries the Subject Prefix, not the complement. For example, amito neno
means 'I want to see' in both Acholi and Lango. Mito has the Subject Prefix,
not neno. There are two modal verbs in Lango and Acholi {romo 'can', and
twero 'to be able') which behave like equi-verbs; they carry the Subject Prefix
 114 Edith L. Bavin

and are followed by the Infinitive form of the verb, not the Subjunctive. These
properties are illustrated with romo in (13).

(13) Acholi
Ka i-tye i-peko i-romo penyo ngati-mo
if 2:SG-be with-problem 2:SG-can ask person-some
'If you have a problem, you can ask anyone.'
The position of the Negative pe shows up another difference between a
complement-taking verb and the modal particle (o)myero. As illustrated
above, pe usually appears between myero and the lexical verb. However, pe
precedes an equi-verb or a modal verb, as in the second clause of (14).
(14) Acholi
Gu-yabo wang ot ka ngiyo ka kwene ma
3: PL-open eye house PROG look place where REL
ogwalpok tye iye; pe gi-romo neno-ne
frog be LOC NEG 3:PL-can see-3 SG
They opened the window looking for the frog to see where it was;
they couldn't see it.'
A change in interpretation is made in both Lango and Acholi when the
Past Imperfective marker precedes the obligation marker. For example, in
(15) the Acholi speaker is indicating that while there was some obligation for
them to play ball they did not, whereas in (16), he is indicating that because it
is the most likely situation, he believes they did play ball. That is, when the
aspect marker follows the obligation marker, there is an epistemic interpreta­
tion. Other interpretations for the obligation constructions will be discussed in
Section 4.
(15) Gin onongo myero gu-tuk odilo4
They PI must 3:SG-play ball
They should/would have played ball.'
(16) Gin myero onongo gu-tuk odilo
They must PI 3:PL-play ball
They must have played ball.'
The Obligation in Western Nilotic 115

3. Lexical sources

3.1 Introduction

The source for the Dhopadhola obligation affix is not evident. Ripo is not a
lexical verb in the language group and it will not be considered further.
However, the lexical sources of the obligation markers in the other four
languages are evident.

3.2 Lango and Acholi

The lexical verb myero is listed by Driberg (1923) in an early description of

Lango with the meaning 'to be fitting for, suitable for, a match for'.
Crazzolara (1955), writing about Acholi, lists myero with the meanings 'to be
suitable, fit, becoming, tolerable' while Kitching (1907), also writing about
Acholi, lists its meaning as 'to be fit', right, worthy' as well as the obligation
meaning 'ought', 'must'. This lexical source is compatible with sources listed
for obligation in Bybee et al. (1994) who list 'be fitting' and 'be proper' as
possible sources. Clearly, myero was a lexical verb in the early part of the
century, given the meanings listed by both Driberg and Kitching; Driberg also
uses one example, which is given in (17). In both Lango and Acholi the lexical
semantics have now been been lost since myero is now used only as a modal
particle, not a verb.

(17) Lango
Oluk o-myer-e jobi
Oluk 3:SG-match-REF buffalo
'Oluk was a match for the buffalo.' [Driberg, 1923: 391]
In form, omyero is the Past Third Person Singular of the verb myero.
Given that Acholi uses omyero or myero, it is hypothesized that that both
Lango and Acholi started with omyero, but Lango has dropped the prefix. This
seems the most likely direction of change, since forms generally become
reduced in form as they become grammaticized (Bybee 1985).
Assuming the Third Singular Past form is the source for the obligation
particle, one change to be accounted for is the change in the function of the
prefix o. In omyero 'obligation', o does not refer to a particular entity, and
 116 Edith L. Bavin

thus the form omyero is impersonal in interpretation. That is, a Third Singular
form with past reference was reinterpreted as impersonal and as an on-going
state or condition ('It is fitting'). As discussed by Bybee (this volume), a
change from past form to present interpretation is common for modals. The
impersonal sense of 'It is fitting' indicates that it is desirable for an existing
situation to be maintained, that is, certain conditions that hold are appropriate
according to the current views of the community, the authority that deter­
mines what is fitting or necessary. To maintain the values of this society, a
person is obliged to conform to these views.
Since 'It is fitting/suitable' has a non-agentive, intransitive meaning, it is
feasible that the short form of myero was first used with the impersonal
meaning. Recall that a distinction is made for a few verbs in the languages
between a short (intransitive) form and a long, transitive form which is
marked with the o suffix. The distinction might have been more common in
earlier stages of the languages, with the o suffix being added to the short stem
of verbs to mark transitivity. Since this is the same form as the Third Person
Subject prefix o, the most likely development is that o was used to mark a
third Person Object but was reinterpreted as a marker of transitivity, and later
as part of the verb stem for all but a few verbs. However, this development
could have taken place prior to myero becoming used as a modal particle.
The impersonal form ('It is fitting') would have propositional scope with
the source of the modality being society in general. However, changes have
taken place to allow the form to become a clause-internal grammatical marker
which predicates the modality onto the agent of a particular activity. One
necessary change is that the impersonal verb takes a verbal complement, as in
'It is fitting to run'. In this way, the impersonal lexical item becomes part of a
larger construction.
In order to personalize an utterance containing omyero and a verbal
complement, the speaker must direct the message to or about a particular
individual. This would necessitate adding a Subject Prefix. Because Lango
and Acholi are pro-drop languages, an overt subject nominal is not required,
but the bound form is. The fact that the Subject Prefix did not replace o on
myero indicates the shift was already in progress, with the complement being
analyzed as the main predicate. Thus there was a change from a propositional
scope lexical modality to a specific predication of the modality onto an agent,
the subject of the new predicate.
There is no evidence for discrete stages in the shift, either semantic or
The Obligation in Western Nilotic 117

syntactic, and it is likely that there was synchronic variability as the changes
took place. Thus the following list does not represent distinct stages in the
path of change. Rather it illustrates the starting point and the current construc­
tion. Myero started as a lexical verb but it can no longer be used as a lexical
verb, only as a modal particle in a clause which has an inflected predicate.
(18) o-myero 'He/she/it fits/matches X' (transitive)
(OR o-myer 'It fits' (intransitive))
omyero 'It is fitting/suitable/appropriate' (impersonal)
omyero + verb 'It is appropriate to V'
omyero prefix+V 'It is appropriate/ necessary for X to V'
One further development has occurred. The prefix on omyero serves no
purpose as a subject marker and has been dropped in Lango, and also by some
Acholi speakers. For some of those Acholi speakers who use both forms, the
forms have developed different functions. For example, one male who used
both myero and omyero explained the difference as follows: omyero has more
force, but with myero the speaker is just telling someone what they must do.
This suggests that omyero is more formal, and this would fit in with the
hypothesis that it is the older of the two forms. Lango, then, is slightly more
ahead in the grammaticalization of the obligation marker in that no speakers
have been found to use the o prefix. Lango also shows more change in the
forms of verb prefixes generally, since o is used for third singular and all
plural persons.

3.3 Alur

The source for the Alur particle omako is the lexical verb mako 'hold, catch'.
The verb form mako is currently a lexical verb with the meaning 'to hold,
catch' in all five languages under discussion. In addition to its active meaning
with an agent subject, it is also used in Acholi in expressions showing a
situation has arisen in which a non-animate is the cause of a current state. For
example, 'He is mad' is literally 'madness holds/ captures him'. In Alur, the
meaning has been extended to the current meaning of obligation.
What is of interest is the form of the verb. As with omyero, omako has the
o prefix, but because the prefix does not refer to a particular person but society
at large, it is an impersonal form. Similar developments seem to have oc­
curred with this verb as with omyero in Acholi and Lango, in that a verbal
 118 Edith L. Bavin

complement with Subject Prefix must have been added, and the form omako
reinterpreted as a clause-internal grammatical marker, which shifts from
having propositional scope to predicating the modality onto the subject of the
predicate. In its lexical use, mako takes a nominal complement; for example,
omako winyo means 'He/she caught/held a bird'. But the verb mako does not
take a verbal complement, as is possible with English 'caught'. Consider the
sentence 'He/she caught a fish swimming in the river'; here 'fish' could be
analyzed as the object of 'caught' as well as the subject of 'swimming'.
However, in the Western Nilotic languages, mako would not be used here; the
verb gwok 'constrain' would be used instead.
The fact that mako (as a lexical verb) does not take a verbal complement
suggests the personal meaning of omako was not used as the source for the
shift to a modal meaning. It is likely that an impersonal meaning developed
for omako, as with omyero, and that as the semantic changes took place so did
the syntactic. Just as with omyero, omako is a Past form of a verb, and its
impersonal meaning 'It was held' has been extended to non-past contexts.
While physical contact is part of meaning of the lexical verb mako,
'catch/hold', the modal meaning of omako implies social constraints, which
are the views held by society as to what is appropriate. For example, the literal
interpretation of (3b) is 'If you want food, it is held you stay with money'.
Omako here does not convey any sense of physical constraint or support;
however it does convey the sense of social constraint. That is, a concrete
lexical item has been used to express a more abstract notion of constraint. A
similar extension has been discussed for English 'have', which once meant 'to
hold in one's hand', the meaning has changed so that the notion of physical
presence of a possession has been lost, although one can have an abstract
possession (Bybee and Pagliuca (1985:72). In Alur, the meaning of omako
'hold' has been metaphorically extended so that one may hold ideas as well as
physical objects. However, the interpretation is not associated with the per­
sonal form of the verb. Since it is the impersonal form which has been
grammaticalized with a modality function, it is through this form that the
semantic change has spread.

3.4 Dholuo

For DhoLuo, Stafford (1967) lists the meaning of the form nyaka as 'compul­
sion/ necessity'. Huntingford (1959), Omondi (1982), Gregerson (1962), and
The Obligation in Western Nilotic 119

Malo (1952) also cite the form nyaka, although they differ in the senses listed.
While Huntingford enters meanings of 'must', 'from' and 'since', Omondi
gives only 'must'. The form nyaka is also currently used in DhoLuo with a
spatial meaning 'up to' and this is a cognate of naka 'up to' in Acholi as in
example (19).
(19) Acholi
Gu-ci-kwalo dyangi naka Patong o
3:PL-go-steal cows up to Patongo
'They went to steal cows as far as Patongo.'
In discussing Lango, not DhoLuo, Driberg (1923) lists nak as the stem
for 'to bear fruit' with nyal or nywalo as 'to be fruitful'. In the related
language, Dhopadhola, nyalo is currently used for the epistemic modality of
Possibility ('might'). While it is beyond the scope of this paper to speculate
the possible relationship between the Dhopadhola form for 'might' and the
DhuLuo form for 'must' and the relationship to a verb meaning 'to be
fruitful', it is conceivable that there is one.
While nyaka is the form generally used in DhoLuo to mark the modality
of obligation, there are alternatives. Malo (1952) gives onego, owinjore and
oromo. Omondi (1982) lists both onego and owinjore with the meaning
'should', while Stafford (1967) lists onego and owinjore as meaning both
'should' and 'ought'. For the Reflexive verb winjore, he gives the meaning 'It
is convenient, fitting'. Thus the lexical source for winjore is similar in
meaning to the source which has developed into the Lango and Acholi
markers of obligation.
Omondi argues that while nego and winjo in DhoLuo are interchange­
able, with the meaning 'fits into', there must be a meaning difference since
only winjo can be substituted by romo 'to deserve'. Writing on Alur, Ringe
(1948) lists the meaning of romo as 'fit'. However, in Lango and Acholi,
romo does not have either meaning ('deserve' or 'fit'); it functions as a verb in
these languages with the meaning 'to be enough, sufficient, equal', and it is
also used as a marker of epistemic modality with the meaning 'can'. This is
not a function for romo in the other three languages in the group.
Although there are alternative forms for marking obligation, one consult­
ant pointed out that he interprets nyaka as more 'forceful' than nego, and
winjore as the least strong, being used in the sense 'It is sensible, reasonable'
or, 'You the listener should see the sense of what I'm saying'.
 120 Edith L. Bavin

3.5 Similarities among the languages

In summary, there are some similarities across the languages in the meanings
of the lexical sources for the obligation markers, even though the forms are
different. These similarities support the notion of a conceptual basis for
diachronic changes. For example, omyero, the impersonal of 'to be fitting' is
the source for the obligation marker in Lango and Acholi. For DhoLuo,
Stafford gives the meaning for winjore, a weak obligation marker, as the
impersonal 'It is fitting', although Omondi lists 'fits into'. The lexical mean­
ing is retained alongside the obligation meaning in Dholuo, unlike the situa­
tion in Lango and Acholi, where the lexical meaning has been lost.
Another form used in DhoLuo as a marker of obligation, romo, is listed
by Ringe (1948) with the meaning of 'fit' for a related language, Alur. This
form has developed into the marker of another modality in Lango and Acholi,
the epistemic modality of Possibility which, in contrast to the agent-oriented
modalities, has scope over the whole proposition.

4. Semantics of the Obligation marker in Lango and Acholi

4.1 Introduction

As is documented in several sources, markers of obligation and necessity in a

language can be extended to other contexts (e.g. Bybee 1985, Traugott 1989).
One often-cited example is the extension of English 'shall', from a meaning of
obligation to a meaning of intention when used with first person. In this
section I examine some of the interpretations of (o)myero in Lango and
Acholi. In addition to the obligation meaning (weak and strong), (o)myero can
also be interpreted as a marker of Necessity, depending on whether the
constraints are social or physical. The relationship between the speaker and
hearer also affects the interpretation; when an authority figure uses (o)myero
in an utterance directed to a younger person, it is the obligation sense that is
interpreted. Another interpretation is that sentences with (o)myero express the
speaker's own attitude, rather than society's in general, and this may reflect
the speaker's desire for a particular outcome. The form has an epistemic
function also. Some of these uses have already been illustrated, and other
examples are discussed below.
The Obligation in Western Nilotic 121

4.2 Uses of the Obligation marker for Necessity and Obligation

In (11), the Lango speaker is tired and wants to sit, so her physical condition
motivates the use of myero; the form here is interpreted as Necessity, as it is in
(20). Necessity 'reports the existence of physical conditions compelling an
agent to complete the predicate action' (Bybee et al. 1994). In (20) the
children need to lift up their clothes so they don't get wet; the motivation to
lift them is created by an external factor, the physical conditions.
(20) Ma peya wa-oo i-school omyero wa-ngol kulu,
REL not yet l:PL-arrived LOC-school must l:PL-cross river
ma peya wa-ngolo kulu omyero wa-lung bongo-wa
REL not yet l:PL-cross river must l:PL-lift clothes-1:PL
'Before we reached the school we had to cross a river; before we
crossed the river we had to lift up our clothes.'

Examples (21) and (22), from Acholi, are extracts from a riddle. The
listener is told that a man has some obligation to get his things across the river.
He has to take three things, a lion, a chicken and some seed, but he can only
take one thing at a time because the boat is too small to take them all. If he
leaves the lion with the chicken, it will it eat the chicken, and if he leaves the
chicken with the seeds, it will eat the seeds. Thus the items must be trans­
ported in a particular order. Here the physical conditions create the necessity
to take the items in a particular order.

(21) Ci omyero en o-cwal jami-ni weng loka kulu

and must he 3 SG-take thing-DEM all across river
'And he needed to take all these things across the river.'
(22) Omyero o-te mené wiati
must 3: SG-take which first
'Which one must he take first?'

However, example (23), from Acholi, expresses an obligation; there are

strong social constraints concerning choosing a wife.
(23) Ma peya dako o-bino bot-i omyero i-cul lirn
REL not yet woman 3:SG-come to-2:SG must 2:SG-pay money
'Before the woman comes to you, you have to pay some money.'
122 Edith L. Bavin

Context affects the interpretation of (o)myero. It is used, together with

the Negative marker, with a sense of strong obligation when a social more is
being violated (as in (24)). There is also an obligation interpretation when a
person of authority speaks, particularly to a child. This is illustrated in
examples (25) - (27) which I have extracted from Alerotek's (1972) Acholi
book about a boy named Kidega. (Glosses have been added to the original.)
The two forms of the particle (myero and omyero) are used interchangeably in
the text: both appear with verbs with singular or plural subjects, and both
appear clause-initially as well as following some other words(s). In the
examples, the school headmaster is addressing Kidega and his father. Kidega
understands the headmaster's language, but his father does not.
In (25), the headmaster warns the family that they must save so there is
enough money to pay the school fees. In (26), the headmaster is directing
Kidega to explain to his father; both instances of myero in this example are
interpreted as markers of obligation; the boy is obliged to translate because
his father does not speak the language. In (27), the boy is being warned; if he
does not improve his behavior he will not be allowed to continue with his
studies.

(24) Acholi
In omyero i-cam mot
YOU(SG) must 2:sG-eat slowly
'You should eat slowly.'
(25) Edimacta lacen o-tito ni pic myero gi-gwok
headmaster elder 3:SG-say COMP schoolfees must 3:PL-keep
pien kare me cul pud tye anyim, wa i-nino
because time for school still be ahead, up to LOC day
dwe 27 me Janwari
month 27 for January
'The headmaster said that they must save their money because
school (and school fees) were due on January 27th.' [p. 12]
(26) A-mito ni myero Hi ki woru myero en
l:SG-want COMP must 2:SG-tell to father must he
o-gwok cente pic maber in i-bi-kelo ne i-nino
3:SG-keep money fees properly you 2:SG-FUT-putthisLOC-day
The Obligation in Western Nilotic 123

dwe 27 me dwe man

month 27 for month this
'I want you to tell your father that he must keep the money
properly; you will bring it on the 27th.' [p. 12]
(27) In Kidega myero i-lub cik me cukul ka i-mito wan.
you Kidega must 2:SG-follow rules for school if 2:SG-want study
' You, Kidega, must follow the school rules if you want to stay in
school.' [p. 13]

4.3 Obligation particle used to express speaker's attitude or speaker's

intention

Sentences with (o)myero often express something about the attitude of the
speaker. In (lb), the Lango speaker wants the listener to convey a message to
someone else; she is really indicating that it would be more convenient for the
other person to travel than her. In the negative sentence (9a), the child's
decision is imposed on the dog; the child has heard a noise and wants to
investigate the source. And in (28), the speaker is letting us know her disap­
pointment with the non-runner. She could have added the equivalent of 'So
why didn't you?'
(28) Lango
Myero onwongo i-ngweci
must PI 2:SG-run
'You (SG) should have run.'
Thus (o)myero) can be interpreted in many contexts as the speaker's
attitude to a particular situation or a desire for a particular outcome. It can also
be used to mark the intention of the speaker, as in (12). The intention reading
is most common when an (o)myero clause is embedded in a matrix clause with
the verb tamo 'think' and a First Person Singular subject, as in (29); the
interpretation is that the predication expressed in the subordinate clause is
what the speaker intends to do, not what he must do.
Example (29) is taken from a passage in which an Acholi man is talking
about his plans for the future. In terms of cultural expectations he should own
land. Because he does not have any land, he intends to acquire it. Compari-
 124 Edith L. Bavin

sons can be made here with English 'have to'. As discussed in Bybee et al.
(1994), the obligation expression 'have to' may be interpreted as 'intention'.
Here omyero is interpretated as 'intention'.
(29) A-tamo ni omyero a-nong kaka-mo ma-ber
l:SG-think COMP must l:SG-find land-some REL-good
ma-ayela o-bi-bedo peke
REL-trouble 3-SG-FUT-stay not
'I intend to find some good land where there is no trouble (i.e. no war).'
In (30), the speaker's intention is conveyed with an embedded omyero
clause, but here the nominal tarn 'idea' is used in the matrix clause rather than
the verb tamo. The speaker produced the utterance in answer to a question as
to whether he would accompany an acquaintance to a social function.
(30) Onongo a-bedo ki tarn ni omyero a-cit ki jo
PI l:SG-stay with idea that must l:SG-go with people
ma a-bedo kwed-gi i-floor-na-ni
REL l:SG-stay with-3:PL LOC-floor-l:SG-DET
T think I should go with the people from my floor.'
(='I intend to go with the people from my floor, not you')

4.4 Epistemic interpretation of the obligation marker

obligation modals in English can be used with epistemic functions, as dis­

cussed in, for example, Coates (1983). The form (o)myero may also be
interpreted as having an epistemic function, indicating a qualification on the
part of the speaker to the truth of the proposition. That is, it has the whole
proposition in its scope. For example, an Acholi speaker answered a question
about some people's location with (31), which states the most likely location
for the people. The interpretation of omyero here is probability, and this is a
similar interpretation to that of the English sentence 'The room must be
empty' (Coates 1983), in which the speaker is noting the most likely condition
of the room.
(31) Omyero gin gu-cit ka-dwar
must they 3: PL-go PROG-hunt
'They must have gone hunting.'
The Obligation in Western Nilotic 125

In Section 2, we discussed the fact that the speakers of Lango and Acholi
may vary the word order of the modality particle and Past Imperfective
marker to mark the speaker's commitment to the proposition. That is, the
order with the Modal Particle following the Aspect marker is used when the
speaker is commenting on the most likely outcome. In English, 'You must
have done it' could be used if the speaker sees evidence that it has been done
already, or if this is the most likely situation (as in 'The room must be empty').
With emphasis on 'must', it could express a sense of disbelief that you did not
do it. In contrast, 'You should have done it' implies you did not, so has a
counter-factual interpretation. In Lango and Acholi, the difference between
expressing what the most likely situation is and what should have been done
(but was not) is marked by the position of the Aspect marker, and listeners use
this variation in ordering to interpret the intended message. Thus linguistic
context affects the interpretation of the form (o)myero.

5. Verb sources for grammatical categories in the languages

Western Nilotic languages frequently use erstwhile verbal forms to mark

grammatical categories. This can be observed in changes that have resulted in
verbal affixes, as well as in constructions which reflect change in progress as
one of the verbs in a combination takes on grammatical functions. The
following examples illustrate some of the grammatical forms which have
lexical verb sources
In Acholi , Lango and Alur bi 'future' is derived from the verb bino
'come' which retains its lexical meaning. The form bi appears between the
subject pronoun and the lexical verb (e.g. a-bi-camo T will eat' from a-bino
camo T come to eat'). In Lango, the long form, bino, can also be used as a
future marker. As illustrated in (5), ripo 'obligation' is used between the
lexical verb stem and the Subject Prefix in Alur. It could well have originated
as a verb, as did bi 'future', but the verb is no longer in use, unlike bino
'come'.
Another example from Lango, Acholi and Alur is the morpheme be, used
to indicate duration of an event; it is derived from bedo 'to stay, sit'. In Lango
and Acholi, both the form be and the long form bedo are used, but in Alur the
difference between bedo and be has been grammaticalized, so that the long
form used in past contexts with an anterior function now contrasts with the
 126 Edith L. Bavin

short, which is used in present time contexts with a Durative function, as in

the following examples.
(32) Alur
a. A-be-sayo gwok para
l:SG-DUR-look-for dog my
'I am looking for my dog'
b. A-bedo ka-sayo gwok para
l:SG-stay PROG-look for dog my
'I have been looking for my dog.'
These examples illustrate that the lexical semantics of a verb may change
over time and the form may become re-interpreted as a marker of a grammati­
cal category. The grammaticalized forms may become reduced (e.g. bi from
bino, be from bedo, and, in Lango, te 'narrative marker' from teko 'to
initiate'). Bybee (1985) and Bybee and Pagliuca (1985) discuss both phonetic
erosion and fusion as features of the grammaticalization process. In Lango,
Acholi and Alur, the forms bi and be are clearly well established as grams; not
only do they have a grammatical function, they are also reduced in form and
are affixed to the verb stem.
There is a difference, however, between developments which result in
reduced verb forms as affixes and the development of (o)myero as a marker of
modality. It is the impersonal form of omyero that is grammaticalized,
whereas in the other examples cited the grammaticalized verb form appears
following a Subject Prefix which varies according to the person and number
of the subject of the clause. When a second verb is added to a prefix + verb
combination, no major change is needed for the structure to develop into one
form: prefix-gram-verb. However, for omyero to have become a grammatical
marker, the following modifications had to be made: the prefix o lost its
meaning as a subject marker, a verbal complement was added, and that verb
carried a Subject Prefix.
A form which functions both as a gram and a lexical verb in Lango and
Acholi is onwongo in Lango and (o)nongo in Acholi. (The cognate for the
lexical verb is nwango 'find' in both Dhopadhola and Alur.)5 There are some
parallels between the development of onwongo as a grammatical marker and
omyero as a modality marker. The form onwongo is the Third Person Singular
Past form of 'find', but it is impersonal in use. It is used in past contexts and
functions as a (past) imperfective aspect marker, marking on-going states or
activities.
The Obligation in Western Nilotic 127

In (33), only the lexical meaning appears, but one sentence may include
both meanings, aspectual and lexical, as in (34) and (35). Note that nwongo
'find' in both (33) and (35) lacks a Subject Prefix because it follows the
narrative linker te; no verb following te has a Subject Prefix. (O)nwongo is
similar to omyero in that it always precedes the lexical verb. However, there is
a difference: while (o)myero has lost its lexical meaning, (o)nwongo has not.
Myero has lost its prefix for Lango speakers and also for some Acholi
speakers, but onwongo has retained the o prefix in Lango and its presence is
variable in Acholi.

(33) Te-nwongo wi nyinyang iyi pii

NAR-find head crocodile in water
'Then he found the head of a crocodile in the water.'
(34) Ikare-moro onwongo tye awobi-moro ame
time-some PI be boy-some REL
onwongo tye kede ogwalogwal-ere a gwok-o o-nwongo
PI be with frog-poss REL dog-DET 3:SG-find
'Once there was a boy who was with his frog that the dog found.'
(35) Bongi ame onwongo tye idi ot o-bobolo tende me
clothes REL PI be middle house 3:SG-throw far for
neno ka ogwalogwal-ca nyo tye i-te-re, ento
see if frog-DET maybe be LOC-under-3sG, but
te-nwongo-ni pe
NAR-found-3:SG NEG
'He threw the clothes that were on the floor to see if the frog was
under them, but he (the boy) did not find it (the frog).'

To summarize, we have given a number of examples to illustrate that

Western Nilotic languages commonly use lexical verbs as the sources for
grammatical markers. However, the sources become reinterpreted in one of
two types of constructions, one with two verbs with the same subject, and the
other an impersonal construction. Sometimes the lexical meaning is retained
and sometimes lost. Both of these possibilities have been discussed as possi­
ble outcomes in grammaticalization processes. Traugott (1989), for example,
argues that when a new function develops, it may exist alongside the old.
Another possible outcome is that the new function takes over and the old
function is lost.
128 Edith L. Bavin

6. Use of the lexical verb for 'want' for agent-oriented modalities

As discussed in Section 4, (o)myero 'must' may also be interpreted as 'need

to'. However, another verb which may be used to express obligation or
necessity is mito 'want', as illustrated in (36). The context for this example is
that the speaker has been told he has to cross the river, so he is not expressing
an internal desire; rather, the speaker is obliged to get to the other side. 6 Note
that mito, like myero 'must', does not carry a Subject Prefix in Lango.
(36) Mito nen ni a-ngola kulu
must see COMP l.SG-cross river
'It seems I have to/must cross the river.'
Similarly, in Alur, which marks obligation with omako, the verb for
'want' is used for 'need to', in the impersonal form omito. Consider the
following Alur sentences in which the use of the lexical verb mito 'want' is
illustrated as well as the modal use of omito. The modal form omito appears in
front of a lexical verb which carries a Subject Prefix. However, as a lexical
verb mito carries a subject pronoun, as in (c). The contrast between (c) and (d)
captures the difference in meaning between the two functions: in (c), the
speaker desires to run, but in (d) there is a need to run (so that he is not late for
class); time is the physical constraint here. In comparison, in (a) the need to
work is motivated by the rewards that will follow.

(37) Alur
a. Omito a-tim matek
need 1 : SG-work hard
'I need to work hard'.
b. Omito nango a-cam
need PAST l:SG-eat
'I should have eaten.'
c. A-mito ringo
l:SG-want run
'I want to run.'
d. Omito a-ringo
need l:SG-run
'I need to run.'
The Obligation in Western Nilotic 129

In DhoLuo, dwaro is the lexical verb for 'want', not mito, and dwaro can
also be used to express Necessity, but it is not grammaticalized for this
function, as in Lango and Alur, in which the impersonal form is used. The
context alone determines whether the interpretation is 'want' or 'need' since
the verb carries a Subject Prefix for both interpretations, as illustrated in (a)
and (b); the prefix is determined by the person and number of the subject of
the predication. In (a), the speaker is not reporting what his friend desires; he
is giving his own view that his friend has few possessions and needs to obtain
a few more basic necessities to improve his quality of life. The speaker in (b)
is not reporting what he wants to do; he has to tell a story because I asked him
to. This was part of a session for which the speaker would be paid for his time;
he had indicated that he did not want to tell a story since he couldn't remember
one.
Note that with both senses (want and need) the verb can have either a
nominal or a verbal complement. In (38a), a nominal complement is used,
while (b) contains a verbal complement. In other contexts, the verb in each
sentence could be interpreted as 'want'.
(38) a. O-dwaro gik mangeny
3:SG-need thing many
'He needs many things.'
b. A-dwaro gano-nu sigana moro ka
l:SG-need tell-2:SG story certain here
T need to tell you this story.'
We can compare the use of English 'want' to express what someone
needs to do. For example, in 'You want to go to the next counter', which is a
response that could be made when someone asks for a particular service at the
wrong counter, the interpretation is that you need to go to the next counter.
Another context in which 'want' is used with the interpretation 'need to' is a
school report on which a teacher might write that a student wants to work
harder, meaning that the student needs to. Similarly, when someone says T
want a pen', it would be interpreted as 'I have need of a pen' in a context in
which that person is required to sign a form. It is context which determines
whether there are physical conditions present which lead to the 'need' inter­
pretation of 'want', or whether the utterance expresses someone's desire and,
perhaps, intention (see Bybee et al. 1994, chapter 6).
 130 Edith L. Bavin

As illustrated above, in the Western Nilotic languages there is evidence

of grammaticalization for the 'need to' sense in two of the languages, Lango
and Alur, but not DhuLuo, in which context alone determines which interpre­
tation to make. This suggests that the 'need to' interpretation has been derived
from 'want', following a regular pattern of change in the language group, a
pattern of deriving grams from lexical verbs (often impersonal in form). The
'want' interpretation might be newer than that of necessity, but there is no
clear evidence which is the older meaning. If 'want' is the newer meaning, it is
the newer meaning which is retained in the lexical verb, while the older
meaning has been grammaticized in two of the languages. Both senses are
retained, but they are formally distinguished.

7. Conclusions

It can be fruitful in a study of grammaticalization to examine diachronic

developments in related languages since changes in meaning and functions
for lexical items and their cognates can be compared. While languages may be
closely related and share a very high percentage of vocabulary items, they
may develop different forms for marking grammatical categories although
there may be similarities in the semantics of the lexical sources for these
categories.
The only languages in the Western Nilotic group to have a verbal affix to
express the modality of obligation is Dhophadola. Three of the other five use
particles which have developed from lexical verbs, while DhoLuo uses a
number of forms which have developed from lexical verbs. In fact, lexical
verbs seem to be the predominant source of grams relating to tense/aspect/
modality systems in the five Western Nilotic languages.
There is common development in Lango and Acholi of the impersonal
form of the verb 'to be fitting, suitable' into a marker of obligation, with the
loss of the lexical semantics of the verb. However, the form is more grammati­
cized in Lango since the impersonal prefix has been dropped, while it is
variable in Acholi. In contrast, the impersonal form of the verb 'to hold, catch'
has developed into a obligation marker in Alur, but its function as a lexical
verb is retained.
The contexts of use for (o)myero reveal that it does not always carry a
Strong obligation sense. It is in those contexts in which an authority figure is
The Obligation in Western Nilotic 131

speaking to a subordinate, or when social customs are being discussed that the
obligation interpretation is made. In some contexts, the form is interpreted as
a marker of Necessity or of Intention, and it can also have an epistemic
interpretation. Such synchronic variation is indicative of change in progress
with the obligation marker gradually taking on other functions as its meaning
generalizes.
Bybee et al. (1994) discuss four semantic notions in the set of agent-
oriented modality: obligation, necessity, ability and desire. In the Western
Nilotic languages there is a close connection between the notions of obliga­
tion, necessity and desire in terms of the forms used to express them. How­
ever, the forms used to express ability in the languages are distinct. In Lango
and Acholi, ability is currently expressed with a subject-inflected verb which
takes an Infinitive complement, as illustrated in (13) and 14), unlike the
obligation marker, which is derived from an impersonal construction.
As illustrated above, several paths of development in a language may
result in different forms having interpretations associated with agent-oriented
modality. The lexical verb for 'want' is also used to express necessity or
obligation in three of the languages. While two of the languages use the
impersonal form, DhoLuo uses a verb inflected with Subject Prefix, and it is
the context alone that determines whether the verb is to be interpreted as
desire or necessity.
Language-specific patterns can be detected in the paths of development.
For example, the grammaticalization of particles is more advanced in Lango,
in that both mito and myero lack prefixes, although the former is still a lexical
verb when used with Subject Prefixes while myero is not.

ABBREVIATIONS

NAR narrative marker PL plural

DEM demonstrative POSS possessive
DET determiner PROG progressive
COMP complementizer REF reflexive
DUR durative REL relative marker
LOC locative SG singular
NEG negative SUBJ subjunctive
PI past imperfective
 132 Edith L. Bavin

NOTES

1 Unless otherwise stated, the data reported in this paper were collected from native
speakers of the languages. Many of the examples are taken from discourse, rather than
from elicited sentences. The data were collected in Uganda in 1992, in Kenya in 1990,
and between 1981 and 1992 from students who have been studying in Australia for short
periods. Speakers have ranged in age from 20-38 years. Funding for the research has
been from an Australian Research Council grant.
2. Note that the Plural forms of the verb in Lango take o as the subject prefix (as does the
third singular) rather than wa '1 PL', wu '2 PL' and gi '3 PL' as used in Acholi. Thus
there is potential ambiguity if there is no overt subject, and Lango speakers often add a
third person plural pronoun plus a definite suffix ('this') following the verb as in the
examples given below. Thus a third option for sentence (8) is given in (c) below:
a. Ka o-poto gin-i i-pii-ni
when 3:PL-fal they-DET LOC-water-DET
'When they fell into the water.'
b. O-tye o-neno gin-i i-wang dirija
3:PL-be 3 PL-see they-DET LOC-eye window
'They were looking out of the window.'
c. Onwongo myero o-bol gin-i odilo
PI must 3:SG-throw they-DET ball
'They would have thrown the ball.'
The prefix o has generally replaced e as the form for third singular in Acholi and Lango
for past contexts (currently e is optionally used in non-switch-reference contexts only).
Non-past has a zero prefix in Acholi, but a is used in Lango. In past contexts, the form
gu is used for Third Plural in Acholi. This is derived from gi+o. For some Acholi
speakers, when the overt third plural pronoun (gin) is used, there is the option of using
the prefix o on the verb instead of gu.
3. Use of the Subjunctive form depends on the syntactic and semantic context. For a
number of lexical items the Subjunctive is used for resulting states (such as 'is stuck', 'is
dry' and 'is broken'), and for others, the short form is used as the intransitive. It can also
function as a nominal, so it is ambiguous as to whether amito cam is T want to eat' or I
want food'. For most verbs the long form is used for both transitive and intransitive
sentences. The final o is most likely an old marker of transitivity since it can be
analyzed, diachronically, as an object marker. Note that it takes the pronominal object
position. Synchronically, however, its function is a verbal marker, and is not restricted to
transitivity.
4. The anterior aspect marker is onwongo in Lango but onongo or nongo in Acholi. Past
tense is marked by three low tones on the verb in both languages.
5. In Dhopadhola nwango is used with past reference, and in Alur, nwango 'find' contrasts
with nango which has past reference. Compare (a) and (b) from Alur:
a. Koth o-daro bino
rain 3:SG-finish come
'The rain has come.'
The Obligation in Western Nilotic 133

b. Nango koth o-daro bino

past rain 3:SG-finish come
'The rain had already come. '
6. Bybee et al. (1994, chapter 6) discuss problems with having both a 'must' and a 'want to'
reading for one form since their hypothesis predicts that no gram should display
obligation and desire as alternative uses.

REFERENCES

Bavin, E.L. 1982. "Aspects of Morphological and Syntactic Divergence in Lango and
Acholi". Studies in African Linguistics 13 (3).231-248
Bavin, E.L. 1990. "The ki-V Construction: Development of a Resultative". La Trobe
University Working Papers 2.29-42
Bybee, J. 1985. Morphology: A Study of the Relation between Meaning and Form.
Amsterdam & Philadelphia: John Benjamins
Bybee, J. & Pagliuca, W. 1985. "Cross-Linguistic Comparison and the Development of
Grammatical Meaning". Historical Semantics, Historical Word Formation ed. by J.
Fisiak. Berlin: Mouton.
Bybee, J., R. Perkins & W. Pagliuca 1994. The Evolution of Grammar: Tense, Aspect &
Modality, Ch. 6. Chicago: University of Chicago Press.
Coates, J. 1983. The Semantics of Modal Auxiliaries. London: Croom Helm
Crazzolara, J.P. 1955, [c 1938]. Acholi Grammar. Oxford: Oxford University Press.
Driberg, J.H. 1923. The Lango, a Nilotic Tribe of Uganda. London: J. Fischer Unwin.
Fleischman, S. 1989. "Temporal Distance: A Basic Linguistic Metaphor". Studies in
Language 13.1-50.
Gregerson, E.A. 1962. "Luo: a Grammar". Yale University PhD dissertation.
Huntingford, G.W.B.1959. Elementary Lessons in Dho-Luo. London: School of Oriental
and African Studies.
Kitching, Rev. A.L. 1907. An Outline Grammar of the Gang Language. London: Society
for the Promotion of Christian Knowledge.
Ladefoged, P., R. Grick & C. Criper 1968. Language in Uganda, Oxford: Oxford Univer­
sity Press.
Malo, S. 1952. Dholuo Without Tears. Kisumu: Anyange Press.
Omondi, L. Ndong'a. 1982. The Major Syntactic Structures of DhoLuo. Berlin: Dietrich
Reimer Verlag.
Ringe, P.C. 1948, A Simple Alur Grammar. Eagle Language Study Series.
Stafford, R.L. 1967. An Elementary Luo Grammar. Nairobi: Oxford University Press.
Traugott, E. 1989. "On the Rise of Epistemic Meaning in English: An Example of
Subjectification in Semantic Change". Language 65.31-55.
The Gestural Expression of Modality in ASL

Sherman Wilcox and Phyllis Wilcox

University of New Mexico

... The essence of language is human activity — activity on the part of one
individual to make himself understood by another, and activity on the part of
that other to understand what was in the mind of the first...
O. Jespersen, The Philosophy of Grammar (1924: 17)

1. Issues in the linguistic description of ASL

For the most part, modern linguistic descriptions of American Sign Language
(ASL) have relied on formalist linguistic theories. The predominant effort has
been to demonstrate that ASL is indeed a language and can be described using
theories developed for spoken languages. Further, it is typically claimed that
the findings of ASL linguistic research lend support to these theories of
language.
These theories incorporate several powerful assumptions about the goals
and expectations of linguistic investigation (cf. Langacker 1991: 507-514).
One assumption is that grammar is independent of meaning. From this as­
sumption follows what Givón (1989: 94-95) calls three "pre-empirical postu­
lates":
(1) Language is a separate module of the mind/brain, not part of 'general
cognition';
(2) Structuralism in the analysis of language; that is, language structure
can be analyzed independently of its communicative function;
(3) The sign-relation between the linguistic code and its mental designa-
tum is arbitrary, unlike the obvious iconicity seen in pre-human com­
munication.
 136 Sherman Wilcox and Phyllis Wilcox

Another assumption is that the mental is independent from the physical. This
leads to the view that linguistic units are mental things that cannot be identi­
fied with a set of articulatory, physical characteristics (Fowler 1986: 9). This
view is stated by Givón (1989: 95) as another pre-empirical postulate:
(4) Some abstract, idealized entity — be it langue or competence — is the
'object'of linguistic analysis.

Little effort has been expended by signed language linguists in questioning

the underlying assumptions of these linguistic theories (S. Wilcox 1990). This
seems especially unfortunate because signed languages would appear to be
unique, natural "laboratories" for testing the validity of these assumptions,
e.g., the obvious presence of iconicity in signed languages needs to be
examined. We will suggest in Section 3 that the predominant strategy has
been either to ignore iconicity in signed languages or to explain it away. The
functional significance of iconicity in ASL has received little if any attention.
The only way for signed and spoken languages to be compared is by
positing structures and rules separate from their physical implementation.
Indeed, a major premise of those who embrace formalist linguistic theories,
such as Poizner, Klima and Bellugi (1987), is that "language is amodal, taking
a form which is independent of channel or of behavioral system" (Kimura
1988: 375). Thus, they conclude that "language, independent of its transmis­
sion mechanisms, emerges in a ... linguistically driven manner" (Poizner,
Klima, & Bellugi, 1987: 23).
One problem with this position, as noted by Kimura (1988), is that it
assumes that the organization of language is independent of its motor pro­
gramming. Another is that it ignores the deep historical and neurological
linkages between oral and manual gestures (cf. Kimura 1976, 1979), and
between human movement, cognition, and language (Edelman 1987, 1989).
Finally, it is difficult to understand how language under such a view —
whether spoken or signed — could have ever developed in the human species.
For example, commenting on the notion that language can be separated from
behavioral systems and transmission mechanisms, Studdert-Kennedy (1986:
101-102) remarks:
Characteristic motor systems have evolved for locomotion, predation, con­
sumption, mating. Matching perceptual systems have evolved to guide the
animal in these activities. The selection pressures shaping each species'
perceptuomotor capacities have come, in the first instance, from physical
The Gestural Expression of Modality in ASL 137

properties of the world. By contrast, these perceptuomotor capacities them­

selves must have played a crucial role in the form of a social species'
communication system. [...] Certainly, specialized neuroanatomical signal­
ing devices have often evolved, but they have typically done so by modify­
ing pre-existing structures just enough for them to perform their new
function without appreciable loss of their old. [...] Language has evolved
within the constraints of pre-existing perceptual and motor systems. We
surrender much of our power to understand that evolution, if we disregard
the properties of those systems.

The viability of approaches which take into consideration language's physical

substance, as constrained by its implementation in human perceptual and
motor systems, has been demonstrated for spoken languages (Browman &
Goldstein 1990; Bybee 1985, 1994; Fowler 1986; Givón 1984; Mowrey &
Pagliuca 1988). We believe such approaches also will prove fruitful in under­
standing signed languages, and ultimately will be a key link in unifying
theories of signed and spoken language on a substantive, i.e. non-formalist,
foundation.
It seems to us that language is not dualistically separated from its physi­
cal realization, but deeply rooted in its bodily basis.1 Neither is grammar
independent of meaning. Such a view is compatible with cognitive or func­
tional theories of language developed in works such as Bybee (1985), Deane
(1991, 1993), Givón (1989), Lakoff (1987), and Langacker (1987, 1991).
For example, Deane (1991) argues for a cognitive, nonmodular theory of
language based on an elaboration of George Lakoff s (1987: 283) Spatializa­
tion of Form Hypothesis. Deane suggests that several predictions regarding
the relation between grammar and cognition follow from the Spatialization of
Form Hypothesis (363-64):
(i) According to the hypothesis, the acquisition of grammatical compe­
tence occurs when linguistic information is routed to and processed by
spatial centers in the brain.
(ii) Specifically, it is claimed that linguistic expressions are processed as
if they were objects with internal structural configurations. That is,
they are processed in terms of certain basic image schemas, namely
part-whole and linkage schemas critical to the recognition of the
configurations which define complex physical objects.
(iii) But as Johnson (1987) argues at length, image schemas are basically
embodied schemas, high level schemas which function as cognitive
models of the body and its interaction with the environment.
 138 Sherman Wilcox and Phyllis Wilcox

In other words, the Spatialization of Form hypothesis treats grammar as a

form of image-schematic thought in which words, phrases, and sentences are
endowed with an abstract structure grounded in immediate bodily experience
of physical objects. It therefore predicts an association between grammar and
such cognitive abilities as object recognition, spatial structure, and body
awareness, especially modeling bodily movement and position in space.
Stokoe (1991), reacting to current formalist approaches to ASL phonol­
ogy, has proposed a view of signed language phonological structure which he
calls semantic phonology.
[Semantic phonology] invites one to look at a sign — i.e., a word of a
primary sign language — as a marriage of a noun and a verb. In semantic
terminology, appropriate here, the sign is an agent-verb construction. The
agent is so called because it is what acts (in signing as in generative
semantics), and the verb is what the agent does. [...] Semantic-phonologi­
cal, or s-p, verbs are, like common verbs, transitive or intransitive. For
example: when a signer of American Sign Language signifies 'yes', the sign
agent (i.e. the signer's active arm including the hand) flexes at the wrist; it is
intransitive, it has no object, it acts on no patient. But if the signer signifies
'stupid', the agent action continues until it strikes the forehead; the s-p verb
in this sign is thus transitive: it has an object (the grammatical term); or it
takes a patient (the semantic term).

Integrating the proposals of Deane and Stokoe, we suggest that human

arms and hands are — both phylogenetically and ontogenetically — proto­
typical and embodied complex objects. Semantic phonology describes the
"internal structural configurations" of these objects as they are used to pro­
duce a natural language — in this case, American Sign Language.
Thus, by using the phrase "gestural expression" in our title, we wish to
propose a theoretical position which makes the following assumptions:
(1) Physical gestures are the means by which all languages are realized
— the substance of both signed and spoken languages lies in their gestural
expression;
(2) Linguistic theory must acknowledge the physically embodied ground­
ing of linguistic competence — in a very real sense, the body is in the mind
(Johnson, 1987), the essence of language is bodily activity;
(3) When physical gestures are used to represent mental designata, there
is the likelihood that an iconic or isomorphic relationship will exist between
the two — not only at the lexical level, but also at the propositional and
pragmatic/discourse levels.
The Gestural Expression of Modality in ASL 139

In the next section, we will examine several modal verbs in ASL. In

section 3, we will tentatively explore the iconic relationship between ASL
modals and their gestural expression.

2. Modality in ASL2

The expression of mood and modality has not been extensively explored in
ASL. Long (1918) in his chapter on auxiliary verbs briefly describes the
production and semantics of the following modals: CAN; CANT; MAY (MAYBE,
PERHAPS); MAY; SHOULD (OUGHT); MUST (NEED, HAVE TO); HAVE (FINISHED); WILL
(SHALL); WON'T.3
Fischer and Gough (1978) briefly mention modals in their discussion of
ASL verbs, but do not deal with them semantically. Padden (1988) also briefly
mentions the modals CAN, WILL, SHOULD, and MUST, but likewise does not
discuss in any depth the semantics of these words, focussing instead on what
types of nominais may precede modals. Finally, the semantics of modals is
typically dealt with in only a cursory way in most ASL textbooks. Questions of
strength, of which modals can express deontic and which can express epis-
temic modality, and other subtle semantic and pragmatic factors are left to
students to discover on their own.

2.1. Root modality

Strong obligation in ASL is expressed with the lexical modal MUST (Figure
la). Weak obligation is expressed with the lexical modal SHOULD (Figure 2b).

(a) MUST (b) SHOULD

Figure 1: Modals of obligation
 140 Sherman Wilcox and Phyllis Wilcox

Long (1918) describes the meanings and production of several ASL words.
This text is a rich source of etymological and historical information about
ASL (cf. Frishberg, 1975). We will cite Long throughout this paper, primarily
as evidence for phonological change.
Long (1918: 26) offers the following descriptions of SHOULD and MUST:
Should, Ought, indicating duty.— Press the crooked forefinger of the right
"G" hand against the lips and then move toward side and downward as in
"must" (see below). Or,
(2) Indicate by the signs "better" and "must." Better is signed as follows:
Place the end of the right open hand pointing toward the left, palm against
the mouth; draw away toward the side, assuming the position of "A" and
lifting it to a level with the head. "Must," see below.
Must, Need, Have to. — Crook the forefinger of the right "G" hand,
pointing it downward, and press the hand down some distance with more or
less force. Sometimes the motion is repeated several times.

The forms described by Long for SHOULD are no longer used. While MUST
is essentially unchanged from this earlier form, SHOULD is now made with
reduplicated flexion and extension of the wrist. This phonological relation­
ship (single, punctual movement versus reduplicated, cyclic movement) will
reappear in other modals and will be explored in more detail in section 3.
According to Boyes-Braem (1981), the forms MUST and SHOULD are
related to the Old French Sign Language 4 form of NECESSARY (IL FAUT), in
which a G hand (extended index finger) pointed firmly down toward the
ground in front of the signer. This could also mean ICI (RIGHT-HERE) or THIS-
GROUND. Boyes-Braem argues that the more abstract concept 'necessary' is
assigned an abstract deictic handshape (the X, or bent index finger) in ASL.
Modals can occur in three positions: at the end of the verb phrase 5 as in
(1), preceding the verb as in (2), or preceding and at the end of the verb phrase
as in (3) (Humphries, Padden & O'Rourke, 1980).
t
(1) TELEPHONE NUMBER, WOMAN SHE-GIVE-ME SHOULD SHE
T h e woman should give me the telephone number.'
t
(2) TELEPHONE NUMBER, WOMAN SHOULD SHE-GIVE-ME SHE
T h e woman should give me the telephone number.'
(3) I MUST WIN RACE MUST I
'I must win the race.'
Necessity is indicated in ASL with the use of the word MUST. 6
The Gestural Expression of Modality in ASL 141

Figure 2: CAN

Figure 3: Old ASL CAN and STRONG

(4) I MUST BOOK I

'I need a book.'
Ability is indicated by the word CAN (Figure 2). CAN is historically related
to STRONG as demonstrated in Figure 3 (Higgins, 1923). Long's (1918: 25)
description of CAN is the same:
Can, expressing possibility, power, etc.— Hold the "S" hands out in front,
elbows against sides, and let the hands drop a little way with a jerk.

CAN can indicate either physical or mental ability:

t
(5) RIDE BICYCLE, I CAN I.
'I can ride a bicycle.'
q
(6) UNDERSTAND ME, CAN YOU HUH?
'Can you understand me?'
142 Sherman Wilcox and Phyllis Wilcox

Figure 4: CAN'T

CAN is also used to indicate root possibility in ASL:

q
(7) TOMORROW YOU CAN DRIVE YOU?
'Can you drive tomorrow?'
y
I CAN I.
'Sure, I can do that.'
Example (8) below shows MUST and CAN being used in these more
generalized root senses.
(8) CULTURE MUST HAVE FIVE . . .
'Culture must exhibit five [characteristics]'
cond
HAVE ALL CULTURE CAN CALL INDEX
'If something possesses all of these characteristics, then we can call
it culture.'
CAN is negated either with NOT and negative markers as described above,
or with the suppletive form CANT (Figure 4).

2.2. Epistemic modality

Epistemic possibility can be indicated in ASL with the lexical items POSSIBLE
(Figure 5a) and MAYBE (Figure 5b). Long (1918) related the word MAYBE
('may,' 'maybe,' or 'perhaps') to the physical act of comparing weights by
using the hands as a balance scale:
The Gestural Expression of Modality in ASL 143

(a) POSSIBLE (b) MAYBE

Figure 5: Modals of weak inference

May, Maybe, Perhaps, indicating probability.— Hold out both open hands
in front straight from the sides, palms up; balance the hands up and down
alternately several times like the balances of a pair of scales.

POSSIBLE is related to CAN; the phonological relationship parallels that of

MUST and SHOULD, MUST and CAN are articulated with strong, single move­
ments; POSSIBLE and SHOULD are articulated with weaker, reduplicated move­
ments. Example (9) demonstrates the use of POSSIBLE and MAYBE.
cond q
(9) HE STUDY "over & over again", PASS TEST WILL HE?
'If he studies really hard will he pass the test?'
POSSIBLE.
'It's possible.'
Example (9) could also be answered with MAYBE: 'He might pass the test.'
The modal MAYBE is negated by using NOT accompanied by the negative
marker, a set of facial expressions or Non-Manual Signals (NMSs) including
a negative (side-to-side) headshake and squeezing the eyebrows together; the
negative marker can also be used alone to negate MAYBE. Thus, the answer to
example (9) above could be either of the following:
n
(10) MAYBE NOT
'He might not.'
n
(11) NOT POSSIBLE
'It's really not possible.'
 144 Sherman Wilcox and Phyllis Wilcox

Figure 6: IMPOSSEBLE

In addition, there is the single lexical item IMPOSSIBLE (Figure 6). IMPOSSI­
BLE is a stronger negation than NOT POSSIBLE; it indicates that the proposition is
not only not very likely but absolutely impossible.
There is also a variant of CAN'T (Figure 4) that means 'impossible'. It is
signed with two small, repeated movements, IMPOSSIBLE (Figure 6) is much
more commonly seen.
Epistemic modalities in ASL are expressed with lexical items such as
7
FEEL, SEEM and OBVIOUS (Figure 7) , in other constructions such as questions,
and with particular NMSs. For example, in discussing a missing item in one's
apartment, the dialogue in example (12) could occur (from Cokely & Baker
1980).
gaze If q
(12) FEEL SOMEONE 1 -CL-rt 'person walks by' NOTICE-TO GO-INTO SWIPE YOU 8
'Do you suppose someone walked by, noticed it, and just went in
and stole it?'
n nodding
NOT-KNOW SEEM+
'I don't know, apparently that's what happened.'
FEEL in the epistemic sense is related to the word FEEL used for the
physical sense. Typically, the latter is produced with a single upward move­
ment, while the former uses shorter movements, often reduplicated, and a
distinct set of NMSs (this relationship is discussed in more detail in section 3).
SEEM is historically related to the words MIRROR and COMPARE, words
related to physical resemblance. In its old form SEEM retained some of its
"traditional" or lexical iconicity: "Bring the open right hand up in front
toward one side with the thumb edge toward self; turn the hand so as to
The Gestural Expression of Modality in ASL 145

(a) FEEL (b) SEEM (c) OBVIOUS

Figure 7: Modals of inference

present the palm toward self and fix the eyes upon if9 (Long, 1918: 63, our
italics). Compare this description of SEEM to Higgins' (1923: 92) almost
identical description of MIRROR: "vertical palm held at side of face and is
gazed into as if into a mirror." As we will see below, Frishberg (1975) claims
that this type of iconicity is eroding.
OBVIOUS is related to the ASL word BRIGHT ('bright' or 'light'). Com­
pared to BRIGHT it is made with less stressed, reduced movement. In its
epistemic sense it appears in constructions such as:
(13) HE RICH, OBVIOUS

It is worth noting that all of these modals are related to ASL words which
denote physical activity and perception:
(i) physical strength (CAN)
(ii) judging the physical weight of an object (MAYBE)
(iii) the physical sense of feeling (FEEL)
(iv) physical resemblance, as reflected in a mirror (SEEM)
(v) the bright light needed for clear visual perception (OBVIOUS)
This relationship has been noted by others (Sweetser 1990), and clearly
relies on a metaphorical mapping between embodied, physically-grounded
source domains and mental, abstract target domains. The widespread use of
conceptual and experiential metaphor in ASL grammar is explored more fully
in P. Wilcox (1993).
Finally, we should point out that the deontic modals MUST and SHOULD are
not used epistemically in ASL. Thus, the sentence:
146 Sherman Wilcox and Phyllis Wilcox

(14) *YOU MUST TIRED YOU

cannot mean 'You must be tired' in the epistemic sense. Likewise, the
sentence, "We should be finished by 5 PM" in ASL would not use the lexical
item SHOULD.
ASL epistemic sentences may also be produced without any of the lexical
items discussed above. One way in which this may be accomplished is with
tag questions:
_q
(15) YOU WORK HARD, YOU TIRED YOU, RIGHT?

In (15), the word RIGHT is accompanied by NMSs which indicate more

than a yes/no question. They carry the implication, "Judging by the fact that
worked so hard, you must be tired now — I am correct in making that
inference, am I not?"
A similar construction can use a generalized tag question marker:
(16) [Two people are looking at a person passing by wearing expensive
clothes.]
wh+y
HE RICH, HUH?
'He must be rich.'
Example (16) is accompanied by at least two distinct NMSs: (1) con­
tinual head nodding, and (2) a set of NMSs similar to those used as the WH
marker, which includes squinting the eyes and squeezing the eyebrows to­
gether.9 Such constructions are like those presented by Givón (1989: 147) in
his discussion of non-discreteness in assertions. He notes that "various tags
may be used to temper the strength of R-assertions. Such tags are often
augmented with intonation, as well as with various irrealis operators."
It may also be possible for epistemic modality to be indicated without any
lexical marker. In such instances, the NMSs used in conjunction with the
epistemics are enough to convey degree of confidence in an asserted proposi­
tion. For example, consider the following situation.
(17) [Two people enter a meeting in a basement room early in the
morning. It is cloudy and cold. At lunch, person A turns to person
B and says, "Do you think it's raining outside?" Person B an­
swers:]
The Gestural Expression of Modality in ASL 147

wh+y
RAIN
'Surely it's raining.'
NMSs (head nodding, eyes squinting, eyebrows squeezed together)
The NMSs that accompany (16) and (17) are quite similar to those that
accompany imperative sentences. For example, consider example (18) in
Figure 8. The imperative NMSs that accompany the verb GIVE include a sharp
head nod, a slight leaning forward of the torso, squinting the eyes, and
squeezing the eyebrow together (Humphries, Padden, & O'Rourke 1980). We

t
(18) TICKET, YOU-GIVE-ME!
'Give me the ticket!'
Figure 8: NMS's in imperatives

t
(19) TELEPHONE NUMBER, WOMAN SHE-GIVE-ME, SHOULD SHE.
The woman should give me the telephone number.'
Figure 9: NMS's in deontic modality
 148 Sherman Wilcox and Phyllis Wilcox

whq
(20) YOUR CAR, WHERE?
'Where is your car?'
Figure 10: NMS's in WH-questions

also see similar NMSs in the expression of deontic modalities such as example
(19) in Figure 9. Clearly related are the NMSs that accompany WH-questions
(see example (20) in Figure 10).
Thus, we find quite similar NMSs used in the expression of the following
functions in ASL:
(A) imperative; obligation; request for information; epistemic prob­
ability
In the first two (imperative and obligation), the NMSs must accompany
lexical material (in imperatives the focus of the NMSs is on the imperative
verb; in deontic modals the entire phrase). In WH-questions, the NMSs may
accompany a phrase containing a WH-word, as in (20), or a generalized
question word usually glossed as "WHAT" (it can mean 'what', 'where', 'who',
'huh' etc.) Given the proper context, NMSs alone (i.e. without a wh-word)
can signal a WH-question (21). As we have seen above, in the expression of
epistemic probability, these NMSs may be used with modals or alone.
(21) A: YESTERDAY ME BUY TICKET ME.
'Yesterday I bought tickets.'
B: TICKET?
'What tickets ... for what?'
We suggest that the relationships described in (A) are an expression of
interpersonal transitivity. This in turn is related to the grammatical expression
The Gestural Expression of Modality in ASL 149

Figure 11:WILL

of certainty, power, and politeness (Givón 1989; Brown & Levinson 1978).
Parallel to (A), the scale of interpersonal transitivity runs from (i) to (iv)
below:10
(i) I command you to do something: "Sit down!"
(ii) I obligate you to do something: "You must sit down."
(iii) I make a polite request (this indirect speech act is accomplished by
means of a request for information in form): "Why don't you sit
down?"
(iv) I infer something about you: "You must be tired."

2.3. Intention, prediction, and volition

The final modal we will discuss is WILL. Long (1918: 27) attributes only a
future sense to WILL:
Will and Shall, indicating future.— Hold the right open hand pointing
straight out and elevated to a level with the should, and push it straight
forward the length of the arm.
WILL can also be used in intentional (22), predictive (23), and volitional
senses11 (24):
(22) SEND-YOU LETTER, WILL ME.
'I will send you the letter.'
(23) HE RECOVER WILL HE.
'He will get well.'
(24) HE ACCEPT RESPONSIBILITY WILL HE.
'He will accept responsibility.'
 150 Sherman Wilcox and Phyllis Wilcox

Figure 12: Modern WON'T

Negation of WILL depends on the sense. The intentional and predictive

senses can be negated with the ASL negation markers described above, or
with expressions of speaker uncertainty or doubt. Thus, negative versions of
(23) are:
(25) HE RECOVER, NOT.
(26) HE RECOVER, DOUBT ME.
The negation of a volitional such as (24), however, requires WON'T:

(27) HE ACCEPT RESPONSIBILITY WON'T HE.

WON'T has undergone extensive phonological restructuring since it was de­
scribed by Long (1918: 27):
Won't.— Hold the right "S" hand out in front from the side, jerk the hand
back, doubling the elbow.
In 1918, WON'T was clearly related to CAN (see Figure 3). The upward
motion of WONT was the result of forearm flexion (the strong contraction of
the bicep muscles) — the opposite motion from that used in CAN (forearm
extension). Today, WONT is signed with a different handshape and in a
different location, beginning with the hand in upper neutral space in front of
the signer's face and ending with the hand over the shoulder (Figure 12).
The change in handshape (from "S" to "open A") has brought modern
WONT into a set of relations with other words denoting negation. Thus, the
open-A handshape of modern WONT parallels the open-A of NOT and DENY/
TURN-DOWN.12
The change in location has brought modern WON'T into a set of relations
with words which use the time line in ASL (Frishberg & Gough, 1973). The
The Gestural Expression of Modality in ASL 151

time line is a line along the horizontal plane near the signer's ear and cheek
used consistently for many signs indicating time other than the present, WILL
and TOMORROW move forward along the time line (see Figure 11); YESTERDAY,
BEFORE, LONG-AGO are examples of signs that move toward the signer's
back.13 This is not to say that WONT has acquired a temporal sense. It has
resulted, however, in WILL and WONT taking part in a nicely iconic relation­
ship: volitional WILL moves forward (as does future WILL) while WONT jerks
backward.
Finally, it should be noted that the gestural component of old WONT and
modern WONT are the same, even though the change in location has resulted in
a change from upward movement to backward movement: both are produced
with forearm flexion (bicep contraction).14

3. Iconicity in ASL

Judging from the literature on ASL, the existence of iconicity in this language
has been viewed as something of an embarrassment to ASL linguists. This is
not hard to understand, given that from Saussure to Chomsky, formal linguis­
tics has maintained that to the extent that a communication system exhibits
iconicity, it is less a human language. More recently, Givón (1984, 1989) and
Haiman (1984) argue against this position for spoken languages, suggesting
instead that linguistic form is often motivated by iconic principles. For exam­
ple, Haiman (1984: 1) writes, "linguistic forms are frequently the way they are
because, like diagrams, they resemble the conceptual structures they are used
to convey."
It is difficult to deny the existence of iconicity in signed languages. Few
sign language linguists would, however, ascribe the deeply significant role to
iconicity that functional and cognitive linguists have in spoken languages.
Rather, many ASL linguists have taken a different approach, the intent of
which appears to be to explain away iconicity in ASL. They claim that
although iconicity may be present in ASL, it is eroding over time; it plays no
role in the acquisition, processing, or storage of language; and grammatical
processes routinely act to override and submerge it.
Frishberg (1975) argues that historical change acts to diminish iconicity
and heighten the arbitrary nature of ASL. We would point out, however, that
Frishberg's study deals primarily with iconicity at the lexical level. Meier
 152 Sherman Wilcox and Phyllis Wilcox

(1981; 1991) and Klima & Bellugi (1979) argue that iconicity plays an
insignificant role in the acquisition of ASL by deaf children. In Meier's study,
two models of iconicity are proposed. One assumes that deaf children are
attuned to verbs that happen to be enactments, or mimes, of an action, the
other that deaf children are attuned to verbs that map the spatial relations of
the actors. It turns out that deaf children follow neither model; deaf children in
Meier's study showed no tendency to use iconic verb forms earlier than
arbitrary verb forms.
It is interesting to note that Meier's second iconic model was tested using
the so-called directional, or person-agreement forms. In ASL, person-agree­
ment in these verb forms is indicated by incorporating spatial loci (which have
typically been assigned to a previously mentioned nominal) into the verb.
Other verb inflections in ASL, such as aspect, involve changes in the tempo­
ral/kinematic structure of the verb's movement. Thus, even if we accept
Meier's conclusions about the influence of iconicity vs. arbitrariness in lan­
guage acquisition, it is worth noting that these verb forms do, nevertheless,
incorporate iconicity: "Prototypically and phylogenetically/ontologically,
topic/participant deixis tends to be spatial, while propositional deixis tends to
be temporal" (Givón, 1984:270). In ASL, this type of deixis is coded by
indexing spatial locations. Coding of information about the temporal structure
of a verb's meaning, however, is accomplished by modifying the verb's
movement (which necessarily involves the sign's temporal structure). There
is no evidence that ASL is becoming more arbitrary in this regard — that is,
ASL is not beginning to code topic/participant deixis in the kinematics of the
verb's movement and aspectual information as spatial locations.
Klima & Bellugi (1979) further claim that iconicity is not relevant in
encoding and remembering processes. Such a claim is at odds with Givón's
(1989: 97) iconic imperative:
All other things being equal, a coded experience is easier to store, retrieve
and communicate if the code is maximally isomorphic to the experience.
Finally, Klima & Bellugi (1979) argue that the grammatical processes of
ASL function to suppress whatever iconicity may exist: "One of the most
striking effects of regular morphological operations on signs is the distortion
of their forms so that iconic aspects of the signs are overridden and sub­
merged" (1979: 30). As an example, they describe the morphological process
which may be added to a signed word to indicate intensity. This process
consists of the addition of an initial hold on the sign followed by a sudden
 The Gestural Expression of Modality in ASL 153

(a)IDEA (b)SUPPOSE
Figure 13: IDEA and SUPPOSE

release. 15 When applied to the sign SLOW, the resulting sign means 'very
slow.' Klima and Bellugi's point is that the sign VERY-SLOW is made with a
very fast movement: "Thus the form of 'very slow' is incongruent with the
meaning of the basic sign" (30). The grammar of ASL seems to have sub­
merged the iconicity of VERY-SLOW.
But if this is true, it is only because iconicity has re-emerged in the new
form, VERY-SLOW has not overridden the iconicity, it has merely shifted it to
the morpheme meaning 'very.' 'Very' can be understood metaphorically in
terms of the build up of pressure (intensity). As pressure increases so does
intensity, until it is suddenly released. In the ASL morpheme 'very' this build
up of intensity is iconically signalled by a build up muscular or gestural
energy. Finally, the metaphoric pressure is released — as is the gestural energy
in the sudden release into a short, rapid movement. It is an iconic process, and
one which, we would suggest, is not unlike a similar process at work in spoken
language when, for example, teenagers say, "That's bad." Typically, 'bad'
will be pronounced with an extended hold on the bilabial, followed by a
sudden release. Of course, the meaning is the same as the ASL example —
'very' (how 'very bad' came to mean 'very good' is another story).

3.1. Iconicity in the expression of modality in ASL

A number of facts about the expression of modality lend themselves to a

discussion of iconicity in ASL. First, we have already noted the relation
between pairs of modals: MUST and SHOULD (Figure 1), CAN and POSSIBLE
(Figure 2 and Figure 5 a). Other words discussed above also take part in this
relationship: FEEL, SEEM, and MAYBE. The relationship can be described in
terms of stress and reduplication. 16 MUST is stressed and unreduplicated;
 154 Sherman Wilcox and Phyllis Wilcox

Table 1. Transitivity Parameters

HIGH LOW
Participants 2 or more participants, A and O a 1 participant
Kinesis action non-action
Aspect telic non-telic
Punctuality punctual non-punctual
Agency A high in potency A low in potency
Affectedness of O O totally affected O not affected
Individuation of O O highly individuated O non-individuated
a. 'A' = Agent and 'O' = Object

SHOULD is unstressed and reduplicated. The same holds true for CAN and
POSSIBLE.
For FEEL, the relevant feature for the present discussion is reduplication.
When FEEL is produced using a single, upward movement it refers to the
physical sense of feeling. When it is produced with multiple, reduplicated
movements (and the appropriate NMSs discussed above) it acquires the
epistemic meaning described above.
The same phonological relationships are seen between the forms for
BRIGHT/OBVIOUS described above, and for IDEA and SUPPOSE (Figure 13).
SUPPOSE can be used to express hypotheticals ("Suppose you and I drive to
California?") and to mark conditionals:17
(28) SUPPOSE TOMORROW SNOW, I GO-AWAY SKIING.
'If it snows tomorrow, I'll go skiing.'
In order to describe how these forms are iconic, we must return to
Stokoe's semantic phonology and propose the notion of phonological transi­
tivity. By phonological transitivity, we mean simply that ASL signs in their
production can be described as phonologically transitive (an agent, a hand,
performs a punctual action on and significantly changes a passive patient, the
other hand) or phonologically intransitive (a single patient acts, non-punctu-
ally). We are intentionally using semantic terms for phonological, or forma-
tional, characteristics of a sign, to bring out the close relation between
meaning and form — the iconic relation.
To further elaborate the concept of phonological transitivity we can
invoke several of the parameters proposed by Hopper and Thompson (1980:
252) for understanding the notion of semantic transitivity. The parameters
listed in Table 1 seem especially pertinent to phonological transitivity.
 The Gestural Expression of Modality in ASL 155

Many of the modals discussed in this paper have only one participant
(even when two hands are forming the words, it is not the case that one is
acting on the other). Reduplication affects a sign's formational characteristics
on the kinesis, aspect, and punctuality parameters: reduplicated signs are less
active, nontelic, and have no obvious transitional phase between inception
and completion (i.e., they are cyclic as opposed to punctual). 18 Finally, note
that phonological transitivity is an expression of phonetic or gestural sub­
stance (this will be elaborated further below).
The iconicity principle at work in these modals can now be summarized
as follows:
(A) Senses which are more physically or interpersonally grounded, or
semantically less abstract, are expressed with more gestural sub­
stance; the more epistemic or semantically abstract senses are
expressed with reduced gestural substance.
Ferreira Brito (1990: 255) comments on the same relationship between
deontic and epistemic modals in Brazilian Cities Sign Language (BCSL).
Deontic modals are expressed with "simple and energetic movements" which,
she suggests, is related to an underlying metaphor of obligation for something
to be done as actions or events in the real world. Epistemics are distinguished
by nonenergetic movement of the hands.
SEEM is more complex. Consider the following: In a discussion of which
of two possible translations would be more appropriate for a particular ASL
utterance, the following conversation took place:
(29) Speaker A: 'Which do you think she said, this [indexi] or that
[indexj]?'
Speaker B: SEEM++ INDEXI
'She probably said this ...'
(The two study the videotape of the utterance under discussion
some more.)
Speaker B: SEEM INDEXJ!
'Yes, this has to be what she said.'
In example (29), the first occurrence of SEEM is slowly reduplicated; the
second is signed once with emphatic stress. The first expresses weak epis­
temic probability, the second stronger epistemic certainty. The iconicity
principle in operation here seems to be:
 156 Sherman Wilcox and Phyllis Wilcox

(a) OWE (b) DEMAND/ (c) MUST (d) SHOULD

REQUIRE
Figure 14

(B) In epistemics, possibility is expressed with reduced gestural sub­

stance; probability or certainty is expressed with strong gestural
substance. 19
The iconicity principle (B) appears to be confirmed by the fact that the
same relationship holds between other related forms in ASL. Thus, MAYBE
may be signed with slow, unstressed, long movements or with stressed, faster,
shorter movements. The former means 'maybe, but barely possible;' the latter
means 'maybe, probably true'.
This relationship extends to MUST/SHOULD, CANT/IMPOSSIBLE, and IDEA/
SUPPOSE, and thus, across this limited set of words, we can generalize (B) to
the following:
(C) Words with more semantic substance, or less lexical generality (cf.
Bybee, 1985: 16), are expressed with more gestural substance;
words with less semantic substance and more lexical generality are
expressed with less gestural substance.
Finally, the notions of semantic phonology and phonological transitivity
expressed in (C) can be used to explore the relationships among four semanti-
cally-related ASL words: OWE, DEMAND/REQUIRE, MUST, and SHOULD (Figure
14). 20
The phonological transitivity of these words is summarized in Table 2.
Note that as the words become semantically less substantive (i.e., OWE >
DEMAND > MUST > SHOULD) they are expressed with less gestural substance.
This process of parallel reduction in the semantic and phonological substance
of words has been reported to extensively occur across the world's spoken
languages (Bybee, Perkins, & Pagliuca, 1994).
The Gestural Expression of Modality in ASL 157

Table 2. Relation of gestural and semantic substance

Parameter OWE DEMAND MUST SHOULD
Participants + + - -
Kinesis + + + -
Aspect + + + -
Punctuality + + + -
Agency + - - -
Affectedness of O + - - -
6 4 3 0

Of course, many other phonological, morphological, historical, and

processing factors have an impact on the changing forms of ASL words.
These iconic principles cannot be expected to operate in a totally predictive
way. As Givón (1989: 90) notes, however, the lack of total iconic systematicity
in no way implies a total lack of such systematicity.
Nor should the extensive presence of iconicity be considered in conflict
with conventionalization. Clearly there is a range of iconicity displayed in
ASL, from the lexical iconicity explored by Frishberg (1974) to the grammati­
cal iconicity discussed above. Conventionality is at work across this range.

4. Conclusion

Much ASL linguistic research has found itself in an unusual, symbiotic

relationship with formalist approaches to linguistics. For many years ASL
was considered to be merely pantomime, and it was quite common to "ex­
plain" ASL words on the basis of speculation about their purely nonlinguistic
origins. This naturally led ASL researchers to distrust any attempt at studying
the historical process of how lexemes come into existence in ASL or how
lexical material is grammaticized. Also, for many years ASL was haunted by
the "spectre of iconicity" (McDonald 1982: vi). Studies which examined
iconicity in ASL were seen as calling into question its status as a "true"
language as opposed to primitive or pre-human communication, and thus
were viewed with suspicion.
It is little wonder, then, that many ASL linguists have been drawn to
formalist theories, which by their nature are ahistorical. The symbiosis is
further fostered by the fact that formalist approaches assume that language
structure can be understood apart from the bodily-based sources of linguistic
structure such as gesture and object manipulation, vision, force dynamics,
metaphor, and so forth.
 158 Sherman Wilcox and Phyllis Wilcox

In turn, ASL research has done its part in furthering this symbiosis. If the
formalist approach merely assumes that it is possible to view some abstract,
idealized entity as the true object of linguistic analysis for spoken languages,
signed languages seem positively to demand this stance. How else, the reason­
ing would suggest, could a unified linguistic theory be developed across such
radically different substances as light and sound? Having made such an
assumption, the conclusion from ASL research that language emerges in a
linguistically driven manner (Poizner, Klima and Bellugi 1987: 23) is taken as
support, and the symbiotic cycle is completed.
Because of these conditions, and because ASL is an unwritten language
which was not seriously described and studied until the late 1950s (Stokoe
1960), there have been few studies of grammatization in this language. We
believe we have shown here, however, that when such studies are attempted,
they demonstrate that the same processes at work in other languages
(Fleischman 1982; Heine et al. 1991; Sweetser 1990) — semantic extension
from concrete to abstract senses, reemergence of iconicity between cognitive
and linguistic patterning, the use of physical world vocabulary to talk about
the mental world — are operating in ASL.

NOTES

1 This view is more fully described in Armstrong, Stokoe and Wilcox (in press).
2 We want to gratefully acknowledge the assistance of our consultants, whose expert
intuitions were invaluable: Dennis Cokely, Antoinette Eidson, Betsy McDonald, Jenny
Singleton, and Anna Witter-Merithew. Figures are reproduced from Baker and Cokely
(1980) and Humphries et al. (1980).
3 ASL words will be glossed in English with small capital letters. The plus symbol (+) is
used to indicate reduplication.
4 It is well-attested that ASL is historically related to French Sign Language.
5 Humphries, Padden, & O'Rourke (1980) state that the modal occurs at the end of the
sentence, but their example #1 (p. 98, given above as example (1)) suggests that our
characterization is more accurate. These examples and many others in this paper are
from the exercises in Humphries et al.
6 Actually, there are four related words: MUST, NEED, HAVE-TO, and SHOULD, distinguished
by changes in stress and repetition. They are commonly grouped by repetition (cf.
Humphries et al.): MUST/HAVE-TO (+stress, -repetition/-stress, -repetition) and NEED/
SHOULD (+stress, +repetition/-stress,+repetition). See Section 3 for further discussion.
7 Figure 7c is actually the word BRIGHT; OBVIOUS is made with less stressed, reduced
movement
The Gestural Expression of Modality in ASL 159

8 In example (12), FEEL SOMEONE ... is signed with the dominant hand, YOU is signed
simultaneously with the non-dominant hand.
9 Of course it must be noted that this does not signal a wh-question.Still, it is interesting
that Givón (1989: 156) notes that "the use of WH forms to code epistemic uncertainty is
wide-spread."
10 The use of these expressions to convey the same speech act is not universal; thus, ASL
cannot use the request of information (#3) to indirectly convey a command.
11 Example (24) can have both predictive and volitional readings
12 There is another word using the " S " handshape which includes the notion of denial or
negation: REBEL (Long (1918: 46) glosses this as DISOBEY).
13 Although TOMORROW and YESTERDAY also use the open-A handshape, they are not related
to the negative signs WONT, NOT, and DENY. Rather, this handshape is a remnant of
French Sign Language (FSL) forms. In FSL, counting starts with the extended thumb as
' 1 ' . In ASL, ' 1 ' is represented with the extended index finger. Thus, TOMORROW and
YESTERDAY appear to be complex forms representing 'one day in the future' and 'one day
in the past' respectively. This also reflects the different ways Americans and Europeans
count on their fingers.
14 Mowrey & Pagliuca (1988) discuss the same predicament for autosegmental versus
gestural phonologies in terms of the features [coronal] and [anterior]; Bybee (1992) for
'plural' and [coronal] in her discussion of formal/structural versus functional/substan­
tive phonology.
15 Klima & Bellugi (1979:30) describe only the change in movement: "an extremely short,
rapid movement." Also note that by characterizing the process in terms of addition of an
initial hold, we are not implying an analysis based on the Movement-Hold phonology of
Liddell and Johnson (1989).
16 Wilbur & Schick (1987) describe several features associated with stress, including non-
manual behaviors, sharp sign boundaries, higher placement in the signing space, faster
movement, and repetition.
17 Conditionals need not be marked with SUPPOSE; a NMS (raised eyebrows) is enough to
signal the protasis. It is interesting that this NMS is also used to mark topics (cf. Haiman
1978).
18 It may seem that reduplicated signs are more, not less active. Note, however, that as a
sign such as MUST is repeated, especially if it is also unstressed, the result is to change it
from a punctual event to a cyclic state.
19 Although we leave the question unexplored at present, we believe that this is an instance
of the iconic expression of force-dynamics in ASL modals (cf. Talmy 1985).
20 Bybee, Perkins, & Pagliuca (1994) note that a common source for obligation are forms
meaning 'owe'. The relationship between the forms in Figure 14 would suggest that they
too are historically related. We have no evidence that they are (and indeed Boyes-Braem
suggests a different lexical source for MUST, as we noted above). Note also that OWE is
used almost exclusively with a financial connotation.
 160 Sherman Wilcox and Phyllis Wilcox

REFERENCES

Armstrong, D.F., W.C. Stokoe & S.E. Wilcox. 1995. Gesture and the Nature of Language.
Cambridge: Cambridge University Press.
Browman, C. & L. Goldstein. 1990. 'Tiers in Articulatory Phonology, with Some Impli­
cations for Casual Speech." In J. Kingston & M.E. Beckman (Eds.), Papers in
Laboratory Phonology I: Between the Grammar and Physics of Speech. Cambridge:
Cambridge University Press.
Brown, P. & S. Levinson. 1978. "Universals in language usage: Politeness Phenomena."
In E.N. Goody (Ed.), Questions and Politeness: Strategies in Social Interaction.
Cambridge: Cambridge University Press.
Bybee, J.L. 1985. Morphology: A Study of the Relation Between Meaning and Form.
Amsterdam: John Benjamins.
Bybee, J. 1994. "A View of Phonology from a Cognitive and Functional Perspective."
Cognitive Linguistics, 5-4, 285-305.
Bybee, J., Perkins, W.D. & W. Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect
and Modality in the Languages of the World. Chicago: University of Chicago Press.
Boyes-Braem, P. 1981. Features of the Handshape in American Sign Language. Unpub­
lished doctoral dissertation, University of California, Berkeley.
Cokely, D. & C. Baker. 1980. American Sign Language: A Student Text (units 1-9). Silver
Spring, MD: TJ Publishers.
Deane, P.D. 1991. "Syntax and the Brain: Neurological Evidence for the Spatialization of
form Hypothesis." Cognitive Linguistics, 2(4), 361-367.
Deane, P.D. 1993. Grammar in Mind and Brain: Explorations in Cognitive Syntax.
Berlin/New York: Mouton de Gruyter.
Edelman, G.M. 1987. Neural Darwinism: The Theory of Neuronal Group Selection. NY:
Basic Books.
Edelman, G.M. 1989. The Remembered Present: A Biological Theory of Consciousness.
NY: Basic Books.
Ferreira Brito, L. 1990. "Epistemic, Alethic, and Deontic Modalities in a Brazilian Sign
Language." In S.D. Fischer & P. Siple (Eds.), Theoretical Issues in Sign Language
research. Chicago: University of Chicago Press.
Fischer, S.D. & B. Gough. 1978. Verbs in American Sign Language. Sign Language
Studies, 18, 17-48.
Fleischman, S. 1982. The Future in Thought and Language: Diachronic Evidence from
Romance. Cambridge: Cambridge University Press.
Fowler, C.A. 1986. "An Event Approach to the Study of Speech Perception from a Direct-
realist Perspective." Journal of Phonetics, 14, 3-28.
Frishberg, N. 1975. "Arbitrariness and Iconicity: Historical Change in American Sign
Language." Language, 51, 676-710.
Frishberg, N. & B. Gough. 1973. Morphology in American Sign Language. Manuscript,
The Salk Institute for Biological Studies, La Jolla, Calif.
Givón, T. 1984. Syntax: A Functional-Typological Introduction Volume I. Amsterdam:
John Benjamins.
Givón, T. 1989. Mind, Code and Context: Essays in Pragmatics. Hillsdale, NJ: Lawrence
Erlbaum.
 The Gestural Expression of Modality in ASL 161

Goody, E. 1978. Questions and Politeness: Strategies in Social Interaction. Cambridge:

Cambridge University Press.
Haiman, J. 1978. "Conditionals are Topics." Language, 54, 564-89.
Haiman, J. 1984. Iconicity in Syntax. Amsterdam: John Benjamins.
Heine, B., Claudi, U., & F. Hiinnemeyer. 1991. Grammaticalization: A Conceptual
Framework. Chicago: University of Chicago Press.
Higgins, D.D. 1923. How to Talk to the Deaf St. Louis, MO.
Hopper, P.J. & S.A. Thompson. 1980 Transitivity in grammar and discourse. Language,
56(2), 251-299.
Humphries, T., Padden, C. & T.J. O'Rourke. 1980. A Basic Course in American Sign
Language. Silver Spring, MD: TJ Publishers.
Jespersen, O. 1924. The Philosophy of Grammar. NY: Norton [1965 edition].
Johnson, M. 1987. The Body in the Mind: The Bodily Basis of Meaning, Imagination and
Reason. Chicago: University of Chicago Press.
Kimura, D. 1976. "The Neural Basis of Language Qua Gesture." In H.Whitaker & H.
Whitaker (eds.), Studies in Neurolinguistics. NY: Academic Press.
Kimura, D. 1979. "Neuromotor Mechanisms in the Evolution of Human Communication."
In H.D. Steklis & M.J. Raleigh (eds.), Neurobiology of Social Communication in
Primates. NY: Academic Press.
Kimura, D. 1988. Book review of What the Hands Reveal about the Brain, by H. Poizner,
E.S. Klima, and U. Bellugi, Cambridge: MIT Press, 1987. Language and Speech,
31(4), 375-378.
Klima, E.S. & U. Bellugi. 1979. The Signs of Language. Cambridge, MA: Harvard
University Press.
Lakoff, G. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the
Mind. Chicago: University of Chicago Press.
Langacker, R. 1987. Foundations of Cognitive Grammar Volume I: Theoretical prerequi­
sites. Stanford: Stanford University Press.
Langacker, R. 1991. Foundations of Cognitive Grammar Volume II. Stanford: Stanford
University Press.
Liddell, S.K. & R.E. Johnson. 1989. "American Sign Language: The Phonological Base."
Sign Language Studies, 64, 195-277.
Long, J.S. 1918. The sign language: A manual of signs. Washington, DC: Gallaudet
College Press.
McDonald, B. H. 1982. Aspects of the American Sign Language Predicate System.
Unpublished doctoral dissertation, University of Buffalo.
Meier, R.P. 1981. "Icons and morphemes: Models of the acquisition of verb agreement in
ASL." Papers and Reports on Child Language Development, 20, 92-99.
Meier, R.P. 1991. Language acquisition by deaf children. American Scientist, 79, 60-70.
Mowrey, R. & W. Pagliuca. 1988. The Reductive Character of Phonetic Evolution. MS.
Padden, C.A. 1988. Interaction of Morphology and Syntax in American Sign Language.
NY: Garland Publishing.
Poizner, H., Klima, E.S., and U. Bellugi. 1987. What the Hands Reveal about the Brain.
Cambridge: MIT Press.
Stokoe, W.C. 1960. "Sign language structure: An Outline of the Visual Communication
Systems of the American Deaf." Studies in Linguistics, Occasional Papers: 8. Buffalo,
NY.
162 Sherman Wilcox and Phyllis Wilcox

Stokoe, W.C. 1991. "Semantic Phonology." Sign Language Studies, 71, 107-114.
Studdert-Kennedy, M. 1986. "Two Cheers for Direct Realism." Journal of Phonetics, 14,
99-104.
Sweetser, E. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects of
Semantic Structure. Cambridge: Cambridge University Press.
Talmy, L. 1985. "Force Dynamics in Language and Thought." In Papers from the Parases-
sion on Causitives and Agentivity, 1, 293-337. Chicago: Chicago Linguistic Society.
Wilbur, R.B. & B.S. Schick. 1987. "The Effects of Linguistic Stress on ASL Signs."
Language and Speech, 30(4), 301-323.
Wilcox, P. 1993. Metaphorical Mapping in American Sign Language. Unpublished dis­
sertation, University of New Mexico.
Wilcox, S. 1990. "The Structure of Signed and Spoken Languages." Sign Language
Studies, 67, 141-151.
 II
The Interaction Basis of Modality
 The Development of Epistemic
Sentence-ending Modal
Forms and Functions in Korean Children

Soonja Choi
San Diego State University

Modality is the linguistic domain that covers the speaker's attitudes and
opinions about a proposition (Palmer 1986), and it can be divided into at least
two semantic types: agent-oriented modality and epistemic modality (Bybee
1985). Agent-oriented modality, which includes deontic modality, expresses
various conditions on the agent, such as ability and obligation, with regard to
a proposition. For example, in the sentence 'John can (or must) lift this stone',
the auxiliary verbs, can and must, express the ability and obligation respec­
tively of the agent toward the predicate. Agent-oriented modality also in­
cludes desire and intention of the agent as well as permission. On the other
hand, the central function of epistemic modality is to express the degree of the
speaker's commitment to the truth of a proposition. For example, in 'John
should (or may) be home by 5 o'clock', the speaker expresses certainty (or
possibility) about the truth of a proposition. Palmer (1986) and others (Bybee
1985, Chafe & Nichols 1986) include evidentiality as a type of epistemic
modality, one which specifies the source of information, e.g., hearsay, direct
experience, inference. By specifying how the information is obtained, the
speaker conveys the kind of warrant he/she has about the proposition, and
therefore, the degree to which the proposition is likely to be true. Thus,
broadly defined, epistemic modality refers to 'the status of the speaker's
understanding or knowledge' of a proposition (Palmer 1986:51).
The semantic distinction between agent-oriented and epistemic modality
often parallels a morphological distinction. In an extensive cross-linguistic
study of morphology, Bybee (1985) calls attention to a strong tendency for
epistemic modality to be expressed inflectionally in bound forms (i.e. morpho-
phonologically fused with the verb stem), whereas agent-oriented modality is
 166 Soonja Choi

frequently expressed in non-bound lexical forms, e.g., auxiliary verbs. Bybee

(1985:169) gives a historical explanation for this phenomenon: as the mean­
ing of a modal form becomes more abstract over time, from agent-oriented to
epistemic, the form itself often reduces phonologically, i.e. from a free lexical
form to a bound inflectional form.
In the area of language acquisition, which is the major focus of this paper,
children's modal expressions have been reported to develop also from agent-
oriented to epistemic modality: Studies of the acquisition of modality in
English (Wells 1979, 1985) and other languages (Stephany 1986; Aksu-Koç
1988) have suggested that agent-oriented modality is universally acquired
earlier than epistemic modality. Specifically, Stephany (1986) reports that the
agent-oriented meaning of desire, e.g., 'I want to/wanna', is the earliest
indirect request to develop for English-speaking children, and 'can/can't' also
develops early to express ability and permission. In contrast, epistemic mean­
ings related to degrees of certainty about propositions (e.g., possibility, prob­
ability) are developed much later in children's language. In her survey,
Stephany found that the development of epistemic modality occurs in large
part between 3 and 5 years.
Stephany discusses an interesting difference between acquisition in
Greek and in English. She notes that her Greek subjects make morphological
distinctions of moods — Imperative, Indicative, and Subjunctive — using
different inflections at around 1 ;9 (one year; nine months). On the other hand,
in English no such distinctions are apparent at this early stage, partly because
the English-speaking children delete the subject in both imperative and de­
clarative sentence and because English does not have modal inflections.
Stephany suggests that inflectional languages, such as Greek, may facilitate
the acquisition of modal forms because inflections are 'a part of tightly knit
lexical forms' (Stephany 1986:398).
In this study of the Korean modal system and its acquisition by children,
epistemic modality will be the focus. Although there are only a few studies on
Korean acquisition, those studies reported an early emergence of an obliga­
tory class of verbal inflections called sentence-ending suffixes (Zoh 1981;
Choi 1991; Kim 1992). These inflections, which are used in informal interac­
tions, include a number of modal forms that relate particularly to the status of
the speaker's knowledge and evidentiality. In Zoh's study (1981) of the
morphological development of five Korean children, he found that sentence-
ending (hereafter SE) suffixes are the first productive morphological category
The Development of Epistemic Sentence-ending Modal 167

to be acquired by Korean children. Both Kim (1992) and Choi (1991) have
shown that productive uses of SE suffixes begin at around 1;9, i.e., before 2
years.
In an earlier study (Choi 1991), I analyzed the semantic functions of the
first three suffixes, -ta, -e, and -ci, acquired by three Korean children. The
data showed that, between 1;8 and 2;2, the children acquired the suffixes in a
consistent order, and with clear semantic distinctions that relate to old and
newly-acquired knowledge (expressed by -e and -ta respectively) and to
certainty of information (-ci). As a first step toward understanding such an
early acquisition of modality, I examined the frequency of the suffixes in
caregivers' speech to the children. The result showed that the order of acquisi­
tion correlated only partially with the input frequency. The study, therefore,
suggested that the acquisition of the modal suffixes in Korean is the result of
an interaction between the child's cognitive development and the language-
specific input.
The present study investigates what roles caregiver-child interactions
and the child's cognitive capacity play in the acquisition of Korean modal
suffixes. The study analyzes discourse interactions between the caregiver and
child, paying attention to both linguistic and extralinguistic contexts in which
the child produces a given suffix. The analysis shows that a given suffix is
consistently used by all three children in specific types of discourse-pragmatic
context. The present study attempts to characterize these types in a systematic
way. This discourse approach to data not only enables us to discover the
modal meanings acquired by the children, but also gives us insight into how
caregiver-child interactions contribute to the child's development of modal
meanings. In fact, it will be argued that the discourse-interactional component
embedded in the modal meaning enhances acquisition, and that this explains
the difference between the early acquisition of modal suffixes in Korean and
the relatively late acquisition of epistemic modality reported for other lan­
guages. To investigate the cognitive aspects of the acquisition, the kinds of
verbs the children use with each suffix will be analyzed, as children's lexical
development often gives us insight into their cognitive development (Gopnik
& Meltzoff 1984; Tomasello 1992). The study will show that both the child's
cognitive understanding of the world and the discourse-interactional aspects
of contexts play an important role in children's early acquisition of modal
suffixes. The analysis of the present paper includes five modal suffixes
acquired between 1;8 and 3;0 by the three children.
 168 Soonja Choi

The organization of the paper is as follows: First, the kinds of evidential

meanings and status of knowledge discussed in recent studies of epistemic
modality will be presented. This will be followed by an overview of the
Korean modal system. In the remaining part of the paper, the acquisition of
modal suffixes by three Korean children will be analyzed.

2. Linguistic coding of evidentiality and knowledge status

A number of studies on epistemic modality have shown that in many lan­

guages there are systematic morphological distinctions corresponding to what
Palmer (1986:51) describes as "the kind of warrant [the speaker] has for what
he says". One type of distinction reported in several studies is the degree to
which knowledge has been integrated in the speaker's mind. In particular,
epistemic modals in some languages express the difference between what the
speaker already knows (e.g., from past experience) and information that the
speaker has just acquired from the present context. DeLancey (1986) and
Akatsuka (1985) note that this type of epistemic modality distinguishes
between old and new knowledge. Specifically, DeLancey (1986) shows that
in Lhasa Tibetan, two suffixes contrast in this way 'dug is used to express
newly perceived information, not yet an "integrated part of the speaker's
knowledge", while yod is used when the speaker expresses old information
which has become part of his or her knowledge system. For example, when
the speaker unexpectedly finds a strange cat wandering around, 'dug is used
as in rja'i k'a nla si-mi 'dug There's a cat in my house'. However, if the
speaker owns a cat, and seeing the cat is expected and is part of already
established knowledge, yod is used instead (i.e. rja'i k'a nla si-mi yod There
is a cat in my house') (DeLancey 1986:212).
Expressions of the source of information, i.e. evidential markers, are
closely related to the distinction between old and new knowledge (Chafe &
Nichols 1986). In Turkish, for example, there are two evidential markers
distinguishing between witnessed and non-witnessed events: -di is used to
report past events that have been directly experienced by the speaker (Ahmet
gel-di 'Ahmed came'), whereas -mts is used to express inference or hearsay
(Ahmet gel-mis 'he/she/it came (apparently, reportedly) (Aksu-Koç & Slobin
1986). What is interesting is that the functions of the two markers are ex-
The Development of Epistemic Sentence-ending Modal 169

tended to the expression of old and new knowledge: -dt is used for informa­
tion that is expected from some background knowledge (i.e. old knowledge),
but -mts is used for an unexpected event (i.e. new knowledge). Although
languages are not identical in terms of how different degrees of integration of
knowledge are semantically categorized and how they relate to specific types
of evidentiality, these studies and others suggest that the old and new knowl­
edge distinction is an important basis for the evidential/epistemic system in
many languages, particularly Asian languages, e.g., Sherpa (Woodbury
1986), Chinese pidgin Russian (Nichols 1986), Korean (H. Lee 1991).
The meanings related to the status of knowledge in this way are often
interrelated with discourse pragmatics. Palmer (1986) notes that, in several
languages, expressions of status of knowledge depend on what the speaker
thinks the listener knows. For example, in Kogi, ni- is used when the speaker
reminds the listener of information that the latter presumably knows already
(e.g., ni-gu-ku-á T did it just a while ago, as you know'), but na- is used when
the speaker informs the listener of an ostensibly new proposition, (e.g., na-gu-
ngú 'I tell you he did it some time ago') (Palmer 1986: 76). Palmer goes on to
note that, in these cases, it is hard to distinguish between modality and
discourse interactional features, since the latter is an integral part of the
expression of the speaker's knowledge status.
It is not easy to draw a clear distinction semantically between discourse
features and modality. For in discourse we often express opinion, draw
conclusions, etc.; and no doubt, in 'evidential' language [sic] speakers
regularly indicate in their conversations and arguments the evidential basis
for what they are saying. [...] what are clearly systems of epistemic modality
may contain some terms that belong more to discourse (Palmer 1986: 9Iff.)

These markers, then, form a distinct type of epistemic modality in which

discourse-pragmatics plays an important role.
The old vs. new knowledge distinction discussed by DeLancey and
others, and discourse pragmatic functions of epistemic modality discussed in
Palmer's work, relate closely to the functions of the morphological class in
Korean to be studied in this paper. The sentence-ending modals in Korean are
always used in interactive discourse, and, as will be shown below, their
meanings relate to the distinctions between old and new knowledge, sources
of information, and different assumptions by the speaker about what the
listener knows.
 170 Soonja Choi

3. The modal system in Korean

Korean has two types of modal forms: sentence-ending (SE) modal suffixes
and auxiliary verbs. Table 1 shows a list of modal forms in the two categories
with approximate meanings in English, SE suffixes occur typically at the end
of the sentence, as in (1) with the suffix -e. Korean is a verb final (sov)
language, and SE suffixes occupy the final position among the inflections on
the predicate: verb or adjective1.
(1) Younghi-ka Seoul-ul ttena-ss-e.
Younghi-SUBJ Seoul-OBJ leave-PAST-SE
'Younghi left Seoul.'
On the other hand, auxiliary modal verbs are free morphemes which can
occur after the main verb. An auxiliary verb must be connected to the main
verb by a specific connecting suffix (= connective) on the latter to express a
given modal meaning. For example, the connective -ya followed by the
auxiliary verb tway-ta, i.e. -ya tway-ta (-ta is the citation form), expresses
obligation, whereas the connective -to with tway-ta expresses permission.
(2) Younghi-ka Seoul-ey ka-ya/to tway-e.
Younghi-SUBJ Seoul-to go-coNN AUX-SE.
'Younghi must/can go to Seoul.'
Other examples include the connective -ko and the auxiliary siph-ta express­
ing desire, and -swu iss-ta expressing ability. Thus, connectives are an inte­
gral part of the whole auxiliary modal system in Korean. A list of combina­
tions of frequently used connectives and auxiliary verbs in Korean is shown in
Table 1B.
All of the auxiliary verbs can be used as main verbs except -ko siph-ta,
which occurs only as an auxiliary. The meanings of these verbs in the main
verb position are as follows: molu-ta 'not to know', kath-ta 'to be the same',
ha-ta 'to do', i-ta 'to be (Copula)', po-ta 'to see', iss-ta 'to exist', tway-ta 'to
become'. As shown in Table 1, when these verbs are used as auxiliaries, some
denote epistemic meanings (i.e. kath-ta, molu-ta, i-ta, po-ta), and others
agent-oriented meanings (i.e. iss-ta, tway-ta, ha-ta). However, note that each
form denotes either agent-oriented or epistemic meaning, and there is no overlap
of the two types of modality on the same verb, except the verb hata 'to do'.
The Development of Epistemic Sentence-ending Modal 171

Table 1. Form and meaning of modality in Korean

A. Sentence-ending (SE) suffixes

Canonical structure: Verb stem-tense/aspect-SE suffix.
Form Meaning
EPISTEMIC
-ta (Type 1) newly perceived information for the speaker
-ta (Type 2) new information to the listener
-e assimilated information, unmarked form
-ci/-cyana* certainty of proposition, shared information
-kwun newly made inference
-toy reported speech, hearsay/ story-telling
-ney Information based on factual evidence
-ni uncertainty and negative bias of proposition
DEONTIC/EPISTEMIC
-llay future, desire
-kkey future, intention, prediction
Note: I have listed those that are relatively frequent in the adult speech and which
children acquire early.
* -cyana is the suffix -ci followed by the negative marker -ana. The two morphemes,
when combined, express a stronger commitment to the truth of the proposition than -ci.

B. Modal auxiliaries
Canonical structure: Main verb-Connective Auxiliary-tense/aspect-SE.
Form Meaning
EPISTEMIC
informal (infinitive)
-ci molla (moluta) possibility (weak)
-kes kathay (katha) possibility (strong)
-tus hay (hata) possibility (strong)
-n/l kkeya (kesita) probability
-na pwa (pota) inference
DEONTIC
-su isse (issta) ability
-to tway (twayta) permission
-ya hay/tway (hata/twayta) obligation
-ko siphe (siphta) desire
mos + Verb negation of ability
-myen an tway (twayta) negation of permission, negation of obligation
-ki silhe (silhta) negation of desire
 172 Soonja Choi

In contrast to the auxiliary verbs, SE suffixes have more abstract meaning,

and their semantic functions have been an area of investigation in recent
studies of Korean linguistics. Several studies (Choi 1991; H. Lee 1991; K. Lee
1986) have suggested that many of the SE suffixes code the differential status
of information in the speaker's knowledge system. H. Lee (1991) states the
functions of SE suffixes as follows:
What is differentiated by sentence-terminal suffixes is various epistemic
modality categories, including the speaker's knowledge status, background
expectation, evidentiary sources of the information conveyed, and the speak-
er's assumption about the addressee's point of view. (H. Lee 1991:471)

Several SE suffixes code degrees of integration of information in the

speaker's mind. The contrasts among the three suffixes -ta, -kwun, and -e
illustrate this point. The suffix -ta is often used by the speaker, when he/she
has just perceived something noteworthy in the present context, "either be­
cause it is about the accomplishment of awaited events or states of affairs at
the very moment of speaking" (H. Lee 1993). Consider the following context:
John and his wife are waiting for a friend at a restaurant. Their friend is late.
Upon seeing the friend finally appear at the restaurant door, John would say:

o-ass-ta
come-PAST-SE
'(He) has come.'
The use of -ta is appropriate in this context in which John has just seen the
friend and registers the information in his mind for the first time. Comments
made with -ta may not be directed to the listener; they can be noteworthy
remarks to the speaker himself. Once the speaker has registered this new
information, in subsequent mentions, e.g., when the speaker repeats the
information again to his wife, or when he tells this experience later to another
friend, he would use the SE form -e:
o-ass-e
come-PAST-SE
'(He) came.'
In this latter context, the use of -ta, i.e. *o-ass-ta is ungrammatical. The use of
-ta is not restricted to perception of the outside world. It can also occur with an
internal state of mind. For example, adult Koreans often say al-ass-ta (know-
PAST-SE) '(I)'ve got (it)' when they have just come to an understanding of
The Development of Epistemic Sentence-ending Modal 173

something. (It should be noted that -ta is not an aspectual marker denoting
perfective aspect. One can use -ta when describing a state of affairs as well as
an on-going activity if it draws the speaker's attention for the first time, e.g.,
ippu-ta [pretty-SE] '(she's) pretty', or no-n-ta [play-PRES-SE] '(They) are playing').
The suffix -kwun is similar to -ta in that it is also used for newly acquired
information. However, the information marked by -kwun is often an inference
based on what the speaker has just seen, as in example (3).
(3) (The mother, upon arriving home, sees her children's school bags
lying around. She says to herself,)
ai-tul-i tolao-ass-kwun.
child-PL-suBJ return-PAST-SE
'The children have returned (home).'
By using -kwun, the mother indicates that the proposition 'the children re­
turned' is inferred from seeing their school bags. As H. Lee (1985, 1993)
argues, both -ta and -kwun are used for knowledge that has not yet been
assimilated into the speaker's knowledge system. The difference between the
two suffixes is that -ta is used with knowledge obtained through direct
experiences, whereas -kwun is often used with knowledge obtained through
inference.
In contrast to -ta and -kwun, -e is used when the information has already
been assimilated into the speaker's knowledge system: the speaker acquired
the information in the past, therefore has known it for some time. As men­
tioned above, once newly perceived information is encoded with -ta or -kwun,
in subsequent mentions -e must be used. In H. Lee's survey (1991), -e is the
most frequent suffix (58%) in spontaneous discourse interactions. This is not
surprising, since in conversations participants often contribute to the topic by
offering information that they already have about the topic. Indeed, one can
conclude that -e is the unmarked suffix in conversations.
In discussing -kwun above, we have seen how a particular status of
knowledge is related to an evidentiary source, e.g., inference. There are a few
other suffixes that differentiate evidential sources in Korean. The difference
between -toy and -ney is a case in point. Whereas -toy marks hearsay or
reported speech, -ney signals that the event was directly witnessed by the
speaker.
All of these suffixes are used only in informal conversational interactions
where participants, familiar to one another, freely and spontaneously ex-
 174 Soonja Choi

change information about the current topic. Probably due to this characteris­
tic, the meanings of many suffixes incorporate the speaker's assumption about
how much the listener knows. That is, the speaker's choice of a specific SE
suffix reflects his or her assumption about status of the listener's knowledge
about the proposition (H. Lee 1991). For example, the suffix -ci (or -cyana) is
used when the speaker commits him/herself to the truth of the proposition; it
therefore expresses certainty (-cyana expresses a stronger commitment than -
ci). At the same time, -ci denotes that the information is also known to the
listener or can be readily inferred by the listener. This is confirmed by the use
of -ci in questions. The suffix -ci? is used in a question when the speaker is
committed to the truth of the proposition and at the same time wants to
confirm that the listener shares this commitment. This is shown in (4).
Considering the context of (4), the meaning conveyed by -ci is that the
speaker is certain that Mary is pretty and also believes that the listener has the
same opinion. The speaker expects agreement from the listener. Compare this
with the suffix -nil in (5). When -nil is used, it conveys the speaker's
assumption that the listener knows more than the speaker about the matter
(since only the latter saw Mary). By using -ni the speaker expresses a desire to
know what the listener knows to be the true.

(4) (Possible context: Both speaker and listener saw Mary and both
liked her. The speaker later asks,)
Mary-ka ippu-ci?
Mary-SUBJ pretty-SE?
Tsn't Mary pretty?'
(5) (Possible context: The speaker heard from someone that Mary was
pretty. The speaker asks the listener, who has seen Mary,)
Mary-ka ippu-ni?
Mary-SUBJ pretty-SE?
Ts Mary pretty?'
Thus, -ci and -ni express different degrees of certainty which also relate to
what the speaker assumes about the listener's knowledge. To this category,
the suffix -ta (Type 2) should be added. This suffix is used when the speaker
has assimilated the information conveyed in the proposition but believes that
the information is new and noteworthy to the listener. In this case, -ta (Type 2)
is typically used with a high pitch (see the acquisition section below). For
example, when the speaker but not the listener has seen Mary, and the speaker
The Development of Epistemic Sentence-ending Modal 175

wants to convey this information that is unknown to the listener, the speaker
may say Mary-ka ippu-ta (-ta with a high pitch).
In summary, I have discussed several examples to show that a number of
SE modal suffixes in Korean express evidentiality and the different status of
knowledge on the part of the speaker and the listener concerning an event or
state. These suffixes are used only in informal interactive contexts (e.g., a
conversation, or personal letter to a friend). They are not used in formal
situations (e.g., a formal speech), nor in contexts where the speaker/writer
does not have a particular listener/reader in mind, e.g., written reports. The
present study will suggest that such discourse-interactional features and non­
verbal contextual cues that accompany them facilitate and enhance the acqui­
sition of epistemic modal suffixes.
How often do adults use these various SE suffixes? As noted above, one
can always use the unmarked form -e, which simply denotes propositions that
are established in the speaker's mind. However, H. Lee (1991) found in his
informal conversation data that, although -e is used most frequently, it occurs
only in just over half of the utterances (58%). For the remaining utterances
(42%), 21 different SE suffixes are used, with varying degrees of frequency.
Among these, -ci was most frequent (15%) followed by -ta, -kwun, -ney, and -
toy, used at roughly equal frequency of 2%. (-ni was used only 1%). Thus,
although Korean speakers use -e most of the time, they use other SE suffixes
spontaneously to express different degrees of certainty about a proposition,
about the source of the information, and/or to check the status of knowledge
of the listener.

4. Development of SE suffixes in Korean children

As mentioned earlier, research on the acquisition of modality in spontaneous

speech has shown that epistemic modality is acquired later than agent-ori­
ented modality (Stephany 1986). In addition, a number of experimental
studies have shown that an understanding of different degrees of certainty
about a proposition develops only after 3 years (Byrnes & Duff 1989; Hirst &
Weil 1982; Moore, Pure & Furrow 1990). It is important to note here that the
epistemic meanings in these studies relate to the status of knowledge that
results from reasoning by the child which is relatively independent of a
particular discourse-interactional context. For example, the distinction be-
 176 Soonja Choi

tween must and may in English in 'He must/may be home by 5 o'clock' has
little to do with the speaker's assumption about how much other participants
know about the proposition. Also, in most of the experimental studies on the
acquisition of epistemic modality, tests have been designed to understand
such context-independent reasoning. For example, in Moore et al. (1990),
children (between 3 and 6 years of age) were asked to guess the location of an
object hidden in a box solely on the basis of one sentence cues which varied in
epistemic modal auxiliaries, e.g, "It must be in the red box" or "it might be in
the blue box." Moore et al. found that ability to find the hidden object on the
basis of the modal meaning was shown only in children older than four years.
However, the results of these studies give little indication of how and when
children understand modal forms that incorporate discourse-interactional
meanings.
A growing body of research on communicative competence in children
shows that children learn to be good participants in conversations from a
young age. Several studies (Shatz 1983, 1984; Pellegrini, Brody & Stoneman
1987) have shown that even two-year-olds are capable of giving enough
information and keeping their linguistic contributions truthful and relevant to
the topic of discourse. Bloom, Rocissano & Hood (1976) have also shown that
before children reach two years, they have learned a basic rule of discourse,
that of conversational turns, and that between two and three years, children
increase the amount of information they contribute to the shared topic.
These studies suggest that the modal functions of SE suffixes in Korean
which relate to discourse interactions may be within children's cognitive
grasp and linguistic capacity from an early age. Furthermore, there are several
characteristic morphological features of SE suffixes that may facilitate early
acquisition of SE suffixes in Korean, particularly when we consider Slobin's
operating principles (1973). First, SE suffixes occur at the ends of sentences
(most often with one syllable consisting of a consonant and a vowel), and
therefore are perceptually salient. Second, the SE suffixes constitute an obliga­
tory category in that all sentences in discourse interactions must end with a SE
suffix. During interactions with children, the caregiver provides a variety of
SE suffixes appropriate to specific discourse contexts. From the acquisition
perspective, this means that children hear different SE suffixes frequently from
their caregivers. Third, there is a relatively high degree of semantic transpar­
ency in that agent-oriented and epistemic meanings are distinguished morpho­
logically, and SE suffixes denote epistemic meanings most of the time. (Table
The Development of Epistemic Sentence-ending Modal 177

1 shows that all SE forms in Korean carry the type of epistemic meanings
discussed above, except -llay and -kkey which denote desire and intention
respectively. It should be noted, however, that, as discussed in a number of
studies, constructions that express desire and intention often develop a use for
prediction, which can be considered epistemic (Bybee 1985; Bybee &
Pagliuca 1985; H. Lee 1991). In fact, the two SE suffixes are always preceded
by the future tense marker -/.) Agent-oriented meanings are expressed pre­
dominantly by auxiliary verbs, which require children to use two verbs in the
sentence, i.e. a main verb and an auxiliary. Also, the SE suffixes have only
modal meanings; they incorporate neither tense nor aspectual meanings. All
of these morphological characteristics, along with the discourse pragmatic
functions of SE suffixes, may enhance children's acquisition of SE suffixes.

4.1 Data

I have examined the development of SE modal suffixes in three children, HS,

PL, and TJ2. The present analysis is based on the children's spontaneous
conversations with their caregivers and/or the investigator during the follow­
ing periods: HS, from 1 (year);8 (months) till 2;8; PL, from 1;8 till 3;0; TJ, from
1;9 till 4;0. All three children are girls.
Table 2 shows the onset of productivity of modal suffixes and auxiliary
forms that the three children used during the study period. In this analysis,
productivity is determined by counting the number of different verbs (i.e.
types, not tokens) used with a given modal suffix. The first session in which
the child has produced the suffix with more than three different verb stems is
considered to be the onset of productivity. Once the child's production of a
given suffix reached this threshold, it was generally the case that the child
continued to use the suffix regularly in subsequent sessions. Another impor­
tant criterion for the analysis is that only spontaneous combinations of verbs
and suffixes by the child are considered in this study. If the child's production
is an exact imitation of the immediately preceding utterance by the caregiver,
it is excluded from analysis. In other words, the data only include verb plus
suffix combinations that are different from the ones used by the caregiver in
the immediately preceding discourse. For example, if the caregiver asks mek-
ess-e? (eat-PAST-SE? 'did (you) eat?) and the child answers with the same verb
and suffix, i.e. mek-ess-e ('(I) ate.'), the child's use of the suffix -e is
considered an imitation and therefore is not included in the analysis. How-
 178 Soonja Choi

ever, if the child answers with either a different verb (e.g., hay-ss-e do-PAST-SE
'(I) did') or a different suffix (e.g., mek-ess-ci eat-PAST-SE '(I) ate'), or both
(e.g., hay-ss-ci do-PAST-SE '(I) did') the child's utterance is included in the
data.
As can be seen from Table 2, the order of acquisition of the modal
suffixes is strikingly similar across the three children. First, note that several
SE suffixes are productive in the children's speech before auxiliary verbs
appear. Second, we also see a consistent pattern across the three children in
the order of acquisition among the SE modals: -ta (Type 1), -e , -ci, -toy, and -
ta (Type 2), in the order mentioned. The two agent-oriented SE modals with
future meaning, -llay and -kkey, are both acquired after -ta, -e and -ci have
been acquired.
The general developmental pattern is that the children acquire the suf­
fixes -ta (Type 1), -e, and -ci before the suffixes indicating future, and then the
other two epistemic suffixes, -toy, and -ta (Type 2) are acquired along with
some agent-oriented modal auxiliaries. These five SE suffixes are acquired
before 3 years in all three children. In what follows, I analyze the modal
meanings encoded by these suffixes in the children's speech.
In order to identify the meanings of these suffixes in the children's
speech, I have examined the linguistic as well as nonlinguistic contexts in
which the suffixes are used. Concerning the linguistic context, I have ana­
lyzed the structure of discourse interactions between the child and the care­
giver. This includes an analysis of discourse contingency (i.e. whether or not
the child's utterance maintains the shared topic) and the effect of preceding
utterances on the child's selection of a particular SE suffix. Since the children's
lexicon is rather limited at this early stage, I have also examined the kinds of
verbs to which a particular suffix is attached. As for nonlinguistic contexts, I
have analyzed the events, states, and entities that the children refer to in their
propositions. I have also examined the types of context which provide the
child with particular information, i.e. the source of information.

4.1.1 Acquisition of the suffixes -ta and -e: distinction between unassimilated
and assimilated knowledge.
The first two suffixes to appear in all three children's discourse were -ta (Type
1) and -e. These two suffixes were acquired at around the same time, i.e.
during the first month of study in all three children. HS used the two suffixes
productively from the first session (1;10). PL acquired -ta at 1;8 and -e one
The Development of Epistemic Sentence-ending Modal 179

Table 2. Onset of productivity of modal forms in three Korean children

HS PL TJ
Form Age* Age* Age*

SE suffixes
EPISTEMIC
-ta (Type 1) 1;10 1;8 1;9
-e 1;10 1;9 1;11
-ci/cyana 2;1 l;ll 2;2
-toy 2;5 2;l 2;3
-ta (Type 2) 2;5 2;9 3;0

DEONTIC/EPISTEMIC
-kkey 2;1 2;1 2;7
-llay 2;3 2;0 2;6

Modal auxiliaries
EPISTEMIC
2;8 3;0
-na pwa (inference)

DEONTIC 2;7 2;6 3;1

-ya tway (obligation) 2;6 3;0 3;4
-ko siphe (desire) - 2;6 3;6
-su isse (ability) 2;7 2;6 3;1
-to tway (permission)
-mos + Verb (Neg + ability) 2;4 2;6 2;9
-myen an tway (Neg + obligation) 2;7 2;3 3;1

* Ages are in years;months.

month later (1;9), and TJ acquired -e at 1;9 and -ta three weeks later.
Whatever the order of acquisition of the two suffixes was, the two forms
served distinct functions in the speech of all three children, who used them to
make the same distinctions. The distinction between -ta and -e went through
two phases. During the first phase, -e was restricted to requests. Examples are
shown in (6) and (7).
(6) HS(1;10)
(Mother and HS open the refrigerator in the kitchen)
→ HS: uyu cw-e.
milk give-SE
'Give (me) milk.'
 180 Soonja Choi

(7) TJ(1;9)
(TJ says to a friend)
→ TJ: ilwu o-a3.
here come-SE
'Come here.'
In (6), HS asks her mother to give her some milk, and in (7), TJ asks her friend
to come closer. These are appropriate uses of -e in that in the Korean adult
grammar, -e is used for imperative as well as declarative sentences. On the
other hand, the children used -ta in all statement sentences as illustrated in (8)
and (9).

(8) TJ(1;9)
(TJ putting one lego block on top of another)
→ TJ: olla ka-ss-ta.
Up gO-PAST-SE
'(It) went up.'

(9) TJ(1;9)
(TJ looks inside her doll house and sees that it's empty.)
→ TJ: eps-ta.
not-exist-SE
There is nothing.'
As shown in (8) and (9), -ta was used in statements to describe events
(example (8)) as well as states of affairs (example (9)). The contrastive use of
the two suffixes, -ta for statements and -e for requests, was shown by all three
children during the first phase. The distinction between statements and re­
quests is parallel to the mood distinction between declarative and imperative:
-ta for the declarative mood, which expresses the speaker's assertion of
propositions, and -e for the imperative mood, which directs the listener to do
something. It seems therefore that at the beginning the two suffixes distin­
guish mood in the children's speech. This supports Stephany's finding (1986)
that mood distinctions occur early.
At this early period of language development, verbs attached to -ta were
limited in number and were restricted to those that reflected the child's
cognitive interests in trying to understand the world. That is, the kinds of
events and states of affairs that the children described (all with -ta) were
The Development of Epistemic Sentence-ending Modal 181

closely related to cognitive concepts that children develop universally at this

developmental stage, e.g., existence of an entity, success/failure, and change
of state/location (Gopnik & Meltzoff 1984; Slobin 1985; Gopnik & Choi
1990). Furthermore, all of the contexts commented on by the children were in
the here and now (see examples (8) and (9), also (10) - (13) below.) The
following types of events/states as well as specific verbs, were used by at least
two children at this period:

- verbs denoting (non-)existence: iss-ta 'exist', eps-ta 'not-exist'.

- verbs denoting success: twayss-ta 'have become',
- verbs denoting particular states: can-ta 'sleeping', kath-ta 'be same'
- verbs marking perfective aspect: hayss-ta 'done', olla kass-ta 'went up'
The three children expressed all four types of events/states with the suffix -ta
repeatedly (whenever these contexts occurred) during the first few weeks,
typically using the specific verbs noted above. This suggests that statements
made by the children with the suffix -ta at this stage reflect the kinds of
concepts they are trying to assimilate into their general cognitive system.
The distinction of mood by the two suffixes, -ta for statements and -e for
requests, lasted only a brief period, however (1 week for HS, 4 weeks for PL,
6 weeks fo TJ). Within 1 to 6 weeks from the time -e was first used for
requests, both -ta and -e were used in declarative statements. (The children
also began to use -e appropriately in questions.) The question then is what
specific functions within the declarative sentences the two suffixes serve, if
any. The following analyses suggest that the two suffixes systematically
contrast different degrees to which a proposition (i.e. event or state to be
encoded in language) is assimilated into the child's knowledge system. Spe­
cifically, I will argue that the children use -ta for newly perceived information
which is not fully assimilated into their knowledge system, whereas they use -
e for information that has been integrated into their body of knowledge.
First, we will discuss the use of -ta. The functions of -ta in the second phase
were similar to those of the first phase. Observe examples from (10) - (13).
(10) (TJ2;0)
(Looking at the picture of Mr. Bump fallen down on the ground)
→ TJ: nemecye-ss-ta.
fall-PAST-SE
'(He) fell down.'
 182 Soonja Choi

(11) PL(1;9)
(Seeing grandfather's glasses)
→ PL: apuci-kke-ta.
grandpa-POSS-SE
'(They're) grandpa's.'
(12) HS(1;10)
(HS puts a lego person in a chair.)
→ HS: tway-ss-ta.
become-PAST-SE
'done'
(13) (TJ1;11)
M: Mickey eti iss-ni?
Mickey where exist-SE?
'Where is Mickey?'
(TJ pointing to the picture of a Mickey Mouse on her doll house)
→ TJ: Mickey yeki iss-ta.
Mickey here exist-SE.
'Mickey is here.'
The events described in (10)-(13) above occur in the children's immedi­
ate context, and are newly registered in their mind. In (10), TJ describes the
scene which has drawn her attention, and in (11), PL points out that the
glasses belong to her grandpa. In (12), HS marks the completion of her goal as
the action was just accomplished, and in (13) TJ has just found the picture of
Mickey. These examples illustrate typical contexts in which -ta is used, and
they can be categorized as follows:
(a) describing a scene in a picture;
(b) describing events/states that the child has just observed:
i. a perfective aspect which results in a particular state, e.g.,
nemecy-ess-ta 'fallen down', ollaka-ss-ta, 'went up',
ii. an ongoing event/state e.g., can-ta 'sleeping',
iii. naming a referent, appa-(i)-ta4 'daddy'
(c) commenting on the existence or non-existence of an event/object.
These three types were the dominant types of contexts for -ta from the first
phase (i.e. when all statements ended with the suffix -ta), and continued to be
the dominant types after -e was also used for declaratives. Table 3 shows the
The Development of Epistemic Sentence-ending Modal 183

Table 3. Context analysis of-ta. during the first six months of its contrastive use of with -e.

HS PL TJ
(l;10-2;4) (l;10-2;4) (l;ll-2;5)
%(N) %(N) %(N)
a. picture scene 5(14) 19 (23) 29 (32)
b. present event/state 45(116) 28 (34) 49 (55)
c. (non-)existence 46(119) 17 (20) 18 (29)

Subtotal (a-c): 96(249) 64(77) 96(116)

d. past event/state 2(5) 1(1) 1(2)

e. non-realized event* 2(6) 0 1(1)
f. request for information 0 35 (42) ** 2(2)

Subtotal (d-f): 4(11) 36 (43) 4(5)

TOTAL: 100 (299) 100(120) 100(112)

* Pretend play, negation, make-believe, intention.
** The use of -ta in questions by PL was idiosyncratic. This is an overextension in that
the adult speaker's use of -ta is restricted to assertions. Apparently, PL used -ta in this
way to check with the caregiver the truth of the newly perceived information (Choi 1990).
This use was most extensive from 1;8 till 1;10, and disappeared completely from 2;0.

Table 4. Context analysis of-e during the first six months of its contrastive use with -ta.

HS PL TJ
(l;10-2;4) (l;10-2;4) (l;ll-2;5)
%(N) %(N) %(N)
a. picture scene 5(8) 19 (22) 27 (28)
b. present event/state 25 (40) 5(6) 12(13)
c. (non-)existence 4(7) 5(6) 7(7)

Subtotal (a-c): 34 (55) 29(34) 46(48)

d. past event/state 17 (26) 42 (50) 32 (33)

e. non-realized event 18(29) 11(13) 10(11)
f. request for information 31 (48) 18 (22) 12(12)

Subtotal (d-f): 66 (103) 71 (85) 54 (56)

TOTAL: 100(158) 100(119) 100 (104)

184 Soonja Choi

frequency of -ta used in these three types during the first six months from the
time -ta and -e were both used for declarative sentences. As Table 3 shows,
the children used -ta most frequently in the three types of context (a), (b) and
(c). (Further categories (d)-(f) will be introduced below.) This suggests that
the semantic content of propositions with -ta was something that the children
had just perceived through direct experience. The children seemed to describe
the events as they became aware of them.
Table 4 shows the frequency of -e in different types of contexts. The
distribution of -e in these contexts is clearly different from that of -ta.
Although -e is used in categories (a)-(c), these amount to fewer than 50% of
all the uses of -e. (The contrastive uses of -ta and -e in (a) - (c) will be
discussed later.) The majority of the -e uses occur in categories (d)-(f):
(d) to give information about a past event/state;
(e) to convince the listener of an event/state of affairs, or to talk about an
event/state which was not occurring at the time of speech (e.g., negation,
actions which the child was about to perform, make-believe events while
playing with a doll or a toy), and;
(f) in questions to verify the truth of a proposition (see Table 4).
These types are illustrated in examples (14) - (17).
(14) PL(1;11)
(Grandmother talks to another adult about the hair salon where PL
has recently had her hair cut. PL, hearing all this conversation, says)
→ PL: polami meli ippukey hay-ss-e.
PL hair prettily do-PAST-SE.
'PL had (her) hair done. It was pretty.'
(15) TJ(2;2)
(TJ is in the middle of a book, but doesn't want to read anymore. TJ
closes the book.)
—» TJ: eps-e
not-exist-SE.
'no more.'
(16) TJ(1;11)
(TJ is in another room. M asks TJ to bring a color book.)
M: ppalli kacko o-a. illwuo-a.
quickly take come-SE here come-SE
'Bring (it) quickly. Come here.'
The Development of Epistemic Sentence-ending Modal 185

—» TJ: an ka-a
not go-SE
'(I'm) not going.'
(17) HS(2;0)
(HS gives a piece of apple to the Investigator (Inv). Inv. doesn't
want any more. HS holding the apple,)
—> HS: yoke-nun acci-kke nwu-ka hay(-e)5?
this-TOP uncle-poss who-suBJ do(-SE)?
'Who will do (=eat) uncle's apple?'
(By this question, HS implies that it is she who wants to eat the apple. In other
words, in her mind 'who' refers to HS herself, and she asks the question to get
permission from the investigator.)
In (14), PL states a past event that is well known to her, and in (15) the
child creates a desired state (i.e. because the book is closed, reading is
finished) and describes it to convince the adult of its truth. In both (14) and
(15), the information that the child conveys has been established in her mind
before actually saying it. Examples (16) and (17) show that -e is used also for
negations and questions. In (16), M has asked TJ to come back to the living
room, and TJ negates the proposition. In (17), the child asks about a proposi­
tion that is on her mind, i.e. she wants to eat the apple. In both cases, the
proposition negated or questioned has already been established in the child's
knowledge system.
Among the categories (d)-(f), there were, however, individual differ­
ences in frequency: HS used -e most frequently in wh-questions, whereas PL
and TJ used it most frequently for past events/states. Nevertheless, the catego­
ries (d)-(f) taken together constitute more than 50% of all the -e uses in all
three children's speech. This is different from the uses of -ta which was rarely
produced in these contexts.
Several further analyses support the interpretation that -ta and -e are used
for unassimilated/new and assimilated/old knowledge, respectively. In Table
4, one notices that, although -e is used in categories (d)-(f) more than half of
the time, it is also used in categories (a)-(c) quite often. Specifically, HS used
-e for picture scenes (category (b)) 25% of the time, PL and TJ used it for
present events/states (category (a)) 19% and 27% of the time, respectively.
Thus, both -ta and -e were used for describing present situations. However,
the two suffixes contrasted in a systematic way. One contrast was related to
 186 Soonja Choi

the way the two suffixes were used in repetitions of a proposition. When -ta
and -e occurred within one conversational turn, the first mention of the
proposition was marked by -ta, and subsequent repetitions then switched to -e.
This is shown in the following examples:
(18) PL(2;0)
(PL finds the balloon she has been looking for.)
PL: yeki iss-ta. yeki iss-e. yeki iss-e.
here exist-SE. here exist-SE. here exist-SE.
'Here (it) is. Here (it) is. Here (it) is.'
(19) TJ(2;3)
(The picture of an elephant in a book and the elephant design on her
pants are similar.)
TJ: ike ttokkath-ta. ike ttokkath-e.
this same-SE this same-SE
'These are the same. These are the same.'
In (18), PL finds the balloon unexpectedly. She gets excited and describes the
event with -ta first. However, at the second mention of the same proposition,
she switches to -e. A similar explanation can be given to TJ's utterance in
(19). As TJ recognizes that the two pictures are the same, she uses -ta (the
concept of sameness interested TJ greatly at this period). But once the infor­
mation has been registered in her mind, she switches to -e. This sequence of -
ta followed by -e occurred a number of times in the speech of all three
children. The reverse order never occurred. Such a fixed order of -ta and -e
within one conversational unit also occurs in adult speech, and apparently the
children learned this order without errors.
Another distinction between -ta and -e had to do with verb types. As
noted above, at the beginning, verbs attached to -ta were limited in number
and restricted to those that reflected the cognitive concepts that the child was
developing at the period of time in question. Most notable verbs were: iss-ta
'exist', ep-ta 'not-exist', tway-ss-ta 'become', kath-ta 'same', ollaka-ss-ta
'went up'. In addition, -ta was also used frequently after a noun to label
entities (e.g., appa-ta '(It's) daddy'). These verbs were used with high fre­
quency and co-occurred only with the suffix -ta during the first few months.
Gradually however, propositions about existence, success, change of location,
or names of things involved in the here and now were encoded with -e,
particularly when an event or state was expected by the child (see examples
(15), (18), and (19) above, also (20) below). As the children's cognitive
The Development of Epistemic Sentence-ending Modal 187

Table 5. Frequency of occurrence of-ta. and -e in relation to the discourse topic during
the first six months, (from Choi 1991)
-ta -e
Related* Unrelated** Related Unrelated
HS 9% 91% 65% 35%
PL 19% 81% 57% 43%
TJ 24% 76% 64% 36%
*Related to the immediately preceding utterance, or to the current topic.
**Unrelated to the topic, introduction of a new topic.

development advanced, those cognitive concepts seemed to be well estab­

lished in their knowledge structure. The use of both -ta and -e for the same
verbs occurred at 2;1 for HS, 2;0 for PL, and 2;1 for TJ. From this period, -ta
was used for new information that relates to a specific discourse context.
(20) TJ(2;1)
(A game of ducks catching fish. TJ is familiar with this game. TJ
pointing to one fish)
TJ: yeki mwulkoki iss-e.
here fish exist-SE
'Here is a fish.'
Another systematic difference between -ta and -e that the children show
relates to discourse contingency, i.e. whether or not the utterance maintains
the current topic and adds new information to it (Bloom et al. 1976). If the
function of -ta is to register a new and noteworthy event/state in the child's
own mind as she begins to understand it, the form would tend to be used in
monologue situations where the child tells herself about the meaning of the
event. (As mentioned earlier, adult speakers also show this phenomenon.) On
the other hand, as H. Lee (1985) notes, -e would be used to exchange
information since the information has become part of the speaker's knowl­
edge system. The analysis of discourse contingency is shown in Table 5.
In the data, the children used -e much more often than -ta as they
responded to adult utterances giving more information about the topic. Table
5 shows that the proposition with -ta was often not related (i.e. non-contin­
gent) to the current topic, but instead, shifted to a different topic that drew the
child's attention. As a consequence, the information which co-occurred with -
ta abruptly interrupted the current topic and often initiated a new topic which
the adult then commented on. An example is (21).
 188 Soonja Choi

(21) TJ(2;0)
(Investigator points to the girl that she colored)
Inv.: i salam ippu-cil
this person pretty-SE?
'Isn't this person pretty?'
TJ: ippu-ci.
pretty-SE
'pretty.'
Inv.: i salam ippu-cil
this person pretty-SE?
'Isn't this person pretty?'
(TJ hears a baby cry upstairs.)
→ TJ: aka wu-n-ta.
baby cry-PRES-SE.
'A baby is crying.'
Inv.: aka wul-e?
baby cry-SE
'Is baby crying?'
In (21), as TJ hears a baby cry, she comments on this, abruptly interrupt­
ing the topic of coloring. This non-contingency occurred in 83% (on average)
of the children's -ta uses. In contrast, the propositions with -e showed a much
lower frequency of non-contingency (39% on average). In fact, more than half
of the time, the propositions with -e maintained the current topic. More
specifially, -e was used to respond to or comment on the preceding adult
utterance. PL's utterance in (14) above is a good example of providing more
information on the current topic (i.e. about PL's hair).
In summary, several types of analyis have shown that the children in the
present study used the modal suffixes -ta and -e for distinct functions. -Ta was
used to encode a new and noteworthy proposition as the child became aware
of its meaning. At the beginning, propositions with -ta also reflected cognitive
concepts that the children were developing. At a later stage, -ta encoded new/
unassimilated information that the child experienced directly in the here and
now. In both cases, -ta was often used to introduce a new topic during
interaction. In contrast, -e was used for propositions that were already estab­
lished in the child's knowledge system, and its function in discourse was to
contribute more information to the current topic.
The Development of Epistemic Sentence-ending Modal 189

4.1.2 Acquisition of-ci: shared knowledge and certainty of the truth of the
proposition
The third modal suffix to be acquired by the children was -ci. (The children
also acquired -cyana around the same time, and the data in this section include
both -ci and -cyana.) With the use of -ci, the interactional function of modal
suffixes becomes even clearer. Initially, -ci was used in the following con­
texts: (a) when reiterating a proposition in the preceding utterances produced
either by the child herself or by the interactant (example 22), or (b) when
redescribing an event or state which had been described several times before
(example 23).

(22) PL(1;11)
(PL pointing to Grandmother's hairpin)
PL: ike hammwuni kke-ya?
this grandma POSS-SE
'Is this grandma's?'
GM: ung, hammwuni kke-ya.
yes, grandma POSS-SE.
'Yes, grandma's.'
—» PL: hammwuni kke-ci?
grandma POSS-SE
'Grandma's?'
GM: ung, hammwuni kke-ci.
yes, grandma POSS-SE
'Yes, grandma's.'
(23) PL(2;2)
(PL remembers seeing a monster on TV.)
PL: kweymwul an mwusew-e?
monster NEG scary-SE
'Monster is not scary?'
GM: kweymwul eps-e.
monster not-exist-SE
'There's no monster.'
→ PL: kweymwul Tibi-ey-na iss-ci?
monster tv-LOC-only exist-SE?
'Monsters exist only on TV?'
190 Soonja Choi

Table 6. Number of-ci utterances in different types of context.

Age period: 1:10-2:0 2;l-2;6 2;7-3;0

HS PL TJ HS PL TJ HS PL TJ
a. reiteration of adult's
preceding utterance 1 1 5 1 1 3 2
b. frequent occurrence - 6 - - 4 2 - 7 5
c. perceptual evidence
in here and now 1 1 - 8 8 1 32 16 18
d. certainty developed
through discourse 2 1 _ 11 7 2 24 4 7
e. shared past event - - - - 1 - 1 5 5
f. in wh-questions- - - - 2 - - 18 5 2
g. cooccurrence with
obligation marker _ _ _ _ 8 _ 27 4 2
h. other - - - - - - - 28* 1*
Total -ci utterances: 4 8 1 26 29 6 105 71 40

* PL and TJ used -ci in the question, al-ass-ci?, know-PAST-SE, to mean

'(you) got (=understood) (it)?'

In (22), -ci is used when PL repeats the grandmother's preceding assertion

that the hairpin belongs to her, and in (23) it is used for information which has
been given several times before during the session. This characteristic use
suggests that at this time children pay attention to large stretches of discourse,
i.e. several turns of conversation, and also remember conversations which
took place in the past. The repetition of the proposition with -ci in this way
allows the child to be certain that the information contained in proposition is
shared with the caregiver.
Table 6 shows an analysis of the contexts of all -ci utterances between
two and three years. As Table 6 shows, reiteration of the preceding proposi­
tion with -ci (example 23) marks the beginning of the development. Between
2;2 and 2;4, the three children begin to use -ci in the following contexts: when
the child is certain about a proposition based on some perceptual support
(category c), and when the conversation progresses in such a way that the
child becomes certain of the proposition (category d).
The following examples illustrate these categories:
The Development of Epistemic Sentence-ending Modal 191

(24) PL(2;6)
(PL and Mother talk about whether neighbor houses have their
lights on or not.)
PL: emma. ce cip-ey-nun pwul-i kke-ss-e
mommy, that house-LOC-TOP light-suBJ turn off-PAST-SE.
'Mommy. Lights are off in that house.'
M: etil
where
'where?'
PL: yeki
here
'here.'
M: ung, ce wuit cip-ey.
yes that above house-LOC
'Yes, that house on top.'
PL: ung, wuit cip-ey.
yes above house-LOC
'Yes, the house on top.'
M: ung, ung.
yes yes
'Yes, yes.'
—» PL: ce alay cip-un pwul khye-cye iss-ci.
that below house-TOP turn-on-PASS be-SE.
'Lights are on in that house below.'
M: kule-kwuna.
SO-SE
'It is so.'
(25) HS(2;9)
(One coin chip is stuck in toy cash register)
Inv.: kelye-ss-cyana
stuck-PAST-SE
'(It's) stuck.'
HS: kelye-ss-e?
Stuck-PAST-SE
'(It's) stuck?'
 192 Soonja Choi

Inv.: ung.
yes
'Yes.'
HS: ike kocang na-ss-nunka pwa.
this obstacle arise-PAST-coNN seem
This seems to be broken.'
Inv.: kocang an na-ss-e
obstacle NEG arise-PAST-SE.
'(It) isn't broken.'
HS: kocang na-ss-e.
obstacle arise-PAST-SE
'(It)'s broken.'
Inv.: tasi ha-l-kkey.
again do-FUT-SE
'(I')H do it again.'
(Inv. tries but it still gets stuck)
HS: an tway.
NEG become
'(It) doesn't work.'
Inv.: an tway.
NEG become
'(It) doesn't work.'
→ HS: kocang na-ss-ci.
obstacle arise-PAST-SE
'(It)'s broken.'
In (24), the child maintains the current topic, and tells whether the lights are
on or off on the basis of what she sees, i.e. perceptual evidence. In this
example, PL first confirms with her mother that the lights are on in the upper
apartment. This discussion leads PL to draw the conclusion that the lights in
the lower apartment are off by contrast, and PL marks her certainty of this
knowledge by -ci. In example (25), being broken becomes more and more
certain to HS as the coin keeps getting stuck. Also, in (25), the meaning
kocang na-ta 'be broken' is expressed several times by the child herself
before marking it with the suffix -ci. This use of -ci is analyzed as certainty
developed through discourse (category d in Table 6). At this developmental
stage, certainty constructed through discourse is often accompanied by per-
The Development of Epistemic Sentence-ending Modal 193

ceptual evidence, as shown in (25). At a later time, -ci begins to be used in

another type of context (from 2;2 for PL and from 2;9 for HS and TJ): it co-
occurs with the obligation connective -ya or with the nominalizer ke, as in
(26) and (27).

(26) HS(2;9)
(HS putting coins one by one into the toy cash register.)
→ HS: yoke ha-ko yoke ha-ya-ci.
this do-then this do-must-SE
'(I) do this, then (I) must do this.'
(27) PL(2;2)
(PL playing doctor.)
→ PL: cinchal ha-myen ei wul-myen nappen ke-ci.
examine do-then ei cry-if bad NOMINAL-SE.
'It is a bad thing to cry ei (sound of crying) when (the
doctor's) examining.'
This represents category (g) in Table 6. The obligation connective and the
nominalizer in Korean denote obligation and normativity, respectively. This
is interesting because the two meanings are closely related in that when one is
obligated to do something or when one normally does something in a given
situation, the predicated event acquires a degree of certainty. In other words,
obligation and normativity often lead to certainty of an event occurring.
Studies of modality in adult language show that there is a close relationship
between obligation and certainty (Bybee 1985). Also, Steele (1975), in a
cross-linguistic study of modality, argues that for a systematic relation be­
tween deontic and epistemic modality. In particular, Steele's data show that in
many languages obligation and certainty of information are expressed by the
same morpheme. In Korean, modal forms for obligation and certainty are
distinct; however, the two forms frequently co-occur. The three children
acquired such co-occurrences from an early stage.
As in adult speech, the suffix -ci was frequently used in yes/no questions
asking the caregiver to confirm the truth of a proposition: yes/no questions
comprised 35% of -ci utterances in HS's speech, 25% in PL's speech, and
38% in TJ's speech (these are incorporated in the appropriate pragmatic
categories in Table 6). In addition, -ci was used in wh- questions. Initially, this
is suprising because wh- questions are asked when some particular informa-
194 Soonja Choi

Table 7. Number of-tay utterances in different types of context f rom its acquisition till 3;0.

HS PL TJ
(2;5-3;0) (2;l-3;0) (2;3-3;0)
a. what X would say or feel 3 44 7
b. reported speech 3 10 7
c. during storytelling - 9 9
d. errors with 1st person 1 1 1
Total -tay utterances: 7 64 24

tion is unknown. However, an analysis of the kinds of information requested

in wh- questions with -ci shows that they are based on certainty of some
relevant information. For example, when HS played an alphabet puzzle game
and was looking for certain letters, she routinely asked the question eti kass­
­i? where go-PAST-SE 'Where did (it) go?'. In this context, she knew that a
particular letter existed and the question was built on the certainty of its
existence, and of finding it.

4.1.3 Acquisition of-tay and -ta (Type 2): source of information

The three children used -ci to express shared information which led to the
expression of certainty. In the next phase of the development of modality,
children acquire modal forms whose primary function is to inform the listener
of something that the latter does not seem to know. The acquisition of two
new modal forms, -tay and -ta (Type 2), shows this further development.
In the adult grammar, the function of -tay is described as reporting
hearsay information or indirect speech (H. Lee 1985). For example, when a
speaker reports information indirectly obtained (e.g., heard from another
person, or read in a newspaper), the suffix -tay is used to mark the indirect
source of the information.
The children used -tay much less frequently than the three suffixes
analyzed above. Although the sample size is small, the data show a consistent
pattern. As Table 7 shows, -tay was used primarily in contexts where the
children relayed information whose source was a third party: (a) telling the
listener what the child believed the third party said or felt, or (b) reporting
what a third party just said. All three children used -tay in these contexts. In
both contexts, the child gives the listener information that the child assumes
the listener may not know. For example, in (28), PL engages in a pretend play
The Development of Epistemic Sentence-ending Modal 195

with a doll as a baby, and reports what the baby might feel. Obviously, the
baby (i.e. the doll) said nothing, but PL infers that the baby felt sick and
reports this to her mother, who may not share this information. A similar
explanation can be offered for (29): TJ infers that Ernie might hurt his feet by
jumping many times. In (30), HS reports to the investigator what her sister has
just said, appropriately with the suffix -toy.

(28) PL(2;1)
(PL plays with a doll with her Mother.)
→ PL: aka-ka aphu-tay.
baby-suBJ sick-SE.
T h e baby is sick.'
(29) TJ(2;4)
(TJ looking at a picture of Ernie jumping)
Inv.: jumphu-ha-e?
jump-do-SE?
'(Is he) jumping?'
→ TJ: pal ayaya ha-n-tay.
foot a y a y a (typical sound made when one is hurt) do-PRES-SE
'He says his feet are hurt.'
Inv.: ung.
yes
'Oh, I see.'
(30) HS(2;5)
(HS and her older sister are coloring.)
HS's sister: nayka saykchil hay cwu-kkey.
I-SUBJ coloring do give-SE.
T will color (it) for you.'
(HS immediately reports to Inv.)
→ HS: enni-ka saykchil hay cwu-n-tay.
sister-suBJ coloring do give-PRES-SE.
'(My) sister says that she will color (it) for me.'
As shown in the above examples, all of the utterances with -tay were reports
about what a third party was feeling or had said. That is, the information with
-tay did not come from direct experience but from inference or other indirect
sources (except for one error in each child's sample which was about the child
 196 Soonja Choi

herself). In addition, PL and TJ sometimes used -tay in narrating a story from

a picture book. This is partly due to the input. PL's and TJ's caregivers spent
much time reading picture books with the children, and they often used -tay in
telling stories.
When the information to be conveyed was about the child herself, an­
other suffix was used, the suffix -ta with a high pitch, -ta was used when the
information was already known to the child by direct experience, i.e. old
information for the child, but new for the listener. This use of -ta always
carried a high pitch (marked by T in the transcript).

(31) TJ(3;0)
(TJ showing the Investigator her new bracelet on the wrist.)
→ TJ: na phalci iss-ta↑
I bracelet exist-SE
'I have a bracelet!'
Inv.: TJ, phalci iss-e?
TJ, bracelet exist-SE
T J has a bracelet?'
TJ: ung.
yes
'Yes.'
(32) TJ(3;0)
(TJ tells the Investigator what she did before the Inv. came.)
→ TJ: akka-nun Cwunwu-lang pizza mek-ess-ta↑.
before-Top Cwunwu (TJ's friend)-with pizza eat-PAST-SE
T ate pizza with Cwunu awhile ago!'
Inv.: kula-e?
SO-SE
Ts that so?'
(33) HS(2;9)
(HS putting her sunglasses in her pocket, tells the Inv.)
—» HS: nay kyowhay ka-lttay yoke kac-ko ka-nunke-ta↑
I church go-when this take-coNN go-NOMIN-SE.
'This is something that I take to church!'
In all three examples above, the information that the child gave to the
investigator was new to the latter. The child marked this assumed knowledge
The Development of Epistemic Sentence-ending Modal 197

gap in the listener's mind with the suffix -ta. It should be noted that the
information with -ta (Type 2) was often something the child wanted to brag
about to the adult, thus it carried a certain affective component.
In summary, both -tay and -ta convey information the child believes the
listener doesn't have. The suffix -tay marks information that comes from an
indirect source: either inference about what another person would say or
would feel or a report of indirect speech. The suffix -ta (Type 2) with a high
pitch expresses something that the child experienced directly and which the
listener might not know. It is interesting to note that the productive uses of the
two suffixes, -tay and -ta (Type 2) all occurred later than -ci. This suggests
that an understanding of differences in knowledge between the child and the
listener develops after the child learns to express shared knowledge.

4.2 Mechanisms for the early acquisition of SE epistemic suffixes

The data of this study have shown that Korean children acquire and use
several epistemic SE suffixes appropriately at an early stage. Specifically,
between 1 ;9 and 3;0 the three children in the study acquired five SE suffixes to
mark distinct meanings. The meanings were distinguished in terms of the
following features: degrees of integration of a proposition in the child's
knowledge system, source of information, and the child's belief about what
the listener knows. In particular, the children in this study distinguished
between newly perceived information that was not yet assimilated into their
knowledge system and old/assimilated information (-ta (Type 1) vs. -e);
between knowledge shared vs. not-shared with the listener {-ci vs. -tay/-ta),
and between direct and indirect sources of information {-ta (Type 2) vs. -tay).
The data suggest that these oppositions are acquired in a consistent order.
Figure 1 attempts to illustrate this development. First, -ta is acquired to denote
newly perceived information that attracts the child's interest at the time of
speech. Then, -e is acquired to contrast the new knowledge with old/assimi­
lated knowledge. The new vs. old knowledge distinction is related to the
child's own knowledge system and does not seem to be related to whether or
not the knowledge is shared with the listener. However, as -ci is acquired, a
new component, the feature of shared knowledge, is added to the SE modal
system. The propositions with -ci in the children's speech predominently
contain information related to the shared topic: 85% of the -ci uses have
discourse contingency. In addition, -ci expresses knowledge shared by the
198 Soonja Choi

Form.- -ta(Type 1) → -e → -ci → -toy → -ta(Type 2)

Meaning: new/ old/ shared non-shared non-shared
unassimilated assimilated knowledge/ knowledge/ knowledge/
knowledge knowledge certainty indirect evidence direct evidence

Development of | >
Modal meaning: Degree of assimilation of knowledge in the speaker's mind
I >
Listener's status of knowledge about a proposition

Source of information
Figure 1. Development of epistemic functions

child and the listener. The nature of this shared knowledge is based on
discourse or perceptual evidence. This function of -ci leads to marking propo­
sition as certain. Later, as -toy is acquired, shared knowledge comes to be
differentiated from non-shared knowledge, i.e. the child informs the listener
of a proposition the latter may not know. Furthermore, -toy denotes informa­
tion that comes from an indirect source. Lastly, -ta contrasts with -toy in
marking a proposition as reflecting the child's direct experience. This devel­
opmental order suggests that as each new modal form and its corresponding
function are acquired, a new domain of epistemic modality is added to the
existing one(s), as shown in Figure 1.
One explanation for these results has to do with children's general
cognitive development and capacity at this stage. We have seen that the three
children acquired their first SE suffixes to express their own knowledge status,
and subsequently to incorporate the listener's knowledge status. This con­
forms to the general view that children's use of language develops from
egocentric to decentered (Piaget 1955). Also, a closer look at the contexts in
which particular propositions were expressed suggests that at the earliest
stage in the children's development of SE modals, propositions were related to
the information in the immediate context. In particular, information with the
suffix -ta (Type 1) was closely tied to what was going on at the time of speech.
Then, the children acquired -e and -ci to convey information about past
events. This parallels the pattern of a child's general cognitive development,
which proceeds from an understanding of present events to an understanding
of events removed from the present (i.e. past or future). Not only across
suffixes, but also with respect to a single suffix, the development of functions
The Development of Epistemic Sentence-ending Modal 199

reflects underlying cognitive capacity at the time of acquisition. For example,

the use of -ci was first limited to repetitions of the preceding propositions and
later expanded to include past events.
Overall, the perceptual context provided an important basis for the chil­
dren's acquisition of epistemic meaning of SE suffixes. For example, -ta was
used with situations that the child experienced at the time of speech, and -ci
was often used when the evidence was clearly present in the extralinguistic
context. Also, -toy was used to relay information about a referent that was
present in the immediate context. Such an evidential component which is
related to the here and now probably influenced the early acquisition of these
forms.
During the period of the study, not all SE suffixes were acquired by the
children. For example, they did not produce -kwun (unassimilated inference
based on newly perceived information). Furthermore, for those suffixes that
the children did acquire, not all the functions expressed in the adult grammar
were present. For example, -ci, which in adult grammar also expresses cer­
tainty of a proposition based on inference, did not appear often in the chil­
dren's data with this function. The indirect speech function of -toy was not
fully developed in the children's speech. Again, these limitations reflect
children's cognitive constraints at this developmental stage.
Within these limitations, however, the children acquired a variety of
epistemic meanings relating to different degrees of assimilation of knowledge
in their minds, different sources of information, and the knowledge status of
the listener. As shown in the present analyses, the contrasts at issue were made
systematically by the children. Such early acquisition of modality may be due
to the morphological saliency referred to above. But more interestingly, it is
argued here that the discourse interactional functions of the suffixes play an
important role in the early acquisition of epistemic modality. That is, under­
standing that the modal suffixes mark contrast concerning the exchange of
information and the construction of shared knowledge among conversation
participants may have enhanced the children's sensitivity to and acquisition of
different kinds of information. In particular, we have seen that the distinction
between new and old knowledge is related to discourse contingency. Also, the
notion of certainty of information is embedded in the expression of shared
knowledge. Example (24'), a fuller version of (24), illustrates clearly how the
functions of three SE suffixes, -ta, -e, and -ci, are related to the way discourse
interaction progresses.
200 Soonja Choi

(24') PL(2;6)
(PL and Mother have been engaged in a pretend play for awhile. PL
plays a student who came to study in Boston.)
M: kongpwuha-le o-si-ess-eyo.
study-coNN come-HON-PAST-SE.
'(You) came (here) to study.'
PL: ney.
yes
'Yes'
(PL sees a light in a neighboring house.)
→ PL: emma, ce, ceki yep cip-ey pwul khye-cye-iss-ta.
mommy there there side house-LOC turn on-PASS-be-SE
'Mommy, the light is on in that neighbor's house.'
M: ung. ce cip pwul khye-iss-kwuna.
yes. that house light turn on-be-SE.
'Yes, the light is on in that house.'
—» PL: emma. ce cip-ey-nun pwul-i kke-ss-e.
mommy, that house-LOC-TOP Iight-SUBJ turn off-PAST-SE.
'Mommy. Lights are off in that house.'
M: etil
where
'where?'
PL: yeki
here
'here.'
M: ung, ce wuit cip-ey.
yes that above house-LOC
'Yes, that house on top.'
PL: ung, wuit cip-ey.
yes above house-LOC
'Yes, the house on top.'
M: ung, ung.
yes yes
'Yes, yes.'
—» PL: ce alay cip-un pwul khye-cye iss-ci.
that below house-TOP turn-on-PASS be-SE.
'Lights are on in that house below.'
The Development of Epistemic Sentence-ending Modal 201

M: kule-kwuna.
SO-SE
'It is so.'
In this example, as PL sees the light, she gets excited and describes what
she sees, abruptly changing the topic from the pretend play to lights in the
neighbors' houses. PL encodes this newly perceived information with -ta.
Once the topic is established and the information registered in her mind, the
child describes other houses with the suffix -e. As the exchange of informa­
tion on lights being on or off progresses, PL changes to -ci to encode shared
information and its certainty. The three suffixes thus mark various types of
knowledge status that develop within a particular discourse interaction.
Early sensitivity to the discourse functions of SE suffixes in Korean
correlates well with the findings on the acquisition of discourse-pragmatic
phenomena for English-speaking children: children use language in request­
ing information and answering questions from the one-word period (Dore
1974; Shatz 1983). Several studies on discourse contingency between care­
giver and child have shown that children imitate caregivers' utterances from
early on. One function of imitation is maintaining a shared topic. Bloom et al.
(1976) shows that imitation is crucial to the development of the discourse skill
of contributing new information to a shared topic. Also, Pellegrini et al.
(1987) demonstrate that two-year-old children show signs of observing the
cooperative principles, e.g., giving informative and truthful information to
construct shared knowledge relevant to the topic of discourse. The present
study suggests that such an ability to follow the progression of discourse
toward more and more shared knowledge between the speaker and the listener
is instrumental to the early acquisition of SE modal suffixes. This would also
explain the late acquisition of pure epistemic modal forms whose meanings
are relatively context-independent (e.g., English epistemic modal auxiliaries).
Essential to children's sensitivity to discourse interactions is the affective
component that exists between the caregiver and the child. First, all of the SE
suffixes are used in informal conversations between participants who are
intimate with one another. For example, by repeating shared information
using -ci, and by checking and making sure that it is shared with the caregiver,
the child builds a bond between herself and the caregiver. Also, the suffix -ta
(Type 2) is used when the child wants to brag about something to the adult.
Such an affective component may enhance the acquisition of abstract con­
cepts in ways similar to what Akatsuka and Clancy (1991) describe in relation
 202 Soonja Choi

to the early acquisition of conditionals in Korean children. Thus, when the

epistemic modal system is deeply embedded in the protocols of discourse
interaction, which, moreover, contains an affective component, children may
grasp abstract concepts of modality from an early stage. The question still
remains, however, whether children learning languages other than Korean
spontaneouly make similar distinctions in their own way, and if so, to what
extent do children learning different languages systematically express the
distinctions we have seen in this paper. In order to answer this question, we
need to look at the acquisition of several different languages, focusing on how
children express different types of knowledge within discourse interactions.

ABBREVIATIONS

ADV - adverbial suffix CONN - connecting suffix

FUT - future HON - honorific suffix
LOC - locative NOMIN - Nominalizer
NEG - negative OBJ - object
PAST - past tense PL - plural
poss -possessive PRES - present
SE - sentence-ending suffix SUBJ - subject
TOP - topic

NOTES

1. In colloquial speech, sov order is not always observed. In consequence, SE suffixes may
occur in the middle of the sentence, as in the following example.
kho ca-n-ta thokki-ka.
well sleep-Present-SE rabbit-suBJ
'The rabbit is sleeping soundly.'
2. I am very grateful to Pat Clancy and Youngjoo Kim for allowing me to analyze HS'S and
PL's data respectively.
3. -e and -(y)a are allomorphs. -e occurs after high and mid-high vowels and after all
consonants. -(y)a occurs following low and mid-low vowels.
4. In this case, the full form is appa-i-ta (daddy-cop-SE), however, the copula verb i 'be' is
often deleted after a vowel.
5. -e is deleted after a vowel.
The Development of Epistemic Sentence-ending Modal 203

REFERENCES

Akatsuka, Noriko. 1985. "Conditionals and the Epistemic Scale". Language 61.625-639.
Akatsuka, Noriko & Patricia Clancy. 1991. "Conditionally and Deontic Modality in
Japanese and Korean". Paper presented at The Second Southern California Conference
on Japanese/Korean Linguistics. Santa Barbara, California.
Aksu-Koç, Ayhan. 1988. The Acquisition of Aspect and Modality: The Case of Past
Reference in Turkish. Cambridge, England: Cambridge University Press.
Aksu-Koç, Ayhan & Dan Slobin. 1986. "A Psychological Account of the Development
and Use of Evidential in Turkish". Chafe & Nichols. 1986. 159-167.
Bloom, Lois, L. Rocissano & L. Hood. 1976. "Adult-Child Discourse: Developmental
Interaction between Information Processing and Linguistic Knowledge". Cognitive
Psychology 8.521-552
Brown, Roger. 1973. A First Language. Cambridge, Mass.: Harvard University Press.
Bybee, Joan. 1985. Morphology: A Study of the Relation between Meaning and Form.
Amsterdam: John Benjamins.
Bybee, Joan. & William Pagliuca. 1985. "Cross-Linguistic Comparison and the Develop­
ment of Grammatical Meaning". Historical Semantics and Historical Word Formation
ed. by J. Fisiak, 59-83. The Hague: Mouton.
Byrnes, James P. & Michelle A. Duff. 1989. "Young Children's Comprehension of Modal
Expressions" Cognitive Development 4.369-387.
Chafe, Wallace & Johanna Nichols "eds." 1986. Evidentiality: The Linguistic Coding of
Epistemology. New Jersey: Ablex.
Choi, Soonja. 1991. "Early Acquisition of Epistemic Meanings in Korean: A Study of
Sentence-Ending Suffixes in the Spontaneous Speech of Three Children". First Lan­
guage 11.93-119■.
DeLancey, Scott. 1986. "Evidentiality and Volitionality in Tibetan". Chafe & Nichols.
1986. 203-213.
Dore, John. 1974. "A Pragmatic Description of Early Language Development". Journal of
Psycholinguistic Research 3.343-350.
Gopnik, Alison & Soonja Choi. 1990. "Do Linguistic Differences Lead to Cognitive
Differences? A Cross-Linguistic Study of Semantic and Cognitive Development".
First Language 10.199-215.
Gopnik, Alison & Andrew Meltzoff. 1984. "Semantic and Cognitive Development in 15
to 21 Months-Old Children". Journal of Child Language 11.594-513.
Hirst, William & Joyce Weil. 1982. "Acquisition of Epistemic and Deontic Meaning of
Modals". Journal of Child Language 9.659-666.
Kim, Young-joo. 1992. "The Acquisition of Korean". Unpublished manuscript.
Lee, Hyo Sang. 1985. "Consciously Known but Unassimilated Information: A Pragmatic
Analysis of the Epistemic Modal Suffix "-kun" in Korean". Paper presented at The
First Pacific Linguistics Conference, University of Oregon.
Lee, Hyo Sang. 1991. "Tense, Aspect, and Modality: a Discourse-Pragmatic Analysis of
Verbal Affixes in Korean from a Typological Perspective." Doctoral dissertation,
University of California at Los Angeles.
 204 Soonja Choi

Lee, Hyo Sang. 1993. "Cognitive Constraints on Expressing Newly Perceived Informa­
tion: With Reference to Epistemic Modal Suffixes in Korean. Cognitive Linguistics
4.135-167.
Lee, Keedong. 1986. "Pragmatic Function of Sentence Enders". Inmunkwahak (Humani­
ties) 56.41-59.
Moore, Chris, Kiran Pure & David Furrow. 1990. "Children's Understanding of the
Modal Expression of Speaker Certainty and Uncertainty and Its Relation to the
Development of a Representational Theory of Mind". Child Development 61.722-730.
Nichols, Johanna. 1986. "The Bottom Line: Chinese-Pidgin Russian". Chafe & Nichols
1986.239-257.
Palmer, F. R. 1986. Mood and Modality. Cambridge, England: Cambridge University
Press.
Pellegrini, Anthony, Gene Brody & Zolinda Stoneman. 1987. "Children's Conversational
Competence with Their Parents". Discourse Processes 10.93-106.
Piaget, Jean. 1955. The Language and Thought of the Child. New York: Meridian Books.
Shatz, Marilyn. 1983. "Communication." Handbook of Child Psychology, Vol 3, ed. by
John Flavell & Ellen Markman, 841-890.
Shatz, Marilyn. 1984. "Answering Appropriately: A Developmental Perspective on Con­
versational Knowledge". Discourse Development ed. by Stan Kuczaj. New York:
Springer-Verlag.
Slobin, Dan I. 1973. "Cognitive Prerequisites for the Acquisition of Grammar". Studies of
Child Language Development ed. by Charles A. Ferguson & Dan I. Slobin, 175-208.
New York: Holt, Rinehart & Winston.
Slobin, Dan I. 1985. "Crosslinguistic Evidence for the Language Making Capacity". The
Crosslinguistic Study of Language Acquisition, Volume 2: Theoretical Issues, ed. by
Dan I. Slobin, 1157-1256. Hillsdale, NH: Lawrence Erlbaum.
Steele, Susan. 1975. "Is it Possible?" Working Papers on Language Universals. Stanford
University. 18:35-58.
Stephany, Ursula. 1986. "Modality". Language Acquisition (2nd Edition), ed. by Paul
Fletcher & Michael Garman. Cambridge, Mass.: Cambridge University Press.
Tomasello, Michael. 1992. First Verbs. Cambridge, England: Cambridge University
Press.
Traugott, Elizabeth. 1989. "On the Rise of Epistemic Meanings in English: An Example
of Subjectification in Semantic Change". Language 65.31-55.
Wells, Gordon. 1979. "Learning and Using the Auxiliary Verb in English". Language
Development, ed. by Victor Lee. New York: John, Wiley and Sons.
Wells, Gordon. 1985. Language Development in the Preschool Years. Cambridge, Eng­
land: Cambridge University Press.
Woodbury, Anthony. 1986. "Interactions of Tense and Evidentiality: A Study of Sherpa
and English". Chafe & Nichols. 1986. 188-202.
Zoh, Myeong-Han. 1981. Hankwuk Atong-uy Ene Whoyktuk Yenkwu [A Study of Lan­
guage Acquisision in Korean Children] Seoul: Seoul National University Press.
 The Interactional Basis of the Mandarin
Modal néng 'CAN'*

Jiansheng Guo
University of California, Berkeley

1. Introduction

The paper explores the relationship between the interpersonal/discourse func­

tions of the Mandarin modal néng 'can' and its semantic meanings from a
functional perspective. The following statement by M.A.K. Halliday suc­
cinctly conveys the essence of functionalism:
A functional approach to language means, first of all, investigating how
language is used: trying to find out what are the purposes that language
serves for us, and how we are able to achieve these purposes through
speaking and listening, reading and writing. But it also means more than
this. It means seeking to explain the nature of language in functional terms:
seeing whether language itself has been shaped by use, and if so, in what
ways — how the form of language has been determined by the functions it
has evolved to serve. (Halliday 1973:7)

Pursuing the program outlined by Halliday, this paper attempts to eluci­

date how communicative function helps structure the semantic content and
influences the semantic change of a grammatical morpheme. Specifically, the
development of the Mandarin modal auxiliary néng in child speech shows that
the interpersonal function constitutes a crucial part of the meaning of the word
and has played an important role in changing its semantic content. In this
paper, I propose to demonstrate that in child discourse the basic function of
the modal néng is not merely to express notions such as ability, permission,
and epistemic possibility. Rather, its most important function is an interper­
sonal one, namely to index the speaker's attempt to challenge the addressee.
At the first stage of development in child speech, each of this modal's
206 Jiansheng Guo

meanings — abilitative, permissive, and epistemic — is deeply embedded in

the interpersonal tension between the speaker and the addressee. With respect
to dynamic modality, néng signals contested ability1; it is used to challenge
the assumptions the speaker attributes to the addressee about the speaker's
ability. With respect to deontic modality, néng primarily codes prohibitions,
challenging the addressee's actions. And with respect to epistemic modality,
néng figures prominently in argumentation, challenging the propositional
content of the addressee's previous statement. Thus, from the child speaker's
point of view, this interpersonal meaning of challenge constitutes an essential,
inalienable part of the meaning of the modal.
Previous studies on the semantics of modal auxiliaries have concentrated
on their referential meanings (Lyons, 1977; Palmer, 1990; Perkins, 1983).2
This tendency is representative of studies of the English modal can. The
meanings of can, notoriously elusive and controversial, are traditionally
incorporated under the three conventional modal rubrics: ability, permission,
and possibility (e.g. Palmer 1990; Lyons 1977, Coates 1983)3, roughly corre­
sponding to the domains of dynamic, deontic, and epistemic modality
(Palmer, 1990). Perkins (1983:35) proposes unified core meanings for the
English modals and defines can as 'C does not preclude X'. This is roughly
interpreted as: 'certain circumstances (C) exist that do not preclude the
occurrence of the event predicated by the main verb or the truth validity of the
proposition (X)'. This is essentially equivalent to the gloss 'it is possible that'.
According to Perkins, this core meaning determines the dynamic, deontic, or
epistemic readings, depending on whether it is applied to natural, social, or
rational laws. However, both the traditional notions 'ability', 'permission',
'possibility' (including derivatives such as 'it is permitted' or 'it is possible')
and the unified meaning proposed by Perkins, have in my opinion, obscured
the interpersonal dynamics essential to the meaning of can.
Researchers in cognitive linguistics point out the unique force-dynamic
nature of modals such as can. Talmy (1988) argues in detail that modals in
their core meanings contain intrinsic force dynamics, i.e., interactions be­
tween barriers and exertions of force. For example, can 't in the sentence He
can't move indicates that the subject has a tendency toward moving, but some
factor opposes that tendency. Applying the force-dynamics analysis to mo­
dals, Sweetser (1982) construes the basic meaning of can as 'enablement
against some potential resisting force'.
The Interactional Basis of the Mandarin Modal néng 'CAN' 207

However, these researchers have not gone far enough in specifying the
nature of the force dynamics of modals. They elaborate on the the forces that
are exerted, e.g. physical, social, or rational enablement. But they do not
clarify what kinds of counter forces are exerted. Sweetser (1982) says nothing
about the nature of the opposing force, while Talmy (1988) says only that the
opposing force is backgrounded. Yet, the very essence of force dynamics is
the tension between the two forces. I will argue that the opposing force is as
important as the initial force. However, the opposing force is more likely to be
found in discourse interactions than in semantic content. The tendency to
consider only the semantic content of the modals may be one reason why the
opposing force is hard to specify or regarded merely as backgrounded. I
suggest that only by considering discourse function as well as semantic
content can we fully understand the force dynamics of modals. Using the
example of children's use of Mandarin néng, I will argue in this paper that the
force dynamics of modals is motivated primarily by interpersonal tension in
interactional contexts.
The paper is organized in three parts. First, it will give a distributional
description of the meanings of néng across three different age groups, 3, 5,
and 7 year olds. The order of development of the different meanings of néng
will be shown to be:

physical ability → permission → epistemic uses

Second, it will show that all three meanings serve a common interper­
sonal function in dialogic discourse: challenge to the addressee. Third, it will
argue that change in the meaning of néng is primarily motivated by this
interpersonal function of challenge. The permission and epistemic meanings
emerge as a result of the absorption of this interpersonal function into the
semantics of néng.

2. Data

Mandarin Chinese néng translates into English roughly as 'can'. Syntacti­

cally, it is a modal auxiliary, distinct from main verbs. 4 Semantically, it is
similar to its English counterpart can, in that it is used both in a physical sense,
as in (1):
 208 Jiansheng Guo

(1) wǒ néng cóng zhèr tiào guò lál

'I can jump over from here.'
or in a social, permission sense, as in (2):
(2) zhèi shî dàifu de döngxi.nǐbù néng dòng dàifu de dōngxi.
'This is the doctor's thing. You can't take the doctor's thing.'
The epistemic meaning of néng is marginal compared with that of its
English counterpart.5 It is primarily used in rhetorical questions with sentence
-final question markers (e.g. ma, or ba), as shown in (3):
(3) gangtíñg néng zài jiā lĭ ma?
'Can a police station be in a home?' (as a rhetorical question)
In (3), it is difficult to distinguish the epistemic meaning from the root
possibility meaning.6
Néng shares elements of meaning with certain other Mandarin forms.
The modal këyï 'permitted, allowed' can be used to express the same physical
abilitative and social permission meanings as néng, but kěyǐ is used only in the
positive form, not in the negative7 or V-not-V question form.8 Only néng can
express negative physical ability meaning. The negative social permission
meaning can be expressed by the modal xǔ 'allowed', but xǔ has a strong
performative sense which marks the speaker as the deontic source. Thus it is
often used by adults to give prohibitions to children but is rarely used by
children. There are two Mandarin modal adverbs that can express so-called
epistemic possibility, kěnéng 'possibly' and yěxǔ 'maybe'. But these have a
specific function of hedging asserted propositions, which is different from the
meaning and function of néng.
The data used for this paper were collected in a full-time preschool and
an elementary school in Beijing, China. Subjects were three 3-year old, three
5-year old, and three 7-year old Mandarin-speaking children. In each age
group, the teacher helped to select three children who were regarded as natural
playmates. The children were of both sexes. The experimenter took each
group to a separate room in the school and instructed the children to play
together with toys provided by the experimenter. The children were engaged
in four different activities: lego construction, doll play, play dough, and puzzle
solving. Each visit lasted about one hour. The data consist of 5 hours of video
recording for each age group.
The Interactional Basis of the Mandarin Modal néng 'CAN' 209

Table 1. Frequeny of néng utterances by age group

Age 3 5 7
Frequenty 92 74 90

The recordings were transcribed, and all utterances by children contain­

ing néng identified for analysis. Table 1 gives the frequency in néng utter­
ances in each age group.
Four uses of néng have been identified according to the semantics of the
word in utterances: (1) unambiguous physical meaning referring to physical
abilities and characteristics; (2) unambiguous social meaning referring to
social permissions; (3) cases ambiguous between physical and social mean­
ing; and (4) epistemic-like meaning. The physical meaning category includes
uses traditionally subsumed under dynamic modality, i.e. uses referring to the
physical ability or conditions enabling an agent to perform certain acts, or to
the physical characteristics that enable an object to have certain functions in
specific tasks. The social meaning category includes uses traditionally classi­
fied under deontic modality; the source of modality is of a social nature, either
the existing social conventions or the participant's authority or wish. The
ambiguous meaning category includes uses in which the modality could be
interpreted as either physical or social. In the epistemic-like category are uses
involving what we generally think of as epistemic modality; the -like suffix is
added to show that this is not a robust or well-established category (see
Section 6 for details). Further elaboration and examples will be given in
Sections 3 through 6. Table 2 gives a breakdown of the different meaning
categories by age.
From this breakdown we see that about two thirds of the 3 year olds' uses
exhibit pure physical meaning, while social meaning is sporadic. In contrast,

Table 2. Breakdown of semantic categories of' néng by age

Physical Ambiguous Social Epistemic-like
Age meaning meaning meaning meaning N

3 72% 19% 9% 92
5 45% 12% 43% 74
7 52% 11% 26% 12% 90

Note: The total percentage exceeds 100 due to rounding in calculation.

210 Jiansheng Guo

the social meaning in the 5 year olds' uses shows a much higher frequency,
making up 43% of total uses. This indicates that by age 5, néng has acquired a
new meaning of social permission. The 7 year olds' uses of néng are compara­
ble to those of the 5 year olds in the physical and ambiguous categories, but
social meaning is used significantly less than among 5 year olds. But the 7
year olds seem to be developing a new meaning, the epistemic-like meaning.
In the next section, I will discuss in detail the discourse functions and
semantic content of each category, and argue for a functionalist explanation of
the semantic changes of the modal in child development. I will focus on how
néng is used in interaction, and on the discourse functions this morpheme is
intended to serve. For this purpose, the néng utterances within each meaning
category will be further categorized according to discourse functions.

3. The physical meaning category

I have divided the physical meaning category into three functional

sub-categories: (1) challenge to the addressee's assumptions, (2) justification
for prohibitions, and (3) reportive uses. In what follows, I will focus on the
first two sub-categories to illustrate the challenge function of néng and then
contrast these with the third sub-category, reportive uses, which does not
serve this function. In consideration of space, only one example will be
presented and discussed for each sub-category, which, nevertheless, are repre­
sentative of other utterances in the sub-category in question.

3.1. Challenge to the addressee's assumptions

The challenge to the addressee's assumptions category may be further classified

into three sub-categories: boasting, new discoveries, and problem reporting.

3.1.1. Boasting Examples of boasting may serve as prototypical examples to

illustrate the challenge function of néng in discourse. In boasting, the child
speaker tries to show off to the addressee (primarily the adult experimenter)
special abilities that adults do not usually expect children to have. A typical
boasting use is shown in (4), where adult experimenter Guo's lack of confi­
dence in the children's abilities, as indicated by his question 'What else can
you make?', prompts a series of boasts from the 3 year olds (C1, C2, and C3):
The Interactional Basis of the Mandarin Modal néng 'CAN' 211

(4) C1: (holds up a lego construction and shows Guo)

zhèige shì wo dä de .
'It is me who built this one.'
GU: shî ma?
'Is that right?'
C1: shì.
'Yes.'
GU: hĭ hái néng dā shénme ?
you still can build what?
'What else can you build?'
C1: wo hái néng dā bǐnggān ne.
I still can build biscuits PRTCL 9
'I can even build biscuits.'
GU: ao!
'Oh, I see.'
C3: wo hái néng dā miántiáo ne.
I still can build noodles PRTCL
'I can even build noodles.'
GU: zhënde a ?
'Really?'
C2: wo néng dā lóufáng.
I can build buildings
'I can build buildings.'
GU: zhënde a ?
'Is it really?'
C1: wo hái néng dā jīdàn ne. ni kán, shūshū.
I still can build eggs PRTCL. you look, uncle
'I can even build eggs. Look, uncle.'
C1: (shows lego construction to Guo)
In order to tease out the meaning and function of néng in these utterances,
it is appropriate at this point to define my use of certain terminology. I use the
term 'semantic meaning' or 'propositional meaning' in the sense of Lyons'
descriptive meaning, which carries 'factual information' that 'purports to
describe some state-of-affairs' (Lyons 1977:50), or Ogden and Richards'
(1923) referential meaning, as presented by Lyons (1977:175), which identi­
fies 'any object or state-of-affairs in the external world'. In this sense, the
212 Jiansheng Guo

Table 3. Features of challenge to assumption uses of néng

Predicated situation: jump over microphone wire
jump from high above by parachute
build a big truck with lego
build biscuits with lego
etc.
Special markers: emphatic adverbs: hái (even)
attention getters : nǐ kán ! (Look) ;
shüshü (Uncle)
gestural indicator: long eye engagement
Speaker action: act out or bring about the predicated situation
Challenging intonation: high-pitched level intonation over the whole utterance.
(vs. the normal falling intonation at the end of the utterance)

semantic meaning of néng in the above utterances evidently refers to the

physical ability of the speaker to perform predicated actions.
The above characterization of the semantic meaning of néng in these
utterances coincides precisely with traditional accounts of the semantics of
the Mandarin néng and English 'can'. Indeed, if the utterances were extracted
from their discourse contexts, nothing more could be said about their mean­
ing. However, in interactive discourse, néng has another layer of signification
at the interpersonal level. The speakers of these utterances are not merely
presenting a neutral claim about their abilities; rather, they are trying to
convince the addressee that they are much more competent than the addressee
thinks. In other words, they are trying to challenge the addressee's assumption
about their abilities. This challenge constitutes the discourse function of néng
and the utterances that contain it. I use the term 'discourse function' in the
sense of interpersonal or affective function, as discussed by Lyons
(1977:50-51), i.e. how the speaker intends to influence the mental state or
behavior of the addressee.
What empirical evidence is there that utterances like (4) indeed contain
challenges of this nature? A challenge implies intense interpersonal interac­
tion and conflicting beliefs. And in effect, there are various indicators in the
néng utterances in my corpus of intense interpersonal engagement and con­
test. Table 3 lists several linguistic, para-linguistic, and behavioral features
associated with the néng utterances to substantiate this claim.
First, the utterances refer to situations that are surprising, i.e. accomplish­
ments not usually assumed of children, such as 'jumping over a microphone
The Interactional Basis of the Mandarin Modal néng 'CAN' 213

wire', or 'parachute jumping from high above'. In addition, many of these

utterances contain the adverb hái, roughly translated as 'even', indicating that
the designated ability is beyond normal expectations, and in particular the
addressee's expectation. Third, the speakers always try to fully engage the
addressee, using attention getters, such as 'uncle' or 'look' to secure the
addressee's attention before making the utterance. Furthermore, the speakers
consistently perform the actions in question while making the néng utter­
ances, as if to give evidence for the claim. In addition, these utterances are all
made using a sustained high-pitch intonation contour over the whole utterance
that indicates challenge or contest. Examples in (5) schematically contrast the
challenging intonation contour to the normal stative intonation contour with­
out affect.
(5) a. Challenging intonation: wo hái néng tiào shéng ne.
'I can even jump rope.'
b. Stative intonation: wo néng tiào shéng.
'I can jump rope.'
All the above indicators suggest that these utterances fulfill a specific
discourse function of challenging an assumption of the addressee. But in order
to validate this claim, one more question needs to be answered: 'What
evidence is there of an assumption on the part of the addressee that the child
speaker deems challenge-worthy?' This challenge-worthiness may some­
times surface in the discourse context, as shown in (6):
(6) GU: (instructs C2, who does not know what to do next)
nǐ kěyǐ bāng zhe biérén.
'You can help others.'
C1 : wo bù yông biérén bāng zhe, wo jiù néng
I not need others help PROG I even can
pin chü lái.
piece-together out come
T can simply piece (them) together without the help of others.'
In this example, C2 finished her portion of the puzzle quickly. Guo then
told her that she could help the other children finish theirs. C1 heard this as an
underestimation of his ability to put together puzzles, and produced the néng
utterance to counter that underestimation. Whether Guo intended his utter­
ance as C1 perceived it is not an issue here; all that matters is that the child
214 Jiansheng Guo

speaker interpreted it that way. More often, however, the challenged assump­
tion is not explicitly verbalized. But the speaker believes that the addressee
holds the assumption. When children say that they can do X with the focus on
ability, they believe that their ability is at issue. In general, these beliefs do fit
with adult expectations of children's abilities. Most cultures regard children
as weak, not fully competent, fragile, and in need of adult protection; and this
view of children inheres in the belief structure underlying the néng utterances
produced by the Mandarin-speaking children in this study. 10 Néng utterances
provide a strategy for signaling contradiction of the assumed adult beliefs,
thereby constituting an explicit situation of force dynamics. The presupposed
addressee's assumption constitutes the force that will presumably continue if
it encounters no obstacle; the néng utterance constitutes the counter-force that
challenges and seeks to block that assumption. I gloss this discourse function
as 'Do not assume X', thereby emphasizing that it is highly dynamic and
transitive at the interpersonal level.
While I am convinced that the discourse function of néng is to challenge
the addressee's assumptions, I am not claiming that this is also its semantic
meaning (in the sense discussed earlier). In (6), a clear sense of physical
ability is referred to by néng, which serves as the focus of the interpersonal
force dynamics. Following Lyons' definition of semantic meaning (his de­
scriptive meaning) as referring to an event or state of affairs in the external
world, I take the semantic meaning of néng in the category of boasting to be
physical ability.
Although the discourse function and the semantic meaning of néng are
separate components, the two are closely interrelated. The obstacle, in the
semantic meaning, does not merely exist in the objective world, but has a
deeper social and psychological origin in the presupposed addressee's expec­
tations, which the child speaker seeks to challenge. It is here that the abilita-
tive meaning and the discourse function of challenge find their connection.
In summary, the semantic meaning of néng in boasting utterances is that
the doer has the ability to overcome an obstacle. In contrast, its discourse
function is to counteract the addressee's assumption concerning this ability.
At the semantic level, force dynamics operates between ability and a diffi­
culty in the physical domain. At the pragmatic level, it operates between the
addressee's expectation and the speaker's challenge to that expectation in the
interpersonal domain. The meanings at the two levels are intrinsically con­
nected. The physical obstacle has social and psychological origins in the
The Interactional Basis of the Mandarin Modal néng 'CAN' 215

presupposed assumption of the addressee(s). Thus the use of néng has to be

understood in terms of both its referential content and its interpersonal function.

3.1.2. New discoveries The second subcategory of challenge to the address­

ee's assumptions is new discoveries. (7) is a typical example of this use:
(7) C1 : (holds up lego car to Guo and rotates wheel)
nǐkàn, wǒ lúnzi néng zhuàn ne.
you look, my wheel can turn PRTCL(=even)
'Look, my wheel can even turn.'
(rotates wheel again and then gives it to C3)
This subcategory is similar to boasting in both semantic meaning and
pragmatic function. But here the focus is on the patient, namely the toys the
child is playing with, rather than on his or her ability. A typical situation is one
in which the speaker plays with a toy, and suddenly discovers an unexpected
feature or an innovative way to play. The speaker then shows off this newly
discovered feature. Thus, the discourse function of these utterances is to chal­
lenge, and consequently change, the addressee's expectations about the toys.

3.1.3. Problem reporting. The third subcategory under challenge to the ad­
dressee's assumptions is problem reporting. Here, the child reports to the
adult addressee on a problem in the current activity. This may be a negative
statement, as in (8):
(8) C1: (tries to make a lego figure stand; it falls; looks at Guo and
shows lego figure)
zhè yàngzi bù néng zhàn zhe
this manner not can stand PROG
'(It) can't stand like this.'
or a question about the characteristics of the object, as in (9):
(9) C2: (doesn't know how to open cotton bundle, holds it up to Guo)
zhè néng dă kāi ma ?
this can make open Q
'Can this (be) opened?'
These utterances are not made merely to report a problem. Their goal is to
solicit the addressee's help. And help will be forthcoming only after a change
216 Jiansheng Guo

in the addressee's knowledge about the toys. Thus, utterances in this subtype
have a double discourse function: (1) to change the addressee's assumptions
about the object, and (2) to get help.
At least two pieces of evidence point to this double function. First, the
children produce these utterances in a matter-of-fact manner while showing
the objects to the addressee, as if to say 'Look, this toy is not working as you
expect. So help me get it to work.' Then, they simply wait for help. Second,
the sorts of problems reported in these utterances always relate to a normally
expected function of the toys in question. Since the toys are provided by the
experimenter, the children naturally hold the latter responsible for the failure
of these normal functions to occur. For the same reason, they assume that the
experimenter expects the toys to function normally. Thus, a challenge to that
assumption is a prerequisite for soliciting help.
I emphasize that these uses are not plain statements or questions; children
use a different construction when they encounter difficulties with their own
possessions. E.g., if they have difficulty untying their shoe laces, they will
produce utterances like (10):
(10) wo jië bù käi zhèi xiédài.
I untie not open this shoe-lace
'I can't untie this shoe lace.'
In (10), the subject is the agent (the speaker), and a Verb+Resultative
-complement construction is used instead of néng. This construction attributes
the failure entirely to the inability of the agent.

3.2. Justification for prohibition

The second physical meaning category is justification for prohibition. A

typical example is shown in (11):
(11) (C3 puts toy banana into mouth and pretends to eat)
GU: bié bă zhège fàng zuǐ lǐ, häo ma?
'Don't put this into your mouth, OK?'
Cl: yào zhèyàngfàng zuǐ lǐ jiù bù néng wár le
if (it) is put into mouth like that, then not can play PRTCL
'If you put it in your mouth like that, it can't be played with any
more.'
The Interactional Basis of the Mandarin Modal néng 'CAN' 217

Uses of this type always occur where the adult forbids one child to do
something, and then another child makes a néng utterance to justify the adult's
prohibition. In this category, the néng utterance is always negative and always
refers to an undesirable state (e.g. loss of an existing ability) that the prohib­
ited action might engender. The semantic meaning of the utterance is 'inabil­
ity of object Z to perform function X'. The interpersonal discourse function of
the utterance is again two-layered. On the one hand, it challenges the address­
ee's assumption that the action has no negative effect on the object. On the
other, it functions as part of a prohibition, a challenge to the addressee's
action.

3.3. Reportive uses

In the 'reportive use' category, néng appears in utterances which impart

information without personal affect. Reportive néng does not actively index
the speaker's stance in interpersonal communication. It may be used in an
utterance that supplements another (but is not necessarily grammatically sub­
ordinate to it), e.g. by providing a reason or motivation for it, as shown in (12):
(12) C2: (no longer wants a lego, is about to throw it to C3)
zhèi bù néng yòng le. gěi nǐ ba.
this not can use PERF give you PRTCL
'This cannot be used any more. (I) give it to you.'
Here the speaker is not trying to change the addressee's belief with the
néng utterance. The utterance simply provides a rationale for her action. In
this use, the speaker does not present the situation as a potential conflict, since
she simply throws the lego to the other child without waiting for response.
Reportive néng can occur in questions, answers, and statements. The basic
criterion is that néng does not privilege an interpersonal message.
As a summary of the discussion on the physical meaning uses of néng, I
present the breakdown of the frequency of each subcategory for each age
group in Table 4.
From Table 4, we see that the majority of néng utterances in the physical
meaning category do not present the agent's ability or the physical character­
istics of an object as a static entity. Instead, they present them primarily in
contexts where an obstacle exists. Furthermore, this obstacle is derived from
the child's assumptions about what the addressee believes, which constitutes
218 Jiansheng Guo

Table 4. Breakdown of discourse functions of the physical meaning category of néng by age
Challenge to Justification for Representational
Age Assumptions Prohibitions Uses N
3 52% 8% 10% 92
5 42% 3% - 74
7 38% - 14% 90
Note: N is the total occurrences of néng, including social and epistemic meanings.
Therefore, the percentages in each age group do not add up to 100%.

the force which the néng utterance is intended to counter at the interpersonal
level. In other words, néng is used to overcome an assumed doubt (on the part
of the addressee) about physical abilities and characteristics.
Essential to an understanding of the modal néng is an understanding of
both its referential meaning and its interpersonal functions. In these néng
utterances, the addressee, although unmentioned, is always implicitly present
in the discourse interaction. To emphasize this presence at the discourse
(interpersonal) level, I gloss the discourse function of this category as 'Don't
assume X', which emphasizes that néng is highly social and transitive.
Maintaining the traditional distinction between semantic and pragmatic
meanings, we can say that in its physical sense, néng refers semantically to
physical ability, but pragmatically it indexes interpersonal challenge in an
active and intentional manner.

4. The social meaning category

Unlike the physical meaning uses, the social meaning uses no longer convey
'social meaning' pragmatically, but express it directly. The physical ability
meaning is no longer relevant. When a speaker tells an addressee that the
latter cannot do X, it is obvious from the situation that the addressee has the
ability to do X, and that this ability is either actualized or about to be
actualized. The only meaning néng has here is social permission. Social
meaning uses of néng are rare among 3 year olds, but very frequent among 5
year olds. The uses may be divided into four groups: (1) prohibitions; (2)
permissions; (3) reportive uses; and (4) self regulations.
The Interactional Basis of the Mandarin Modal néng 'CAN' 219

4.1. Prohibitions

Prohibition is the major social meaning use, especially among the 5 year olds.
A typical example is shown in (13):
(13) C1: m bu néng dă rén.
you not can beat people
'You can't beat people.'
ba rén tóu da huài le,bătāde tóu kăn xiàlái, gëi rénjiā, duî ma?
'(if) her head is hurt, (then) his head will be chopped off and
given to her. Right?'
The form of the utterance differs here from that of physical meanings. In
social meanings, the subject is the second person and the sentence is negative.
Utterances in this category are of a specific form: 'You cannot VP.' In
contrast to the physical uses, for social uses, semantic meaning and discourse
function are identical: preventing an action from taking place, which is
glossed as 'Don't do X.' This identity between semantic meaning and dis­
course function may explain why uses of this category are referred to as
discourse-oriented uses (Palmer, 1990). Since the 3 year olds seldom use néng
with this social meaning while the 5 year olds do so quite frequently, we
should regard this use category as a new development in the meaning of néng
(see Table 2 for a breakdown).

4.2. Permissions

Seeking permissions is a marginal use in the current corpus; only a few néng
utterances from the 7 year olds can be thus categorized.11 An example of this
use is shown in (14):
(14) C2: (wants to get water from restroom for toy tea set. asks Guo)
néng jië diâr shuǐ qù ma ?
can get little water go Q
'Can (I) go get some water?'
GU: mm, bù néng jiē diănr shut däi huǐr nöng shï le , jiù gāi.
'mm, (you) cannot go get water. You'll get (the floor) wet.'
220 Jiansheng Guo

With this meaning, the néng utterance presupposes a social constraint or

obstacle that will block the speaker's action, in this instance the children's
understanding that once they start the play session, they are not allowed to
leave the room without permission. The discourse function of néng utterances
of this type is an attempt to remove the social constraint. Its semantic meaning
refers to the same goal. But note that no interpersonal challenge is involved,
since the child does not attempt to defeat the constraint, but simply inquires
whether the authority will temporarily lift it.

4.3. Reportive uses

In reportive utterances, néng is clearly used in the sense of social permission,

but it merely reports an event involving a social constraint without attempting
to regulate the participants' behavior. In this category, néng is usually used in
the descriptive part of utterances, as in (15):
(15) C3: (The teacher has told C1, C2, and C3 to stay in the classroom
after the experiment session if the rest of the class is gone to
outside. C3 sees the class leaving for outside)
ei ya , tarnen zŏu le.
'Oh uh, they are all leaving.'
zánmen wár wán bù néng xià qù le,
we play finish not can down go PERF,
zěnme bàn ya?
how do Q?
'(When) we finish playing, we cannot go down any more.
What should we do?'
A report of a third person's directive is shown in (16):
(16) C1: (pretends to kiss a doll)
wǒ gēn tā qīn zuī ne.
'We are kissing.'
C2: lăoshǐ shuō bù néng shuö qïn zuǐ de huà.
teacher say not can say kiss mouth REL talk
'The teacher said that (we) cannot talk about kissing.'
From the context in (16), the illocutionary force of the néng utterance
appears to be to criticize C1 or stop him from talking about kissing. It looks
The Interactional Basis of the Mandarin Modal néng 'CAN' 221

rather like prohibition uses. However, the prohibition is conveyed by report­

ing a norm, i.e. what the teacher said before. The modal néng is embedded in
the quotation from the teacher, and thus does not privilege the interpersonal
message from the speaker. For this reason, these utterances are categorized as
reportive uses.

4.4. Self regulations

This category, too, is only used by the 7 year olds. Structurally, it is similar to
the prohibition category, but differs in that the (explicit or implicit) subject of
the néng sentence is the first-person speaker. The prohibition is self-directed.
An example of this category is given in (17):
(17) C3: (gives C2 a toy, then changes her mind, shyly withdraws toy
from C2)
bù, bù néng gel nǐ zhèige.
no, not can give you this.
'No, (I) can't/shouldn't give you this.'
In this example, the child speaker has internalized the force dynamics
(i.e. the two opposing forces come from the same person), which elsewhere is
socially distributed. In other words, the social opposition that elsewhere exists
between two people has here become a psychological opposition in which the
self is split, representing two opposing forces simultaneously.

5. The ambiguous case category

Ambiguous cases include those néng utterances in which a physical-ability

interpretation and a social-permission interpretation are equally reasonable.
The physical-characteristics meaning of néng may in fact license the social
permission meaning, or, conversely, social permission is presented as if it
were part of the physical characteristics of the object in question. Thus the
meaning in this category is intrinsically ambiguous.12 There are four
sub-categories of ambiguous utterances, which differ in their discourse func­
tions: (1) challenge to prohibition; (2) permission seeking; (3) permission
giving; and (4) prohibition.
222 Jiansheng Guo

5.1. Challenge to prohibition

The use of néng to challenge a prohibition typically occurs in explicit con­

frontations. The experimenter utters a prohibition or makes a comment about
an inappropriate action, and the child responds with a challenge containing
words like gànmâ 'how come', wèishénme 'why', or the challenge question
marker ya (roughly 'what you just said is highly questionable and totally
against my expectation'). The following is an example:
(18) C2: (does not know how to open a cotton bundle, holds out bundle
to Guo)
zhège zěnme dă käi ya ?
'How to open this?'
GU: zhège bù néng dă käi.
'This can't be opened.'
C2: bù néng da käi gànmâ ?
not can make open how-come
'How come it can't be opened?'
In the physical-ability meaning category discussed earlier, the challenge
is generally presupposed by the child and thus implicit. But in this category,
the confrontation is on record and represents a head-on defiance of the
addressee. In terms of semantic meaning, néng here is a direct copy of the
previous adult use. In the adult use, the sentence structure OBJ+not+réng+V
makes the physical-ability interpretation and the social-permission interpreta­
tion equally possible. It may refer to the physical characteristics of the object;
thus, in (18) the cotton bundle cannot physically be opened. Or it may refer to
the adult's authority, i.e. he doesn't allow it to be opened. This ambiguity
cannot be resolved, since both interpretations are licit, and the physical
condition is often the reason for the social prohibition.

5.2. Permission seeking

In this category, néng is used to ask for permissions to act. The sentence
structure is again in the OBJ+néng+V+Q form. The focus is on the object,
while the agent is backgrounded. Although from the context it appears that
néng is basically concerned with social permission, the structure of the
sentence closely associates the predicated meaning with the object, as if
The Interactional Basis of the Mandarin Modal néng 'CAN' 223

permission were part of the intrinsic characteristics of the object, similar to its
physical characteristics. An example of this category is shown in (19):
(19) C1: (helps Guo put lego into the box during clean-up, picks up a
lego car, asks)
zhè néng fàng ma ?
this can put Q
'Can this (be) put in?'
GU: á? néng.
'What? Yes, it can.'
This category is essentially the same as the permissions category, with a
clear social meaning. The difference is that here néng is not completely
dissociated from its source in the physical world.

5.3. Permission giving (compliance)

This category is a special type of assenting, consisting only in compliance

with the addressee's previous request. Here the actor is the speaker, while in
the general sense of permission giving, the actor will be the addressee or a
third party. The following is an example of the compliance use of néng:
(20) GU: (toys dropped onto floor, Guo to C1)
néng gěi jiăn qï lái ma?
'Can (you) pick it up (for me)?'
C1: néng ya.
can PRTCL(=of course)
'Sure, (I) can.'
In this use, although the speaker complies with the request, it is not clear
whether néng refers to the physical ability which makes compliance possible
or to the speaker's consent itself. The social permission meaning is not clearly
differentiated from the physical meaning.

5.4. Prohibition (focus on objects)

This category has the same discourse function as the prohibition category
discussed in 4.1., but the referential meaning of néng is ambiguous. A typical
example of this category is given in (21):
224 Jiansheng Guo

(21) C2: (C1 plays daddy role. C2 takes Cl's playdough)

gěi wǒ.
'Give (it) to me.'
C3: (to C2)
nèi shì baba de dōngxi, bù néng yào.
that is daddy 's thing not can take
'That belongs to daddy and (it) can't (be) taken.'
From the sentence structure, we can see that néng is closely associated
with the object rather than the agent, suggesting that the social convention
whereby one should not take daddy's belongings is an inherent characteristic
of the object rather than an inappropriate act. Since the social meaning is not
clearly differentiated from the physical meaning, these utterances are classi­
fied as ambiguous.

6. The epistemic-like meaning category

This category is called epistemic-like because the epistemic meaning of néng

in Mandarin is marginal. Tsang (1981) claims that néng cannot be used
epistemically at all. However, Lü (1980), a more comprehensive and authori­
tative source, cites examples showing néng used in root possibility and
epistemic possibility senses in questions and negative utterances. In the
current corpus, the 7 year olds produced some néng utterances with an
epistemic-like sense, all in the form of questions. Néng itself may be positive
or negative, and the epistemic meaning is not clearly differentiated from root
possibility or sometimes even from the general root meaning. Thus we cannot
treat these as having full-fledged epistemic meaning. However, as will be
clear in later discussion, néng does communicate the speaker's evaluation of
conclusions in relation to evidence. Thus we are justified in speaking of
epistemic uses. Since this meaning is only used by the 7 year olds, we will
treat it as a new semantic development. A typical example is shown in (22):
(22) C2: (looks for pot and cannot find it)
zhèi guō méi le.
'This pot has disappeared.'
C3: (turns around to look for it, to C2)
bù néng ba ?
not can Q(=tentative)
'(That) can't be, (can it)?'
 The Interactional Basis of the Mandarin Modal néng 'CAN' 225

In this example, néng is used clearly in an epistemic sense. C2 makes an

assertion, and C3 casts strong doubt on C2's conclusion. A more controversial
example is (23):
(23) C2: (puts lego on board)
gängting , zhèige .
'This is the police station.'
C3: (turns to C2, challengingly)
gängting néng zài jiä lǐ ma ?
police station can at home inside Q ?
'Can a police station be in a home?'
In this example, néng can be interpreted as referring to a social conven­
tion, that is, whether police stations can be located in a home. However, in this
use, C3 is not trying to argue that it is right or wrong to locate police stations
in private homes; rather, she is questioning the correctness of C2's claim. The
issue is not one of social appropriateness, but rather the validity of C2's
assertion given C3's knowledge of the cultural convention at issue. Thus the
discourse function of disvalidating the addressee's previous statement gives
rise to the interpretation of the semantic meaning 'A police station cannot
(possibly) be in a home'. As the examples in (22) and (23) show, epistemic
uses of néng occur in a discourse frame of confrontational argumentation.
Their discourse function of challenging the addressee's previous claims gives
rise to the epistemic-like semantic meaning.
The question arises whether the challenge comes from the rhetorical
question form or from the modal néng. Admittedly, a part of the challenge
does come from the rhetorical question, since challenge is the primary illocu-
tionary function of that sentence form in Mandarin. Rhetorical questions also
include an important prosodic feature: sentence stress on the verb, or modal, if
there is one. Thus, in (23) the main stress is on néng. It cannot be placed on
other sentence constituents without changing the meaning. Since modals
always receive the main stress in rhetorical questions like (23), and since
rhetorical questions are typically used in arguments or challenges, the dis­
course function of challenge is carried over into the meaning of the modal.
This hypothesis is particularly compelling for néng since it specializes in the
challenge function in other meaning domains.
Thus, the argumentative sentence form and the interactional context
together provide the semantic source for the epistemic meaning of néng. It is
important to note, however, that this epistemic meaning is not always depend-
226 Jiansheng Guo

ent on the rhetorical question form. In the context given in (22), C2 goes on to
answer C3's challenge by asserting néng ya 'It can (be the case), of course', as
shown in (24):
(24) C3: (turns around to look for it, to C2)
bù néng ba ?
not can Q (=tentative)
'(That) can't be, (can it)?'
C2: (matter of fact manner)
néng ya !
can PRTCL (=of course)
'Of course (it) can.'
Thus, although the epistemic meaning is elusive, it does not necessarily
depend on a particular sentence form. What it depends on at this stage of
development is considerable contextual framing, i.e. challenges and counter
challenges in argumentation.
To conclude the discussion of the data, all the meanings and functions of
the Mandarin modal néng are shown in Figure 1, And Table 5 gives a break­
down of the frequency of each subcategory across all three age groups.

Table 5. Breakdown offrequencies for all subcategories of néng by age

Physical Meaning Ambiguous Social Meaning Epistemic
Age 3.1 3.2 3.3 5.1 5.2 5.3 5.4 4.1 4.2 4.3 4.4 6.0

3 52% 8% 10% 3% 5% 3% 7% 6% - 4% - - 92
5 42% 3% - 3% - 3% 7% 39% - 4% - - 74
7 38% - 14% 1% - 2% 8% 13% 2% 2% 7% 12% 90

Physical meaning category Social meaning category

3.1 Challenge to addressee's assumptions 4.1 Prohibition (focus on agent )
3.2 Justification for prohibition 4.2 Permissions
3.3 Reportive uses 4.3 Reportive uses
4.4 Self regulations
Ambiguous case category
5.1 Challenge to prohibition Epistemic-like meaning category
5.2 Permission seeking 6.0 Epistemic-like uses
5.3 Permission giving (compliance)
5.4 Prohibition (focus on object)
The Interactional Basis of the Mandarin Modal néng 'CAN' 227

Boasting
3.1 Challenge to addressee's assumption New discoveries
Physical meaning 3.2 Justification for prohibition Problem reporting
3.3 Reportive uses

4.1 Prohibition (focus on agent )

4.2 Permission seeking
Ambiguous meaning
4.3 Reportive uses
4.4 Self regulations

5.1 Challenge to prohibition

5.2 Permission seeking
Social meaning
5.3 Permission giving (compliance)
5.4 Prohibition (focus on object)

Epistemic-iike meaning 6.0 Epistemic-like uses

Figure 1. Summary of semantic and functional categories of néng

7. Discussion

7.1. The first aim of this paper is to argue that the Mandarin modal néng not
only has a set of referential meanings — ability, permission, and possibility —,
but it also serves important interpersonal discourse functions. In the physical
abilitative meaning, néng challenges what the speaker believes to be the
addressee's assumptions about the speaker's abilities. In the social permission
meaning, néng challenges the addressee's actions. And in the epistemic-like
meaning, néng challenges the addressee's immediately preceding assertion.
This interpretation of a common interpersonal function for the modals (i.e.
challenge) differs from Perkins' formulation of the core semantic meaning of
the English modals in terms of possibility and necessity as applied to the
physical, social, and rational domains (Perkins 1983). Perkins focuses on the
common features that can be extracted from the static semantic meaning of the
English modals. But what I am arguing here is that a common denominator for
the different semantic meanings of néng, — i.e. dynamic, deontic, and epis-
temic — arises from the interpersonal discourse function néng serves, namely
that of challenge. Each of néng's semantic meanings serves primarily to issue
a challenge rather than to make a statement about abilities, permissions, and
possibilities. Discourse analysis of néng prompts us to pose the broader
228 Jiansheng Guo

question of what modals are and what they do in language. Why is there a
grammatical category 'modal auxiliaries' that expresses possibility and ne­
cessity?
In order to answer this question, let us look at two of the many different
ways of representing modal meanings. In English, physical ability can be
expressed either by the modal auxiliary can or by the adjective able, as in be
able to. Similarly, social permission can be expressed by can or be permitted
to. In each case the two options are referentially equivalent. But they differ in
grammatical status. Modal auxiliaries belong to a closed grammatical class,
while adjectives and verbs belong to open lexical classes. Along the con­
tinuum between lexical words and grammatical morphemes, modal auxilia­
ries are more grammaticalized than adjectives and verbs. This grammatical
difference has significant consequences with regard to the meanings ex­
pressed. With lexical forms such as able or permitted, the speaker presents a
fact without any personal involvement. We interpret the utterance as 'I'm
stating X to you'. But when modal auxiliaries are used, the resulting utter­
ances are colored by speaker involvement in the form of opinion, affect, or
personal dynamics. We interpret such utterances as 'I'm challenging/object­
ing to/arguing with you by stating X to you.' This division of labor fits
Traugott's (1982) formulation of trends in grammaticalization. When a lexi­
cal form is grammaticalized, its meaning may also evolve from propositional
to expressive/interpersonal. Similarly, when discussing the Tense, Aspect,
and Mood system, Givón (1984:269) points out that "as discourse-pragmatic
features, they play a crucial role ... in indicating their time/truth/certainty/
probability modalities vis-a-vis the speaker-hearercontract"(emphasis mine).
Child language research has shown that modals are employed by children to
convey their different social stances in terms of interpersonal commitment in
play with other children (Gerhardt 1985,1990 for English; Shepherd, 1981 for
Antiguan Creole). All this points to a special function for modals in express­
ing interpersonal meanings.
The claim that modals indicate the speaker's subjective evaluation of the
modalized proposition is hardly novel (e.g. Lyons 1977; Halliday 1973).
However, previous approaches have failed to recognize the crucial interper­
sonal function of speaker involvement. Concern has traditionally been with
the relationship between speaker and proposition, rather than with that be­
tween speaker and addressee, with the proposition serving as a means of
actualizing the latter relationship. Lyons (1977), for example, glosses the
The Interactional Basis of the Mandarin Modal néng 'CAN' 229

deontic modality meaning as 'let it be so', and the dynamic modality meaning
as 'it is so'. Although the constitutive power of language that he attributes to
deontic modality is social in nature, he fails to recognize that his gloss
represents only part of the social interaction. Speakers are not simply playing
with the creative power of language. Rather, they are participating in a social
exchange, and the creative power of language is used to regulate the interper­
sonal relationship between the speaker and the addressee. In this sense,
Palmer's (1990) characterization of deontic modality as 'discourse oriented'
comes closest to the position argued in this paper. But his analysis still focuses
on a single participant's involvement in the discourse. What's more, his
analysis is limited to the deontic modals, which are generally regarded as
having performative force. This paper argues, in contrast, that 'discourse
orientation' is the fundamental property for modals in all the three domains —
dynamic, deontic, and epistemic. Language has developed the grammatical
category of modal auxiliaries to serve the function of regulating interpersonal
relations in social interaction.

7.2. The second goal of this paper is to argue not only that our understanding
of the discourse functions of modals is indispensable to our understanding of
their meanings, but also that their discourse functions form an important
source of semantic change. Semantic bleaching and abstraction, strengthening
of inferential meanings, and metaphorical transfer have been recognized as
the basic processes of semantic change involved in grammaticalization (e.g.
Bybee 1988; Traugott 1988; Sweetser 1988). However, given their methodo­
logical limitations, historical studies have seldom investigated the role played
by interpersonal discourse functions in semantic change. Studies of develop­
mental change in child language can give us insightful revelations in this
respect. In particular, the development of the social permission meaning in the
5 year olds' uses of néng, and of the epistemic-like meaning in the 7 year olds'
uses of this modal, reflects a process of semanticization of the interpersonal-
discourse function that is associated with earlier uses of néng. The discourse
function of néng gradually saturates its semantic content, resulting in a new
semantic meaning.
The change from the 3 year olds' abilitative meaning to the 5 year olds'
social permission meaning is revealing. For the 3 year olds, the physical
-meaning uses of néng all share a discourse function, roughly glossed as
'Don't assume X' (see Section 3). 13 In addition, the justification for prohibi-
230 Jiansheng Guo

Semantic Meaning Discourse Function

I can X
Physical Meaning
It can X (you) Don't assume X
(Typical of 3 year olds' uses)
It cannot X

Process of semanticization
of discourse function

You cannot X Social Meaning

(=You don't do X) You don't do X (Typical of 5 year olds' uses)

Figure 2. Process of semanticization of discourse function

tions category has the dual function of 'Don't assume X' and 'Don't do X'.
For the 5 year olds, the newly developed prohibition meaning has 'Don't do X'
as its semantic content and discourse function. What is the connection? I
suggest that the interpersonal function of challenge has been semanticized,
i.e., what began as a contextual meaning frequently associated with the form
has become conventionalized and incorporated into the semantic content of
that form.14 This process of change is illustrated in Figure 2.
By this process, néng's discourse function of challenge, pragmatically
interpreted in the 3 year olds' uses, becomes its semantic content in the 5 year
olds' uses. In this semantic change, a challenge to the addressee's assumption
becomes a challenge to the addressee's action. Two crucial components of the
challenge function are involved in the semanticization process. 'Don't' as in
'Don't assume X' contains both a social regulative meaning and a negative
meaning. In the 3 year olds' physical-meaning uses of néng, the social
regulative meaning is absent from the semantic content. Through semantici­
zation, it ultimately becomes part of the semantic content of néng, and with it,
the negative meaning comes to the surface.
Admittedly, longitudinal data are needed to reveal the step-by-step proc­
ess of semanticization of a discourse-interactional function. But for the time
being, this analysis seems more satisfactory than other possible explanations
for this semantic change in children's use of néng.

7.3. There are at least two alternative explanations for this semantic change.
The first is imitation of adult uses. But imitation cannot explain why children
should wait until age 5 before starting to learn the adult social-permission
The Interactional Basis of the Mandarin Modal néng 'CAN' 231

meaning. The 3 year olds, it seems, hear as many social-permission uses from
adults as do the 5 year olds. 15

7.3.1. Delayed imitation cannot be explained by the social power differential

between the 3 and 5 year olds either. Even in Chinese society where children
are supposedly trained to be obedient and respectful of authority, they never­
theless seem to want power and control. This is especially true in play with
peers, as in the current corpus. During the play sessions, the 3 year olds
frequently tried to impose their will on the other children. When they wanted
to thwart the others, they would simply take the toys. The children also knew
how to act as if they had adult authority, offering lessons to other children. For
example, when the children used néng in the physical meaning as justification
for a prohibition, their intonation changed to a higher and leveled pitch, as if
they were adults lecturing other children. Children also challenged the adults'
prohibitions explicitly as shown in the challenge to prohibition uses. All this
indicates that 3 year old Chinese children are hardly powerless and timid
creatures.
Even if we accept the notion that 3 year olds do not use néng for
prohibitions because they lack social power, we cannot explain why they fail
to use it for permission seeking, which should fit their allegedly weak status.
From Table 5, we see that permission-seeking uses are rare in the ambiguous
category. In the unambiguous social-meaning category, even 7 year olds
rarely use néng for seeking permissions. I suggest that this differential usage
indicates that challenge is central to néng. If the general deontic modal
meaning consists of prohibitions (permission denying), permission seeking,
permission giving, and their reportive derivatives, then our developmental
data show prohibition to be the first basic stage in children's acquisition of the
general deontic modal meaning.

7.3.2. The second reason why the simple imitation explanation is unsatisfac­
tory comes from the 5 year olds' use of prohibitions. A careful look at their
sentence forms in the social-meaning category reveals that these differ from
adult usage. Adults' prohibitive uses of néng appear consistently in sentences
with no subject, as shown in (25):
(25) GU: bù néng qiäng.
not can grab
'(you) can't grab (it).'
232 Jiansheng Guo

Physical meaning: Enablement to overcome physical obstacles

Physical Domain
Physical Intended/
Physical
Enablement: Anticipated
Resistance
Ability Resulting State

Social meaning: Enablement to overcome social obstacles

Social Domain
Social Intended/
Social
Enablement: Anticipated
Resistance
Permission Resulting State

Figure 3. Semantic structure transfer account

But as illustrated earlier, many of the 5 year olds' social meaning uses of
néng include a second-person subject, as shown in (26):
(26) Cl: nǐ bù néng dă rén.
you not can hit people
'You can't hit people.'
Simple learning from adult uses cannot account for the addition of an
explicit second-person subject by the 5 year olds.
7.4. The second alternative explanation is that the semantic change results
from a transfer of the basic semantic structure from the physical to the social
domain. According to the model put forth by Sweetser (1982) to explain the
historical change in the English modals, this potential alternative account
would look like Figure 3.
Néng, in its physical abilitative meaning, refers to a physical force which
enables the actor to overcome physical obstacles. In its social meaning, néng
refers to the social force which enables the actor to overcome social obstacles.
In the change from the 3 year olds' abilitative uses of néng to the 5 year olds'
social-meaning uses, the semantic structure of the earlier physical meaning is
The Interactional Basis of the Mandarin Modal néng 'CAN' 233

metaphorically extended to the social domain. However, this account again

fails to explain why the new social deontic meaning in the 5 year olds' uses
starts with the discourse function of prohibition, using the sentence form 'You
cannot VERB'.
Since the predominant form in the 3 year olds' abilitative uses is 'I/It can
X', the semantic-structure transfer explanation predicts that a social permis­
sion meaning will first surface in forms indicating social enablement, like
'I/you/he/she can do X', either as permission seeking or permission granting.
Many transitional sentence forms and discourse contexts where the physical
meaning and social meaning are not clearly differentiated should also appear,
e.g., sentences such as 'You can read and talk, but you cannot get out of bed
for a week' (said by a doctor to a patient who has just received a heart
transplant), where the two cans have a physical as well as a social permission
meaning. However, the majority of social-meaning uses by the 5 year olds are
unambiguous prohibitions which take the surface form 'You cannot do X'.
The semantic-structure transfer account cannot explain why physical enable­
ment in the form of 'I can X' becomes social disenablement in the form 'You
cannot X'.

7.5. In contrast to the two alternative explanations, my explanation of the

semantic development, i.e. as involving semanticization of earlier discourse
functions, explains why this semantic change takes the particular course it
does. The discourse functions and sentence form of the social-permission
meaning are consistent with the discourse function that néng serves in the
earlier physical meaning, i.e. challenging the addressee's expectations. The
sentence form of the prohibitive uses of néng is nothing but an explication of
the challenge function pragmatically interpreted in the 3 year olds' uses. Thus,
we are led to the conclusion that the semanticization of challenge is the crucial
factor in the semantic development of néng that takes place between ages 3
and 5.
A similar argument explains the development of the epistemic-like
meaning in the 7 year olds' uses. The epistemic meaning of néng by the 7 year
olds is still quite elusive and context-dependent, allowing us to observe the
ongoing process through which the context gives rise to a new meaning. Thus,
in example (23), 'Can a police station be in a home?', néng is not emptied of
its root meaning. A strong sense of root possibility is present, i.e. it is
conventionally not the case that police stations are located in people's homes.
 234 Jiansheng Guo

However, in the discourse context of an argument, the primary concern of the

speaker focuses on refuting the addressee's previous assertion, in this case,
the addressee's claim that the lego structure is a police station. Thus the root
possibility meaning is backgrounded only to serve as the premise for the
conclusion that it is not the case that the lego construction in question is a
police station. It is the speaker's focus on the truth validity of the addressee's
claim in the context of contradiction that shifts the meaning of néng from
cultural-conventional appropriateness to logical possibilities, providing the
structural basis for the epistemic modality, i.e. arriving conclusions on the
basis of given premise. From the perspective of ontogeny, the apparently
individual-based psychological process of reasoning emerges from highly
social and interpersonal contexts.

7.6. Although I have emphasized in this paper the dynamic interpersonal

functions néng serves, I do not claim that it is never used for reportive
purposes. Néng does, in fact, get used in reportive ways, with both its physical
abilitative meaning and its social-permission meaning. However, the main
claim of this paper is that, from the developmental perspective, the primary
function of modals is to convey interpersonal meaning. Reportive usage
should be viewed as secondary to, and derivative from, the discourse func­
tions. This proposal was first made by Lyons, who offers a profound insight
into the understanding of modality when he says:
Languages are learned and used in contexts which are in part determined by
the variable assumptions and presuppositions of the people who use them;
and these assumptions and presuppositions are not necessarily representable
in terms of a set of determinable propositions. The objectification of both
epistemic and deontic modality is something that we have here taken to be
secondary in the acquisition of language; and it may very well be that not all
languages, but only those that have been long used in literate societies for
the specialized purpose of academic discussion, provide the means for this
kind of objectivization. [...] Modality, as it operates in a good deal of
everyday language-behavior, cannot be understood, or properly analyzed,
otherwise than in terms of the indexical and instrumental functions of
language, to which its descriptive function is, at times if not always, subordi­
nate. (Lyons 1977:849)

Lyons's wisdom of 25 year ago finds empirical confirmation today. I

hope that this study is only the beginning of programmatic research into the
functional origins of modality.
The Interactional Basis of the Mandarin Modal néng 'CAN' 235

To conclude, let me recapitulate my main findings. Developmental data

indicate that the meaning of the Mandarin modal néng cannot be fully under­
stood unless we consider its interpersonal functions. In all meaning catego­
ries, néng is primarily used to convey a challenge to the addressee. This
discourse function is the primary factor motivating the semantic change of
néng from the 3 year olds' physical abilitative meaning through the 5 year
olds' deontic meaning, to the 7 year olds' epistemic-like meaning. The
process involved in this semantic change is hypothesized as semanticization
of discourse function.

NOTES

* This paper was inspired by joint research on children's use of the English modal can
carried out by Julie Gerhardt and the author. Many of the ideas originated in discussions
related to that project. Thanks are due to Julie Gerhardt for detailed comments on the
earlier versions of the paper, to Mary Erbaugh for comments on theoretical issues and
insights into the Chinese language and culture, and to Suzanne Fleischman for her
thorough and painstaking editing which has clarified and sharpened many crucial ideas
and made this paper readable. I would also like to thank Dan Slobin, Eve Sweetser, and
Leonard Talmy for reading an earlier version of the paper and helping me to clarify and
sharpen the concepts and arguments. Susan Ervin-Tripp, Amy Kyratzis, and several of
my fellow graduate students also gave helpful feedback, for which I am grateful. Of
course I assume full responsibility for errors of fact or interpretation.
The data used in this study were originally collected for a research project on
children's use of modals. International travel for data collection was generously sup­
ported by the Institute of East Asian Studies, UC Berkeley. Purchase of tapes was funded
by a graduate student research grant by the Institute of Human Development, UC
Berkeley. The research was supported by the Chancellor's Dissertation Year Fellow­
ship, UC Berkeley, Ma Hsiang Fang Research Fellowship, and NSF grant No. NSF-BNS-
8919569 to Susan Ervin-Tripp for research on pragmatic factors in syntax development
at the Institute of Cognitive Studies, UC Berkeley, which also provided computer,
copying, and videotape laboratory facilities.
1. The ideas of interpersonal challenge and contested ability were first proposed by
Gerhardt (personal communication) in her analysis of children's use of English can. Her
proposal was that the child-speaker contests what he or she assumes to be the address­
ee's belief that the child is unable to carry out the action predicated in the utterance. Thus
can is used to challenge and fix a state of doubt about the ability in question. For
example, when a child says I can put the poncho on, there is an expectation that putting
on the poncho is a difficult task, and thus it is questionable that the child has the ability to
do it. Evidence supporting this claim is also found in the use of the Spanish modal poder.
Silva-Corvalán (this volume) reports that while both saber and poder may refer to
ability, saber simply makes a neutral statement, while poder implies the existence of
difficulty or an obstacle in accomplishing the predicated action. Thus the use of saber in
contexts like I can swim the butterfly stroke for hours is ungrammatical.
 236 Jiansheng Guo

2. However, Robin Lakoff (1972) offers extensive and interesting discussions on the
pragmatics of modals. Unfortunately, Lakoff draws a sharp distinction between the
pragmatics and the semantics of modal meanings, and fails to recognize their interplay.
Bybee (this volume) illustrates the role contextual inferences play in diachronic changes
of modal meanings. She proposes that the 'present' reading of the past forms of the
English modals has resulted from the conventionalization of pragmatic inferences.
3. See Sweetser (1990:56-57) for a brief survey of traditional treatments of modality.
4. For more discussion on the grammatical and semantic status of néng, see Li & Thompson
(1981) and Tsang (1981).
5. However, Sweetser (1982) notes that epistemic uses of can in English are also limited,
with the affirmative form being rarely used. See also Palmer (1990) and Coates (1983).
6. For discussion on root possibility, see Coates (1983:93-99). I use the term epistemic as
defined by Sweetser (1990), i.e. as reflecting our understanding of the world of reason­
ing in terms of premises and conclusions.
7. However, there is a dialectal difference between Beijing Mandarin and Taiwan Manda­
rin, in which kěyǐ can be used in the negative form.
8. The V-not-V question in Mandarin is similar to the English yes-no question. For more
discussion, see Li & Thompson (1981:535-45).
9. PRTCL = Sentence-final Particles; PERF= Perfect Marker; PROG = Progressive Marker; REL
= Relative Clause Marker; Q = Question Marker.
10. This idea of children's conception of adults' expectations of their abilities as manifested
in child-adult discourse comes from Julie Gerhardt's analysis of children's use of
English can (personal communication).
11. The marginality of the permission use of néng in the data is unrelated to the existence of
a competing form kěyǐ (see Section 1), also a low-frequency form in the corpus.
12. This meaning category represents a relatively small proportion of the uses of néng for all
age groups (see Table 2).
13. With the exception of the reportive uses, which make up only a marginal proportion.
14. Hopper (1982:16) uses the term semanticizing in discussing Russian aspectual func­
tions; Traugott and König (1991) use the term strengthening of informativeness in a
similar sense.
15. No studies have been done on the amount of input of modal auxiliaries children get in
Mandarin. My estimate that 3 year olds and 5 year olds get essentially the same input is
based on the frequency of modal usage by the adult experimenter, i.e. the author, in the
current data corpus. Although this tells nothing about what the children hear at home or
in school, it does suggest that adult use of modals in the different meanings may not vary
according to the age of the addressee.
 The Interactional Basis of the Mandarin Modal néng 'CAN' 237

REFERENCES

Bybee, Joan. 1988. "Semantic Substance vs. Contrast in the Development of Grammatical
Meaning". Proceedings of the Berkeley Linguistics Society Meeting 14: 247-64.
Bybee, Joan. This volume. "The Semantic Development of Past Tense Modals in English."
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries. London: Croom Helm.
Gerhardt, Julie. 1985. "On the Use of Will and Gonna: Toward a Description of Activity-
types for Child Language". Discourse Processes 8: 143-75.
Gerhardt, Julie. 1990. "The Relation of Language to Context in Children's Speech: The
Role of hafta Statements in Structuring 3-year-old's Discourse". IPrA Papers in
Pragmatics 4:1/2: 1-57.
Givón, T. 1984. Syntax: A Functional Typological Introduction, Vol 1. Amsterdam: John
Benjamins.
Halliday, M.A.K. 1973. Explorations in the Functions of Language. London: Edward
Arnold.
Hopper, Paul. 1982. "Introduction". Tense-Aspect: Between Semantics and Pragmatics.
Amsterdam: John Benjamins.
Lakoff, Robin. 1972. "The Pragmatics of Modality". Papers from the Eighth Regional
Meeting of the Chicago Linguistics Society 8: 229-46.
Li, Charles & Sandra Thompson. 1981. Mandarin Chinese: A Functional Reference
Grammar. Berkeley, CA: University of California Press.
Lyons, John. 1977. Semantics, 2 vol. Cambridge: Cambridge University Press.
Lü, Shuxiang. 1980. Xiàndài hànyǔ bäbäi cí (Eight Hundred Words in Modern Manda­
rin). Beijing: Shangwu yinshuguan (The Commercial Press).
Ogden, Charles & Ivor Richards. 1923. The Meaning of Meaning: A Study of the Influence
of Language upon Thought and of the Science of Symbolism. London: Routledge &
Kegan Paul.
Palmer, Frank. 1990. Modality and the English Modals, 2nd ed. London: Longman.
Perkins, Michael. 1983. Modal Expressions in English. London: Frances Pinter.
Shepherd, Susan. 1981. "Modals in Antiguan Creole, Child Language Acquisition, and
History." Unpublished doctoral dissertation, Stanford University.
Silva-Corvalán, Carmen. This volume. "Contextual Conditions for the Interpretation of
'poder' and 'deber' in Spanish".
Sweetser, Eve. 1982. "Root and Epistemic Modals: Causality in Two Worlds". Proceedings
of the Berkeley Linguistics Society Meeting 8: 484-507.
Sweetser, Eve. 1988. "Grammaticalization and Semantic Bleaching". Proceedings of the
Berkeley Linguistics Society Meeting 14: 389-405.
Sweetser, Eve. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. Cambridge: Cambridge University Press.
Talmy, Leonard. 1988. "Force Dynamics in Language and Cognition". Cognitive Science
12: 49-100.
Traugott, Elizabeth. 1982. "From Propositional to Textual and Expressive Meanings:
Some Semantic-Pragmatic Aspects of Grammaticalization". Perspectives on Historical
Linguistics, ed. by Winfred Lehmann & Yakov Malkiel. Amsterdam: John Benjamins.
238 Jiansheng Guo

Traugott, Elizabeth. 1988. "Pragmatic Strengthening and Grammaticalization". Proceed­

ings of the Berkeley Linguistics Society Meeting 14: 406-16.
Traugott, Elizabeth & Ekkehard König. 1991. "The Semantics-Pragmatics of Grammati­
calization Revisited". Approaches to Grammaticalization, Vol. I. ed.by Elizabeth
Traugott & Bernd Heine. Amsterdam: John Benjamins.
Tsang, Chui-Lim. 1981. "A Semantic Study of Modal Auxiliary Verbs in Chinese."
Unpublished Doctoral Dissertation, Stanford University.
 The Discourse and Interactive Functions of
Obligation Expressions

John Myhill
Laura A. Smith
University of Michigan

0. Introduction.

The field of functional-typological linguistics can be thought of as having two

focuses: detailed functional descriptions of phenomena in individual lan­
guages and cross-linguistic generalizations based upon these descriptions.*
The latter endeavor is dependent upon the former, in that cross-linguistic
generalizations must be based upon a cross-linguistic database. The impor­
tance of an extensive empirical base has been clearly illustrated by develop­
ments in the study of, e.g. word order; the large amount of empirical data on
word order variation which appeared in the 1980's has fueled considerable
theoretical speculation on the universal functions of word order variation
(Givon 1983, Sun and Givon 1985, Myhill 1986, 1992, Payne 1987, Herring
1990). The same may be said about voice and aspect (Hopper 1979, Myhill
1984, Cooreman, Fox, and Givon 1984, Cooreman 1988, Thompson 1989).
No such base of empirical studies exists for many other linguistic
phenomena, including modality marking. This is unfortunate, because a more
detailed framework for describing modality would be useful for making
typological generalizations. For example, if we consider the problem of
grammaticalization, it has been observed (Bybee and Pagliuca 1987 among
others) that some words associated with obligation meaning become markers
of future tense (e.g. shall), others come to be used for inference (e.g. must),
while still others develop both functions (e.g. the Spanish Synthetic Future,
diachronically derived from a construction literally translating as 'have to').
240 John Myhill & Laura A. Smith

We do not as yet know why some obligation markers develop in one direction
and others develop in another. This may be simply arbitrary, but a more
interesting possibility is that what we now call obligation subsumes a number
of subfunctions, each associated with a distinct historical development. At
present, however, we have no way of telling whether this is the case, because
existing descriptions of the functions of obligation markers are insufficiently
detailed.
The main purpose of the present paper is to begin to develop a framework
for describing the discourse and interactive functions of obligation marking.
In the short run, this will serve a useful descriptive purpose by providing a
categorization of obligation constructions in different languages. In the long
run, this categorization system will allow for more powerful cross-linguistic
generalizations about universal form-function relationships.
We will consider and contrast the expression of obligation in four lan­
guages: English, Chinese, Hopi, and Biblical Hebrew. These were chosen
because they are genetically unrelated, are available in text form, and are
familiar to us. The main questions we want to address are: In what contexts do
speakers of a given language choose to use an obligation marker? How do
these contexts differ across languages? In a language with more than one
obligation morpheme at the speaker's disposal, what factors motivate the
choice of one or another of these morphemes? In English, for example, when
do we prefer have to to gotta, or vice versa? Do speakers of Chinese use the
same criteria to distinguish between, say, dei and yao? In addressing these
questions, we will be working toward a typology of obligation contexts across
languages.
Interestingly, the languages we will investigate differ not only in how
they use obligation markers but also in how frequently they use them: English
and Chinese have a number of commonly-occurring obligation markers,
while Hopi has only one productive (and not particularly common) lexical
marker of obligation. Biblical Hebrew, on the other hand, has none at all (we
will suggest that word order serves a related function in this language). Both
Hopi and Biblical Hebrew have a large number of clauses in which, from the
English point of view, there is an 'implied' obligation, without there being an
explicit marker.
Our approach in this study will be to analyze and categorize naturally
occurring tokens associated with obligation. We will not give a dictionary-
style 'definition' of obligation, because we do not believe that such a defini-
Functions of Obligation Expressions 241

tion can be meaningfully or objectively applied to naturally occurring data.

Instead, we will use a list of obligation markers as traditionally understood in
each language. We will then take as the goal of our investigation a characteri­
zation of the circumstances under which these markers of obligation are used,
that is, the discourse function of the expression of obligation. We will show
that certain obligation markers are used in certain circumstances and others
are used in other circumstances; thus, rather than 'defining' these markers, we
will systematically characterize the circumstances in which they are used.
This approach is consistent with the general methodology of functional lin­
guistics; as noted above, research into such phenomena as word order, voice,
and aspect has shown that the functions involved are too abstract to allow a
dictionary-style definition, and characterization of usage offers the most
objective account of function.
In section 1, we will discuss preliminary considerations. We will con­
sider previous efforts to categorize types of obligation meaning. We will then
briefly review research on grammaticalization, which will be relevant when
we discuss which forms to include in our study. In section 2 we will turn to our
empirical studies of English, Chinese, Hopi, and Biblical Hebrew.

1. Preliminary considerations.

1.1. Previous categorizations of obligation meaning.

Linguists have distinguished between strong and weak obligation. In English,

must and have to are associated with 'strong' obligation while should and
ought to are associated with 'weak' obligation. In distinguishing between
'strong' and 'weak' obligation, Palmer (1986:100) states that for 'weak'
obligation
the speaker admits the possibility that the event may not take place. This is
seen in [the difference between] He ought to/should come, but he won't
(and) *He must come, but he won't. The second of these is most unlikely, if
not anomalous; if the speaker thinks that the obligation may not be fulfilled,
ought to/should would be used.

We have found it difficult to apply this criterion objectively to naturally

occurring data. If we apply it strictly, we find practically no cases where the
speaker can guarantee that the projected event will take place.
242 John Myhill & Laura A. Smith

A speaker can say You have to leave now, realizing that the hearer may in
fact not leave. Such is the case with the great majority of naturally-occurring
tokens of 'strong' obligation markers such as have to. If we change the
criterion to mean that the speaker is acting as though the action will definitely
take place, then we have to find independent evidence that the speaker is
acting this way. In the great majority of naturally-occurring cases, there is no
evidence other than the presence of a morpheme which has been designated as
marking 'strong' obligation; the argument for calling these 'strong' obligation
markers is therefore circular.
Coates 1983 makes a distinction between obligation which can be para­
phrased as 'it is obligatory/absolutely essential that...' and obligation para-
phrasable as 'it is important that...', the first type being 'stronger'. However,
as Coates notes (pg. 34), the parameters for distinguishing between these are
'indicative rather than definitive,' and 'between these two extremes there is
considerable fuzziness.' For example, she suggests that (1) is paraphrasable
with 'obligatory' and (2) is paraphrasable with 'important,' but it seems to us
that the reverse is just as plausible:
(1) They were told by the Chairman, Mr. Jos. D. Miller, "You must
have respect for other people's property. "
(2) If you commit murder, Charlotte, you must be punished.
In attempting to apply Palmer's and Coates' tests for 'strong' and 'weak'
obligation to naturally occurring data, we found too many cases for which
there was no evidence of the strength of the obligation other than use of a
'strong' vs. 'weak' obligation marker. We therefore have not attempted to
apply independent tests for the 'strength' of the obligation.
Another parameter used to distinguish between different types of obliga­
tion function is referred to variously as 'discourse orientation' (Palmer 1974),
'subjectivity' (Lyons 1977), and 'speaker involvement' (Coates 1983).
Coates (p. 33) considers the important parameter distinguishing must and
have to to be whether the "speaker is interested in getting (the) subject to
perform the action." This distinction is consistent with our own intuitions
about some of the English and Chinese data, but we were hard-pressed to code
for it objectively. Often, we simply could not determine the degree to which
the speaker was interested in having the action performed. Indeed, it is not
clear why a speaker would mention an obligation to perform an action unless
s/he had some interest in having that action carried out.
Functions of Obligation Expressions 243

Palmer (1986:103) argues similarly that "The most natural distinguishing

feature [between must and have to] is that with [...] have to the speaker is
disassociating himself from the obligation [...], and, by implication therefore,
associating himself with it in using must [...]". Here too, we found it fre­
quently, if not usually, impossible to determine objectively the degree of
speaker association with the obligations.1 However, there were some func­
tional subcategories for which something resembling this distinction could be
objectively coded and was relevant to the choice of marker. We also found
that we could include in our characterizations of certain types of obligation
function something about the attitude of the speaker towards the obligation
(although it is not clear how similar this is to what Coates calls 'speaker
involvement'); this will be described in Section 2 when we discuss the
categories used in our analyses.
Coates (1983) notes a number of other criteria which can be useful in
categorizing types of obligation constructions: person/animacy of the subject
and speaker authority over the subject. We have found these parameters to be
much more straightforward and objectively applicable than 'strong/weak' and
'involved/uninvolved', and we will make use of some of them in the discus­
sion in Section 2.

1.2. Grammaticalization.

Cross-linguistic research on grammaticalization (Givon 1975; Ultan 1978;

Bybee and Pagliuca 1987) has shown that the diachronic development from
lexical to grammatical form follows certain routes. For example, markers of
future tense (inflections or auxiliaries) often diachronically derive from verbs
or constructions expressing desire (e.g. English will, Indonesian mau), motion
(e.g. English go, Spanish ir), or obligation (e.g. English shall, the Spanish
Synthetic Future). This research has also demonstrated that there are universal
form/function relationships, such that lexical forms are associated with cer­
tain types of functions and grammatical forms are associated with different
types of functions; those functions which are consistently associated with
more grammatical form can be considered to be more 'grammatical' functions
(see Bybee 1985). For example, the English auxiliaries used to be main verbs
with more 'lexical' meanings such as obligation and desire, and in the course
of becoming auxiliaries (developing structural properties which distinguish
them from main verbs, e.g. contracting, not taking do-support) they also
244 John Myhill & Laura A. Smith

developed more 'grammatical' functions, such as future, inference, etc. The

distinction between lexical and grammatical is of course not discrete for either
forms or functions, but particular forms and functions may still be designated
as more lexical or more grammatical. At any point in time, a form can have
both more lexical and more grammatical functions.
Bybee and Pagliuca (1987) present evidence that functional changes in
the course of grammaticalization are not random, but rather that a grammati-
calizing morpheme develops functions similar to the functions it already has,
e.g., shall as a future marker retains shades of its lexical meaning 'owe' in
contexts such as the lessee shall give notice at least 30 days in advance. Once
a morpheme has developed these functions, it can go on to develop other new
related functions. Thus in the course of a grammaticalization process through
which a lexical morpheme with an obligation meaning becomes a grammati­
cal morpheme with, e.g. future meaning, the grammaticalized function differs
from the earlier lexical function but is related to it in a non-arbitrary way. In
other words, the more grammaticalized function retains some elements of an
earlier lexical function. Bybee and Pagliuca (1987:115) note that, for exam­
ple, "[...] the contemporary modal nuances of shall and will are direct continu­
ations of their original lexical meanings, those of shall are all related to
obligation and those of will are related to desire." They support this argument
with data from Coates' 1983 empirical study of the use of modal auxiliaries in
present-day British English.
Because obligation markers commonly develop other functions which
are related to obligation, it is important to the study of grammaticalization to
include those other, more grammaticalized functions. We will use the term
obligation to refer to the set of obligation functions in the traditional sense,
and we will use 'obligation' (with scare quotes) to refer to this set of obliga­
tion functions and also its related, more grammatical functions (e.g. certain
types of inference and future meaning) which morphemes associated with
obligation commonly develop. In our usage in this paper, the future functions
of, e.g. shall (in a dialect productively using this word) are considered a type
of 'obligation' function, because they relate to the obligation functions which
shall used to have; on the other hand, the future functions of go are not
generally associated with 'obligation'.
Functions of Obligation Expressions 245

1.3. Delimitation of the database.

In any text-based study, it is necessary to decide which data to include and

which to exclude. We have stated that this paper will analyze obligation. What
do we mean by obligation? One possible approach to this problem is to give a
dictionary-style definition of obligation and include in our study every token
in the languages investigated which meets this definition. However, after
wrestling with this problem, we found little merit to this approach; any
definition we used would necessitate another definition (e.g., if we say that an
obligation is a requirement, responsibility, or necessity to do something, then
we must define 'requirement,' 'responsibility,' or 'necessity').
Instead, we will delimit our database by designating certain morphemes
in certain languages as obligation markers on the basis of a consistent transla­
tion relationship (in texts and bilingual dictionaries) with English morphemes
commonly recognized as associated with obligation. Although reliance on
translations is not an ideal solution to the problem, it appears to us to be better
than using one or another abstract definition of obligation in general which
might be controversial and which would definitely be difficult or impossible
to apply objectively to naturally occurring data.
Delimiting the database is complicated by the fact that obligation mark­
ers commonly also have other functions. As noted above, obligation markers
often develop inference or future meaning. Thus, English must, have to, have
got to, and should can also have the meaning of inference {That must be Joe
knocking on the door). However, in the great majority of our data, it is
generally easy to distinguish those with obligation meaning from those with
inference meaning, and we have excluded the latter from our analysis. The
situation is more complicated for obligation markers with future meaning,
where the two meanings often merge (Coates 1983, among others), with
individual tokens showing both future and obligation meaning; in such cases,
it is frequently difficult or impossible to say how much obligation meaning a
token has. This creates a problem in analyzing the function of forms which
have or have had both functions, such as shall,2 the Spanish Synthetic Future,
and Chinese yao. The only such form we will discuss is yao; we will describe
our approach to this problem in Section 2.2.
In determining which data to include, we also need to consider the
problem of cross-linguistic comparability. If we produce a list of the tokens of
the Hopi obligation marker nawus, divided up into its different functions, and
246 John Myhill & Laura A. Smith

another list of tokens of English have to, divided up into its different func­
tions, we have no way of telling whether the differences in frequency of these
functions are artifacts of the respective texts. For example, we found that in
our English texts gotta is used for biological needs (e.g. I gotta take a leak) but
in our Hopi texts nawus is never used for this function; however, there was no
context of this type in the Hopi texts, and we therefore cannot conclude that
nawus lacks this function.
In order to limit the effect of this problem, i.e. that the distribution of
morpheme functions is an artifact of the texts used, and to facilitate cross-
linguistic comparison, we will make use of English translation data in our
studies of Chinese, Hopi, and Biblical Hebrew. Translations are not an ideal
source of data, but they are objective, as we have used someone else's
translations rather than doing it ourselves. Translations sometimes retain
more of the structure of the original than would be colloquial in the target
language. Thus, our claims about obligation functions in English will be based
on original English sources; the translation data serve only as a supplement
for purposes of comparison, and it is important not to overestimate the
similarities between expressions of obligation in two languages on the basis of
translation. However, the tendency to retain the structure of the original in
translation makes cases in which the structure of the original is not retained
all the more striking as evidence for functional differences between the
languages.
Biblical Hebrew presents a special problem for the study of obligation
marking. In our database for this language, there is no commonly-occurring
lexical marker of obligation, and so it is not immediately obvious which data
to include.3 As stated above, we will argue that something resembling 'obliga­
tion' function in this language is marked by word order alternation: the
language is normally verb-initial, and non-verb-initial word order has as one
of its functions something similar to 'obligation' marking. As this type of
'obligation' marking (if it can be called that) is grammatical rather than
lexical, we would expect it to be associated not only with the more specific
lexical types of obligation function considered in other languages but also
with the more grammaticalized functions which obligation markers com­
monly develop. Therefore, in comparing Biblical Hebrew and English, we
will consider not only more basic obligation functions such as the obligation
meaning of must and should, but also more grammatical functions, such as
inference and those associated with shall.
Functions of Obligation Expressions 247

2. Empirical studies.

In this section, we will discuss the results of empirical studies we have done of
obligation marking in English, Chinese, Hopi, and Biblical Hebrew. We have
used original texts from each language (supplemented by English translations
for the last three languages); these will be briefly discussed in the relevant
sections. As will be seen from the following discussion, the situation regard­
ing usage of the various obligation markers is extremely complicated; to keep
confusion to a minimum, we will limit the data we analyze to one genre,
dialogue in written texts; future work will of course have to investigate other
genres.

2.1. English.

The English data were collected from three plays by the modern American
playwright Lanford Wilson. The plays are The Hot L Baltimore (1973), The
mound builders (1976), and Burn this (1987). Since coding of data was often
highly context dependent, brief plot summaries are useful: The Hot L Balti­
more is the story of downtrodden but often witty hotel residents whose
decaying home is up for demolition. In The mound builders, archaeologists
uncover a rare Indian burial site whose treasures are lost forever when the
landowner's son, no longer able to sell the land, avenges his lost profits. Burn
this portrays a group of New York artists who insightfully and often cynically
reflect on their lives after the accidental death of a close friend. These plays
have the advantage of using very colloquial language, which is especially
important for gathering tokens of (have) gotta. We chose plays by a single
author to avoid admitting author preference (rather than function) as a factor
in morpheme choice; this will serve to minimize the number of complicating
variables in this pilot study. For practical reasons, we narrowed down English
obligation markers to five: have to, gotta, should, oughtta, and supposed to.4
A total of 153 occurrences were found in the three plays.
For reasons described above, traditional distinctions such as strong vs.
weak obligation are not useful to the study of naturally occurring data because
they cannot be objectively applied; therefore, we have developed alternative
means of categorization. After extensive analysis of the data, the criterion
which most clearly suggested itself was whether or not someone is negatively
affected by the action expressed in the verb phrase, and, if so, which person is
248 John Myhill & Laura A. Smith

negatively affected. We found that the English obligation expressions are

often used to reflect the speaker's negative evaluation of an event. Consider,
for example, the following clause about a friend of two accident victims:
(3) He had to go out there and identify Dom and Robbie, notify their
families. (Wilson 1987:9)
Here the speaker uses had to to express the unpleasantness of the action.
Without had to, there would be no explicit reflection of the speaker's negative
evaluation of the action as unpleasant. Furthermore, if the action were some­
thing we typically think of as positive, then have to would imply something
quite different, as shown in (4)-(5):
(4) Ed had to go to this wonderful party, the lucky bastard.
(5) Ed had to go to this wonderful party, and no one could talk him out
of it.
In (4), the speaker uses had to to convey disappointment at missing out
on some fun, while in (5) the speaker uses had to to convey annoyance at Ed's
unbridled determination (i.e., Ed was hell-bent on going). Note that in (3) the
unpleasantness is experienced by the subject, while in (4) and (5) it is
experienced by the speaker. The use of had to in (5) could also imply that the
subject is negatively affected (i.e., Ed went against his will), but this interpre­
tation is inconsistent with the clause no one could talk him out of it. Such
examples as (3)-(5) demonstrate that markers such as have to, traditionally
labeled 'obligation' markers, do not simply report obligation; rather, they
convey evaluations by the speaker of the effect of the event. An interesting
non-obligation use of have to clearly illustrates this function of negative
evaluation:

(6) Why did he go out? Why didn 't someone hear him... WHY DID HE
HAVE TO HEAR NOISES IN THE NIGHT? (caps in original)
(Wilson 1976:146)
We can see that this is not an obligation usage because the subject has no
control over the action, hear. Rather, have to expresses the speaker's negative
evaluation of a grievous situation (her husband's disappearance). We have
classified a number of different types of negative effect in Table 1 below (a-g),
and will discuss each in turn, after which we will examine tokens with no clear
negative effect (h-k). We note here that coding tokens is a difficult task with
Functions of Obligation Expressions 249

Table 1. Discourse functions of five English obligation markers supposed

supposed
have had should is/ was/
Functional category to to gotta should have are were oughtta TL
a. Negative effect on subject 1 8 8 4 0 0 0 0 0 30
b. Mitigated inconvenience
to listener (1st pers. sj) 2 4 9 0 0 0 0 0 1 5
c. Negative effect on listener,
speaker unsympathetic
2nd pers. sj. 6 0 3 0 0 0 0 0 9
d. Negative effect on speaker 0 0 0 0 0 0 0 0 0
e. Reprimand (2nd pers. sbj) 2 0 0 1 0 5 1 3 12
f. Counterfactual
(non-2ndpers. sj.) 0 0 0 1 6 7 2 5 21
g. Help not needed 5 0 0 0 0 0 0 0 5
h. Facetious suggestion 0 0 1 2 0 0 0 3 6
i. Advisable action
1st pers. subj. 0 0 0 7 0 0 0 0 7
2nd pers. sj 1 0 7 8 0 0 0 2 17
Generic subject 0 0 0 1 0 0 0 0 2
j . Procedures/routines 7 1 3 0 0 0 0 0 11
k. Biological/physical need 0 2 9 0 0 0 0 0 11
1. Uncodable 1 0 4 2 0 0 0 0 7
TOTAL 42 15 40 22 6 12 3 13 153

modals, which reflect a speaker's inner mental state. It is difficult to get at

what a speaker is thinking, because people often say the opposite of what they
mean; they can be sarcastic, facetious, or insincere. Interpretation of utter­
ances can easily become quite subjective if care is not taken. Keeping this in
mind, we have tried to code the data consistently, making liberal use of overall
discourse context and extra-linguistic factors (stage cues/descriptions) to
guide our interpretations. We will refer to these factors throughout the paper.
Since have to, gotta, and should are the most common obligation mark­
ers, comprising 78% of the data base, we will focus on these, commenting
briefly on the others.
We have separated the had to tokens from the have to tokens because in
the past tense the distinction between have to and gotta is neutralized (*ƒ did
gotta, *I gotta-ed), and had to is the past tense form for gotta. This particu­
larly affects the categories 'Mitigated inconvenience' and 'Biological/physi­
cal need', which are normally represented with gotta rather than have to but
250 John Myhill & Laura A. Smith

which have a number of had to tokens. Hence, there is some evidence that had
to occurs in functions not normally associated with have to.
When there is another auxiliary preceding, only have to and not gotta
appears in our data (*He might gotta leave, *He's gonna gotta leave, *He'll
gotta leave, *Does he gotta leave?).5 However, it is generally grammatically
possible to substitute gotta alone for AUX+ have to (e.g. You're gonna have
to leave vs. You gotta leave). On the basis of our data, it appears that in this
situation the usual functional distinction between have to and gotta is main­
tained; in other words, the AUX+ have to tokens pattern like the AUXless
have to tokens, not like the gotta tokens. For this reason, we include the
AUX+ have to tokens with the other have to tokens in Table 1.
Turning now to discussion of the categories in Table 1, the 'Negative
effect on subject' category includes those usages illustrated in examples (7)-
(9) below, where the action expressed in the verb phrase is emotionally or
physically unpleasant to the subject.6 Table 1 shows that this function favors
have to over gotta (18 vs. 4).
(7) This glowering older brother had to go get my clothes...
(Wilson 1987:22)
(8) They're tearing down the whole building, so we all have to move.
(Wilson 1973:48)
(9) / gotta get Horse's things.
(Wilson 1973:46)
We have evaluated 'negative effect' on the strength of various cues either
in the utterance or in the broader context. In (7), for example, glowering sig­
nals the brother's view of the action as unpleasant (as perceived by the
speaker, of course). As for (8), earlier comments (e.g. the speaker's refusal to
help distribute eviction notices) reveal that the speaker is upset about the
impending demolition. Lastly, in (9), a mother is resigned to moving her son's
belongings out of the hotel after unsuccessfully imploring the manager to let
him stay.
The second category listed in Table 1 is 'Mitigated inconvenience to
listener.' All tokens of this type have 1st person subjects, and so the speaker is
in effect excusing doing something against the interests of the listener by
referring to an obligation requiring this action, as in (10):
(10) I've got to talk money with some St. Louis real-estate men this
evening. This isn't going to be much of a vacation for you, Jean.
(Wilson 1976:13)
Functions of Obligation Expressions 251

The speaker in (10) is apologizing to his assistant's wife, who will

presumably see little of her husband during the excavation (her 'vacation').
We have no evidence that the speaker/subject himself is unhappy about
talking money; in fact, he may well be enthusiastic about it.
The 'Mitigated inconvenience to listener' function is particularly com­
mon in ending (or avoiding entering into) conversations, as in (11):
(11) Dan: Drink your beer.
Chad: I gotta get my ass home. (Wilson 1976:76)
The obligation in such cases can easily be vague or invented, since the
listener will not normally challenge its existence or importance. However, our
database does contain two tokens where the listener does not accept the
priority of the ostensible obligation:
(12) Paul: I gotta be somewhere tomorrow.
Jackie: Come on, everyone's gotta be somewhere tomorrow.
(Wilson 1973:97)
(13) Anna: Actually, I've got to get back to—
Pale: Actually, would you just hold it a second, okay?
(Wilson 1987:29)
Gotta is the preferred way to mark 'Mitigated inconvenience' ; it occurs 9
times with this function. Had to occurs four times (presumably because gotta
lacks a past tense form) while have to only occurs twice, including (14):
(14) I have to work; you have work to do. (Wilson 1987:81)
One possible explanation for the use of have to in (14) is that the speaker
is already at wits' end with the listener and is more interested in getting rid of
him than in veiling her disinterest. In fact, he has just done something which
warrants little courtesy from her; he has picked up a second phone in her
apartment and intimidated her boyfriend into hanging up. The distribution of
tokens for this category thus suggests that gotta is the preferred morpheme for
mitigating an inconvenience to one's listener, while have to in this context
implies a reason rather than an apology.
In categories (a) and (b), 'Negative effect on speaker' and 'Mitigated
inconvenience', we see a split in the functions of have to and gotta: have to is
associated with cases where the subject evaluates (or could be expected to
evaluate) the action negatively, while gotta is associated with cases where the
subject (who is the speaker, since all tokens have 1st person subjects) implic-
 252 John Myhill & Laura A. Smith

itly apologizes for the negative effect the action has on the listener. There is an
interesting illustration of this contrast in two excerpts from the narrative of a
single speaker. In (15), she uses gotta to portray herself trying to weasel out of
an invitation, while in (16), she uses have to to describe the unpleasant
consequences of failing to do this.
(15) His mother's on the bed with a washcloth on her forehead. I'm
trying to tell them how I've got to get a bus back to civilization...
(Wilson 1987:20)
(16) So then it's midnight and the last bus has left at ten, which they
know, I'm sure, damn them, and I hadn 't checked, like an idiot. So I
have to spend the night in Robbie's little nephew's room in the attic.
(Wilson 1987:21)
Note that (15) is humorous precisely because it mixes the offensive
phrase back to civilization with the courtesy of justifying a departure. Have to
(historic present) in (16), on the other hand, is consistent with the subject/
speaker's obviously negative view of the ordeal.
The third functional category in Table 1 is 'Negative effect on listener,
speaker unsympathetic'. In the English data, there is only one subtype for this
category, with 2nd person (or 1st person inclusive, which includes the lis­
tener) subjects. For this subtype, the listener is to perform the action and be
negatively affected by it, and the speaker is unsympathetic or even hostile;
this unsympathetic attitude distinguishes these tokens from tokens of 'Nega­
tive effect on subject' with 2nd person subjects (which are in any case very
rare in the English database). This type is exemplified in (17)-(18):

(17) It's a damn shame you 're going to have to find yourself some other
field of operation. (Wilson 1976:128)
(18) We gotta go when I need it, damnit. (Wilson 1976:61)
In (17), the speaker owns some property which he intends to sell to a
developer, while the listener is an anthropologist committed to protecting an
excavation site on the property. In (18), the speaker is trying to persuade the
listener to do as he requests, and the listener is resisting.
This type has 6 tokens of have to and three of gotta. It appears that an
important factor distinguishing have to from gotta here is whether the speaker
regards this unpleasant obligation as being imposed by factors beyond the
immediate control of the interactants. When there is a legal or quasi-legal
Functions of Obligation Expressions 253

basis for the obligation, have to is used 5 of 6 times (this is related to the
'Procedures/routines' function of have to, which we will discuss below) 7 .
Thus in (17) the speaker believes that negotiations have already been made to
sell the land, and is under the mistaken impression that the anthropologist has
no legal recourse to prevent the sale. On the other hand, when the basis for
imposing the obligation is purely personal, as in (18), gotta is used two of
three times.
As we have noted, the only subtype of 'Negative effect on listener'
represented in the English data is the one with 2nd person subjects. Obligation
markers can also be used in this category with 1st person subjects (If you don 't
keep quiet, I'm gonna have to break your nose) and 3rd person subjects (He's
gonna have to teach you a lesson), but there are no tokens of this in the
English database (we will see in Section 2.2 that there are such tokens in the
Chinese database).
The fourth functional category is 'Negative effect on speaker'; this type
is defined as having a non-1st person subject, which distinguishes it from
'Negative effect on subject' with 1st person subject. There are no tokens of
this category in the English database, although such uses are possible, e.g. in
Why did he have to do that? the obligation marker can indicate the speaker's
disapproval (of course, it can also indicate that the subject could be expected
to view the obligated action negatively). We will see in Section 2.2 that
obligation markers are used in the Chinese database with this function.
The fifth functional category, 'Reprimands', also implies a negative
effect on someone. 8 There are 12 tokens of this type, exemplified in (19)-(20):
(19) Hey. Paul. You're supposed to be doing this with me.
(Wilson 1973:136)
(20) You ought to be ashamed of yourself robbing Mr. Morse.
(Wilson 1973:102)
This type is most commonly represented in our data with supposed to (6
tokens) or oughtta (3 tokens). Supposed to implies a tacit understanding
between the subject and a second party (e.g., compare You're supposed to be
meeting him tomorrow with You ought to be meeting him tomorrow). In (19),
the speaker is looking through hotel records to help Paul find his missing
grandfather; the speaker assumes an understanding between herself and Paul
that he will help in response to her gesture, but he has not done so. Oughtta, on
the other hand, conveys the speaker's personal and often detached viewpoint,
254 John Myhill & Laura A. Smith

an on-the-spot opinion, negatively evaluating a situation. Hence, in (20), the

speaker is disgusted with the listener's behavior, but hardly feels rebuffed. In
this latter utterance, supposed to would be odd because it would imply that the
speaker expected the listener to exhibit shame, and was somehow disap­
pointed when the listener didn't; the reprimand would then appear to be not
for the action committed, but for the failure to show remorse. 9 In all the
utterances under this category, as with all the categories we have examined so
far, the obligation morpheme expresses a negative evaluation of a situation. 10
'Counterfactuals' (statements which are positive about events which do
not occur, or statements which are negative about events which do occur)
differ from 'Reprimands' in terms of subject differ from 'Reprimands' in
terms of subject type, as noted above; the former are non-2nd person, while
the latter are only 2nd person. We include counterfactuals among tokens
implying a negative effect because the speaker, in noting an error, has pre­
sumably recognized some negative consequence of it.11 As shown in Table 1,
each of the 'weak' obligation markers is represented in the Counterfactual
category — 9 supposed to, 7 should {have), and 5 oughtta — a total of 21
tokens. As with 'Reprimands', we find that supposed to, unlike the other
markers, implies an understanding between the subject (or agent) and another
party, while oughtta implies the speaker's on-the-spot viewpoint. Compare
the following:

(21) I'm supposed to be home with Frank 'cause he's on morphine and
God knows what condition he might be in. (Wilson 1973:12)
(22) He ought to sleep with it (the window) open anyway; you can't talk
to him about it. (Wilson 1973:34)

In (21), supposed to implies some previous understanding (e.g. between

speaker and doctor) which calls for the subject's presence at home; oughtta
would imply that the obligation has just occurred to the speaker, and sounds
oddly detached. In (22), on the other hand, oughtta characterizes the recom­
mended action as advisable in the eyes of the speaker and evaluates the
current situation negatively; supposed to would imply that the speaker pre­
sumes some previous arrangements between the subject and some party (e.g.
the speaker herself) 12
Although should is possible in (21) and (22), it is rare in our database in
the 'Counterfactual' category (only one token). Should have, on the other
hand, is quite common, and is exemplified in (23):
Functions of Obligation Expressions 255

(23) No, I should have come with you. (Wilson 1987:20)

The 'Help not needed' category is made up of utterances using You don't
have to.... Here, the speaker is (ostensibly at least) relieving the subject of
some burden:
(24) No, that's okay, you don't have to do that. (Wilson 1987:90)
(25) You don't have to go through them, I'll go through them. (Wilson
1973:65)
We have thus far looked at 7 categories (a-g in Table 1) in which an
obligated action (or a failure to fulfill an obligation) has a negative effect on
someone — listener, speaker, or third party. We will now turn to the remain­
ing categories (h-k), where no such effect is clearly inferrable.
One of these is 'Facetious suggestions', which has only 6 tokens and
hence will be illustrated briefly. In this category, the suggestions are not
expected to be taken seriously:
(26) Mr. Morse, you gotta throw away the mustard plaster and take
something worthwhile to bed with you. (Wilson 1973:27)
(27) Hell, probably ought to celebrate. (Wilson 1976:125)
We know from context that the speakers do not really expect their
suggestions to be carried out. Mr. Morse is senile and hard-of-hearing; the
comment is a joke. (27) is also a half-hearted proposition: the speaker has just
learned that the married woman whom he has been pursuing is pregnant and
as uninterested in him as ever, and he is hardly disposed to celebrate.
'Advisable action' tokens can be divided into three subtypes, depending
upon the person of the subject. With 1st person subjects, there are 7 tokens, all
using should and all couched in terms of personal opinion and/or uncertainty;
four are in subordinate clauses under I think, one is a question, one is under /
decided, and one follows maybe, so that all are in one way or another couched
in terms of personal opinion and/or uncertainty. This type is exemplified in
(28)-(29):
(28) I think I should go straighten my room, (to give the listeners some
privacy) (Wilson 1987:83)
(29) Should I set an alarm or something for him, so he doesn 't miss
work again? (Wilson 1987:75)
 256 John Myhill & Laura A. Smith

Moving now to 'Advisable action' with 2nd person subjects, the most
common markers used were gotta (7 tokens), e.g. (30), and should (8 tokens),
e.g. (31):
(30) You gotta get up to your girl (Wilson 1987:20)
(31) There are some cost accountants and some professional architects
down at Memphis you should talk to. (Wilson 1976:129)
The difference here is probably at least partially due to the relatively
colloquial nature of gotta. We also suspect that something such as 'speaker
involvement' favors the use of gotta; however, this parameter is extremely
difficult to measure, especially with few tokens.
There is one example of 'Advisable action, generic subject':
(32) It's the night of nights! People should be up. (Wilson 1976:79)
Our next category, 'Procedures/routines', consists of obligated actions
which have no clear negative effect on anyone. As Table 1 shows, have to and
gotta are used for this function, with have to favored (7 vs. 3). 13 In this
category, actions are often habitual, as in (33)-(34)14:
(33) She took her stuff— / have to take my own, though. That's the way
we do it. (Wilson 1973:116)
(34) They had to hang around the house and tend the fields — sacrifice
to the gods of harvest and whatnot. (Wilson 1976:46)
There are only three instances of gotta in this category. It is not clear why
gotta is sometimes used here, and we would need a larger database to
demonstrate anything conclusive.
The last of our categories, 'Biological/physical need', is straightforward.
In the present tense, it is only expressed with gotta', in the past, it is expressed
twice with had to, presumably because gotta lacks a past tense form. Exam­
ples are (35)-(36):
(35) / had to piss so bad — about Fifteenth Street and Eighth or Ninth
Avenue... (Wilson 1987:62)
(36) I gotta have another beer. (Wilson 1976:120)
Perhaps one factor favoring gotta in these contexts is informality: we are
more inclined to express physical needs around people we know well and with
whom we are more likely to use the more colloquial gotta.
Functions of Obligation Expressions 257

To summarize, in our English data, we have divided tokens based on

whether the action can be said to negatively affect anyone. The English
obligation morphemes analyzed here are associated with negative effect in
more than half (63%) of the utterances coded. Have to is preferred over gotta
when the speaker is attempting to portray an event as having negative conse­
quences for the subject and/or the speaker; it can even be used in this function
in non-obligation utterances (not gathered here, e.g., Why did he have to hear
noises in the night?), where the verb is non-volitional and the action is beyond
the control of the subject. Have to is also preferred to gotta in routines and
procedures. With the form You don't have to..., the speaker is either express­
ing a reprimand or (at least ostensibly) relieving the listener of an obligation.
Gotta, on the other hand, is the choice when mitigating an imposition, where
the speaker often portrays an intention as an obligation in order to avoid
offending the listener. Gotta is also used in reference to biological/physical
needs, and its use with this function may be motivated in part by the informal­
ity which makes such mentions acceptable in the first place.
The remaining obligation expressions — should, supposed to, and oughtta
— are less frequent in our data and have a different set of functions. The major
functions of should are to express the speaker's wish that something else had
happened (counterfactuals), to frame cooperative suggestions, and to refer to
actions which the speaker has thought about taking. Supposed to and oughtta
are both most commonly used in reprimands and counterfactuals in our data.
Supposed to suggests action obligated by previous agreement or by tacit
understanding between the subject and another party; oughtta occurs in
situations where the speaker expresses personal distaste for a situation, al­
though s/he is often not directly affected by it.
As we have noted, informality undoubtedly plays a role in the choice of
obligation marker, particularly with gotta. Having described the general
functions of gotta in the present study, it will be easier for future studies using
data with broader stylistic variation to identify deviations from these general
functions which are motivated by stylistic considerations.

2.2. Chinese.

The Chinese data were gathered from dialogue in two plays by Cao Yu,
Beijingren ("Peking Man," 1954) and Leiyu ("Thunderstorm," 1961)15. The
first portrays a once-prosperous family in decline, the second a pair of ill-fated
258 John Myhill & Laura A. Smith

lovers from different social classes who turn out to be long-separated siblings.
There were several reasons for selecting these plays. First, they are by the
same author and hence keep author-preference and dialect variables con­
stant.16 Second, the plays are recent enough to contain dialogue fairly typical
of modern colloquial Chinese; Cao Yu's works are recognized as representa­
tive of the new genre huaju (spoken drama), which contrasts sharply with the
lyrical, highly formulaic style of traditional Chinese drama. Third, Beijingren
is available in a reasonably idiomatic English translation and is long enough
to yield a sufficient number of English tokens. This is important for Chinese-
English comparison (at the end of this section).
We examined three Chinese obligation markers, the preverbal auxiliaries
dei, yao, and gai.17 We have gathered only their preverbal and (where there
was an overt subject) post-subject occurrences, excluding tokens with nomi­
nal or sentential complements (e.g., Zhe dei ta guan 'this dei he handle'='This
has to be handled by him').
All of these morphemes also have epistemic functions, which are ex­
cluded here. The morphemes are illustrated in the following examples:18
(37) Ni qian le wo de, ni dei huan (Cao 1954:369)
you owe ASP me PRT you pay-back
'You have to pay back what you owe me.' 19
(38) Yihou shenme dou yao gaosu ma! (Cao 1954:395)
from-now-on something all tell Mother
'From now on you must tell your mother (=speaker) everything.'
(39) Nimen ye gai gei nimen fuqin songxing nal
you(pl) also to your(pl) father bid-farewell PRT
(Cao 1954:335)
'You two should also say goodbye to your father.'
The status of dei and gai as obligation markers is supported by descrip­
tions in several grammars (e.g., Liu 1983:116, Chao 1968:741-42, Li and
Thompson 1981:182-83). Yao, however, is problematic because it can also
express desire/intention and prediction/future meanings, as exemplified in
(40)-(41):
(40) Ke ni daying le wo. Wo yao fang, wo yao fang!
but you promise ASP me I fly I fly
(Cao 1954:296)
'But you promised! I want to fly it (the kite), I want to!'
Functions of Obligation Expressions 259

(41) Tarnen shuo Dujia Laotaiye... jiu yao duanqi...

they say Du-family old-master just cease-breath
(Cao 1954:383)
'They say Old Master Du is going to die anytime now...'
Additionally, when preceded by the negative marker bu, yao has impera­
tive function (i.e. bu yao+verb='don't+verb). The question is thus: Which yao
usages should be counted as obligation? There are four possible strategies for
addressing this problem. One is to include all yao occurrences; however, this
would mean including a large number of utterances which do not appear to
mark anything resembling obligation meaning (e.g., Wo yao jiabei huan ni
'I'll pay you back double'). Another strategy would be to exclude all yao
tokens, but that would eliminate numerous examples where yao conveys a
clear sense of obligation. A third strategy would be to include only those yao
tokens which are translated as one of the English obligation morphemes.
However, some yao utterances have more idiomatic translations, such as (42)
(Cao 1954:423):

(42) Zenme yang de rizi dou shi yao guo de.

any type PRT day all are experience PRT
(Cao 1954:423)
'Everything will be all right, no matter what happens.'
(42) is more literally translated as 'All kinds of days have to be experi­
enced', i.e., one has to take the bad times with the good. If we used transla­
tions as a criterion for inclusion, we would be forced to exclude data such as
(42); we could of course ignore the translation data under such circumstances,
but we prefer to use more objective criteria for inclusion.
We ultimately decided upon the following strategy: we included or
excluded tokens of yao based upon whether they fit into any of the categories
of obligation meaning described in Section 2.1. We first gathered all non-
negated occurrences of yao from Beijingren, which yielded 189 tokens (this
was enough for our purposes, so we did not gather tokens of yao from Leiyu).
Of these 189, we eliminated 15 in which yao is unequivocally a future-
marker, that is, in which there is an inanimate subject and/or a non-agentive
verb (e.g. (41) above). For the remaining 174 tokens, we categorized 68 as
belonging to one of two general types, one (59 tokens) where someone is
negatively affected by the action expressed in the verb phrase (i.e. types a-g
 260 John Myhill & Laura A. Smith

described in section 2.1), and the other (9 tokens) comprised of the different
subtypes of obligation function without negative effect (i.e. types h-k).
The remaining 106 yao tokens are typically either assertions of intended
action (with 1st or 3rd person subjects), unsolicited promises (with 1st person
subjects), or detached inquiries as to the listener's intentions (with 2nd person
subjects); we set these tokens aside. The 68 yao tokens included in our
database do not invariably translate into English with obligation morphemes,
and in some cases such a translation is not even plausible. However, we do not
regard this as a problem, because it is our ultimate goal to develop characteri­
zations of types of obligation functions which are language-universal and not
dependent upon whether one language or another happens to represent them
with an obligation marker.
We should note that, by adopting this strategy, we are proposing not a
solution, but an approach. Exactly what constitutes an obligation usage of yao
is still an open question, partly because the morpheme conflates what from an
English viewpoint are two distinct meanings, obligation and intention.
Our database of Chinese obligation markers thus consists of 29 tokens of
dei, 68 tokens of yao, and 26 tokens of gai. Dei and gai tokens were gathered
from both plays, yao tokens from only one. Table 2 compares the raw
frequencies of the three morphemes by showing their distribution in the one
play from which yao tokens were gathered:
Table 2. Obligation dei, yao and gai in 'Beijingren'
Dei 10 11%
Yao 68 73%
Gai 15 16%
Total 93 100%

Table 3 shows the distribution of these morphemes in terms of the functional

categories described for English in section 2.1. Note that while some of the
obligation functions represented in the English data are not represented in the
Chinese data (and vice versa), the same functional categories are in general
useful for both languages.
As in English, we distinguish between those functional categories for
which we can infer that someone would be negatively affected (a-g), and
those for which we cannot (h-k). In the former group are the bulk of dei and
yao tokens, in the latter the bulk of gai tokens.
Functions of Obligation Expressions 261

Table 3. Functional categories for dei, yao, and gai

dei yao gai Total

a. Negative effect on 7 14 0 21
subject
b. Mitigated inconvenience to 5 4 0 9
listener (1st pers. subj.)
c. Negative effect on
listener, unsympathetic
1st pers. subj. 0 8 0 8
2nd pers. subj. 10 0 0 10
3rd pers. subj. 0 7 0 7
d. Negative effect on speaker 1 24 0 25
e. Reprimand (2nd pers. subj.) 0 2 3 5
f. Counterfactual (non-2nd pers. subj.) 0 0 7 7
g. Help not needed 0 0 0 0
h. Facetious suggestion 0 0 0 0
i. Advisable action
1st person subject 0 2 8 10
2nd pers. subj. 6 4 6 16
Generic subject 0 2 1 3
j . Procedures/routines 0 1 0 1
k. Biological/physical need 0 0 0 0
1. Uncodable 0 0 1 1
Total 29 68 26 123

To express 'Negative effect on subject', both dei and yao are used, as in (43)-(44):
(43) ...chi renjia de qianliang, jiu dei shou
eat somebody PRT money-grain then be-subject-to
renjia de guan. (Cao 1961:7)
somebody PRT discipline
'...if somebody's paying your wages you've got to take orders from
them.'
(44) Keshi tarnen shuo zhe zhang xianzai yao fu...
but they said this account now pay
(Cao 1954:277)
'But they said the bill had to be paid right now.'
The distinction between dei and yao in this category appears to be
related to the nature of the obligation. Most of the dei tokens (6/7) deal with
262 John Myhill & Laura A. Smith

the subject's prescribed behavior toward another party (who is mentioned),

and the behavior is obligated by virtue of their social relationship. These
obligations include accepting the authority of one's employer (as in (43)),
being filial to one's father-in-law, being mindful of one's family's needs, etc.
Note that these dei obligations are constant to the extent that the social
relationship holds; where the relationship is permanent, as in say, child-to-
parent obligations, the obligations are permanent, and hence in some sense
generic. In contrast, most of the yao tokens (13/15) refer to actions that are
either marked as inherently unpleasant independent of social relationships
(e.g., having to suffer in life, having to worry) or necessary under temporary
circumstances (e.g., paying a bill, as in (44), or selling a prized possession) 20 .
The second category, 'Mitigated inconvenience to listener', also occurs with
both dei and yao, illustrated in (45)-(46):

(45) Ma, nin bie shuo zhexie hua le, wo hai

mother you-honorific not say these word ASP I still
dei zou ne. (Cao 1961:123)
gO PRT
'Mother, please don't say these things now, I've got to get going.' 21
(46) Wo yao zou, wo yijing deng le liang nian le.
I leave I already wait ASP two year ASP
(Cao 1954:362)
'I have to leave this place, I've already waited two years.'
As in the English data, utterances in this functional category have 1st
person subjects; the speaker is acting against the wishes of the listener while
mitigating the inconvenience in some way. The distinction between these dei
and yao usages is as follows: In four out of five of the dei clauses, the speaker
needs the cooperation of the listener to carry out the action. For example, in
(45), the speaker is a son who cannot in good conscience leave without his
mother's blessing. In contrast, three of the four yao usages occur when the
speaker's action (e.g. leaving after declining an invitation to stay) is not
contingent upon anyone else's permission. Yao's function here is clearly
related to its intention function, since the speaker can be expected to carry out
the action without the listener's acquiescence. The intention to perform an act
negatively affecting the listener is mitigated by other expressions in the
clause; e.g., (46) above is preceded by a clause translating as 'I can't wait any
longer', demonstrating that the speaker has no choice. Our next category,
Functions of Obligation Expressions 263

'Negative effect on listener, speaker unsympathetic*, is divided into three

types depending upon the person of the subject. Clauses with 1st person
subjects (47) are all threats using yao, while clauses with 2nd person subjects
(48) are all hostile commands using dei:
(47) Kai men! Wo yao shao fangzi la! (Cao 1954:367)
open door I burn house ASP
'If you don't open the door, I'll burn the house down!'
(48) Zhe shi ni qian wo de qian, ni dei huan...
this is you owe me PRT money you pay-back
(Cao 1954:369)
'...that's what you owe me, and that's what you're going to pay me
back for.'
In examples with 3rd person subjects, the speaker characterizes inten­
tions carried out independent of and presumably against the listener's wishes;
there are 7 tokens of this type in the database and all 7 use yao. For example,
the listener in (49) is negatively affected by the obligation because he is in
love with the subject and arrangements are being made for her to marry
someone else:
(49) Renjia yao jiaren... (Cao 1954:299)
she get-married
'She has to get married...'
Note that, in the English data discussed in 2.1, there are also utterances
with 2nd person subjects in the 'Negative effect on listener, speaker unsympa­
thetic' category; these are hostile predictions or commands. There are, how­
ever, no tokens in this category with 1st or 3rd person subjects. It is our
impression that such cases are rarely expressed with obligation markers in
English; 'I have to burn the house down' is not a plausible translation of (47),
and the listener-affected interpretation of the translation of (49) is impossible
in English (indeed, the translation 'She has to get married' gives the impres­
sion that it is 'she' and not the listener who is negatively affected by the
obligation, which is not the sense of the original). To be sure, it is possible in
English to use an obligation marker in this functional category with a 1st
person subject — always, it seems, in combination with a future marker (e.g.
I'm gonna have to burn the house down). However, no such examples occur in
our English database, and there is a clear tendency in English to emphasize
 264 John Myhill & Laura A. Smith

either the intentional meaning (with will or gonna and no obligation marker),
or the 'Mitigated inconvenience' meaning, where the obligation marker 'ab­
solves' the speaker of responsibility (e.g. I'm sorry I have to do this). In
Chinese, on the other hand, yao conflates the intention and obligation mean­
ings and is commonly used when the listener is negatively affected, with or
without mitigation on the part of the speaker.
Yau is the preferred morpheme in our next functional category, 'Negative
effect on speaker', occurring in 96% (24/25) of the cases. This category
includes only cases where the subject is non-1st person, as in (50) below,
which distinguishes it from 'Negative effect on subject' with 1st person
subjects:

(50) Lin zou, lin zou, hai yao exiongxiongdi dui

about-to leave about-to leave still fiercely to
wo fa yi dun piqi. (Cao 1954:300)
me emit one CL anger
'Here he is getting ready to leave and he's ranting and raving at me
like this.'
10 of the 25 tokens of this type are questions (generally with 2nd person
subjects) in which the speaker questions the motivation of the subject, as in
(51M52):
(51) Suyi, ni weishenme yao gaosu Yuanjia Popo ne?
Suyi you why tell Yuan-family uncle PRT
(Cao 1954:363)
'Auntie Su, why did you have to tell Uncle Yuan about it?'

(52) Ni yao gan shenme? (Cao 1954:363)

you do what
'What are you doing?' (i.e. 'What do you think you're doing?')
In all of these cases, the action has a negative effect on the speaker (this is
not obvious from the English translation of (52), but it is clear from the
context in Chinese, hence our alternative English translation 'What do you
think you're doing?').
There are no tokens of 'Negative effect on speaker' with obligation
markers in the English database, although this usage is certainly possible, as
in the translation of (51), or (6) (Why did he have to hear noises in the night?).
Functions of Obligation Expressions 265

Table 4 Functions of dei and yau

dei yao
Negative effect on subject Negative effect on subject
('X has to do/be Y toward Z') ('X has to do (awful thing) Y
under circumstance Z')

Mitigated inconvenience to listener Mitigated inconvenience to listener

('Please, I've got to X') ('Sorry, I'm gonna have to X')

Negative effect on listener, Negative effect on listener,

unsympathetic, 2nd pers. unsympathetic, 1st or 3rd pers.
subj. (e.g. 'You have subj. (e.g. 'I'm/he's gonna have
to X, and you won't like it') to X, and you won't like it')

Negative effect on speaker

('Why did you have to X?')

For reasons which are not clear, English does not use an obligation marker
when there is a question word other than why. For example, What do you have
to do? does not occur when the obligation marker indicates the speaker's
disapproval of the action (parallel to the speaker-affected interpretation of
Why do you have to do that?); rather, the obligation marker depicts the action
as having a negative effect on the subject. Although obligation markers are
sometimes possible with 'Negative effect on speaker' in English, this usage is
obviously less common than with yao. The functions discussed thus far have
been represented in the data by dei or yao, but not gai, and in this respect differ
from the remaining functions, which generally favor gai. We summarize the
functions of dei and yao discussed thus far (categories a-d) in Table 4 (the
parenthetical English examples capture the gist of the Chinese contexts).
The contrast between dei and yao is clear from Tables 5-6, which show
(on the x-axis) the subject type, and (on the y-axis) the negatively affected
party, for the functional categories discussed so far (a-d). Most of the dei
clauses in our data (16/23, or 70%) have the listener negatively affected.
These are exclusively 1st person subjects (mitigated discourtesies) or 2nd
person subjects (hostile commands). With yao, on the other hand, listener-
affected clauses comprise only 35% (20/57) of the tokens, while speaker-
affected clauses comprise 58% (33/57). Furthermore, with yao, the subjects of
these clauses are often 3rd person.
266 John Myhill & Laura A. Smith

Table 5. dei—subject type (x-axis) and negatively affected party (y-axis)

Speaker Listener 3rd party Generic Total
1st pers. 2 5 0 0 7
2nd pers. 0 11 0 0 11
3rd pers. 1 0 4 0 5
Generic 0 0 0 0 0
Total 3 16 4 0 23

Let us now look at the remaining obligation functions in which a negative

effect on someone can be inferred.
Among these are 'Reprimands' and 'Counterfactuals'; like English, Chi­
nese uses obligation morphemes to express both meanings. 'Reprimands' here
have 2nd person subjects, which distinguishes them from 'Counterfactuals'.
In our data, there are five reprimands, two with yao and three with gai. In
the former (unnegated) category, the speaker expresses dissatisfaction with
the listener's failure to do something, as in (53), while in the latter (all three
negated), the speaker reprimands the listener for having taken action deemed
wrong, as in (54):
(53) Yihou shenme dou yao gaosu ma! (Cao 1954:395)
from-now-on what all tell mother
'From now on you must tell your mother (=speaker) everything!'
(54) Feng, ni yiwei wo zheme zisizili ma? Ni bu
F. you think I this selfish PRT you not
gai zhenme kan wo. (Cao 1961:40)
thus view me
'Now, Feng! You don't imagine I'm that selfish, do you? You
mustn't think I'm that sort.'
'Counterfactual' meaning in our data is represented only with gai, usu­
ally with negation (5/7 tokens), as in (55):
(55) Wo zao jiu gai xue naxie xinpai de
I long-ago then learn-from those modern PRT
taitaimen... (Cao 1954:290)
wives
'I should have learned from those modern wives a long time ago...'
Functions of Obligation Expressions 267

Table 6. yao—subject type and negatively affected party

Speaker Listener 3rd party Generic Total
1st pers. 9 12 0 0 21
2nd pers. 10 1 0 0 11
3rd pers. 14 7 2 0 23
Generic 0 0 0 2 2
Total 33 20 2 2 57

We will now turn to Chinese usages in which the action of the verb
suggests no clear negative effect on anyone. The first two such categories
('Help not needed' and 'Facetious suggestion') are not represented in our data
by the Chinese morphemes studied here.
We will next discuss the 'Advisable action' category. With 1st person
subjects, gai is usually used here (8/10 times), as in (56):
(56) Wo shuo ye gai wenwen Su biaomei de yijian ba.
I say also ask Su cousin PRT idea PRT
(Cao 1954:407)
'...I think that we should talk to Cousin Su about it first.'
With 2nd person subjects, on the other hand, 'advisable actions' are
expressed using all three obligation markers (6 dei, 4 yao, 6 gai). Dei is used
when the speaker cannot be assumed to have authority over the listener;
indeed, in 4 of the 6 dei clauses here, there is independent evidence that the
speaker is showing deference. In (57), deference is indicated by the polite
form of the 2nd person pronoun; the speaker is a servant, consoling a bereaved
houseguest.
(57) Zhe shi meifazi de shi, —keshi nin dei
this is no-way PRT matter but you (honorific)
kuku... (Cao 1961:133)
have-a-cry
'...there's nothing you can do now — now come on, you must have
a cry.'
In contrast to dei, all four yao clauses in this category are uttered by an
authority figure (e.g. parent, elder), as in (58), where a woman is admonishing
her pregnant niece:
 268 John Myhill & Laura A. Smith

(58) Ni yao dui wo shuo shihua. (Caol954:317)

you to me tell truthful-word
'(You've got to) tell me the truth.'
With both dei and yao, the speaker is strongly interested in seeing the
action performed, usually out of commitment to the listener's well-being, or
because of some personal agenda, or both.
Unlike the dei and yao tokens in this category, the 6 gai tokens have
subjects showing neither deference nor authority. Consider (59), where the
speaker appears to make a friendly suggestion to two would-be lovers (one of
whom is her husband):

(59) Lin zou le, ye gai liu dian jinian. (Cao 1954:310)
about-to leave ASP also keep a-bit remembrance
'Now that he's leaving, you should have something to remember
each other by.'
There are also three tokens of 'Advisable action' with generic subjects,
two using yao, as in (60), and one using gai:22
(60) wo shuo yige ren yao you dian liangxin. (Cao 1954:330)
I say a person have a-little conscience
T say a person should have a little conscience.'
Gai is typically translated as 'should', the 'weak' obligation marker in
English, and in our data it shares several features with that morpheme. Over
half of the gai tokens (15/26) occur in the 'Advisable action' category, so that
its use with this function is much more common than dei (6/29) or yao (8/68).
Similarly, most of the tokens of should (16/22) occur in this category, but very
few tokens of have to (1/42) or gotta (7/40) do. As in English should clauses,
many of the gai tokens in this category (6/15) occur with a hedge, e.g. with the
particle ba meaning 'Don't you think so?' 2 3 or with something translating as
'I think'. Such hedges (see (60) above) never occur with dei or yao in these
categories.
The last two functional categories, 'Procedures/routines' and 'Biologi­
cal/physical need' are only minimally represented in our Chinese data, and
hence will not be discussed here. Returning now to categories e-k in Table 3,
we can summarize the 'No negative effect' functions of dei, yao, and gai as in
Table 7 (again, the English examples are intended to convey the gist of the
categories).
Functions of Obligation Expressions 269

Table 7. Functions of dei, yau, and gai

dei yao gai
Reprimand Reprimand
(' You must X, (' You shouldn't have Xed
but you haven't yet.') but you did.')
Counterfactual
('X should(n't) have Yed.')
Advisable action Advisable action Advisable action
('Maybe I/we should X.')
('With all due respect, ('You've got to X; ('Maybe you should X.')
you've got to X.) I know what's
best for you.')
('People should X.')

The only noteworthy overlap of the three morphemes is with 'Advisable

action, 2nd pers. subj.', and here the distinction is linked to the relationship
between speaker and listener: In general, dei marks deference, yao marks
authority, and gai is neutral. Gai, like English should, tends to have hedges in
this category, and is the only morpheme used in counterfactuals.
To compare obligation usages in English and Chinese, we tabulated
obligation morphemes in Beijingren against an English translation. 24 For this
purpose, we included from each text all clauses which had a (non-epistemic)
obligation marker in Chinese, English, or both. For yao clauses, this included
the 68 tokens discussed above, plus an additional four which translated with
an English obligation morpheme. Since gotta, supposed to, and oughtta are all
extremely rare in the translation (only five tokens), we will exclude them here.
The translation correlations are shown in Table 8. Where only one
language has an obligation marker, we indicate whether the other has a
productive alternative way of encoding obligation meaning ('other oblig.
marker'), or whether the clause is unmarked for obligation ('no marker'). This
is especially relevant for clauses translated with should, where the Chinese
has several means of expression other than a preverbal modal (e.g., clause-
final particle ba, meaning roughly 'Why don't....'). 25
The data in Table 8 suggest that, of the three Chinese morphemes, yao
occurs in places where English is least likely to use an obligation morpheme;
most 'non-obligation' translations in the English (41/47) take yao in the
Chinese. This is not surprising, given yao's intentional meaning; about a third
of these yao clauses (14/41) translate with gonna, will, or want, while nearly
270 John Myhill & Laura A. Smith

Table 8. Chinese-English
comparison
havel
(Beijingren+English other no
translation)
must had to should marker marker TL
dei 0 1 3 2 4 10
yao 2 22 5 2 41 72
gai 1 0 11 1 2 15
other oblig 2 1 13 X X 16
no marker 1 20 8 X X 29
TL 6 44 40 5 47 142
X=no data gathered
another third (12/41) use the Present Progressive form to express unilateral
action already underway (e.g., ...and he's ranting and raving at me). We only
included tokens of yao corresponding to one or another of our categories of
obligation function, generally having a negative effect on someone;26 this
negative effect frequently makes it at least possible to translate yao tokens
into English with obligation markers, but the data in table 8 show that this is
relatively uncommon.
Of the English morphemes, have to/had to marks the contexts least likely
to take an obligation expression in Chinese. 20 of the 29 'non-obligation'
contexts in Chinese take have to in the English, and 14 of these 20 fall into two
categories: they refer either to past events, and translate as had to or did have
to (6 tokens), or to habitual actions or states (8 tokens), usually with a time
adverbial like meitian 'every day'. In our Beijingren data, Chinese mor­
phemes which generally share the functions of have to — dei and yao — occur
almost exclusively in either future or present contexts: dei never translates
with any past tense English form, while yao does so only 10% of the time (7/
72 times). It appears that past obligations in Chinese are rarely expressed with
either dei or yao. As for habitual actions or states, English seems more likely
than Chinese to explicitly mark a negative evaluation by using an obligation
marker. In (61), for example, have to makes it clear that the speaker does not
perform her routine willingly, while the Chinese translation uses no obligation
marker:
(61) ...lian wo bu shi ye xia chufang bangzhe
even I not be also go-down-to kitchen help
Zhang Shun... (Cao 1954:295)
'Even I have to help Zhang Shun (=servant) in the kitchen.'
Functions of Obligation Expressions 271

In many of these clauses, have to reinforces the speaker's negative

evaluation, even when the unpleasantness of the action is quite obvious (e.g.
with the verb suffer). We suspect that in English, this function is important in
habituais because a repeated action might otherwise imply the speaker's
willingness.
In cases where a Chinese obligation marker is translated into English
with an obligation marker, there is a strong tendency for yao to be translated
with the 'strong' obligation marker 'have to' and gai with the 'weak' marker
should; this is consistent with the patterns in the original data. There are,
however, a few cases where yao is translated with 'should'. These are gener­
ally cases where some kind of abstract morality is being discussed, as in 60 (T
say a man should have a little conscience'). In English, the use of a weak
obligation marker here suggests that if the subject fulfills this obligation it is
because s/he is a good person rather than because it is an absolute obligation;
strong obligation markers do not seem appropriate here.

2.3. Hopi.

Hopi has only one word which we can consider to be a productive marker of
obligation, nawus.21 Basic word order in Hopi is SOV, and nawus normally
comes between the subject and the object or complement PPs. Unlike the
obligation markers in English and Chinese, which have syntactic and/or
phonological properties of auxiliaries, nawus is a free, fully lexical form,
which has the syntactic distribution of an adverb. Although it is the only
productive way of marking obligation in Hopi, it is relatively uncommon;
must, have to, should, and be to each occur more frequently in the English
translation than nawus occurs in the Hopi.
Our Hopi data are taken from Malotki and Lomatuway'ma (1987a and
1987b), and Geertz and Lomatuway'ma (1987), collections of Hopi texts with
English translations. The texts consist of personal narratives as well as tradi­
tional stories about the god Maasaw and his interactions with Hopis and
various anthropomorphized animals. As in the other languages, we include
only data from conversational (as opposed to narrative) parts of the texts.
Nawus occurs 27 times in the conversations in these texts, always with an
obligation function (even in two cases where it is not translated into English
with a typical obligation form).
For the purposes of delimiting our database and comparing Hopi and
English, it is necessary to determine when a type of obligation meaning can be
272 John Myhill & Laura A. Smith

Table 9. Obligation expressions in Hopi and English

Nawus No nawus
Have to 12 29
Must 8 33
No choice (but to) 4 0
Need (to) 1 7
Should 0 29
Be to 0 26
Supposed to 0 9
Better 0 3
Got to 0 1
No obligation word 2 X
Total 27 137

inferred from the text even if it is not marked with nawus. The strategy
employed in Section 2.2 in selecting tokens of yao to include in the data base,
namely, including tokens which correspond to an obligation category de­
scribed in Section 2.1, is not feasible in Hopi. As we will see, the categories
are not really appropriate for the Hopi data, and additionally we would have to
apply them to every single clause in the data, with nawus or without, which
would be prohibitively time-consuming. Under these circumstances, the only
realistic course of action is to use the English translations as our criteria for
inclusion. Thus we include in our database for this section all sentences which
have nawus in the Hopi and/or a word expressing obligation in the English
translation. As mentioned in Section 2.2, this strategy has the disadvantage of
excluding tokens where some kind of obligation meaning appears to be
present but an obligation marker is not used in the English translation;
however, the alternative approaches have stronger disadvantages. The transla­
tion relationships in the database are shown in Table 9.28
The first observation which we can make from Table 9 is that nawus
appears to be associated specifically with 'strong' obligation, as it is only
translated with the 'strong' obligation forms must, have to, and no choice, but
never with the 'weak' markers should, be to, supposed to, or better. For
reasons noted above, it is problematic to develop independent criteria to test
the 'strong/weak' distinction. However, it is worth mentioning that in all but
one of the 27 cases, the obligation is fulfilled; in the one exception, a child
disobeys his parents' command, but it is clear that the parents intended a
command rather than a recommendation. This pattern is consistent enough to
support the view that nawus is specifically associated with 'strong' obligation.
Functions of Obligation Expressions 273

We will now turn to the problem of characterizing the functions of

nawus. One strategy for doing this is to apply the categories for obligation
types used in Sections 2.1 and 2.2, which could be applied to the English and
Chinese data in a more or less straightforward fashion. However, when we
attempted to do this with the Hopi data (or even the English translations), we
found that very frequently these data did not fit neatly into the categories.
In the English database, there is relatively little difficulty distinguishing
between obligation constructions involving 'Negative effect on subject'
(which almost never have 2nd person subjects), 'Negative effect on listener',
'Advisable action' with 2nd person subjects (which are basically for the good
of the 2nd person), and 'Procedure/routine'. There is rarely overlap between
these categories.
In Hopi, on the other hand, this categorization system generally does not
work. Many of the tokens in the obligation-function database involve people
in a superior position stating things which have to be done by people in an
inferior position (a rare situation in our English database, where there is a
more egalitarian relationship between the characters). It is often unclear
whether to categorize such cases as 'Procedure/routine'; when the basis of the
instructions seems to be more ritualistic the initial temptation is to code them
as 'Procedures', whereas when the leader seems to think out and improvise a
course of action on the basis of the specifics of the situation, these might be
coded as something else (e.g. 'Negative effect on subject', 'Advisable ac­
tion'). However, this essentially amounts to categorizing the instructions of a
Hopi leader on the basis of whether or not they can be rationally explained in
a Western frame of reference. For example, it is not clear how to categorize
instructions on conducting a raindance. While there are some general rules,
there is also considerable scope for improvisation, and so this sort of situation
lies somewhere in between 'Advisable action' and 'Procedures'.
Another difficulty with applying the categories of Section 2.1 to the Hopi
data lies in defining 'Negative effect' and determining who is negatively
affected. In the Hopi data, subjects are often required to perform unpleasant
tasks necessary for the good of their social group. These actions may be
'negative' for the subject but are at the same time 'Advisable actions', since
the listener, as a member of the group, will in some sense benefit (if only in the
long run); commanding such actions might even be considered hostile, since
the speaker is essentially imposing an unpleasant task on the listener. Tokens
of this type are common and difficult to code. In the English data, on the other
hand, this problem essentially does not come up; if the speaker imposes or
274 John Myhill & Laura A. Smith

Table 10. Use of nawus

Nawus No nawus
Decision at moment of crisis 6 1
Subject explicitly shows disinclination 11 5
Subject has been doing opposite 3 0
Specific word showing unpleasantness 6 3
Permanent separation 1 2
Other 0 126

suggests an obligation to the listener, it is almost invariably either clearly

hostile and against the interests of the listener, or else clearly cooperative and
supportive of the interests of the listener.
For these reasons, we felt that the categories used in the descriptions of
English and Chinese were inappropriate for description of the Hopi data; there
were simply too many cases where it was not clear how to apply them. We
have therefore developed other categories of obligation function to use in
analyzing the Hopi data. The database will be the same as that used in table 9,
consisting of all sentences in dialogues which have nawus in the Hopi and/or
an obligation marker in the English translation. Table 10 summarizes the
frequency of nawus for the different types of obligation functions.
We will describe and exemplify these functions below, drawing parallels
where appropriate with some of the categories used in Sections 2.1 and 2.2.
The first category is 'Decision at moment of crisis'. Nawus is used in this
situation 6/7 times. This often corresponds to 'Advisable action, 2nd pers.
subj.'; there is no clear evidence that the action is viewed as negative. This
generally occurs at the beginning of a story when a crisis comes up and the
people have to change their normal way of life to deal with it. For example, at
the beginning of one story about a boy and his mother, winter is coming on,
they are running out of firewood, and, unlike previous years, no one has
volunteered to get firewood for them. The mother uses nawus twice in telling
her son that they must take action:
(62) urn as nawus itamungem komoktoni. Itam hapi son
you for-us get-wood-FUT we really not
nawus kohot qa na'sastani(Malotki and Lomatuway'ma 1987a:12)
wood not prepare-FUT
'You must go gather some firewood for us. We really have to stock
up with wood...'
Functions of Obligation Expressions 275

In this case, as in others, once a course of action is decided upon, nawus is

not used again. The boy immediately agrees to go, and the mother then gives
him a series of instructions without using nawus. The English translation also
tends to avoid expressions of obligation in favor of imperatives, but there is
one use of the obligation marker need (You need only cut it into pieces about
this long (Malotki and Lomatuway'ma 1987a: 15)). Later, talking to someone
else, the boy says 'because there is no one to bring wood for us, I have had to
come here myself (Malotki and Lomatuway'ma 1987a: 15); here again, the
Hopi does not use nawus. Similarly, in another story, a group of Hopis learn
that they are about to be attacked by a group of Paiutes. Immediately, one of
them states: Pay itam tur nawus itaatumalay haak maatatve, 'We will have to
let go our work for the time being (to prepare for war)', using nawus (Malotki
and Lomatuway'ma 1987a:54). Once this decision has been made, various
leaders give instructions on how to deal with the conflict, but now that they
have embarked upon a course of action, nawus is not used again.
This pattern is consistent with the discussion of the distinction between
obligation and imperative function in Palmer 1986. Palmer hypothesizes that
imperatives present commands in a semantically neutral way, while obliga­
tion expressions are used in social contexts where giving commands is
marked behavior. Thus imperatives are appropriate when the speaker is in a
position of authority (e.g. officer to private) and/or the speaker and hearer
have temporarily determined to enter into a cooperative relationship (e.g.
preparing for a battle). Expressions of obligation with 2nd person subjects, on
the other hand, are most strongly associated with situations in which the
speaker wishes to give a command to the listener but no such command
relationship can be presupposed.29 The use of nawus is therefore associated
with a moment of crisis, when the subject must be convinced to reorder
priorities. Once this has been done, the obligation is presupposed rather than
asserted, and nawus is no longer used.
The next functional category for Hopi is 'Subject explicitly shows disin­
clination' to do the proposed action. Nawus is used 11 times in 16 tokens in
this situation (which often corresponds to 'Negative effect on subject', 'Ad­
visable action, 2nd pers. subj.', or 'Help not needed'). In one case, a man is
putting a repulsive mask on his daughter-in-law for ceremonial purposes. The
English translation of the relevant section is 'As he was about to place the
mask over her head, however, she shrank back from him', and then he says to
her (Malotki and Lomatuway'ma 1987a:234):
276 John Myhill & Laura A. Smith

(63) Ta'ay, pay urn son nawus it qa ang pakini

come PRT you not PRT not on put
'Come, you'll have to put this on.'
In another story, a Hopi leader has sent some men to make an offering to
the god Maasaw, but when they see how disgusting he looks, they run back
without making the offering. Their leader tells them Is ohiy, pay uma son
nawus qa piw awyani Too bad, you'll have to go back', with nawus (Malotki
and Lomatuway'ma 1987b:50). This category has some resemblance to the
first in that both are used specifically when the speaker is attempting to
convince the listener that an action must be undertaken.
In the third category, 'Subject has been doing opposite of what the
speaker wants', the subject is instructed to change his/her behavior; here
nawus is used three out of three times. This type is generally similar to
'Advisable action, 2nd pers. subj.'. In one case a man is performing a
ceremonial dance in an egotistical fashion, believing that the dance will
succeed specifically because of him. Others realize this and tell him that he
must confess to it (confession being required by Hopi ethics in this case). One
says to him:
(64) Urn oovi nawus itam umi nahostani
you so us to confess
(Geertz and Lomatuway'ma 1987:348)
'You must confess to us'
This usage of nawus is similar to the one exemplified in (63). The next
category is 'Specific word showing unpleasantness'; this was only counted if
the 'unpleasant' word is in the same clause (this category often corresponds to
'Negative effect on subject'). Nawus is used 6/9 times for tokens of this type.
For example, the god Maasaw informs a man:
(65) Noq oovi urn hiitawit uutiw'ayay, maanat, pas aw
PRT so you one your-niece girl very to
unangway'taqey put urn nawus tavini
beloved her you sacrifice
(Malotki and Lomatuway'ma 1987b:94)
'You'll have to sacrifice the niece who is most dear to you (to
obtain fat to keep the sun burning)'
Functions of Obligation Expressions 277

Here the unpleasantness of this obligation is emphasized with the term

unangway'taqey 'beloved', and nawus is used. Similarly, nawus is used in the
following sentence: Noq oovi itam yang utuhu'puva nawus qa oo'onakyangw
yang tutskwava hiita noonovaniqey put a' aniwnaya 'For this reason we Hopis
must suffer through hot weather and by toiling hard raise the crops in the
fields' (Malotki and Lomatuway'ma 1987a: 132). Here there are also specific
words emphasizing the unpleasantness of the obligation (utuhu'puva 'hot'
and oo'onakyangw 'get tired').
In the fifth category, 'Permanent separation,' the speaker informs the
listener of the necessity of a death or other permanent separation. In this
situation, nawus is used in one of three tokens. This type generally corre­
sponds to 'Negative effect on subject'. For example, the god Maasaw informs
the Hopi that they cannot all live in the same place and says, using nawus:
(66) Uma son nawus qa naanahoyyani
you not not go-in-different-ways
(Malotki and Lomatuway'ma 1987b:61)
'You'll have to migrate in different directions'
To summarize, we have seen that nawus is basically limited in its uses to
a subset of the uses of the English obligation markers. These can be grouped
into two types:
(i) At the moment of speech, there is a conflict between what the
subject has been doing or wants to do and the exigencies of the
situation. The speaker states the obligation to motivate the subject
(which may be 1st person plural) to change his/her priorities. This
includes the first three types discussed above.
(ii) The action undertaken is particularly unpleasant (or is at least
represented as such), but must be done anyway. This includes the
last two types discussed above. The great majority of the tokens
with obligation markers in the English translation do not fit either
of these types; in such cases, nawus is never used.

2.4. Biblical Hebrew.

Thus far, we have discussed markers in English, Chinese, and Hopi with a
variety of functions which can be considered to represent types of obligation.
Our analysis of Biblical Hebrew differs in that we will not attempt to catego-
278 John Myhill & Laura A. Smith

rize different types of obligation functions but rather to demonstrate that these
and related functions can be expressed through purely grammatical rather
than lexical means.
As argued in Section 1.2, obligation markers can develop functions
which are more grammatical, such as inference and future meaning, and we
would expect these more grammatical functions to be related to the more
lexical obligation functions in a non-arbitrary way. There will then be a
network of related functions consisting of different types of obligation func­
tion and also certain types of more grammatical functions like inference and
future. We will refer to this network of related functions as 'obligation' (with
scare quotes). Note that 'obligation' does not include all types of inference/
future functions, but only those related to obligation and likely to develop
from an erstwhile obligation marker.
This view of 'obligation' as representing a network of related functions
(ranging from strongly lexical to highly grammaticalized) is supported in an
interesting fashion by data from Biblical Hebrew. This language has no
lexical marker of obligation similar to English have to, Chinese dei, or Hopi
nawus. However, something resembling 'obligation' meaning is expressed
through word order alternation. Biblical Hebrew generally has verb-initial
order (69% of the time in our database), and the use of non-verb-initial order
has as one of its functions the expression of something resembling 'obliga­
tion'.
Before demonstrating this, we should note that we are not claiming that
'obligation' meaning is expressed through non-verb-initial order in Biblical
Hebrew. It is more accurate to say that the use of non-verb-initial order
expresses an abstract function which, in certain contexts, resembles 'obliga­
tion', and which, from the English point of view, is interpreted as 'obligation'.
For this reason, sentences using non-verb-initial order in these contexts are
consistently rendered in languages such as English with 'obligation' markers.
We have gathered data from all the conversations in the first 30 books of
Genesis and an English translation (Plaut 1981). Our database includes all
clauses with functions which might be considered future, subjunctive, opta­
tive, imperative, and/or modal, a total of 450 Hebrew clauses and their
translations. We then coded the data according to the Hebrew word order
pattern and the English translation.
Because Hebrew lacks a lexical obligation morpheme, our strategy for
analyzing must be different from that used for the other languages in this
Functions of Obligation Expressions 279

Table 11. Hebrew word order and English 'obligation ' markers
Non-verb-initial Verb-initial Non-verb-initial%
Shall 46 27 63%
Should 5 0 100%
Beto 4 1 80%
Must 2 1 67%
Total 'obligation' 57 29 66%
Other 81 283 22%
Chi-square=63.42 p<.001

study. Attempting to apply an abstractly defined, universal meaning of 'obli­

gation', or even more specific meanings such as those used in Sections 2.1 and
2.2, to a Biblical Hebrew text results in too many indeterminate cases where
the Hebrew does not unequivocally tell us whether 'obligation' function is
present. The alternative is to use translation data. English has a number of
words associated with 'obligation' function, and consistent use of these words
in translations of a construction indicates that that there is a connection
between that construction and 'obligation' function. Of course, the degree of
consistency will vary depending on which translation is used; however, as
will be seen, the pattern is so strong that we believe that it would appear no
matter which translation was used. Our results are shown in table l l . 3 0
We include here should, must, and be to, which mark obligation function
{must and should also have related functions such as inference), and shall,
which previously marked obligation and which now has future function while
retaining some of its old obligation meaning (Bybee and Pagliuca 1987); have
to and gotta did not occur in the translation. All of these words represent
'obligation' functions (functions associated with lexical obligation markers
and the related functions which these later develop). The coding of the
translation data was entirely on the basis of the lexical form used in the
English translation; no distinction was made between different usages of
shall, should, be to, and must.
Table 11 shows that non-verb-initial order is three times as common
when an 'obligation' marker is present in the English translation as when the
English has no obligation marker (66% vs. 22%). There is thus evidence of a
strong relationship between non-verb-initial order and 'obligation' meaning.
This relationship is by no means categorical; 22% (81/364) of the non-
'obligation' clauses are non-verb-initial and 34% (29/86) of the 'obligation'
280 John Myhill & Laura A. Smith

clauses are verb-initial. However, a large number of these exceptions can be

explained as due to other functions of word order alternation in Hebrew and
auxiliary usage in English. Non-verb-initial order in Hebrew is associated not
only with 'obligation' but also with clauses with contrastive topicalization or
focusing (Fox 1983), as in (67)-(68) (these and following translations are
from Plaut 1981):
(67) ten-li hanefesh vharxush qakh-lax. (Genesis 14:21)
give-me the-persons and-the-possessions take-for-yourself
'"Give me the persons, and take the possessions for yourself.'"
(68) shvu-laxem poh 'im-hakhamor va?ani vhana'ar
stay here with-the-ass and-I and-the-boy
nelxah 'ad-koh. (Genesis 22:5)
go up-there
"'You stay here with the ass. The boy and I will go up there.'"
In such contrastive pairs, non-verb-initial order is normally only used for
the second contrastive clause, not the first (Myhill and Xing 1992); these
clauses account for a large number of non-verb-initial constructions not
translated as 'obligation' markers. In presenting these examples, we are not,
of course, implying that there is any tendency to avoid 'obligation' markers in
English in sentences with contrastive function, but rather that contrast pro­
vides another motivation for using non-verb-initial order in Hebrew; this
accounts for many of the 81 non-verb-initial constructions in our data which
are not translated with 'obligation' markers.
We turn now to another exception to the correlation of non-verb-initial
clauses in Hebrew and 'obligation' markers in English. Of the 29 exceptional
clauses using verb-initial order and translating with an 'obligation' auxiliary,
22 share the following three features:
(a) God is talking;
(b) There is a 2nd or 3rd person subject;
(c) The clause does not occur in a plausible context for using the
imperative.31
When these three features are present, there is an overwhelming tendency
for the English translation to use 'obligation' markers (particularly shall)
regardless of the word order in Hebrew: out of 63 tokens with these three
features, 56 use shall, 3 use be to, 1 uses must, 2 use will, and 1 uses can; in
similar cases which are plausible contexts for imperatives, of course, the
Functions of Obligation Expressions 281

translation simply uses an imperative (e.g. 'Go from your land...'). It appears
that the translators are following a stylistic convention of using shall in God's
speech about someone else's future actions (excluding imperatives); this
occurs 89% of the time. The most likely explanation for this is that shall has
acquired an archaicizing function independent of its' obligation' function and
so is particularly suitable for conveying God's prognostications and pro­
nouncements about the fate of humans.32
Of the remaining 7 tokens (out of 29) of verb-initial constructions excep­
tionally translated with English 'obligation' markers, all follow the equivalent
of if or when clauses in Hebrew, e.g. (69):
(69) v?im-lo? to?veh ha?ishah lalexet akhareixa
and-if-not willing the-woman to-go after-you
vniqatah mishvu'ati. (Genesis 24:8)
you-will-be-free from-my-oath
'And if the woman is not willing to go with you, you will be free
from my oath.'
Such apodosis clauses are invariably verb-initial in our Biblical Hebrew
database (in 25 tokens), regardless of whether they translate with 'obligation'
words. We have shown, then, that one function of non-verb-initial order in
Biblical Hebrew is related to 'obligation'. Since this is not the only function of
this order (it also marks contrast), we are not claiming a categorical correla­
tion between this order and translation with 'obligation' markers, but 'obliga­
tion' is a major function of this order. The use of word-order alternation to
mark 'obligation' function presumably compensates to some extent for the
fact that the language does not have a lexical obligation marker.
It is not at this stage entirely clear why Biblical Hebrew should mark
'obligation' function with word order. Cross-linguistic research will be neces­
sary to determine whether the relationship between non-verb-initial order and
'obligation' is widespread and based on general cognitive principles or simply
restricted to Biblical Hebrew and more or less coincidental. We might tenta­
tively hypothesize that this relationship may be related to the notion of
continuity.33
Givón (1977) and others have argued that in narrative discourse, the
unmarked order of presentation is 'iconic sequence' (the order of clauses
mirrors the order in which the events reported in these clauses took place);
more specifically, temporally sequenced clauses are continuous and unse-
282 John Myhill & Laura A. Smith

quenced clauses are discontinuous. Givón shows that in Biblical Hebrew,

verb-initial order is associated with continuous (temporally sequenced)
clauses, while non-verb-initial order is associated with clauses which are
discontinuous (unsequenced) clauses (Myhill 1992 argues that this pattern is
in fact universal in languages with predominant VS word order).
The continuous/discontinuous dichotomy applies only to narrative por­
tions of a text, where past actions are reported; however, a related distinction
emerges in other types of text. For example, we can say that apodosis clauses,
as in (69), are 'continuous' because they have a similar function in irrealis
time (chronologically following the protasis) as sequenced clauses have in
past time. The fact that apodosis clauses are categorically verb-initial in
Biblical Hebrew is then consistent with the general pattern of verb-initial
clauses being continuous in this language. Clauses involving contrast (as in
(67), on the other hand, are not verb-initial and can be construed as discon­
tinuous, because they interrupt a narrative or a chain of logical reasoning for
the purpose of making a comparison between two entities.
If we apply the concept of continuity to conversations about future time,
we can associate it with cooperation, while discontinuity is associated with
'obligation'. We hypothesize that the unmarked assumption in a conversation
about the future is that the interlocutors will talk about it cooperatively.
Myhill 1993 provides quantitative data showing that verb-initial order in
future clauses in Biblical Hebrew is associated with situations involving some
type of cooperation, e.g., the speaker proposes a course of action to the
listener, one event enables another (a purpose clause), or the speaker hopes
that something good will happen to the listener. On the other hand, 'obliga­
tion' meaning and non-verb-initial order are associated with contexts lacking
'cooperation' of this type, e.g., the speaker predicts something bad for the
listener, or the speaker makes a unilateral statement about the future which
does not consider the actions and feelings of the listener.
The distinction between cooperation and 'obligation' also appears to be
expressed by the contrast between verb-initial and non-verb-initial order in
commands. When the speaker can presuppose the listener's cooperation (be­
cause the speaker is in a higher-ranking position or because the interlocutors
have already agreed upon a plan), the Imperative form is used, and it is almost
always verb-initial (Myhill 1993). On the other hand, when the speaker cannot
presuppose the listener's cooperation (because the speaker is not in a position
to command the listener), non-verb-initial order is usual, as in (70):
Functions of Obligation Expressions 283

(70) ?elay tavo?. (Genesis 30:16)

to-me you-come
'"You are to sleep with me."'
The speaker here is Leah addressing her husband Jacob. She has just
bought him for the night from her sister, his other wife Rachel. Normally,
wives do not command their husbands in this text, so the social context for
presupposing the right to command is not present here. Thus, the Hebrew uses
a non-verb-initial construction with the Imperfect form of the verb (rather
than the Imperative) and the English translation uses an obligation expression.
This distinction between imperative and 'obligation' function is in line with
Palmer's (1986) argument (mentioned in Section 2.3) that imperatives are
semantically neutral, used in situations where the speaker is in a position to
command the listener, while obligation marking is appropriate for commands
when the speaker is not in a position to give orders. Stating the obligation is
thus a justification for giving the command; this is presumably also related to
the morphologically unmarked status of imperatives as opposed to obligation
expressions.
This account of why non-verb-initial order is associated with 'obligation'
function in Biblical Hebrew is of course speculative, and the matter will have
to be researched in other languages with a high incidence of verb-initial order.
For the moment, it seems to be at least a reasonable way to approach the
question of why Biblical Hebrew should have a means of expressing obliga­
tion so radically different from that of the other languages discussed in this
paper.

3. Conclusion.

The main purpose of this study has been to develop a framework for the cross-
linguistic description and comparison of obligation functions in natural usage.
We have discussed and exemplified types of obligation functions in four
languages and correlated these with obligation markers in particular lan­
guages. We have gone to some length to describe and exemplify the functions
identified in our database so that linguists working in other languages might
test them against their data. The functions generally correlate with particular
markers in the languages we have investigated, and when a given function
284 John Myhill & Laura A. Smith

occurs commonly with more than one marker we have attempted to divide it
into subfunctions which might account for the distribution of markers. Given
the relatively small size of our database, we cannot expect to have entirely
determined the factors governing the choice of forms, nor do we expect our
list of obligation functions to be exhaustive.
It is important to note here what we have not attempted to do. We have
not attempted to define exactly what obligation means in general. After
analyzing a large number of expressions of obligation in the languages
investigated here, it is clear to us that obligation expressions do not simply
report objectively necessary actions. Rather, they are devices used by speak­
ers to evaluate and justify actions, so that the only way to analyze their
function is in terms of usage rather than 'meaning'. We have of course used an
intuitive notion of obligation to delimit our database, but this has been done
purely for convenience; neither our functional categories nor the morphemes
we have chosen to analyze are intended to exhaustively represent or define
obligation.
We have also not attempted to organize the categories of obligation
function into 'core' and 'peripheral' types, along the lines of Coates 1983. It
would of course be possible to choose one or another function as 'core'
obligation and rank the others according to the number of shared semantic
features. However, there appears to be no objective basis for choosing one
function over another as 'core', nor any obvious motivation for organizing the
data in this fashion.
It is probably the case that some of obligation functions are more lexical
while others are more grammatical. Research such as Bybee 1985 has demon­
strated that some functions are consistently associated with more lexical
representation while other functions are consistently associated with more
grammatical representation, and it is clear that the future and inference
functions which obligation markers can develop are more grammatical than
the obligation functions we discuss here. However, among the obligation
functions we identify here, it is not clear which are more lexical and which
more grammatical. To determine this, it would be necessary to gather data
from a wide variety of languages and to look for cases in which some of the
functions are expressed through more lexical forms and others through more
grammatical forms. Alternatively, we might study in detail the diachronic
development of a single marker in a single language to see how this marker
has progressed from one function to another. If there is a change in terms of
Functions of Obligation Expressions 285

the predominant functions of the marker, we might hypothesize that the

earlier functions are more lexical and the later functions more grammatical.34
This line of research is likely to produce interesting results; however, it is
premature at this stage to speculate on what they might be.
We have not developed a single list of obligation functions which can be
applied to all texts from all languages; the obligation functions which we used
for the English and Chinese texts turned out not to be the most appropriate or
revealing ones for the Hopi and Biblical Hebrew texts. The use of obligation
markers is obviously closely connected to the organization of a society, and
the societies associated with these languages differ in a number of ways. This
does not invalidate the search for a 'universal' list of obligation functions.
However, such a task is beyond the scope of a single paper; additionally,
making such a list would require data from more than four languages. Rather
than attempt such an ambitious project here, we have analyzed the data from
each language in a way which seemed most appropriate to that language. This
study has therefore not resulted in a comprehensive list of obligation functions
perfectly suited to texts in all languages, but rather has identified a variety of
functions which may or may not prove useful in the (long-term) goal of
constructing such a list.
We suspect, however, that even if such a comprehensive list of obligation
functions were to be constructed, it would not necessarily solve the problem
of cross-linguistic comparability. For example, on the basis of the Hopi data,
we could make a category like 'Instructions for performing a ritual spiritually
beneficial to the community as a whole', which might be objectively applica­
ble to data from any language. However, this context would occur rarely if
ever in original texts in languages such as English and Chinese. This difficulty
can be alleviated somewhat through the use of translation data (e.g. English
translations of Hopi texts); additionally, we might look for data from certain
sections of a society where this context would be most likely to come up. This
may sound like a formidable task, but it seems to us the only way to compare
obligation functions in widely differing languages in an empirically responsi­
ble fashion. The alternative would be to have a vague understanding of
obligation as a general universal concept, and to say that speakers of indi­
vidual languages interpret this concept in accordance with their social struc­
tures and world view; however, it seems to us that this approach will not lead
to much in the way of solid empirical findings either in descriptions of
individual languages or in cross-linguistic comparisons.
286 John Myhill & Laura A. Smith

The present paper, then, has sought to show that obligation function can
be productively analyzed with quantitative methods. Largely due to historical
accident, quantitative discourse-oriented research has focused primarily on
word order and voice alternations, and it is important that linguists working in
this field expand the range of phenomena under investigation. Quantitative
methods produce results which cannot be obtained from introspection alone,
and they force the researcher to think in terms of objectively defineable
criteria. We do not deny the value of introspective research; however, we see
few if any linguistic phenomena which cannot benefit from quantitative as
well as introspective investigation.

NOTES
* We thank Joan Bybee and Suzanne Fleischman for their extensiveand helpful comments
on an earlier draft of this paper.
1 Both Palmer (1979:91) and Coates (1983:33) admit that it is often difficult to categorize
tokens according to the criterion of speaker involvement.
2 Shall is very rare in the original English texts we used and in the English translations of
the Hopi and Chinese texts. It is, on the other hand, quite common in the translation we
used of the Biblical Hebrew text. However, this is not a problem, because in Biblical
Hebrew we are analyzing 'obligation' function (including the more grammaticalized
functions which obligation markers develop) rather than obligation function, so there is
no need to distinguish the obligative shall from its future-oriented uses.
3 Lambdin (1971:129) notes that the expression 'al + NP + / + infinitive is used to express
obligation or responsibility, but there were no tokens of this construction in our data­
base.
4 Except where otherwise indicated, gotta will include have got to, got to, and gotta, and
oughtta will include ought to . Must occurs only once in Wilson's dialogue, need to only
four times. Hence, we have not included them in this study.
5 Does (subj) gotta? is acceptable to some speakers, but it does not occur in our data. We
use the asterisk to indicate non-occurrence in our data base, not to indicate that we or the
playwright would reject a form as ungrammatical (although in some cases we would).
6 In one token, the subject is actually avoiding something emotionally or physically
unpleasant (No, you do that 'cause you got to, 'cause you'd be embarrassed not to
(Wilson 1976:117)).
7 This speaker-external source of authority is probably related to the low 'speaker involve­
ment' or low 'subjectivity' associated with have to (see Section 1.1).
8 Conceptually, of course, reprimands are a type of counterfactual, e.g., You shouldn't
have done that is counterfactual because the auxiliary is negated but the addressee must
have done the action. We distinguished these categories in our coding by including as
reprimands only counterfactual clauses with 2nd person subjects.
Functions of Obligation Expressions 287

9 Two of the supposed to tokens may also be partly motivated by negation (...you aren't
supposed to be here rather than you ought not be here) and/or past tense; we get neither
* ought not nor *didn 't ought in our data.
10 Our oughtta reprimands are somewhat borderline obligation usages, since their verbs are
Stative (be ashamed, be used to). You cannot oblige someone to 'perform' a state; on the
other hand, you can oblige someone to perform an action that avoids or effects a state,
and on this reasoning, we have included such tokens.
11 This discussion of counterfactuals is not intended to be exhaustive; a much larger
database is needed for a thorough and systematic analysis of natural usage.
12 The association of supposed to with arrangements by unmentioned agents is likely due to
its origin as a passive construction (we thank Joan Bybee for suggesting this).
13 One have to token is negated with don't, which may explain the speaker's not using gotta.
14 Coates (1983:54) observes that, in its 'root' or obligation meaning, have got to is never
habitual while have to can be. She cites the distinction between / have to get up at 7 a.m.
every day and *I've got to get up at 7 a.m. every day.
15 Beijingren was written in 1940 and slightly revised by the author in 1954; in the text, we
cite it as Cao 1954. Leiyu was written in 1934 and revised twice by the author before its
1961 printing, and we cite it as Cao 1961.
16 This is important because dei is commonly recognized as a Beijing colloquialism. As a
Tianjin native (see Rand et al 1980), Cao Yu is not a representative speaker of the
Beijing dialect. It would be interesting to compare the dei vs. yao distinction in Cao's
work to that of a Beijing native, where some of the yao functions described here would
presumably be filled by dei.
17 All of these can also be used as main verbs, often translating as 'take, require', 'want',
and 'owe', respectively. Under gai we have included 7 tokens of yinggai, the latter often
described as a more formal variant. Though there are no doubt some differences in their
functions, we will not attempt to distinguish them here.
18 Our abbreviations are: ASP=aspect, PRT=particle, FUT=future, CL=classifier.
19 Unless otherwise indicated, translations of Beijingren examples are from Lo 1986 and
translations of Leiyu examples are from Wang and Barnes 1954.
20 The two exceptions refer to obligations to father-in-law and family, respectively. But
even these obligations are to some extent also specific to the circumstances: The
speaker's husband, who usually assumed part of this responsibility, has left.
Another feature distinguishing dei and yao tokens in the 'Negative effect on
subject' category is word order. In our data, most transitive verbs with yao (10/12, or
83%) have preverbal patients (see (44)). This is significant given that none of the four
transitive dei constructions in this category have preverbal patients, and yao construc­
tions with preverbal patients using are rare (less than 10%) in our other functional
categories and in the language overall (Myhill and Xing 1992). It is not clear why yao
utterances which are 'Negative effect on subject' should have this unusual word order
pattern, but we did notice that patient-verb order is always used in the 8 instances where
the action is not marked as inherently negative. It may be that the patient-verb order is
used here to emphasize that the subject is obligated to do the action and to discourage the
intentional reading.
288 John Myhill & Laura A. Smith

21 This is our translation; Wang and Barnes 1958 omit this sentence in their translation.
22 This translation is our own and directly reflects the structure of the Chinese. The
translation in Lo 1986 is much looser: 'Shouldn't we listen to our consciences?'
23 See Li and Thompson 1981:307 for a description of the function of this particle.
24 Unfortunately, the translators appear to have had a stylistic bias against (have) got to;
they used it only once, in contrast with Wilson, who uses it frequently.
25 We have excluded 'Don't have to' clauses from this table, since these are not expressed
with any of the Chinese morphemes studied here.
26 Except for the four other tokens translated into English with an obligation morpheme.
27 We have decided to analyze only nawus after extensive study of Hopi texts and Albert
and Shaul's 1985 Hopi-English lexicon. We investigated several other words as possible
obligation markers. Albert and Shaul translate paapu as 'admonition (modal), should do
it (reminder)', but in the texts we consulted, the English translations of sentences with
paapu almost never have a word with obligation meaning. Albert and Shaul list 'should
do' as one of the functions of as, but this is an extremely common discourse particle with
a wide variety of functions, and only a fraction of the tokens using it have any obligation
meaning. They refer to naapas as a 'modal of disapprobation' ('you shouldn't do that');
naapas is used with this function in our database, but there are only three tokens of it.
Sonqa sometimes appears to function as a marker of obligation. However, it really
means 'definitely' (it is composed of son and qa, both negative markers). It translates
considerably more often as some non-modal marker of certainty than as an obligation
marker. It is of course an interesting question why certain markers which are not
basically obligation markers occur consistently (though infrequently) in sentences trans­
lating into English with explicit obligation; this is relevant to the present study, but
beyond its scope.
28 We include here only the obligation meanings of these English words. This contrasts
with the translation data from Biblical Hebrew, where we include the usages of the
English words associated with obligation, whether or not the particular token appears to
represent obligation. The motivation for this distinction, as discussed in Section 1.2, is
that in Hopi we are dealing with a lexical, literal expression of obligation (nawus), which
we want to compare with literal obligation in English, while in Biblical Hebrew, as we
will see, we are dealing with a grammatical expression of obligation (signaled by a word
order alternation). We therefore want to include both concrete (literal) expressions of
obligation in English and the more abstract (grammatical) meanings related to obligation
which words associated with the expression of obligation later develop.
There are two tokens of nawus which we have marked as being translated with 'no
obligation word'. One is translated as 'It will take you several days to reach your goal'
(Malotki and Lomatuway'ma 1987a: 116), more literally 'You will have to arrive at your
goal in several days'. The other is translated as 'It's Maasaw only who rests during the
day and goes around at night' (Malotki and Lomatuway'ma 1987b:89); this is a repri­
mand to someone who is sleeping too late, and it is more literally translated as 'It should
be only Maasaw who...' Both of these clearly convey some type of obligation meaning.
However, the fact that they are represented with an obligation form in Hopi but not
English shows that it is not totally accurate to say that nawus marks a subset of the
functions marked by English obligation markers.
Functions of Obligation Expressions 289

29 It is of course the case that mothers are generally in a social position which sanctions
giving commands to their children; thus mother-to-child commands are generally rela­
tively unmarked. However, in the Hopi texts even this type of relationship calls for
nawus when the speaker is trying to orient the listener towards a new goal in order to deal
with a crisis which has arisen.
30 The data here also include subjects possessed by humans (e.g. 'His soul shall be cut off
from his people'). They exclude sentences with negatives and question words (which are
obligatorily clause-initial) and subordinate clauses (which have subordinators in initial
position).
31 Implausible imperative contexts are:
(a) The listener can't control the action because the verb doesn't take a volitional subject
(e.g. 'You shall be the father of many nations').
(b) The listener can't control the action because it is addressed to him and his descend-
ents (e.g. 'You shall keep my covenant').
(c) The listener can't control the action because it is addressed to other people not
present (e.g. 'You shall come into the ark, you and your sons and your wife and your
sons' wives with you.').
(d) The listener can control the action, but it is contingent on another action which the
listener cannot control (e.g. 'Your wife Sarah shall bear you a son, and you shall call
his name Isaac').
32 It appears that a number of the non-imperative constructions with (a) God speaking;
(b) a 2nd or 3rd person subject; (c) non-verb-initial order in Hebrew; and (d) shall in the
English translation are cases in which the Hebrew text portrays God as speaking in a
more cooperative, human fashion (which is not uncommon), but the English translation
nevertheless represents God in a transcendent manner. Examples are 'you shall be buried
at a ripe old age' (Genesis 15:15), where God makes a personal statement about
Abraham's future, and 'your descendants shall seize the gates of their foes' (Genesis
22:17), where God is rewarding Abraham for his willingness to sacrifice his son Isaac,
and so there is an element of cooperation. This may be compared with the more typical
case, where God simply makes a promise independent of any particular human action,
and non-verb-initial order is normal (e.g. 'Kings shall come from her' (Genesis 17:16)).
Scholars such as Heschel (1975) and Rosenberg and Bloom (1990) have argued, for
completely different reasons, that there is no evidence for a purely transcendent concep­
tion of God in the Old Testament but that interpreters of the text influenced by Greek
thinking have overemphasized the transcendent aspect in their translations and interpre­
tations; this would account for tokens where shall is used to translate verb-initial clauses.
33 Another possible explanation is that the verbal form typically used in non-verb-initial
constructions with 'obligation' function (the Imperfect) is historically derived from a
modal marker of obligation plus a verb which subsequently fuss together. The Imperfect
is characterized by personal prefixes agreeing with the subject — possibly derived from
fusion of a pronoun + obligation marker which eventually became prefixed to the verb.
This account is purelyspeculative, and we do not know of any historical evidence for it.
Additionally, any such account would be complicated by the fact that the Imperfect is
largely syncretic with two other forms, the Va-consecutive and the Jussive, which are
invariably (for the Va-consecutive) or almost invariably (for the Jussive) verb-initial.
290 John Myhill & Laura A. Smith

34 It is tempting to speculate that, since Hopi has the most restricted overt marking of
obligation of the languages investigated, the functions associated with nawus are the
most lexical, while other functions are more grammatical. This may turn out to be true.
However, even if nawus does represent a more lexical type of obligation function, it does
not necessarily represent the only such function; obligation markers are themselves
diachronically derived from words with other meanings, and presumably the function of
a given obligation marker will depend upon the meaning of its source.

DATA SOURCES

Cao, Yu. 1962 (1934, rev 1961). Leiyu ('Thunderstorm'). In: Caoyu Xuanji (Selected
Plays by Caoyu). Beijing: People's Literary Press.
Cao, Yu. 1962 (1940, rev 1954). Beijingren ('Peking Man'). In: Caoyu Xuanji (Selected
Plays by Caoyu). Beijing: People's Literary Press.
Geertz, Armin W. and Michael Lomatuyway'ma. 1987. Children of cottonwood: Piety
and ceremonialism in Hopi Indian puppetry. American Tribal Religions, vol. 12.
Lincoln: University of Nebraska Press.
Lo, Leslie Nai-Kwai (et al), transl. 1986. Cao Yu Peking Man. New York: Columbia
University Press.
Malotki, Ekkehart, and Michael Lomatuway'ma. 1987a. Stories of Maasaw, a Hopi god.
American Tribal Religions, vol. 10. Lincoln: University of Nebraska Press.
Malotki, Ekkehart, and Michael Lomatuway'ma. 1987b. Maasaw: Profile of a Hopi god.
American Tribal Religions, vol. 11. Lincoln: University of Nebraska Press.
Plaut, W. Gunther. 1981. The Torah: A modern commentary. New York: Union of
American Hebrew Congregations.
Wang, Tso Liang and A. C. Barnes, transl. 1958. Thunderstorm (Translation based on
minor 1954 revisions). Beijing: Foreign Language Press.
Wilson, Lanford. 1973. The Hot L Baltimore. New York: Hill and Wang.
Wilson, Lanford. 1976. The Mound Builders. New York: Hill and Wang.
Wilson, Lanford. 1987. Burn This. New York: Hill and Wang.

REFERENTIES

Albert, Roy, and David Leedom Shaul. 1985. A Concise Hopi and English Lexicon.
Amsterdam: John Benjamins.
Brown, Penelope, and Stephen C. Levinson. 1987. Politeness: Some Universals in Lan­
guage Usage. Cambridge: Cambridge University Press.
Bybee, Joan. 1985. Morphology: A Study of the Relation between Meaning and Form.
Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan, and William Pagliuca. 1987. "The Evolution of Future Meaning." Papers
from the 7th International Conference on Historical Linguistics, ed. by A.G. Ramat et
al., 109-122. Amsterdam & Philadelphia: John Benjamins.
Functions of Obligation Expressions 291

Chao, Yuen Ren. 1968. A Grammar of Spoken Chinese. Berkeley: University of Califor­
nia Press.
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries. London: Croom Helm.
Cooreman, Ann. 1988. "Ergativity in Dyirbal Discourse." Linguistics 26.717-746.
Cooreman, Ann, Barbara Fox, and T. Givón. 1984. "The Discourse Definition of
Ergativity." Studies in Language 6.343-374.
Fox, Andrew. 1975. "Topic continuity in Biblical Hebrew narrative." Givón 1983. 215-54.
Givón, T. 1975. "Serial Verbs and Syntactic Change: Niger-Congo." Word Order and
Word Order Change ed. by Charles N. Li, 47-112. Austin: University of Texas Press.
Givón, T. 1977. "The Drift from VSO to SVO in Biblical Hebrew: The Pragmatics of
Tense-Aspect." Mechanisms of Syntactic Change ed. by C.N. Li, 181-254. Austin:
University of Texas Press.
Givón, T. (ed.). 1983. Topic Continuity in Discourse. Amsterdam & Philadelphia: John
Benjamins.
Heschel, Abraham J. 1975. The Prophets, Vol. II. New York: Harper and Row.
Herring, Susan. 1990. "Topic and Focus Positions as a Consequence of Word Order Type:
What is Universal?" Paper presented at the International Pragmatics Conference.
Hopper, Paul. 1979. "Aspect and Foregrounding in Discourse." Discourse and Syntax
{Syntax and semantics 12), 213-241. New York: Academic Press.
Lambdin, Thomas O. 1971. Introduction to Biblical Hebrew. New York: Charles
Scribner.
Li, Charles N. and Sandra A. Thompson. 1981. Mandarin Chinese: A Functional Refer­
ence Grammar. Berkeley: University of California Press.
Liu, Yuehua. 1983. Shiyong Xiandai Hanyu Yufa (Chinese Grammar). Beijing: Foreign
Language Teaching and Research Press.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Myhill, John. 1984. "A Study of Aspect, Word Order, and Voice" University of Pennsyl­
vania Ph.D. disseration.
Myhill, John. 1986. "The Two VS Constructions in Rumanian." Linguistics 24.331-350.
Myhill, John. 1992. "Word Order and Temporal Sequencing." The Pragmatics of Word
Order Flexibility ed. by Doris Payne, 265-278 Amsterdam & Philadelphia: John
Benjamins.
Myhill, John. 1993. A Verb Position in Future Clauses in Biblical Hebrew. Language
Variation and Change 4.289-309.
Myhill, John, and Zhiqun Xing. 1992. "Towards an Objective Definition of Contrast."
University of Michigan manuscript.
Palmer, Frank. 1974. The English Verb. London: Longman.
Palmer, Frank. 1979. Modality and the English Modals. London: Longman.
Palmer, Frank. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Payne, Doris. 1987. "Information Structuring in Papago Narrative Discourse." Language
63.783-804.
Rand, Christopher C. and Joseph S. M. Lau, transl. 1980. The Wilderness (Yuan-Yeh), by
Cao Yu. Bloomington: Indiana University Press.
Rosenberg, David, and Harold Bloom. 1990. The Book of J. New York: Grove
Weidenfeld.
292 John Myhill & Laura A. Smith

Sun, Chao-Fen, and T. Givón. 1985. "On the So-Called SOV Word Order in Mandarin
Chinese." Language 61.2.329-51.
Thompson, Chad. 1989. Voice and Obviation in Athabaskan and other Languages.
University of Oregon Ph.D. dissertation.
Ultan, Russell. 1978. "The Nature of Future Tenses." Universals of Human Language,
vol. 3, ed. by J. Greenberg. 83-123, Stanford: Stanford University Press.
 Apprehensional Epistemics

Frantisek Lichtenberk
University of Auckland

1. Introduction1

Epistemic modality, one type of which is the focus of this study, is character­
ised by subjectivity, by pragmatic force. It has to do with "belief, knowledge,
truth, etc. in relation to proposition" (Palmer 1986:96), and with "the degree
of commitment by the speaker to what he says" (Palmer 1986:51).
Many languages exhibit a historical relation between epistemic modality
on the one hand and other types of modality on the other. Bybee and Pagliuca
(1985:63) refer to the latter group collectively as 'agent-oriented modalities',
including the notions of ability, obligation, desire, and intention. The results
of their extensive study suggest a unidirectional process of development of
agent-oriented modalities into epistemic modalities (Bybee and Pagliuca
1985:66). Developments in the opposite direction do not occur. To say that
agent-oriented modalities develop into epistemic modalities rather than vice
versa is, of course, not to say that epistemic modalities can only develop from
agent-oriented modalities. It remains an empirical matter to determine what
other sources of epistemic modalities there are; the type of epistemic modality
to be discussed here does not (typically) develop from agent-oriented modali­
ties.
This study is concerned with a type of modality that is both epistemic and
attitudinal: it has to do with the speaker's degree of certainty about the factual
status of a proposition and also with his or her attitude concerning the
desirability of the situation encoded in the clause. This is a case not of
polysemy but of mixed modality (von Wright 1951): both types of modality
294 Frantisek Lichtenberk

are present simultaneously. For reasons that will become apparent later, I will
refer to this mixed modality as 'apprehensional-epistemic'.
The development of apprehensional-epistemic modality is of relevance
to the study of semantic/functional change, specifically to the issue of the
directionality of change in terms of degrees of subjectivity and pragmatic
strength. Traugott (1989 and elsewhere (see Section 5 for references)) has
suggested that semantic change is characterized by an increase rather than a
decrease in subjectivity, by pragmatic strengthening rather than weakening.
This is true of semantic change in lexemes and also of the early stages of
grammaticalization. Although the data discussed here by and large support
Traugott's hypothesis, they at the same time suggest a refinement.
I will start by discussing in some detail data from To'aba'ita, an Austro-
nesian language spoken in the Solomon Islands. Although the information on
the present-day uses of the element in question is reasonably detailed, there is
no direct historical evidence concerning the development of the modality. To
overcome this problem, I will first consider some indirect, structural and
historico-comparative evidence, and, following that, data from several other
languages where similar independent developments have taken place. In the
case of two of those languages, there is some direct historical evidence
bearing on the issue. From the various kinds of evidence available, a general
pattern in the development of the modality emerges even though there are
some differences of detail among the languages.

2. Apprehensional-epistemic modality in To'aba'ita

To'aba'ita has a modality marker of the form ada, which is formally a

complementizer. As demonstrated in examples (1), (3), (5), and (6), ada is
used to signal relative uncertainty on the part of the speaker about the factual
status of the proposition; in other words, ada functions as an epistemic
downtoner. For convenience, I will gloss ada and similarly functioning forms
in other languages as 'LEST' ; this does not, however, mean that the uses of
these elements are necessarily exactly the same as those of English lest.

(1) Ada 'oko mata'i.2

LEST you(SG):SEQ be sick
'You may be sick.'
Apprehensional Epistemics 295

Compare:

(2) 'Oe 'o mata 'i.

you(SG) you(sG):FACT be sick
'You are sick.'
(3) Ada dani ka 'aru.
LEST rain it:sEQ fall
'It may rain.'
Compare:
(4) Dani kai 'aru.
rain it:NONFACT fall
'It will rain.'
(5) Ada keka fanga sui na'a.
LEST they:SEQ eat COMPL PERF
'They may have finished eating.'
(6) Ada bii na 'i ka a 'i si
LEST food in stone oven this it:SEQ NEG.VB it:NEG
'ako ba -na.
be cooked LIM-its
'The food in the stone oven may not be done yet.'
While ada functions as an epistemic downtoner in the examples above,
this is not its only function there. Besides signaling less-than-full certainty on
the part of the speaker about the factual status of the proposition, ada also
signifies that the possible situation is in some way undesirable. Following
Dixon (1977), I will refer to this function as 'apprehensional': there is appre­
hension — typically on the part of the speaker — that a potential (undesirable)
situation may turn out to be so. The undesirability of the situation encoded in
(1) is obvious; (3) could be used to express fear about being caught in the rain;
(5) might be used if, for example, the speaker was worried about not getting
any food; and (6) could be used as a warning against premature opening of a
stone oven (both the closing and the opening of a stone oven are laborious
processes).
In the examples above, ada is a mixed-modality marker: it has an
epistemic function, as a downtoner, and it has, what, following Palmer (1986),
296 Frantisek Lichtenberk

one may call a 'volitive-modality' function, more specifically an apprehen-

sional function: there is fear that an undesirable state of affairs may obtain. 3
As will be seen later, ada has some other uses besides apprehensional-
epistemic. Apprehensional-epistemic modality is highly 'subjective' in the
sense of Traugott 1989: it reflects the speaker's uncertainty about whether a
state of affairs does, did, or will obtain, together with his/her own evaluation
of that state of affairs as undesirable (for example, rain is not necessarily
undesirable, as after a period of drought).
The examples above demonstrate that ada can be used with propositions
of any temporal status: present (1), future (3), or past (5). Nevertheless, the
subject/tense marker is always the same: Sequential. The Sequential subject/
tense markers are used elsewhere to indicate temporal sequence: the situation
encoded by a verb with a Sequential subject/tense marker follows the situation
encoded in the preceding clause:

(7) Buriana kera oli Siriaoa ka lae mai

after they:FACT return S. it:SEQ go hither
ka dau i maa-na 'aba tha Bari'i.
it:SEQ alight on eye-its arm ART B.
'After they [people] had gone back, Siriaoa [a bird] alighted on
Bari'i's wrist.'
Since ada sentences require Sequential, rather than other (Factative or
Nonfactative), subject/tense markers, the temporal status of the proposition
being expressed is not overtly signaled; thus (1) could be used to express not
only 'You may be sick' but also 'You may have been sick' and 'You may get
sick'. 4
Ada is used not only with an apprehensional-epistemic function, as in the
examples above. It is also used as a complementizer with clauses embedded
under a verb of fearing; there the apprehension about an undesirable situation
is expressed by the higher verb:
(8) Nau ku ma 'u 'asia na'a ada laalae to'a baa
I : F A C T be.afraid very LEST later people that
ki keka lae mai keka thaungi kulu.
PL they:SEQ go hither they:SEQ kill US(INCL)
T am scared the people might come and kill us.'/'I am scared lest
the people should come and kill us.'
Apprehensional Epistemics 297

I will refer to the use of ada (and the LEST elements in other languages)
with clauses embedded under predicates of fearing as the 'fear' function. In
the fear function also, the embedded clause introduced by ada must have a
Sequential subject/tense marker, regardless of the temporal status of its propo­
sition vis-à-vis the temporal status of the proposition of the higher clause. In
(9) the situation encoded in the ada clause necessarily precedes the time of
fearing:
(9) Nau ku ma 'u 'asia na'a ada to'a na'i ki
I I:FACT be.afraid very LEST people this PL
keka lae mai keka thaungi kulu.
they:SEQ go hither they:SEQ kill US(INCL)
'I am scared the people may have come to kill us.'
Compare the parallel example (8) above, where the situation encoded in the
ada clause is subsequent to the time of fearing; in both cases the subject/tense
marker in the ada clause is sequential.
Even though ada clauses require the Sequential/subject tense markers,
regardless of the temporal status of the situation encoded in the clause, there is
an important distinction between present-time and past-time situations on the
one hand and future-time situations on the other (including future in the past).
The distinction has to do with the possibility of precautions against a possible
undesirable state of affairs, whether in the sense of averting it altogether, or in
the sense of somehow alleviating its (anticipated) effects. One cannot take
precautions against past and present situations, but one can take at least some
precautions against possible future situations. In some cases, one may be able
to avert a potential future situation; one may, for example, hide in order not to
be seen:

(10) Ada wane 'eri ka riki nau.

LEST man that he:SEQ see me
'[I fear] the man might see me.'
Even if a future situation cannot be averted, one may be able to take
measures to alleviate its effects, for example, by taking an umbrella when it
looks like rain:
(11) Ada dani ka 'arungi kulu.
LEST rain it:SEQ fall on US(INCL)
'We might get caught in the rain.'/'We might get rained on.'
298 Frantisek Lichtenberk

A distinction can then be made between a 'precautionary situation' and

an 'apprehension-causing situation': a precautionary situation is (to be) brought
about because of the possibility of an apprehension-causing situation.
This brings us to the third and last use of ada, which is to connect a clause
encoding an apprehension-causing situation to a preceding clause encoding a
precautionary situation. Sentences (10) and (11) above express only appre­
hension-causing situations. In the corresponding sentences (12) and (13)
respectively, both the apprehension-causing and the precautionary situations
are expressed:

(12) Nau ku agwa 'i buira fau ada wane 'eri

I I.FACT hide at behind rock LEST man that
ka riki nau.
he:SEQ see me
'I hid behind a rock so that the man might not see me.'/'I hid
behind a rock lest the man see me.'
(13) Kulu ngali-a kaufa ada dani ka
we(INCL) take -them umbrella LEST rain it:SEQ
'arungi kulu.
fall on US(INCL)
'Let's take umbrellas in case we get caught in the rain.'/'Let's take
umbrellas lest we should get caught in the rain.'
There are two basic kinds of relation between an apprehension-causing
situation and a precautionary situation. In one, there is a direct causal link
between the two: if no precaution is taken, the apprehension-causing situation
will take place: if not X, then Y. In other words, the precautionary situation
serves to avert the apprehension-causing situation: X in order that not Y. I will
refer to the use of a LEST element under these circumstances as 'avertive'. In
the other type, there is no causal link; whether an apprehension-causing
situation takes place or not is independent of the precautions being taken. 5 The
precautions are taken in case the apprehension-causing situation should oc­
cur: X in case Y. I will refer to the use of a LEST element under these
circumstances as 'in-case'.
The question to be considered is whether or not the avertive (negative-
purpose) and the in-case uses of ada (and of the LEST elements in other
languages) represent two distinct functions. In other words, is this a case of
polysemy or (perhaps fuzzy) monosemy?
 Apprehensional Epistemics 299

There is something to be said for the monosemy thesis. It could be argued

that although two uses — avertive and in-case — are distinguishable, they
nonetheless do not represent distinct functional categories. Rather, ada could
be said to have a fairly general function, something like signaling that there is
a possibility, likelihood of an undesirable situation coming about, on account
of which a precautionary situation is brought about. Whether the interpreta­
tion is avertive or in-case depends on extralinguistic, pragmatic factors,
specifically on the presence or absence of a causal link between the two
situations. If the apprehension-causing situation cannot be averted by the
precautionary situation, only the in-case meaning is possible. If the apprehen­
sion-causing situation can be averted, both meanings may be possible (see
note 5). The following sentence may be given both an avertive (a) and an in­
case (b) interpretation:
(14) ... wane kai too i laala kai lio
man he:NONFACT stay at inside he.NONFACT look
ma'asi -a maelimae ada ka lae mai
wait.for -him enemy LEST he.SEQ go hither
ma ka ili -a ta -si doo.
and he:SEQ do-it some-CLASS thing
a. '...the man would stay inside and look out for the enemy so
that he [the enemy] might not come and do something.'
b. '...the man would stay inside and look out for the enemy in
case he came and did something.'
A proponent of the monosemy approach would have to argue that (14)
exhibits pragmatic rather semantic ambiguity, depending on whether the
man's looking out was supposed to prevent the enemy (coming and) doing
something. (For the concept of 'pragmatic ambiguity' see Horn 1989 and
references therein.) On the other hand, a proponent of the polysemy thesis can
argue that since in many sentences only one interpretation (either avertive or
in-case) is possible, this suggests that two distinct — though obviously related
— semantic categories are involved.
There are other, stronger arguments in favor of the polysemy approach.
Ada is not the only means of signaling negative purpose in To'aba'ita. Besides
ada, To'aba'ita also has a Purpose marker fasi, which, however, is used to
encode not only negative but also, predominantly, positive purpose. When
fasi encodes positive purpose, the purpose clause is grammatically positive;
300 Frantisek Lichtenberk

when it encodes negative purpose, the purpose clause is negative. See exam­
ples (15) and (16), respectively:
(15) Fale-a ta-si fanga 'ana wela na'i
give-it some-CLASS food to child this
fasi(-a) ka bona.
PURp(-it) he:sEQ be quiet
'Give some food to the child so that he is quiet.'
(Fasi, which ultimately derives from a transitive verb 'leave, forsake' (see
further below), optionally indexes the purpose clause by the object marker -a.)
(16) Nau ku agwa 'i buira fau fasi-a wane 'eri
I I:FACT hide at behind rock PURP-it man that
'e a'i si riki nau.
it:FACT NEG.VB he:NEG see me
'I hid behind a rock so that the man might not see me.'
Although both fasi and ada can mark negative purpose (compare 16 and
12), ada is preferred in this function. However, unlike ada, fasi does not have
the in-case function. The following examples demonstrate the similarities as
well as the differences between ada and fasi. Examples (17) and (18) demon­
strate the avertive, negative-purpose uses:
(17) Ngali-a kaleko 'aa'ako fasi 'osi
take -them clothes warm PURP you(SG):NEG
gwagwari 'a-fa rodo.
be cold at-CLASS night
'Take warm clothes so that you are not cold at night.'
(18) Ngali-a kaleko 'aa'ako ada 'oko gwagwari
take -them clothes warm LEST you(sG):SEQ be cold
'a-fa rodo.
at-CLASS night
'Take warm clothes so that you are not cold at night.'
It is even possible for fasi and ada to be used jointly to mark negative
purpose (although this is not common), in which case it is ada, not fasi, that
determines the structure of the clause encoding the situation to be averted (the
clause is grammatically positive). Compare in this regard (19) below with
Apprehensional Epistemics 301

(17)and (18) above:

(19) Ngali-a kaleko 'aa'ako fasi(-a) ada
take -them clothes warm PURP(-it) LEST
'oko gwagwari 'a-fa rodo.
you(SG):SEQ be cold at-CLASS night
'Take warm clothes so that you are not cold at night.'
Since fasi, unlike ada, has only an avertive function, a sentence like (20)
is incongruous (indicated by '#'), while (21) with ada is fine:
(20) #Ngali -a kaleko 'aa'ako fasi-a fanu 'eri
take -them clothes warm PURP-it place that
'e a'i si gwagwarila ba -na 'a-fa rodo.
it:FACT NEG.VB it:NEG be cold LIM-its at-CLASS night
#'Take warm clothes so that it may not be cold at night.'
(21) Ngali-a kaleko 'aa'ako ada fanu 'eri ka
take -them clothes warm LEST place that it:SEQ
gwagwarila ba -na 'a-fa rodo.
be cold LIM-its at-CLASS night
'Take warm clothes in case it's cold at night.'
Taking warm clothes may avert the person's feeling cold, but it cannot
prevent the weather being cold. (When presented with (20), my consultant
said the sentence was nonsensical.)
Ada appears to be the only element with a simultaneously apprehensional
and epistemic-downtoning function. To'aba'ita has a variety of other means
of epistemic downtoning, but they do not necessarily have an apprehensional
connotation. Compare the following two examples, the first with ada and a
necessarily apprehensional force, the second with another epistemic down-
toner which has no evaluative force:

(22) = (5) Ada keka fanga sui na'a.

LEST they:SEQ eat COMPL PERF
'[I fear] they may have finished eating.'
(23) Mad-e a'i kera fanga sui na'a.
or -it.FACT NEG.VB they:FACT eat COMPL PERF
'They may have finished eating.'
302 Frantisek Lichtenberk

In (23) epistemic downtoning is signaled by mad-e a'i, literally 'or it is not'.

Whether the situation encoded in (23) is undesirable depends on the context.
What was said above about the replaceability of the To'aba'ita LEST
element by the purpose marker is true also of English lest, which likewise has
an avertive and an in-case function. As consideration of the glosses of (17)
and (20) above will show, so that... not is semantically equivalent to lest
when the latter has a negative-purpose function but not when it has an in-case
function. Czech has a LEST element used with an apprehensional-epistemic
function and a fear function. The same etymon also functions as a purpose
marker regardless of positive/negative polarity, but it is not used with an in­
case function. And in Martuthunira (Australia), there is an overt formal
difference between LEST clauses of purpose and in-case LEST clauses. (English,
Czech, and Martuthunira will be discussed in Section 3.)
The various states of affairs mentioned above — (a) that an element used
to encode negative purpose need not have an in-case function; (b) that there is
a formal difference between negative-purpose and in-case clauses in at least
one language; and (c) that there are differences in paraphrase possibilities
between negative-purpose and in-case clauses — serve as evidence that the
avertive and the in-case functions are conceptually distinct from each other.
Consequently, the use of one element in both functions is a case of polysemy,
not monosemy. (This means that sentence (14) above is semantically rather
than pragmatically ambiguous.) However, even though the avertive and the
in-case functions are distinct, they are nonetheless related. I will use the cover
term 'precautioning function(s)' for both when there is no need to distinguish
between them.6
To'aba'ita ada, then, has three basic functions: precautioning (avertive
and in-case), fear, and apprehensional-epistemic. As will be seen in Section 3,
this kind of polysemy is not at all unusual cross-linguistically, and at this stage
it is useful to compare and contrast the three functions. When a LEST element is
used with a precautioning function, there are two clauses in a complex
sentence, a LEST clause encoding an apprehension-causing situation and a
clause encoding a precautionary situation (as in examples 12 and 13 above).
Both situations are prominent; both are encoded. When a LEST element is used
with a fear function, the LEST clause, encoding an undesirable situation, is
embedded under a predicate of fearing (as in 8). Apprehension is signaled
explicitly by the higher predicate. When a LEST element is used with an
apprehensional-epistemic function, it itself expresses the meanings of possi-
Apprehensional Epistemics 303

bility as well as apprehension, and the clause can function as an independent

sentence (as in 1). However, even though the semantic/pragmatic distinctions
among the three functions are reasonably clear in principle, the task of
determining which function a given LEST element fulfills in a particular sen­
tence is not always clearcut, depending as it does on certain other considera­
tions; see the discussion of the Martuthunira data in Section 3.
The question to be addressed next is the development of the polysemy of
To'aba'ita ada. In order to find an answer, we have to move outside of
To'aba'ita and look at some closely related languages. In To'aba'ita, ada
functions only as a grammatical element, but in at least some of its close
relatives the etymon is found functioning lexically, as a verb: Kwaio aga
'look (at)', agasi 'see' (Keesing 1975), Kwai ada 'see' (Tryon & Hackman
1983), and Lau a da 'open the eyes, use the eyes, see, look, wake (intr.)', adasi
'look at, watch, expect, wait to see' (Fox 1974). In Kwaio, the etymon appears
to be used only in the verbal function. The information on the Kwai form
comes from a word list, and it is unclear whether the form has any other
functions. Of the three languages with cognates of ada, Lau is genetically
closest to To'aba'ita (Tryon & Hackman 1983:27, note 1). And in Lau, ada is
used not only as a verb, but also grammatically. Although the relevant
information on Lau is rather limited, it is evident that Lau ada can be used
grammatically (at least) with the avertive function. The following examples
from South Lau illustrate the lexical as well as the grammatical uses of ada:

(24) Alifii -laa ada wela ka ada.

be noisy -NOMI LEST child SEQ wake up
'Quiet, (or) the child will wake up.'
(Lit.: 'being noisy, the child will wake up')
(25) Ada suli 'oe ada 'oe ka 'asía.
Look after you(SG) LEST you(SG) SEQ fall
'Look out, or you'll fall down!.'
In what follows, I will sketch out the likely development of the apprehen-
sional-epistemic function of To'aba'ita ada. Because of the lack of historical
records, there is no direct evidence of the steps to be postulatéd, but there are
several kinds of indirect evidence all pointing in the same direction. There is
structural evidence permitting some internal reconstruction, comparative evi­
dence from closely related languages, and the existence of similar semantic
shifts in other, unrelated languages.
304 Frantisek Lichtenberk

At an early stage, the etymon functioned as a verb meaning 'see, look at,
watch'. Later the verb acquired a 'warning' meaning, 'look out, watch out';
see the South Lau example (25) above. The use of verbs meaning 'see, look at,
watch' with a warning meaning is not unusual crosslinguistically. It is found
in, for example, English (look out, watch out), Basque (begira 'look at',
begira gero! 'be careful!' (gero 'later')), Zulu (beka 'look, look at, watch,
take precautions against, be careful'), and also in present-day To'aba'ita
(compare the uses of riki in (27) and (28) below). Sweetser (1990) points out
a common type of polysemy in Indo-European languages that unites mean­
ings having to do with vision and meanings having to do with guarding,
keeping control, and monitoring (although this kind of polysemy can appar­
ently arise in either direction).
Ada was used as a verb of warning against possible future undesirable
situations: *ada + clause 'look out, Y might/will happen', 'look out so that Y
may not happen'. The use, in present-day To'aba'ita, of the sequential sub­
ject/tense markers in ada clauses embedded under verbs of fearing and in
independent ada sentences regardless of the temporal status of the proposition
encoded in the ada clause is evidence that when ada was used as a verb of
warning it was so used with reference to situations subsequent to the act of
warning. Most likely, ada was used in this function in exhortations and
commands ('Look out,...!'), which in To'aba'ita are normally expressed by a
simple verb, without a subject/tense marker, as in (26) (and in 28 below):

(26) 'Ono 'i ninima -kul

sit at beside -my
'Sit (down) beside me!'
At this stage, negative purpose was most likely marked by the element
*fa 'asi, of which To'aba'ita fasi is a descendant. Positive purpose was marked
by a different element, *uri. It was only later that *fa'asilfasi acquired a
positive-purpose marking function. (For more detailed discussion see
Lichtenberk 1991.)
At some stage in the history of To'aba'ita, ada ceased to be a verb; today
ada cannot form a predicate and cannot occur with a subject/tense marker. To
express the concepts of 'seeing' and 'looking at', the verb riki is used:
(27) ... ma keka riki-a ba'u 'eri...
and they:SEQ see-it banana that
'... and they saw the banana tree ...'
Apprehensional Epistemics 305

And just as once was the case with ada, riki also has the meaning 'look out,
watch out'. It can thus cooccur with ada:
(28) Riki -a ada 'oko dekwe-a kwade'e kuki 'ena.
see -it LEST you(SG):SEQ break -it empty pot that
'Look out; you might break the empty pot.'/'Mind you don't break
the empty pot.'
When ada still functioned as a verb, it was used to encode a general
precautionary event: 'look out, watch out'; the specific nature of the precau­
tionary event depended on the situation. After ceasing to function as a verb,
ada continued as a grammatical element with a warning function signaling
that a precaution was to be taken in view of a possible adverse future situation.
In this function, ada began to encroach on the territory previously occupied by
the negative-purpose marker, present-day fasi. As a result, fasi has to a large
degree ceded negative-purpose marking to ada; today it is used primarily to
signal positive purpose. The details of the historical process whereby ada has
nearly displaced fasi in marking negative purpose may be unrecoverable, but
the fact that fasi and ada may cooccur in present-day To'aba'ita (see 19
above) can be taken as evidence of the process.
The apprehensional-epistemic function of ada is a later development
from the precautioning function, most likely through an intermediate stage of
the fear function. An undesirable future situation is likely to be feared.
Through this metonymy, ada clauses began to be embedded under predicates
of fearing. When one fears a situation, one fears a potential situation: T am
afraid that Y might happen/have happened/be happening'. In a discussion of
LEST clauses subordinated to predicates of fearing in Ngiyambaa (Australian),
Donaldson (1980:286) says that such sentences express '"fear of the possibil­
ity of [e.g.] falling', not 'fear of [e.g.] falling' as such" (original emphasis).
That the fear function is a later development, postdating the warning and
precautioning functions is suggested by the use of the sequential subject/tense
markers in ada clauses embedded under verbs of fearing even if the situation
encoded in the ada clause is not posterior to the time of fearing, as in (9)
above. It is conceivable that at an earlier stage ada was used to express
warning only about potential situations against which precautions could be
taken, i.e. situations subsequent to the act of warning, and that by the time ada
developed the fear function, the subject markers had become devoid of the
sequentiality-marking function there.
306 Frantisek Lichtenberk

The final step (thus far at least) in the history of ada is the development of
the apprehensional-epistemic function. At the 'fear' stage, apprehension was
signaled by the predicate dominating the ada clause; the ada clause encoded a
possible situation. Later, through metonymy the notion of apprehension came
to be associated with ada. That is, since ada was used to introduce clauses that
encoded apprehension-causing situations (as signaled by the higher verb), it
acquired the connotation of apprehension. After this had happened, a predi­
cate of fearing was no longer necessary. The notions of possibility and
apprehension were both signaled by ada; as an apprehensional-epistemic
complementizer, ada could now introduce independent sentences.

3. Apprehensional epistemics in other languages

Polysemy uniting an apprehensional-epistemic function with fear and precau-

tioning functions is not unique to To'aba'ita; and in at least one language a
further development has taken place beyond the apprehensional-epistemic
stage.
LEST forms appear to be common in Australian languages, Diyari being
one of them. Austin (1981:225) says this about LEST clauses in Diyari:
'Lest' clauses basically serve to indicate some situation which the speaker
considers to be unpleasant and which should be avoided. They are most
commonly used as admonitions or threats but they are also used in giving
warnings ...

The LEST element is a verbal suffix -yati. -yati has a precautioning

function, both avertive (29) and in-case (30):
(29) napu -ri -ya -mayi natu yina nanda-yati
quiet -INCH -IMP-EMPH I:SUBJ you(SG):DO hit-LEST
'Be quiet or I'll hit you.' (Austin 1981:225)
(30) makita padaka-Ø -mayi wanku yundu
gun:ABS take -IMP-EMPH snake:ABS you(SG):suBJ
wa¡a nayi -yati
soon see -LEST
'Carry a gun in case you see a snake.' (Austin 1981:225)
Apprehensional Epistemics 307

LEST clauses can also be embedded under a predicate of fearing:

(31) nani yapa-li nana -yi nana tutu -yali
I:SUBJfear -ERG be -PRES I:DO reptile-ERG
matamata-tadi-yati
bite:RED -DUR-LEST
'I'm afraid that some reptile might bite me.'(Austin 1981:227)
And the LEST suffix can also be used in independent sentences with an
apprehensional-epistemic function:
(32) nulu -ka kintala-li yinana mata-yati
it(NONFEM):SUBJ-TOKEN dog -ERG you(SG):DO bite-LEST
'This dog might bite you.' (Austin 1981:229)
According to Austin, "in all the examples of this type of construction it is clear
from the context that an 'understood' imperative, warning or suggestion is
implicit", and such sentences "may be regarded as structurally subordinate
because it is always possible to add a main clause before them, although
context may make it unnecessary" (Austin 1981:229, original emphasis). In
the situation encoded in (32), "the context would be clear and make it
unnecessary to say 'Look out!' or 'Be careful!' and so on" (ibid.). But there is
no reason to appeal to an 'understood' main clause; this is precisely what the
apprehensional-epistemic function of LEST elements is all about: there is a
possibility of a situation coming about, and there is apprehension about that
possibility.
Another Australian language, Martuthunira, has a LEST form, which is
used with precautioning — avertive and in-case — functions and possibly
also with an apprehensional-epistemic function (Dench 1988). (Dench gives
no examples of the LEST element with a fear function.) What is interesting
about Martuthunira is that it makes an overt distinction between those cases
where an apprehension-causing situation directly results from lack of precau­
tion and those cases where there is no such causal link. In other words, there is
a formal distinction between the avertive and the in-case functions. The LEST
element is a suffix on the verb; in addition, the verb may carry a nominal
inflectional suffix, either accusative or locative. If the apprehension-causing
situation results from the absence of a precautionary situation, the verb carries
accusative marking:
308 Frantisek Lichtenberk

(33) Mir.ta wantha-rninyji murla-a yakarrangu-la

NEG leave -FUT meat -ACC sun -LOC
puwa -npa -wirri -i.
rotten-INCH-LEST -ACC
'Don't leave the meat in the sun or it'll go rotten.'
(Dench 1988:109)

When there is no implication of a causal link between the absence of a

precautionary situation and an apprehension-causing situation, the verb in the
LEST clause carries either locative marking, as in (34), or no nominal marking,
as in (35).

(34) Kartu wankuma-lalha paju, walyi -lalha ngurnu

you(SG):NOM check -PAST very uncover-PAST that:ACC
punkurrimarnu-u purntura-a wanti -lha -a wayil
blanket -ACC rolled -ACC lie -PAST-ACC maybe
wii nhartu -ngara, parralhara-ngara wiiy wayil wii
if what -PL centipede -PL if maybe if
wanti -wirri-la mirtungkura -la punkurrimarnu -la?
lie -LEST -LOC underneath -LOC blanket -LOC
'Did you check carefully and undo that rolled up blanket in case
there was anything, perhaps a centipede, lying underneath it?'
(Dench 1988:109-10)

Dench says that in Martuthunira a LEST clause may function as a main

clause. This would suggest an apprehensional-epistemic function of the LEST
element; however, in both examples of LEST sentences that Dench gives, the
LEST sentence is preceded by another sentence that encodes a precautionary
situation, as in (35):

(35) Mir. ta thaawu -rninyji. Pawulu puni -wirri kayulu-la -rru

NEG let go -FUT child go -LEST water -LOC-NOW
nyuni -lu -rru.
drown -PURP:SS -NOW
'Don't let (him) go. The child might go in the water and drown.'
(Dench 1988:110)
Taking the LEST sentence in isolation, the LEST element may be said to
have an apprehensional-epistemic function: '[I fear] the child might go in the
water and drown'. However, if one takes the preceding sentence into consid-
Apprehensional Epistemics 309

eration as well, the LEST element can be said to have a precautioning function.
Similarly, in the following (made-up) example the LEST element could be seen
as having not just a fear function but also an avertive function: 'Don't let him
go. I fear lest he should go in the water and drown.' Although the distinctions
among the precautioning, fear, and apprehension-epistemic functions are
clear in principle, in some cases the identification of the function of a LEST
element depends on whether or not one takes wider, extrasentential context
into consideration.
Note the absence of nominal (accusative or locative) marking on the verb
'go' in (35). According to Dench (1988:110-111),
the speaker apparently chooses not to imply that the unpleasant situation
will come about as a consequence of the addressee's actions or inaction. The
marking options — accusative, locative, or unmarked — allow a range of
interpretations about the responsibility on the part of the participants for the
situation described by the lest clause.

An apprehensional-epistemic construction is also found in Hua (Papuan);

the same construction is also used with precautioning — avertive and in-case
— functions (Haiman 1980). From the examples given, it is not clear whether
or not the construction is also used with a fear function.
Apprehensional epistemics are not restricted to languages where the
marker in question can only signal negative, not positive, purpose. Czech has
a Purpose marker aby, used to express both positive and negative purpose.
The following examples are from colloquial Czech, but the same is true,
mutatis mutandis, of standard Czech. Aby carries a suffix indexing the subject
of its own clause. When the purpose is positive, the purpose clause is gram­
matically positive:

(36) Spěcifiai sem domu aby -ch zastih televizní noviny.

I.hurried AUX home PURP-1SG I.caught T.V. news:ACC
'I hurried home to catch the T.V. news.'
When the purpose is negative, the purpose clause is grammatically
negative:
(37) Spëchal sem domu aby -ch ne -propás televizni
I.hurried AUX home PURP-1SG NEG-I.missed T.V.
noviny.
news'.ACC
'I hurried home so I wouldn't miss the T.V. news.'
310 Frantisek Lichtenberk

Aby can be used to express negative purpose, but it cannot be used with
an in-case function. For that reason, (38) is incongruous:
(38) #Vem si deštník aby
you(SG).take:IMP MID:DAT umbrella:ACC PURP:3SG
ti ne -prselo na cestu.
you(sG):DAT NEG-it.rained on road:ACC
#'Take an umbrella so that it does not rain while you're on the
road.'
(38) cannot be used to mean 'Take an umbrella in case it rains while you're on
the road.' 7
While aby is decidedly anomalous in (38), it can be used to introduce
parallel clauses that are embedded under a predicate of fearing or that are
independent; see (39) and (41) below.
Negative clauses introduced by aby may be embedded under predicates
of fearing:
(39) Bojím se aby nám ne -pršelo na
Lam.afraid MID:ACC LEST:3SG US:DAT NEG -it.rained on
cestu.
road:ACC
'I'm afraid it might rain while we're on the road.'
The feared situation need not be subsequent to the time of fearing:
(40) Bojím se aby ne -byl nemocnej.
Lam.afraid MID:ACC LEST:3SG NEG-he.was sick
'I'm afraid he might be sick.'
Finally, aby, introducing a negative clause, may also be used with an
apprehensional-epistemic function to encode a possible undesirable situation,
future or not:
(41) Aby nám ne -prselo na cestu.
LEST:3SG US:DAT NEG-it.rained on road:ACC
'[I fear] it might rain while we're on the road.'

(42) Aby ne -byl nemocnej.

LEST:3SG NEG-he.was sick
'[I fear] he might be sick.'
Apprehensional Epistemics 311

Aby has an avertive, negative-purpose use, but not an in-case use. Never­
theless, in the fear and the apprehension-epistemic functions aby can be used
in encoding situations that cannot be averted, either because they are of the
non-avertible type (39, 41) or because they may exist already (40, 42). This
suggests that the apprehensional-epistemic function developed from the fear
function rather than directly from the avertive, negative-purpose use.
Another example of the development of the apprehensional-epistemic
function comes from Classical Greek. The LEST element had the form mé:. An
early function of mé: was that of a negator used in certain contexts. (There
was another negator ou, used elsewhere.) Mé: was used in prohibitions, which
can be thought of as a kind of avertive function: a prohibition is issued in order
to avert an undesirable situation. Unlike the cases considered thus far, in
Greek prohibitions the precautioning situation is not encoded in a clause of its
own; instead, it consists of the speech act of issuing a prohibition:

(43) mé: poié:se:is toûto

NEG you(SG).do:IMP this:ACC
'do not do this' (Goodwin 1929:89)
Another use of mé: in independent clauses was to express "apprehension,
coupled with a desire to avert the object of fear, both ideas being inherent in
the construction" (Goodwin 1929:90):
(44) Homer, Iliad XVI. 128:
mè: dè: nê:as hélo:si...
LEST indeed ships:ACC they.seize:suBJUNCT
'may they not (as I fear they may) seize the ships ...'
(Goodwin 1929:90)
This apprehensional-avertive use of mé: is characteristic of Homeric Greek
(Goodwin 1929).
At a later stage, apprehensional-avertive clauses introduced by mé: came
to be used under predicates of fearing:
(45) Xenophon, Anabasis III. 2, 25:
dédoika mè: epilathó:metha tê:s oíkade hodoû
I.fear LEST we.forget:SUBJUNCT ART:GEN homeward road:GEN
'I fear lest we may forget the road home'(Goodwin 1929:24)
312 Frantisek Lichtenberk

The feared situation is typically, though not necessarily, posterior to the

time of fearing.
Mé: also acquired the function of marking negative purpose, another type
of avertive use. At an earlier stage, to express negative purpose one of several
purpose markers had been used, with mé: as a negator in the purpose clause:
(46) apérchetai, hína mè: toûto íde:i
he.departs PURP NEG this:ACC he.see:suBJUNCT
'he is departing that he may not see this' (Goodwin 1929:106)
Later, the original purpose markers may be dropped, and mé: becomes a
"connective" expressing negative purpose (Goodwin 1929:108, 109):
(47) machoúmetha mè: nê:as hélo:si
we.will.fight LEST ships:ACC they.seize:suBJUNCT
'we will fight that they may not seize the ships'
(Goodwin 1929:90-91)
Mé: had other uses that are not of direct concern to this study. However,
there is one function of mé: that is relevant. At a later stage still, mé: came to
be used to express "cautious assertions", expressing "a suspicion that some­
thing may be (or may prove to be) true" (Goodwin 1929:92, original empha­
sis):
(48) Plato, Gorgias 462 E:
mè: agroikóteron e:i to ale:thés eipeîn
LEST rude thing it.is:suBJUNCT ART:ACC truth:ACC say
'I am afraid the truth may be too rude a thing to tell'
(Goodwin 1929:92)
As Goodwin (1929:92) puts it:
In these cautious assertions and negations, although no desire of the speaker
to avert an object of fear is implied, there is always a tacit allusion to such a
desire on the part of some person who is addressed or referred to, or else an
ironical pretence of such a desire of the speaker himself.

What Goodwin says about "ironical pretence" of a desire of the speaker

to avert the undesirable situation suggests that this construction was used even
when the undesirable situation was not under anybody's control, when there
was no question of it being averted. In other words, this use of mé: is
apprehensional-epistemic, not precautioning. That this was indeed the case is
Apprehensional Epistemics 313

shown by the fact that the cautious-assertion use of mé: is not restricted to
future situations. Mé: may be used with possible present or past situations (in
which case the verb is in the appropriate indicative rather than the subjunctive):
(49) Aristotle, Ethics X. 1, 3:
mé: pote dè ou kalô:s légetai
LEST at some time but NEG well it.is.said:IND
'but it may be that this is not well said' (Goodwin 1929:93)
The next set of examples comes from English. English has the form lest,
which is usually said to have two functions, which the OED identifies as
follows: First, it is used "as a negative particle of intention or purpose,
introducing a clause expressive of something to be prevented or guarded
against". This is a precautioning function, avertive, as in (50), and in-case, as
in (51):
(50) He was put in a cell with no clothes and shoes lest he injure
himself. (Collins COBUILD English Grammar, p.355)
(51) He remained inside the fortress lest the enemy should arrive in
great numbers.
The following sentence is ambiguous between an avertive and an in-case
interpretation of the LEST clause, depending on whether the medicine is to
prevent the illness or cure it:
(52) When you go to Africa, take this medicine with you lest you
(should) get ill.
The following sentence also is open to both an avertive and an 'in-case'
interpretation; in fact, the two interpretations are not mutually exclusive — an
instance of what Coates (1983) calls 'merger':
(53) Hole, C , English Shrines and Sanctuaries:
And lest future generations should wonder why St. Withburga
herself is not enshrined here, a tablet informs the visitors that the
monks stole the relics. (Scheurweghs 1959:381)
The tablet is there in case the visitors should wonder why St. Withburga is not
enshrined there (absence of the tablet would not necessarily bring about the
visitors' wondering), and its presence does remove the visitors' (further)
wondering.
314 Frantisek Lichtenberk

Second, lest is used "after verbs of fearing, or phrases indicating apprehen­

sion or danger, to introduce a clause expressing the event that is feared" (OED):
(54) I trembled lest they should see us. (Jespersen 1933:349)
The feared situation need not be subsequent to the time of fearing:
(55) I fear lest he may have deceived me.
Historically, lest derives from an Old English comparative construction
py lœs pe, earlier pӯ lœs, 'by that/which (i.e. whereby) less' (py, also spelled
pi, pe, 'by which', lœs 'less',pe Relative particle). Later, py dropped out, and
lœs pe became les te and then lest (OED).
Both of the main functions of lest, precautioning (56) and fear (57), are
found in Old English:
(56) Beowulf, XXVIL1917-19:
sœlde tō sande sïdfœpme scip oncerbendum fœst, py lœs hym ypa
drym wudu wynsuman forwrecan meahte
'He tethered to the beach the roomy ship, held fast with anchor-
ropes, lest the waves' force should drive the joyous craft away
from them.'(Klaeber 1950; translation from Hall 1950)
(57) King Alfred, Pastoral Care 461.30:
Ac him is donne micel ðearf ðœt he hine hrœdlice selfne gewundige
mid dy ege dœt he him ondrœde ðylœs he weoröe upahœfen for his
wordum.
'But there is then to him a great need to feel the pangs of terror,
fearing lest he should become arrogant on account of his own
words.' (Mitchell 1985:16)
Although both functions are attested in Old English, it appears that the
precautioning function historically preceded the fear function. According to
Mitchell (1985:16), although py lœs could be used to introduce clauses
embedded under verbs of fearing, such clauses were usually introduced by
pœt. Secondly, Mitchell (1985:495), following Shearin (1909), speaks of the
development of py lœs (pe) "final" clauses, i.e. LEST clauses with a precaution­
ing function, into "object" or "noun" clauses, i.e. clauses that functioned as
complements of verbs of fearing. In both main functions that have been
identified, lest introduces dependent clauses. However, lest may also be used
to introduce an independent sentence:
Apprehensional Epistemics 315

(58) Sir, — To those of Yugoslav origin who keep urging New Zealand
involvement in the affairs of their homeland, I would like to remind
them of the terrible cost to the world, and New Zealand, the last
time someone acted violently in Sarajevo. Lest we forget.(A letter
to the editor, New Zealand Herald, 11/20/1991.)
In (58) lest has an apprehensional-epistemic function: '[I fear] we might
forget'. It is lest itself that expresses the apprehension about a potential
undesirable event.
Lest is not common in modern English; a recent grammar characterises it
as formal, old-fashioned (Collins COBUILD English Grammar). Given the
relative obsolescence of lest, it is unlikely that its apprehensional-epistemic
use will ever become well established. (The few examples that I have come
across of lest introducing an independent sentence all contain the verb forget.
This usage is most likely ultimately modeled on Rudyard Kipling's use of lest
in his poem Recessional: Lord God of Hosts, be with us yet, Lest we forget —
lest we forget... Judge of the Nations, spare us yet, Lest we forget — lest we
forget.)
The last set of examples comes from Fijian, an Austronesian language,
but only relatively distantly related to To'aba'ita; there is no reason to assume
that the developments in the two languages are related to each other.8 What is
interesting about Fijian is that a further development of the LEST element has
taken place there, beyond the apprehensional-epistemic stage. The examples
below come from Standard Fijian (essentially the Bauan dialect) and from the
Boumaa dialect; the LEST element appears in several different forms, depend­
ing on the dialect, on its linguistic environment, and on the source of the data:
dē or de in Standard Fijian and dee in Boumaa.
The LEST element is used with a precautioning function:
(59) Standard Fijian
kua ni driva vakatotolo de -o na qäi
PROH COMP drive fast LEST-you(SG) FUT SEQ
coqa
run into s.t.
'don't drive fast lest you run into something' (Schütz 1985:442)
A LEST clause can be embedded under a predicate of fearing:
316 Frantisek Lichtenberk

(60) Standard Fijian

sä rere ko koya de kasa na waqa
PERF be afraid ART he LEST:IT run aground ART canoe
'he is afraid lest the canoe should run aground'
(Churchward 1941:24)
The LEST element may also be used with an apprehensional-epistemic
downtoning force:
(61) Standard Fijian
de sega beka ni dina
LEST NEG perhaps COMP be true
'it may not perhaps be true' (Churchward 1941:24)
(62) Standard Fijian
äü nanu -m -a sara gā de ko na sega
I think -TRANS -it INT LIM LEST you(SG) FUT NEG
ni māi wā -raki äü
COMP hither wait for -TRANS me
'I thought that you might not wait for me'(Schütz 1985:443)
Even though as an epistemic downtoner dē is typically used with an
apprehensional force, it may in fact be used as a neutral downtoner, without
any connotation of apprehension. In (63) below, dē introduces a clause
embedded under the verb 'think', and according to Schütz (1985:443), "there
seems to be no afflictive meaning to the situation represented by the dē
phrase". (Compare (62) above, where also the LEST clause is embedded under
'think'.) In private communication (November, 1991) Schütz has confirmed
that, given the broader context, the interpretation of the LEST clause in (63) is
indeed non-afflictive:

(63) Standard Fijian

eräü nanu -m -a gä na vëitinani,
they(DU) think-TRANS -it LIM ART mother, and.child
dē sä vodo tiko māi ko Rejieli,
maybe:he PERF ride CONT hither ART R.
e na dua vēi ira
it ART one ABL them
'the two of them, mother and son, thought that Rejieli might be
riding in one of them' (Schütz 1985:443)
Apprehensional Epistemics 317

According to Paul Geraghty (personal communication, December, 1991),

the non-apprehensional epistemic-downtoning use of dē in Standard Fijian is
"not at all rare". (64) is another example:
(64) dē sā noqu na kalougata
maybe:it PERF my ART luck
'maybe I'm the lucky one!' (Geraghty, p.c., D e c , 1991)
Similarly in the Boumaa dialect, where, according to Dixon, the dee
clause [typically] refers to some unpleasant possibility, and the other clause to
something that can be done either to avoid that possibility or to compensate
for it, ... However, a dee clause need not refer to anything unpleasant. The
other clause may describe something which, if accomplished, might yield a
welcome result (referred [to] by the dee clause) (Dixon 1988:260, 261,
original emphasis)
(65) Boumaa Fijian
au vaa'aasama -ta'ina tVo dee o -na la'o
I think -TRANS CONT maybe you(SG) -FUT go
mai (se sega)
hither or NEG
'I' ve been thinking that you might come here (or you might not)'
(Dixon 1988:269)
However, the Fijian element has a further function: in Boumaa, accord­
ing to Dixon (1988:269, 270), dee is also used in "the most polite speech-
style" to express requests, or "to make a strong point in a polite way" (Dixon
1988:269, 270). In saying (66), the speaker "adopted an extremely reverential
tone" (Dixon 1988:269):
(66) Boumaa Fijian
au 'ere-a ti'o dee rawa ni o
I ask -TRANS CONT POLITE can COMP you(sG)
va'amacala -ta'ina a -ituvatuva ni veiqaravi
make clear -TRANS ART -order of presentation
va 'avanua va 'atuuraga
traditional chiefly
'I ask that it might be possible for you to explain the order of
traditional, chiefly presentations' (Dixon 1988:269-270)
318 Frantisek Lichtenberk

With (67), according to Dixon (1988:270), "what the speaker means (and
is understood to mean) is 'I feel that the motion should be changed'":
(67) Boumaa Fijian
au aa rai -ca dee rawa ni veisau -ta 'i
I PAST see-TRANS POLITE can COMP change-TRANS
a moosoni yai
ART motionthis
'I felt that the motion might possibly be changed'
(Dixon 1988:270)
The politeness function of the etymon is also found in Standard Fijian:
(68) Standard Fijian
dē rawa ni tou cakacaka vata
POLITE can COMP we(PAUC,INCL) work together
'perhaps we could work together' (Geraghty. p.c., July, 1993)
According to Geraghty (personal communication), (68) expresses a po­
lite suggestion: the speaker desires to work together with the others.
What has happened in Standard and Boumaa Fijian is that the LEST
element is losing the necessarily apprehensional connotation and is develop­
ing into a neutral epistemic downtoner; furthermore, on its way to becoming a
neutral epistemic downtoner, the element has come to be used as a politeness
marker. 9 Crosslinguistically, epistemic downtoning is a common feature of
polite speech, a marker of deference (see Givón 1990:822 for some refer­
ences). 10
Since epistemic downtoning is one of the features of politeness, how can
we be certain that the Fijian LEST element has acquired a new pragmatic
function, that of a politeness marker? There are two pieces of evidence that
this is the case. First, in polite speech, the element is (normally) followed by
rawa 'can' (Dixon 1988:269, Geraghty, personal communication, July,
1993); this does not appear to be the case with the other uses. Second, and
more importantly, in the polite-speech use there need not be any epistemic
uncertainty: "dee rawa can be used to make a strong point in a polite way"
(Dixon 1988:270; emphasis added): in (67) the speaker is expressing his
belief (and is understood as such) that "the motion should be changed" (ibid).
Similarly, in (68), according to Geraghty, the speaker makes a suggestion in a
polite way. In one function, the element signals epistemic uncertainty without
Apprehensional Epistemics 319

necessary politeness; in the other, in conjunction with rawa, it expresses

politeness rather than uncertainty.

4. The development of apprehensional epistemics

A number of languages contain an element that has an apprehensional-

epistemic downtoning function. (In some languages, e.g. Czech, the clause
encoding the potential undesirable situation must be grammatically negative.)
The same element may also be used with other functions: precautioning (in­
case, avertive (negative-purpose)) and fear. Consideration of the data pre­
sented above suggests the following general historical development:
(69) precautioning > fear > apprehensional-epistemic
Of the three functions, the precautioning one is historically primary; it may
(but, of course, need not) give rise to the fear function. In the precautioning
function, LEST elements signal that there is a potential undesirable situation
against which a precaution is to be taken. The implication of a situation being
undesirable, or indeed feared, becomes explicit by virtue of clauses encoding
such situations as subordinate to predicates of fearing. What one fears is a
potential situation. This might suggest that it is the in-case variety of the
precautioning function that motivates the development of the fear function.
However, recall that in Czech the element that has a fear function (as well as
an apprehensional-epistemic function) is used only with an avertive (nega­
tive-purpose) function, not with an in-case function. This can be taken as
evidence that the rise of the fear function is motivated — metonymically —
by the undesirability component of apprehension-causing situations, (This
does not preclude the possibility of the in-case function acting as a contribu­
tory motivating factor.) Over time, through metonymy based on cooccurrence
of linguistic forms, the notion of apprehension comes to be associated with the
LEST element, and the predicates of fearing become expendable. When LEST
clauses come to be used without being embedded under predicates of fearing,
the LEST element becomes an apprehensional epistemic. It signals both the
possibility of an undesirable situation and apprehension.
For two languages in the sample (Martuthunira and Hua), the sources of
the data give examples of the precautionary and the apprehensional-epistemic
functions, but no examples of the fear function. At this stage it is not clear
320 Frantisek Lichtenberk

whether these are merely lacunae in the data or whether the apprehensional-
epistemic function may indeed develop directly from the precautioning func­
tion.
Crosslinguistically, apprehensional epistemics are used typically with
respect to future situations, but non-future uses are possible as well, at least in
some languages. (This is also true of the fear uses of the LEST elements.) The
typically future use of apprehensional epistemics is to be ascribed to 'persist­
ence' (Hopper 1991) from the avertive, negative-purpose use (purpose being
necessarily future-oriented).11
An apprehensional epistemic may lose the connotation of apprehension
and thus become a neutral epistemic downtoner, even though the apprehen­
sional connotation may still be the typical case. This kind of development has
taken place in Standard and Boumaa Fijian.
The category of the precautioning function subsumes the avertive and the
in-case functions. At this stage, because of lack of relevant data, it is impossi­
ble to tell whether there is a general unidirectional historical process whereby
one of the two functions develops from the other, avertive from in-case, or in­
case from avertive. Example (53) above suggests that the development of one
function from the other proceeds through a state where both meanings may be
present or available simultaneously in a sentence (see also 52).
As far as the sources of the precautioning function, as a category, are
concerned, little can be said at present. In To'aba'ita, the precautioning
function derives from a verbal warning meaning, 'look out, watch out', itself
from 'see, look at, watch'. English lest derives from a comparative construc­
tion, 'by which less', while the Classical Greek apprehensional epistemic can
be traced back to a negator. I do not have relevant information on any of the
other LEST elements discussed above. Wayan Fijian has a LEST element with
avertive, in-case, and fear functions, but apparently not an apprehensional-
epistemic function (Pawley and Sayaba, forthcoming; Pawley, personal com­
munication, October, 1992). The Wayan LEST element (unrelated to the
Standard and Boumaa Fijian and the To'aba'ita LEST elements) derives from a
verb meaning 'survey, watch, see s.t., take a close look at s.t., examine or look
carefully at s.t.'; this is reminiscent of the development of the To'aba'ita LEST
element. Central Siberian Eskimo has a Volitive-of-Fear mood, which ex­
presses the notions of 'for fear that Y', 'lest Y'. According to de Reuse (1991),
the mood marker (a suffix on verbs) derives from the verb 'think'; there is no
mention of an apprehensional-epistemic function.
Apprehensional Epistemics 321

At present there is no evidence that apprehensional epistemics (typically)

derive from agent-oriented modalities (cf. Bybee and Pagliuca 1985, referred
to in Section 1). In fact — judging from the small amount of historical
evidence available — it is the notions of apprehension and precaution that are
the main motivating factors in the development of LEST elements, with the
epistemic-downtoning function arising as a by-product.

5. Directionality in semantic/functional change

In a number of studies of semantic/functional change, Traugott has argued

that there is a general unidirectional process whereby meanings become more,
rather than less, subjective, and that semantic, relatively objective meanings
develop into pragmatic, relatively subjective meanings/functions rather than
vice versa (see, for example, Traugott 1982, 1989, 1990, and also Traugott &
König 1991). Traugott (1989:34,35) identifies three unidirectional tendencies
(T) in semantic/pragmatic change:
(70) Tl : Meanings based in the external described situation > meanings
based in the internal (evaluative/perceptual/cognitive) de­
scribed situation.
T2: Meanings based in the external or internal described situation >
meanings based in the textual and metalinguistic situation.
T3: Meanings tend to become increasingly based in the speaker's
subjective belief state/attitude toward the proposition.
Tl can feed T2 and either one can feed T3.
Tl and T3 have to do with the subjectification of meaning; T2 is con­
cerned with the development of textual cohesive functions and functions
having to do with performing speech acts. Semantic/pragmatic change in
grammaticalization, at least in the early stages, is characterized not by seman­
tic weakening, bleaching, but by a change in the nature of the meaning: from
relatively more objective meanings to relatively more subjective meanings/
functions, and from (relatively) objective meanings to textual and speech-act
functions.
As for the factors motivating the kinds of semantic/pragmatic change
characterized by the three tendencies, Traugott (1989) suggests that a distinc­
tion needs to be made between the development of epistemic meanings, which
322 Frantisek Lichtenberk

are subject to T3, on the one hand, and the development of other, non-
epistemic meanings, which are subject to Tl and T2, on the other. The
development of non-epistemic meanings is typically metaphor-based, while
the development of epistemic meanings is based on metonymy, in particular
on the conventionalization of conversational implicatures/invited inferences.
Now, assuming that the scenario concerning the development of appre-
hensional epistemics presented above is essentially correct, it is evident that
— where relevant — the developments in question do in many respects follow
the three tendencies identified by Traugott. The change in the meaning of
To'aba'ita ada from 'see, look at, watch' to 'look out, watch out' exemplifies
both Tl and T2. The change is from a relatively objective meaning to a
relatively subjective meaning, and the lexeme comes to be used to perform a
speech act: a situation is interpreted by the speaker as undesirable, and the
lexeme is used to issue a warning. It is the speaker's own interpretation of a
situation as one calling for a warning (others need not interpret the same
situation in this way).
In the absence of historical evidence, the details of the development
cannot be determined, but it is not implausible to assume that after ada
acquired the warning function, ada clauses began to be preceded by clauses
encoding precautionary situations, and ada eventually became a conjunction
with new, precautioning — avertive and in-case — functions. What we get is
development of a textual element from a speech-act element. 12 When ada
clauses began to appear embedded under predicates of fearing, ada acquired a
new textual function, that of a complementizer. Finally, the development of
the apprehensional-epistemic function from the fear function is an instance of
T3: ada comes to signal both apprehension and less-than-full certainty. 13
Traugott (1989:35) says that "meanings tend to become more subjec­
tive". While it is true that changes from less to more subjective meanings can
be observed in the developments of the apprehensional epistemics, there are
also developments that do not fit this pattern. First, although it is true that the
development of apprehensional epistemics involves pragmatic strengthening
(the LEST element coming to signal uncertainty and apprehension), this is not
necessarily the end of the process of semantic/pragmatic change. As we have
seen in the case of Fijian, a LEST element may lose its necessarily apprehen­
sional force and come to be used as a neutral epistemic downtoner. When an
Apprehensional Epistemics 323

erstwhile apprehensional epistemic comes to be used without the apprehen­

sional force, the result is pragmatic weakening.
Second, in Standard and Boumaa Fijian the (neutral) epistemic element
has acquired a new function, that of a politeness marker. What we get is a shift
from one pragmatic function — epistemic — to another pragmatic function —
politeness marking. One would be hard put to argue whether this involves
pragmatic strengthening or weakening
Clearly, not every instance of semantic/functional change in grammati-
calization results in greater subjectivity, in pragmatic strengthening; gram­
matical functions do not keep on becoming more and more subjective, bearing
more and more pragmatic significance. One finds both pragmatic strengthen­
ing and weakening, and also shifts from one type of pragmatic function to
another. Traugott and König (1991:212) say that it is the early stages of
grammaticalization processes that are characterized by pragmatic strengthen­
ing, by a change from less to more informativeness (see also Traugott 1982).
But this raises the problem of defining the early stages of a grammaticaliza­
tion process: when does 'early' stop and 'late' begin?.
On the basis of studies of pragmatic shifts in grammaticalization, includ­
ing the present one, the following principle of directionality of semantic/
pragmatic change in grammaticalization can be formulated:
(71) In a grammaticalization chain, pragmatic strengthening standardly
precedes pragmatic weakening, not vice versa.
The principle does not claim that in every single chain of grammaticali­
zation processes both pragmatic strengthening and weakening are manifested
(they are not likely to be both manifested if the chain is relatively young), but
it does claim that if both are manifested the changes that involve weakening
are more likely to follow rather than precede the changes that involve
strengthening. The principle also subsumes shifts from one type of pragmatic
function to another.
Although the evidence concerning the empirical status of the principle of
directionality of semantic/pragmatic change in grammaticalization is encour­
aging, there is clearly need for many more case studies of semantic/functional
change to test the validity of the principle.
 324 Frantisek Lichtenberk

ABBREVIATIONS

ABL - ablative MID - middle

ABS - absolutive NEG - negative, negator
ACC - accusative NOM - nominative
ART - article NOMI - nominalizer
AUX - auxiliary NONFACT - nonfactative
CLASS - classifier NONFEM - non -feminine
COMP - complementizer NOW - now
COMPL - completive PAST -past
CONT - continuative PAUC - paucal
DAT - dative PERF - perfect
DO - direct object PL - plural
DU - dual POLITE - politeness
DUR - durative PRES - present
EMPH - emphatic PROH - prohibitive
ERG - ergative PURP - purpose
FACT - factative RED - reduplication
FUT - future SEQ - sequencer, sequential
GEN - genitive SG - singular
IMP - imperative ss - same subject
INCH - inchoative SUBJ - subject
INCL - inclusive SUBJUNCT - subjunctive
IND - indicative TOKEN - token
INT - intensifier TRANS - transitive
LIM - limiter VB - verb
LOC - locative S.t. - something.

NOTES

1. I am grateful to Reuel Riianoa and Lawrence Foanaota for the To'aba'ita data, to
Athanasius Faifu for the South Lau data, to Paul Geraghty and Albert Schütz for data and
information on Standard Fijian, to Andrew Pawley for information on Wayan Fijian, and
to Frantiska Lichtenbergová for her help with the Czech data. In revising this paper, I
have greatly profited from discussions at the Mood and Modality Symposium and from
detailed comments by Suzanne Fleischman.
2. The other conventions and abbreviations used in glossing the examples are given above.
3. Besides fear, Palmer includes hopes and wishes under volitive modality, "fear [being]
essentially the counterpart of hope (not of wish)" (Palmer 1986:119).
Apprehensional Epistemics 325

4. Besides Sequential subject/tense markers, To'aba'itahas 'Factative' subject/tense mark­

ers, used to encode past and certain non-past situations (see kera in 7), and 'Non-
factative' subject/tense markers, used to encode future and certain non-future situations
(see kai in 14).
5. It is, of course, possible for an apprehension-causing situation not to be connected by a
causal link to certain precautions but to be so connected to certain other precautions.
6. Note that the term 'precautioning' identifies types of function of LEST elements (avertive
and in-case), while the term 'precautionary' has to do with situations (the precautions (to
be) taken in view of an apprehension-causing situation).
7. The notion of 'in case' can be expressed by, for example, v pripadě ze/kdyby 'in case
that/if' and kdyby náhodou 'if by chance', in which case the clause encoding the
potential situation is grammatically positive. The potential situation is not necessarily
undesirable.
8. LEST elements are common in the Oceanic subgroup of Austronesian, of which
To'aba'ita and Fijian are members. A LEST element is also found in Melanesian Pidgin,
where its presence is most likely due to Oceanic-substrate influence. The Pidgin LEST
element has the form nogut, from English no good.
9. In another Fijian dialect, Nabukelevu, the cognate of the Standard and Boumaa Fijian
LEST elements has (at least) an avertive and an in-case function; there is no mention of a
fear, an apprehensional-epistemic, or a politeness-marking function in Pawley's (1982)
brief sketch of Nabukelevu grammar.
10. Givón speaks of 'epistemic deference'.
11. Although Hopper uses the term 'persistence' specifically with reference to changes from
lexical meanings to grammatical functions, basically the same phenomenon is found in
changes from one grammatical function to another. Of course, a component of the
source grammatical function may itself be due to persistence of a component of the still
earlier lexical meaning.
12. It remains to be seen whether this development in To'aba'ita is accidental or whether
there is a general tendency for speech-act functions to give rise to textual functions
rather than the other way around.
13. Sweetser (1990) gives a number of examples of polysemy uniting content-expressing,
epistemic, and speech-act meanings/functions in English. The development of
To'aba'ita ada from the content-expressing meaning 'see, look at, watch' to the speech-
act, warning function 'look out, watch out' and ultimately to the apprehension-epistemic
function is a diachronic counterpart of this kind of synchronic polysemy.

REFERENCES

Adamson, Sylvia, Vivien Law, Nigel Vincent & Susan Wright, eds. 1990. Papers from the
5th International Conference on English Historical Linguistics, Cambridge, 6-9 April
1987. Amsterdam/Philadelphia: John Benjamins.
Aronoff, Mark, ed. 1991. Proceedings of the LSA Arizona Institute Workshop on Mor­
phology. SUNY Press.
 326 Frantisek Lichtenberk

Austin, Peter. 1981. A Grammar of Diyari, South Australia. Cambridge Studies in Lin­
guistics 32. Cambridge: Cambridge University Press.
Austin, Peter, ed. 1988. Complex Sentence Constructions in Australian Languages. Ams­
terdam/Philadelphia: John Benjamins.
Bybee, Joan L. & William Pagliuca. 1985. "Cross-Linguistic Comparison and the Devel­
opment of Grammatical Meaning." Fisiak 1985. 59-83.
Capell, A. 1973. A New Fijian Dictionary (4th ed.). Suva, Fiji: Government Press.
Churchward, C. Maxwell. 1941. A New Fijian Grammar. Suva, Fiji: Government Press.
Coates, Jennifer. 1983. The Semantics of the Modal Auxiliaries. London and Canberra:
Croom Helm.
Collins COBUILD English Grammar. 1990. London and Glasgow: Collins.
Dench, Alan. 1988. "Complex Sentences in Martuthunira." Austin 1988. 97-139.
Dixon, R. M. W. 1977. A Grammar of Yidin. (Cambridge Studies in Linguistics, 19.)
Cambridge: Cambridge University Press.
Dixon, R. M. W. 1988. A Grammar of Boumaa Fijian. Chicago and London: University of
Chicago Press.
Donaldson, Tamsin. 1980. Ngiyambaa: The Language of the Wangaaybuwan. (Cam­
bridge Studies in Linguistics, 29.) Cambridge: Cambridge University Press.
Fisiak, Jacek, ed. 1985. Historical Semantics: Historical Word-Formation. Berlin: Mou­
ton.
Fox, Charles E. 1974. Lau Dictionary. Pacific Linguistics C-25. Canberra: Australian
National University.
Givón, T. 1990. Syntax: A Functional-Typological Introduction, Vol. II. Amsterdam/
Philadelphia: John Benjamins.
Goodwin, William W. 1929. Syntax of the Moods and Tenses of the Greek Verb. London:
Macmillan and Co.
Haiman, John. 1980. Hua: A Papuan Language of the Eastern Highlands of New Guinea.
(Studies in Language Companion Series 5.) Amsterdam: John Benjamins.
Hall, John R. C. 1950. Beowulf and the Finnesburg Fragment (new ed.). London: George
Allen & Un win.
Hopper, Paul J. 1991. "On Some Principles of Grammaticization." Traugott & Heine
1991. 17-35.
Horn, Laurence R. 1989. A Natural History of Negation. Chicago and London: University
of Chicago Press.
Jespersen, Otto. 1933. Essentials of English Grammar. London: George Allen & Unwin.
Keesing, R. M. 1975. Kwaio Dictionary. Pacific Linguistics C-35. Canberra: Australian
National University.
Klaeber, Fr. 1950. Beowulf and the Fight at Finnsburg (3rd ed.). Boston: D. C. Heath.
Lehmann Winfred P. and Yakov Malkiel, eds. 1982. Perspectives on Historical Linguis­
tics. Amsterdam: John Benjamins.
Lichtenberk, Frantisek. 1991. "On the Gradualness of Grammaticalization." Traugott &
Heine 1991. 37-80.
Mitchell, Bruce. 1985. Old English Syntax (Vol. II). Oxford: Clarendon Press.
Oxford English Dictionary (2nd ed.). 1989. Oxford: Clarendon Press.
Palmer, F. R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Apprehensional Epistemics 327

Pawley, Andrew. 1982. "A Sketch Grammar of the Nabukelevu Language of Kadavu." Te
Reo 25.35-93.
Pawley, Andrew and Timoci Sayaba. Forthcoming. Wayan Dictionary: Describing the
Wayan Dialect of the Western Fijian Language. (Pacific Linguistics.) Canberra:
Australian National University.
de Reuse, Willem J. 1991. "The Role of Internal Syntax in the Historical Morphology of
Eskimo". Aronoff 1991.
Scheurweghs, G. 1959. Present-Day English Syntax: A Survey of Sentence Patterns.
London: Longmans.
Schütz, Albert J. 1985. The Fijian Language. Honolulu: University of Hawaii Press.
Shearin, Hubert G. 1909. 'The Expression of Purpose in Old English Poetry." Anglia
32:235-252.
Sweetser, Eve. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects
of Semantic Structure. (Cambridge Studies in Linguistics 54.) Cambridge: Cambridge
University Press.
Traugott, Elizabeth Closs. 1982. "From Propositional to Textual and Expressive Meanings;
Some Semantic-Pragmatic Aspects of Grammaticalization." Lehmann & Malkiel
1982.245-271.
Traugott, Elizabeth Closs. 1989. "On the Rise of Epistemic Meanings in English: An
Example of Subjectification in Semantic Change." Language 65.31-55.
Traugott, Elizabeth Closs. 1990. "From Less to More Situated in Language: The Unidirec-
tionality of Semantic Change." Adamson et al. 1990, 497-517.
Traugott, Elizabeth Closs & Bernd Heine, eds. 1991. Approaches to Grammaticalization,
Vol. I: Theoretical and Methodological Issues. Amsterdam/Philadelphia: John
Benjamins.
Traugott, Elizabeth Closs & Ekkehard König. 1991b. 'The Semantics and Pragmatics of
Grammaticalization Revisited." Traugott & Heine 1991. 189-218.
Tryon, D. T. & B. D. Hackman. 1983. Solomon Islands Languages: An Internal Classifi­
cation. Pacific Linguistics C-72. Canberra: Australian National University.
von Wright, Georg H. 1951. An Essay in Modal Logic. Amsterdam: North-Holland.
 Moods and MetaMessages.
Alienation as a Mood

John Haiman
Macalester College

0. Introduction

Many discussions of mood as a motivated general category describe it as "the

grammaticalization of the speaker's attitude to the propositional content of his
message" (cf. Palmer 1986: 16, 51, 96). Belief in its truth is indicative;
suspended judgment as to its validity, subjunctive; the wish for its realization,
imperative or optative; ignorance as to its truth, coupled with curiosity, is
interrogative; fear of its realization, avolitional or apprehensive. But there are
many other propositional attitudes which are not so universally codified even
as words, to say nothing of grammatical mood markers.
In this paper, I want to describe some of these sub-canonical moods: the
quasi-modal marker I will deal with is that of negation or rejection, which,
like the modalities of ability and obligation, has deontic, epistemic, and de
dicto meanings. I will focus neither on the deontic (= "I don't want this"), nor
the epistemic (= "This is not the case"), but on the de dicto or metalinguistic
meaning of negation (= "I reject these words"). At the end of an unsystematic
survey of some markers of metalinguistic negation, primarily in English, I
want to offer a guess as to why — with one significant exception — they have
never achieved ritual grammatical canonization as grammaticalized mood
markers in any language I have ever heard or read about.
330 John Halman

1. The sarcastive

One might (for example) call the sarcastive that modality (or, more plausibly,
that collection of related modalities) which expresses the speaker's belief that
the content of the ostensible message is not only false, but ridiculous. Like
many other propositional attitudes, sarcasm is easily recognized, and we have
a folklore about sarcasm and an extensive vocabulary for talking about it. But
there is no spoken language familiar to me in which the sarcastive modality is
marked with a clearly recognized and stable separate morpheme of the sort
that gets written up in sober grammatical descriptions.
There are a small number of very widespread and iconically motivated
suprasegmental markers of disgust (e.g. nasalization), non-commitment (e.g.
monotone), contemptuous mimicry (e.g. falsetto) and alienation (e.g. sing­
song) which signal lack of belief, identification, or enthusiasm in one's
message. I have described some of these in greater detail elsewhere (Haiman
1989, 1990,1991).
But there are segmental (and, in the written language, orthographical)
realizations of these attitudes as well. Here are a couple of segmentally
realized sarcastive modality markers which are current in colloquial Ameri­
can English (current, at least, as I write these words):

1.1 Like

In the examples of (1), the whole utterance following like (or its predictable
synonym as if) is in the scope of this marker. The hearer is warned that what
follows like is what the speaker may be expected to believe, but does not.
(1) a. Like I care. (= I don't care.)
b. Like there's any difference. (=There's no difference.)
c. Like I haven't heard that one before. (=I've heard that one
before.)
d. As if TV advertising weren 't intrusive enough already. (= TV
advertising is too intrusive already.)
Possible equivalents with invariably sarcastic meaning in French and
Russian are si tu crois que "if you think that" and mozhno podumat' "(it is)
possible to think":
Moods and MetaMessages: Alienation as a Mood 331

(2) a. Si tu crois que je l'ai regardé!

'Like I was looking at him.'
b. Mozhno podumat' ja bespokojus'.
'Like I care'

1.2 ....not

This utterance deflater signals that the preceding utterance (often a compli­
ment) was meant facetiously. Until it is uttered, the hearer may believe that
the prior utterance was sincere.
(3) a. He's a snappy dresser.....not!
b. Guess I'll hit the books in time for that quiz not!
c. Real Life with Jane Pauley... not.
à. Bush will kill that program. Not.
e. That's a fabulous science fair project. Not!
According to Larry Horn (e-mail circular of December 21, 1991), this
prototypically generation X (Wayne's World) locution surfaced not on the TV
show Saturday Night Live about three years ago (as is commonly supposed),
nor even (pace William Safire) ten years before that, in 1978, when Steve
Martin, on the same show, uttered (3e), but at least a couple of generations
before then. Horn credits his correspondent Richard Piepenbrock with un­
earthing a Little Nemo cartoon of 1910 in which this locution appears (McCay
1990), and notes, I think quite correctly, that it is highly unlikely to have been
an innovation there. Clearly, this totally motivated construction has come
around often in American speech — without ever becoming firmly estab­
lished. A British English equivalent with a longer history is — I don't think,
now almost entirely obsolete (Suzanne Kemmer, pc).
Sarcastive retroactive ..not is clearly an example of what Horn (1985) has
called metalinguistic negation, and it differs from garden-variety epistemic or
deontic negation only in position. It is interesting that the formal relationship
between de dicto negation and the ordinary de re negation is parallel to the
difference that Horn extensively discusses in cases like
(4) a. I'm not happy: in fact, I'm downright miserable.
b. I'm not "happy" : I'm ecstatic.
The latter is metalinguistic. And it is the metalinguistic not which cannot
appear as a derivational prefix:
332 John Halman

(5) a. I'm unhappy: in fact, I'm downright miserable.

b. *I'm unhappy: I'm ecstatic.
In each case, the "metalinguistic" negator contrasts with its garden variety de
re opposite number in that it is further separated from the constituent or the
utterance which it negates. I've argued that separation from its embedding
context is one of the ways in which a speaker marks the peculiar other-worldly
status of quoted or mentioned material (Haiman 1989): perhaps the external
position of the negator in iself serves as a kind of iconically framing quotation
mark.

1.3 or anything

The clause introducer not that is generally used to signal that the speaker's
interest in what follows is minimal:
(6) a. Not that I value my sanity so highly.
b. Not that I care about the money.
The motivation is clear. A reconstituted "full" paraphrase would be something
like:
(7) a. It's not that I value my sanity so highly. (What terrifies me are
screams of my fellow lunatics, the curses of my keepers, and
the clanging of the chains.) (with apologies to Pushkin)
b. It's not that I care about the money. (What I do care about is
the principle of the thing.)
Like many other things, however, a meaning can be mocked through ostensi­
ble exaggeration or emphasis. (It seems to me that this is the essence of all
caricature.) When strengthened by the tag or anything, the belittling constitu­
ent not that is apparently strengthened: but in practice, the speaker's ostensi­
ble act of trivialization is mocked.
(8) a. Not that I care about the money or anything. (=1 care passion­
ately about the money.)
b. Not that it's cold out or anything. (= It's real cold out.)
c. Not that you annoyed me or anything. (=You annoyed the hell
out of me.)
But for the fact I have a teenager in the family, I wouldn't have heard this
marker.
Moods and MetaMessages: Alienation as a Mood 333

1.4 Quote

This is of course a spoken rendition of the orthographic quote sign. Although

in dictation quote...unquote is a discontinuous constituent with arbitrarily
long strings within its scope, as a sarcastive quote (unquote) is continuous and
only the following word (or, at most, short phrase) is within its scope.

(9) Your quote (unquote) principles are nothing but snobbism.

Quote is the only sarcastive morpheme which has lasted for any continuous
length of time in English. It is also the only one which has grammaticalized
congeners that I have been able to find in other languages.
The quotative is an evidential category in a large number of the world's
languages (cf. Chafe & Nichols, 1986). In at least two of these, its meaning
has been extended to convey irony as it does in English. Thus, Turkish:
(10) Her gün ko§- uyor- mu§
every day run PROG evid 3SG
'He reportedly runs every say'

"can convey not only hearsay, but doubting scorn when predicated of a well-
known exercise hater" (Aksu-Koç & Slobin 1986:162; the authors also cite
Kononov (1956:232) who notes that the evidential in Turkish conveys "an
ironical attitude toward the carrying out of an action").
Similarly, the Albanian admirative, which is 'traditionally
defined as the mood expressing surprise' is 'also used to express irony, doubt,
and reportedness' (Friedman 1986:180):

(11) E na i dashka bullgaret.. ai e..

and to-us them loves(adm.) Bulgarians he ha!
e pse keshtu u pritka nje Bullgar?
and why thus is-met a Bulgarian
'And he 'likes' Bulgarians. Him? Hah! After all, is that how you
treat a Bulgarian?'
Friedman (ibid. 183) notes that in this sentence, the speech event "I like
Bulgarians" is mocked (rather than simply reported). In the written transla­
tion, the quotation marks seem to capture the admirative perfectly.
334 John Haiman

In Rumanian, lexical equivalents of the quotative morpheme are the

sentential adverbs parca 'seemingly' and adica 'that is to say, in other words',
as in (Mallinson 1986:6).
(12) a. parca noi am jurat masina?
seemingly we have stolen car
'I suppose that you're now going to say we stole the car?'
b. adica, sa fiu eu cel care a murit?
i.e. SUBJ am I DEM=MS who HAS=3SG died
'I suppose you're going to say I'm the one that died?'
Note that Mallinson's felicitous English translation of the sarcastive as 'I
suppose you're going to say' is also a periphrastic quotative.
Another possible congener is Japanese post-utterance datte 'said', but
only if it is separated from the preceding quotation by a slight — framing —
pause (Adachi, ms.):
(13) a. "Watashi wa ii mono shika kawanai kara " datte, (sincere)
b. "Watashi wa ii mono shika kawanai kara" ...datte, (sarcastic)
I (TOP) good things only not buy because says
'"I only buy good quality things" (...) she says.'
Natural as the extension from quoting to mocking mimicry may seem to
us, it is by no means universal. Robert Oswalt, for example (p.c.), emphasizes
that in Kashaya neither the quotative evidential, nor the actual direct quote
followed by /nihcedu/ 'say' can ever be used to signal irony, sarcasm, or doubt
of the veracity of the source; nor does this seem to be possible in Pomo
generally. I would speculate that those languages in which the quotative
cannot be sarcastic may be among those fabled languages and cultures immor­
talized by Douglas Q. Adams in The Hitchhikers ' Guide to the Galaxy where,
as in Betelgeuse, "they don't have sarcasm" at all.

2. The guiltive

In the following pages, I will discuss a more elusive (though instantly recog­
nizable) related propositional attitude which I'll call the guiltive, for which
not even an adequate pretheoretical folk vocabulary exists. Consider the fol­
lowing three examples from Philip Roth's Portnoy's Complaint:
Moods and MetaMessages: Alienation as a Mood 335

(14) "Don't ask what kind of day I had with him yesterday. " So I don't.
"Alex", sotto voce still, "when he has a day like that you don't
know what a difference a call from you would make." I nod. "And,
Alex" — and I'm nodding away, you know — it doesn't cost
anything, and it may even get me through— "Next week is his
birthday.
That Mother's Day came and went without a card, plus my
birthday, those things don't bother me." (39)
(15) ...my father carries himself to the kitchen table, his head sunk low,
as though he has just taken a hand grenade in his stomach. Which
he has. Which I know.
"You can wear rags for all I care, you can dress like a peddler,
you can shame and embarrass me all you want, curse me,
Alexander, defy me, hit me, hate me —" (70)
(16) Yes, she will give me the food out of her mouth, that's a proven
fact! And still I will not stay five full minutes by her bedside.
"Run ", says my mother, while Mrs. Re-ver-ed, who in no time at all
has managed to make herself my enemy, and for the rest of my life,
Mrs. Re-ver-ed says, "Soon Mother will be home, soon everything
will be just like ordinary...
"Sure, run, run, th ey all run these days," says the kind and
understanding lady — oh, they are all so kind and understanding, I
want to strangle them! — "walking they never heard of, God
bless them." (75)

The sarcastive attitude is recognized — is meant to be recognized — when the

speaker's ostensible message is accompanied by a derived meta-message
"this message is bogus". All it takes to be sarcastic is a single speaker, who
produces both the message and the denigrating commentary on it.
The guiltive, however, is a cooperative venture, a linguistic pas de deux
in that essentially the same derived metamessage ("this message is bogus") is
not produced by the speaker but is rather left to be supplied by the addressee,
who is thereby made to feel like a worm.
As I have tried to show, the sarcastive is overtly marked. Cues for the
sarcastive meta-message (literally 'I don't mean this message') include 'spit­
ting it out with a sneer', nasalization, deadpan monotone, sing-song, carica­
tured courtesy, formality, and sympathy, such indices of fakery as mark
336 John Haiman

rhetorical questions and commands (e.g. "Don't ask me what kind of day I had
with him yesterday"), and a small repertoire of the indices of mimicry,
including most obviously (written, gestured, or pause-marked) quotation
marks. Colloquial English includes as well a handful of segmental signs. All
of these signs more or less iconically convey the framing metamessage This
message is counterfactual' or 'I don't mean this', whence the inference
follows that 'this message is bogus'.
The guiltive speech act is entirely different, in that, in order to properly
trash her interlocutor, the guilter (typically but not always the stereotyped
Jewish mother) has to sound perfectly sincere (and hence cannot be heard to
broadcast the metamessage 'I don't mean this', which must therefore be
covert). This suggests a kinship between the guiltive and polite language, in
which the speaker also suppresses his/her own emotions, is known to be
suppressing them, and still sounds "sincere".
Nevertheless, it is instantly obvious that the bolded passages from
Portnoy's Complaint above are in the guiltive mood, rather than polite: while
the illocutionary force of politeness is (presumably) to avoid aggression, the
illocutionary force of guiltive utterances is clearly a kind of passive-aggres­
sive one-upsmanship whose purpose is to make the hearer feel bad. How is
this clear? How do the passages above do this work, and how do they contrast
with the same passages uttered with genuine politeness or with genuine
sincerity, assuming that such a sincerity is even possible?
Some obvious suggestions that Dear Abby might provide for a non-guilt-
producing illocution include the following:
(14') Don't exaggerate your selflessness. ( If you want to be sincere, say:
" Of course it bothered me that you forgot Mother's Day and my
birthday. " If you want to be polite, don't mention the incident at
all.)
(15') Don't exaggerate your pain. (If you want to be sincere, say: "It
really hurts my feelings that you won't dress up for synagogue. " If
you want to be polite, don't show you 're upset at all.)
(16') Don't exaggerate your forbearance. (Sincerity might impel you to
say: "You wretched little beast, running off to play baseball!"
Politeness, once again, would impel you to keep quiet.)
What's common to all of the ostensibly benign messages from Portnoy's
elders is that the speaker's ludicrously exaggerated selflessness, hurt, and
Moods and MetaMessages: Alienation as a Mood 337

generosity, all express the explicit meta-message that "I the utterer of this
message am a saint/martyr".
This metamessage, then, is not so much a commentary on the message
itself, as on the nature of the person who can utter it sincerely. Implicit for
Portnoy to elucidate are the derived metamessage "this message is therefore
objectively bogus", and the accompanying inference that "you, Portnoy, are a
heartless swine".

3. The mass-productive

I have already suggested that emphasis may have contradictory effects: osten­
sibly it increases the import of a message, while in practice it undermines it.
The same is true of repetition, the simplest kind of emphasis imaginable.
Paradoxically, replication (both of objects and of signs) validates in some
cases, invalidates in others: see for example Moravcsik's exemplary cross-
linguistic discussion of reduplication (Moravcsik 1978).

3.1 Replication validates

3.1.1 The "big lie" repeated is believed. This much is a truism, but it entails
an equally plausible corollary: not even the truth is believed if uttered only
once in the context of a culture of incessant repetition. If, on TV, we saw
George Bush streaking through Washington just once, and the clip were never
repeated, we would soon tend to think we had only imagined it. Incessant
repetition breeds a kind of if-it's-true-Fll-get-a-chance-to-see-it-again-later
philosophy. In other words: replication is a prerequisite for truth.

3.1.2. Replication produces wealth, visibility, and glamour for mass media
stars. It confers "seriousness" on political candidates. The validating power of
the prototypical replicating medium, TV, is so much a given, that the humour
of the Calvin and Hobbes cartoon included here as Figure 1 lies not in the
recognition that "TV validates existence" ( a point for me as a plodding
linguist, but a self-evident cliché for the cartoonist) but in the pomposity of
the kid's diction, and in the irony that Calvin and Hobbes have themselves
achieved the status of "cultural icons" precisely through the replicating me­
dium of the comics.
338 John Haiman

Figure 1

3.2 Replication invalidates

History repeats itself: the first time as tragedy, the second (and all subsequent)
times as farce.

3.2.1. Irrespective of its source (whether hand-made or mass produced) any

individual object loses its value in a context of mass produced items: any
single thing can be replaced by something nearly identical.

3.2.2. The philosophically once problematic notion of similarity becomes

more and more natural, familiar and plausible. Forever obsolete are uncon­
vincing homespun similes based on identities from nature: "like two peas in a
pod". Universal Categories become objectively, as well as psychologically,
real. Possibly, no two peas, no two dogs, no two people are completely alike:
but two identical cans of Campbell soup, two Fords, two copies of a paper­
back, two pieces of china, two sofas, two cameras, two suits, are much less of
a problem. Borges' nominalist Funes complained that we had the same word,
not only for dogs in general, but for the same animal, viewed head on at 3:14,
and in profile at 3:15 (Borges 1963). Mass production suggests Funes had it
wrong, and that Andy Warhol was right: the artificial universe is populated
with mass produced articles which, as far as the human eye and ear can judge,
are as nearly identical as electrons, certainly as alike as photocopies of the
same original. Your Campbell Soup can is indistinguishable from the Platonic
Moods and MetaMessages: Alienation as a Mood 339

Campbell Soup Can laid up in Heaven. Each of the glamorous zombies in the
videos of Robert Palmer is conceptually and perceptually identical with all the
others.

3.2.3. It becomes necessary to express personality and originality (in dress; in

a living room) primarily through idiosyncratic combinations of mass pro­
duced items: through collage (as in art) or sampling (as in rap music), rather
than through original production. (We may observe a parallel to the phenom­
enon of double articulation: individual items, by virtue of their mass produc­
tion, are on the way to becoming as meaningless as phonemes.)

3.3. Repetition as a diachronic paradigm

3.3.1. The meaning of a morpheme, or of an utterance, is determined not only

by the extralinguistic circumstances under which it is uttered, all the other
utterances with which it contrasts (synchronic paradigmatic context), and by
the utterances which precede and follow it (synchronic syntagmatic context).
It may also be determined, or at least affected, by speakers' awareness of the
sheer number of times it has been uttered before (by themselves or by others).
Jingles, clichés, hackneyed platitudes, sayings, tags, maxims, aphorisms,
slogans, mottoes, proverbs, saws, adages, boilerplate —whatever we repro­
duce as "blah, blah blah" (Ong 1971:303) —are the Campbell Soup cans of
discourse, and mean something different from freshly minted expressions.
The William Tell Overture, the injunction "Don't leave home without it", the
question "Who you gonna call?", the exclamation "I love what you do for me"
and hundreds of others are both enriched or impoverished through being
recognized as mass produced.
A speaker consciously quoting "X" is taking a different stance towards
his ostensible message than one who utters "X" (or believes himself to be
uttering it) for the first time. (My favorite example of the resultant ambiguity
of any text is Borges' profound short story "Pierre Menard, Author of the
Quixote", whose hero undertakes to write portions of the Quixote, but with
the mindset of a twentieth Century French symbolist poet rather than a
sixteenth century Spanish veteran. The different context in which the same
words are uttered invests the text with totally unfamiliar meanings). For a
more homely example, consider the act of reclamation, whereby oppressed
minorities preempt pejorative epithets generally hurled against them for ironi­
cal use amongst themselves: "nigger", "bitch", "dyke" and "fag" can be terms
 340 John Halman

Indeed, the Republican establish­

ment denounces Duke at every turn
'--Bush has called him "an insincere
charlatan." But Democrats gleefully
keep putting Duke at the Republican
doorstep and GOP officials often
have difficulty drawing the line be­
tween Duke's carefully chosen rheto­
ric, and established party positions,
such as opposition to racial quotas.
Republicans praised Kirk Fordice
when he was elected governor of Mis­
sissippi last year. His campaign
themes included criticism of affirma­
tive action programs, warnings
against racial quotas in hiring and
welfare overhaul. Duke is raising
those issues too.
Republican Party Chairman Clayton
Yeutter concedes that much of what
Duke says is similar to what Fordice
and Bush say. "The fundamental
economic views of these two gentle-
men and the president of the United
States are similar," Yeutter said.
"The difference comes in the types of
human beings who are involved. Da­
vid Duke is a charlatan and a fraud."

Figure 2

of jocular endearment. For another example of a related phenomenon, con­

sider the difference between received perceptions of identical racist senti­
ments uttered by George Bush and David Duke, reproduced as figure 2.
The self-conscious quoter's stance may be positive: affectionate, an
attempt at asserting a phatic bond with others who recognize the quote; or it
may be negative: an expression of mockery or boredom, and of a lack of
personal commitment. It may be both.

3.3.2 Grammaticalization
The form of an utterance is also affected by these three variables. When both
form and meaning are affected by repetition, grammaticalization has oc­
curred. The most familiar semantic aspect of grammaticalization is the ero­
sion of meaning. But repetition may flat-out destroy meaning:
 Moods and MetaMessages: Alienation as a Mood 341

Brian: You are all individuals!

Multitudes (in unison): "We are all individuals."
(Monty Python's Life of Brian)
Consider a commercial invitation which many of us have heard more
than once: to "buy one, and get another one free". This familiar offer was
recently (Minneapolis Star Tribune Dec.11, 1991) ruled false advertising in
Minneapolis, where one company, Pearle Vision, had been using it to adver­
tise their eyeglasses . Precisely because the offer had stood for awhile, the
Better Business Bureau correctly reasoned, the regular price had in fact been
halved. Here — literally — a true statement became false through repetition.
There are in general two ways in which this may happen.
Any word like "sale", "individual", or "special" which has "unique" or
"unusual" somewhere in its definition or its presuppositions becomes literally
false (by definition) simply by virtue of being repeated. This is what happened
with the Pearle Vision come-on, and this is what happens when the messiah
instructs the multitudes to chant in unison "we are all individuals" in Life of
Brian. Something like this has happened with the reflexive pronoun in Eng­
lish, which is still homophonous with the emphatic pronoun. (In fact, we now
signal suprising coreference through the oblique non-reflexive pronoun, as in
"the prosecuting attorney believed me, and I believed me." cf. Haiman ms.)
More generally, any text becomes pale — if not false —through repeti­
tion. We tend to become aware of this only when we realize that the words of
a familiar unattended ritual have become practically a practically meaningless
formula. The word dear in salutations, when applied not only to strangers, but
to anonymous committees, is a good example of this trend. And, of course,
this has happened to the closing salutation also:
For a while, we simply closed all letters, "Love". It didn't matter if we were
writing to parents, lover, friend, or dentist [...] Finally, we wore the word
out. (Richard 1992)

This brings me finally to the point of this section. If the utterance of a repeated
expression may express the speaker's lack of commitment to it, then whatever
formally marks an utterance as a repetition can count as a sort of mood
marker. Given how often what we say is recycled, it is amazing how undevel­
oped are our means for labelling our clichés or standard routines as such,
thereby safely distancing ourselves from them (if that is what we want to do).
For example, it is entirely clear what the quotation marks are signalling in the
342 John Haiman

Doonesbury / By Garry Trudeau

Figure 3

wonderful Doonesbury cartoon of Figure 3, but it is by no means clear how

they are to be pronounced.
The most codified of the self-conscious repetition marking devices in
English (but I have also noted it in Russian, Turkish, and Oromo) is a weary
descending singsong chant (with stylized intonation and rhythm), often pref­
aced by "I know: ", prefaced or followed by "Yeah, right: ", uttered on the
same descending melody. This melody typically occurs where the speaker is
ironically quoting an interlocutor or the wisdom of the silly ancestors.
(17) a I know. 'The check is in the mail.'
b Yeah, yeah. 'Read my lips.'
c Sure, sure. 'Look both ways before you cross the street.'
d He's retiring. (Yeah, sure). (Time June 18,1990, in an article
on the "Joe Isuzu" character.)
Another is the hen-pecked husband's eternally weary pre-interrogative
now, or post-interrogative marker this time, both heavily stressed, as in:
(18) a. Now what have I done ?
b. What is it this time ?
In indicating that he has asked the question many times before, the Henpecked
Husband is signalling his lack of the concern or curiosity which attend a
normal question.
Another way of indicating that an utterance has been often repeated is to
repeat it yourself. While it is generally true that repetition serves the vaguely
Moods and MetaMessages: Alienation as a Mood 343

iconic function of "emphasis" (cf. Wierzbicka 1991:262), it is nevertheless

true that repetition may directly signal weariness, boredom, and alienation (cf.
Haiman 1991). It is (I think) no accident that the weary sarcasm of the bored
repeater of other people's threadbare clichés is preceded by an almost obliga­
tory repetition — "yeah, yeah", "sure, sure" or the like.
Japanese does this also. Ume 'good' is appreciative; but umai, umai,
umai is almost invariably sarcastic (Adachi, ms.).
Larry Gorbet points out that Yiddish "schmo reduplication" has much the
same kind of pejorative function: justice-schmustice, in reduplicating, but
also distorting, the word, conveys the speaker's attempt to trivialize the
concept the word represents.
In a cryptic passage on reduplication in Ewe, Ansre (1963:132) suggests
that reduplication of an entire sentence in this language has the same kind of
pragmatic function: the repeated sentence is usually " [an] alleged saying,
giving notions that need to be corrected " or, possibly, notions which are
treated as "cheap talk":
(19) maalee maalee, Di wodOna
'(I shall bathe)' 2 makes one go to bed dirty

4. Proper names > common nouns

The Holy Roman Empire, as everyone knows, was neither holy, nor Roman,
nor an Empire. Although many proper names originate as definite descrip­
tions, once congealed as proper names, they are no longer required nor
expected to mean what they say. Perhaps this could be viewed as a special
case of grammaticalization. In any case, the device is self-consciously used as
an artful disclaimer of commitment in at least one recent case of false
advertising, Ragu foods vs. the Food and Drug Administration (Advertising
Age, May 28, 1990). The company marketed "Ragu fresh Italian pasta
sauces", and the FDA took exception to the word "fresh", which was inappro­
priate for processed tomatoes. After a nine-month suit, the company capitu­
lated: they announced they were marketing "Ragu fresh Italian brand pasta
sauces". The term "fresh" is now "part of the product's brand name, and not a
product description".
Arguably, this example is one of a purely legalistic trick (and one that
actually failed, as Larry Horn has informed me, his best efforts as hired gun
344 John Haiman

for Ragu notwithstanding). But personal experience offers many parallels,

particularly in the transformation of descriptive nicknames into proper names.
A fat boy I knew (in Onondaga Camp, 1952) was called Fatso as an insult, but
once everyone called him that as a matter of course, the insult dissipated. I still
recall my amazement as a six-year old ESL student overhearing the question
"What's the matter, Fatso?" voiced by another kid in the unself-conscious
accents of genuine solicitude and concern.
A visible change which probably originates through repetition is the
conversion of definite descriptions into common nouns: a gentleman (with
compound stress) is not necessarily a gentle man (with nuclear stress), and an
oldboy literally cannot be an *old boy. But there is no case I know of where a
speaker has ever used this structural property to signal lack of commitment to
the words s/he utters.

Conclusion

My major question in this discussion is: why have virtually none of these
devices for signalling lack of speaker's commitment achieved the grammati­
cal status of the subjunctive or the interrogative? (The only ones that come
close are quote and stylized intonation, both of which have other, more central
functions.)
I suspect that this has not happened because the indices of sarcasm, like
those of anger and other personal emotions, belong to the same realm: they are
not really signs but symptoms of these emotions. Raising one's voice is
probably a symptom of anger in whatever language that one chooses to speak:
nevertheless, no language has grammaticalized increased amplitude as a sign
of anger. Why? Maybe because it's too universal, too natural, and too self-
evident. Signs don't get codified until they're at least a little bit arbitrary.
And maybe sarcasm has failed as yet to get grammaticalized (this sounds
paradoxical, speaking of sarcasm), because — being a symptom — it's too
sincere. I regard it as an essential defining property of a sign that we can
choose to use it, and thus be able to use it insincerely. We have at our disposal
a number of metalinguistic devices for signalling insincerity of the first order:
"this message is bogus". But perhaps as humans at the present stage of our
evolution we are limited by a purely performance-like constraint (similar to
that on center-embedding) which inhibits us from signalling insincerity of the
Moods and MetaMessages: Alienation as a Mood 345

second or higher orders: * "this (meta)metamessage is bogus", which we

could utter once the symptomatic "this message is bogus" was harnessed as a
sign, is simply not sayable by your normal average human, even in postmod­
ern America.

ACKNOWLEDGEMENTS

I'am very grateful to Larry Horn, Suzanne Fleischman and Michele Emanatian
for their comments and suggestions on this paper.

REFERENCES

Adachi, T. ms. "Sarcasm in Japanese". Macalester College.

Aksu-Koç, A. and D. Slobin. 1986. "Developments in the Use of Evidentials in Turkish."
Chafe and Nichols 1986: 159-67.
Ansre, G. 1963. "Reduplication in Ewe". Journal of African Languages, vol. 2, part 2:128-32.
Borges, J. 1963. "Funes the memorious". in his Labyrinths. New York: New Directions.
Chafe, W. and J. Nichols, (eds) 1986. Evidentiality: The Linguistic Coding of Epistemology.
Norwood, NJ: Ablex.
Friedman, V. 1986. "Evidentiality in the Balkans: Bulgarian, Macedonian, and Albanian."
Chafe and Nichols 1986: 168-87.
Haiman, J. 1989. "Alienation in Grammar." Studies in Language 13.121-70.
Haiman, J. 1990. "Sarcasm as Theater." Cognitive Linguistics 1.181-206.
Haiman, J. 1991. "The Bureaucratisation of Language." Wolfart 1991: 45-70.
Haiman, J. ms. "Grammatical Signs of the Divided self."
Horn, L. 1985. "Metalinguistic Negations and Pragmatic Ambiguity". Language 61. 121-74.
Kononov, A.N. 1956. Grammatika Sovremennogo Turetskogo Literaturnogo Jazyka.
(=Grammar of contemporary literary Turkish). Leningrad: Akademia Nauk.
Mallinson, G. 1986. Rumanian. London: Croom Helm.
McCay, W. 1990. The Complete Little Nemo in Slumberland. Vol. 4. Richard Marschall,
ed. New York: Remco Worldservice books.
Moravcsik, E. 1978. "Reduplicative Constructions". In Universals of Human Language,
vol. 3: Word Structure, ed. J. Greenberg et al., Stanford: Stanford University Press,
297-334.
Ong, W. 1971. Rhetoric, Romance, and Technology. Ithaca: Cornell University Press.
Palmer, F. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Richard, J. 1992. "The Distance from '67 to '92: Would You Hug a Stranger?" Baltimore
Sun April 28, 1992.
Roth, P. 1970. Portnoy's Complaint. New York: Dell.
Wierzbicka, A. 1991. Cross-Cultural Pragmatics. The Hague: Mouton.
Wolfart, H. ed. 1991. Linguistic Studies Presented to John Finlay. Winnipeg: Algonkian
and Iroquoian Linguistics Memoirs.
 III
Irrealis Modality and Subjunctive
 The Realis-Irrealis Distinction in Caddo, the
Northern Iroquoian Languages, and English

Wallace Chafe
University of California, Santa Barbara

0. Introduction

There are many ways in which language sheds light on the nature of the mind,
and conversely ways in which an understanding of mental phenomena can
further our understanding of language. The realis-irrealis distinction illus­
trates the usefulness of combining linguistic and cognitive insights in ways
that will contribute to an improved understanding of both. In Chafe (1994) I
explored various paths by which ideas enter consciousness, among them the
direct perception of current states and events, the remembering of previously
experienced states and events, and the imagining of states and events that are
judged not to accord with current objective reality. Of interest here is the
evidence languages provide for a cognitive distinction between what may be
called realis and irrealis ideas, and how recognizing such a distinction can
help us understand what we find in languages.
I will begin with a discussion of the Caddo language, which obligatorily
expresses reality or irreality in all its verbs. Caddo also suggests one of many
diachronic paths by which the distinction may come to be grammaticized, and
helps us understand how, in the course of such a development, there may
develop inconsistencies. I will then turn to the Northern Iroquoian languages,
whose morphology reflects a three-way reality distinction. Finally, I will
mention the fact that English, which does not mark reality overtly, neverthe­
less shows its influence indirectly in the semantics of specificity.
350 Wallace Chafe

Realis expressions are typified by the following, where I have included

constructed English examples for easy comparison:
(1) past states I had a toothache.
past perfective events I got the car fixed.
present states I've got a toothache.
present imperfective events The car's getting fixed.
We can suppose that states and events like these are believed by the invented
speaker to accord with objective reality. The first two examples express
memories of a state or event that was directly experienced by the speaker at an
earlier time. The third expresses a state that is being experienced at the time of
speaking. The fourth expresses a somewhat more inferential kind of knowl­
edge (the speaker may not be directly perceiving the relevant events at the
moment of speaking), but memory combined with normal expectations have
led this speaker to believe that the event accords with reality.
Realis expressions like those in (1) contrast with irrealis expressions like
the following, illustrated with a constructed event (the idea of getting the car
fixed) that in each case is imagined rather than directly perceived or remem­
bered:
(2) yes-no questions Did you get the car fixed?
negations I didn't get the car fixed.
futures I'm going to get the car fixed.
necessities I need to get the car fixed.
possibilities I might get the car fixed.
imperatives Get the car fixed.
prohibitions Don't get the car fixed.
conditions If I get the car fixed...
There is nothing in the overt grammar of English that unites expressions
like those exemplified in (2) and opposes them to expressions like those in (1).
This lack of overt marking in some languages along with the apparent incon­
sistency of marking in others has led some investigators to doubt whether the
realis-irrealis distinction has any general validity at all. Trask (1993:147), for
example, in his dictionary of grammatical terms defines irrealis dismissively
as "a label often applied in a somewhat ad hoc manner to some distinctive
grammatical form, most often a verbal inflection, occurring in some particular
The Realis-Irrealis Distinction 351

language and having some kind of connection with unreality. Palmer (1986)
recommends that this term should be avoided in linguistic theory on the
ground that it corresponds to no consistent linguistic content."1 Bybee,
Perkins and Pagliuca (1994) express similar qualms. These reservations are
understandable, given the diverse and often inconsistent ways in which reality
and irreality manifest themselves. As we will see, however, it is fair to
imagine that if linguistics in its present guise had been developed by the
Caddo and not by speakers of the languages of Europe, the reality distinction
would be regarded as having a central importance to language and thought.
Bybee, Perkins, and Pagliuca specifically question the functional basis of the
distinction, they are troubled by its seemingly heterogeneous manifestations
in languages, and they find it unhelpful in understanding diachronic develop­
ments. Rather than responding immediately to each of these areas of doubt,
largely repeating points made in Givôn (1994), I will proceed to a description
of the manifestations of reality in Caddo, the Northern Iroquoian languages,
and English, returning in the end to the general issue of whether and how
reality is relevant to an understanding of language and the mind.

1. Caddo

The Caddo people once lived in a large number of towns that were spread over
what is now eastern Texas, northern Louisiana, and southern Arkansas. Euro­
pean encroachment into that area, which began with the De Soto expedition in
1541, was devastating, and after a series of tragedies and removals, the few
remaining Caddo settled in southwestern Oklahoma in the nineteenth century.
In spite of four and a half centuries of subservience to other languages, the
Caddo language is still spoken, although only by a rapidly diminishing
number of elderly people. I would hope that one of the lessons of the
following discussion would be the urgent need to learn as much as we can
from such rapidly dying languages during the few years in which they will
still be with us.
Caddo is a prototypically polysynthetic language, and one of the promi­
nent features of its verbs is the inclusion of a prefix that refers to one or more
of the participants in an event or state. Such a "pronominal prefix" distin­
guishes person, case, and — of most interest here — reality. (Number is marked
 352 Wallace Chafe

elsewhere in the verb.) Persons include 1st, 2nd, 3rd, and, in addition, what I
have called a Defocusing person which functions rather like German man or
French on (Chafe 1992). Case includes Agent, Patient, and Beneficiary.
Reality includes Realis and Irrealis. The reality distinction is an obligatory,
clearly marked, and unambiguous feature of every pronominal prefix (with
one minor exception), and thus of every verb. Realis and Irrealis can be
thought of as contrasting semantic components that combine with the four
persons and the three cases in a three-dimensional array that establishes the
semantics of the pronominal prefix system.
The Realis and Irrealis prefixes have the following underlying forms:
Realis Prefixes:
Agent Patient Beneficiary
1st person ci- ku- ku-
2nd person yah?- si- si-
3rd person — — nu-
Defocusing yi- ya- yu-
Irrealis Prefixes:
Agent Patient Beneficiary
1st person t'a-lt'i- ba- ba-
2nd person sah?- sa?a- sa?u-
3rd person sa- sa- ?u-
? ?
Defocusing ?a- ?a?a- a u-
Several points are worth noting. It can be seen that 3rd person partici­
pants are not always marked. Specifically, they are missing for Realis Agents
and Patients, which are actually the most frequently occurring of all 3rd
person referents. They are marked, however, when they function as Realis
Beneficiaries or when they are Irrealis, regardless of case. Thus, even though
there is not always an overt 3rd person marker, it is always clear whether a 3rd
person verb is Realis or Irrealis. It can also be seen that the Realis and Irrealis
forms have virtually nothing in common. Realis ci- contrasts with Irrealis fa-
ox t'i-, Realis ku- with Irrealis ba-, and so on. There is thus no reason to think
that the Irrealis forms had their origins in modifications of the Realis forms.
The historical origins of the two sets appear to be distinct.
In addition to the prefixes listed above, there are also various combina­
tions of 1st and 2nd person Agents with 1st and 2nd person Patients and
Beneficiaries:
The Realis-Irrealis Distinction 353

Realis Prefixes:
1st person Agent with 2nd person Patient t'a-
lst person Agent with 2nd person Beneficiary t'u-
2nd person Agent with 1st person Patient yahku-
2nd person Agent with 1st person Beneficiary yahku-
Irrealis Prefixes:
1st person Agent with 2nd person Patient fa­
lsi person Agent with 2nd person Beneficiary t'a?u-
2nd person Agent with 1st person Patient sahku-
2nd person Agent with 1st person Beneficiary sahku-
Here it can be seen that the combination of a 1st person Agent with a 2nd
person Patient is the only environment in which the Realis-Irrealis distinction
is neutralized. The fact that this single form, t'a-, is also homophonous with
one form of the Irrealis 1st person Agent prefix (see above) appears to be only
a coincidence.
While person and case are properties of the referent expressed by the
pronominal prefix, the domain of reality is not the referent, but rather the state
or event expressed by the entire clause (a clause being typically coextensive
with a verb). One may wonder why the expression of a clause-level meaning
should ride along on a pronominal prefix, but we may never know the
circumstances under which the language developed this way of doing things.
There are a variety of contexts that condition the use of the Irrealis
prefixes. All but one of them involves the simultaneous presence of a
so-called prepronominal prefix: a verb-initial prefix whose function is com­
patible with irreality. The fact that these prepronominal prefixes participate
less than fully in the complex phonology of the verb suggests that some if not
all of them were, at an earlier stage of the language, separate particles, and that
they have become attached to verbs more recently.

1.1 Yes-No Questions.

The one environment in which an Irrealis prefix is used without an accompa­

nying prepronominal prefix is the yes-no question. In Caddo, then, a yes-no
question can be regarded as the unmarked use of the Irrealis. Because the
boundaries of Caddo morphemes have been obscured through numerous
sound changes, it is useful to cite examples in an interlinear format whose
 354 Wallace Chafe

second line provides reconstructed forms in which morpheme boundaries are

more clearly evident.2 In the third line each morpheme is glossed, and the
fourth line provides a translation of the entire word.
(3) sàyybáwnah
sah?-yi=bahw-nah
2ND.AGENT.IRREALIS-See-PERFECT
'have you seen him?'
Unlike yes-no questions, question-word questions are expressed with
Realis prefixes. For example:
(4) dikadàyybâwnah
dikat-yah?-yi=bahw-nah
WHAT?-2ND.AGENT.REALIS-see-PERFECT
'what have you seen?'
It is easy to interpret the difference between (3) and (4) as motivated by the
fact that a yes-no question implies a lack of knowledge as to whether the event
actually occurred (you may not have seen him), whereas a question-word
question presupposes the event and asks only about the identity of a partici­
pant. Thus, (4) presupposes that you have seen something, and the speaker
wants only to know what it was.
The fact that all the remaining uses of the Irrealis pronominal prefixes
involve the attachment of a prepronominal prefix to a form which, by itself,
expresses a yes-no question suggests a way in which the marking of irreality
may have entered the language. It may well be that the Irrealis forms of the
pronominal prefixes arose first as a way of marking questions, to which the
prefixes described below, perhaps in their earlier incarnation as particles,
came to be attached. If this was indeed the path of development, it would have
amounted to a tacit recognition by speakers that all these uses share a common
property of irreality.

1.2 Negations.

An obvious site for irreality is negation (but see Mithun in press). Simple
negations in Caddo are expressed with the Negative prepronominal prefix
kúy- 'not':
The Realis-Irrealis Distinction 355

(5) kúyt' áybah

kúy-t'a-yi=bahw
NEGATIVE-1ST.AGENT.IRREALIS-see
'I don't see him'
Simple negations may be modified in various ways. For example, the Nega­
tive prefix may be immediately followed by -c'í- to indicate an event that is
not real at the present time, but might be expected to become real in the future.
In other words, the combination kúy-c'í- means 'not yet':

(6) kúyc' ít' áybah

kúy-c'í-t'a-yi=bahw
NEGATIVE-YET-1ST. AGENT.IRREALIS-see
'I haven't seen it yet'
Or the Negative prefix may be followed by -?ní- to express the idea of an event
that happened in the past, but is not happening now. In other words, the
combination kúy-?ní- means 'not any more'. It is followed by the alternative
1st person Irrealis Agent form t'i- rather than the fa-:
(7) kúy? nít' ííbah
kúy-?ní-t'i-yi=bahw
NEGATIVE-ANY.MORE-1ST. AGENT.IRREALIS-see
'I don't see it any more'
An additional possibility is to supplement the combination just described with
the morpheme -káy- 'anything':
(8) kúy?níkáyt 'áybah
kúy-?ní-káy-t 'a-yi=bahw
NEGATIVE-ANY.MORE-ANYTHING-1ST. AGENT.IRREALIS-see
'I don't see anything any more'
An alternative to the simple Negative prefix kúy- is the Temporal Negative
prefix nus- 'not for a long time'. It too is followed by the 1st person Irrealis
Agent form t'i-:
(9) nust' íibah
nus-t'i-yi=bahw
TEMPORAL.NEGATIVE-1ST. AGENT.IRREALIS-see

'I haven't seen it for a long time'

All of these negative options condition an Irrealis pronominal prefix.
356 Wallace Chafe

1.3 Prohibitions.

Prohibitions, or negative imperatives, also condition an Irrealis pronominal

prefix. They are expressed with the prepronominal prefix kas-:
(10) kassáy? bah
kaš-sah-yi-bahw
PROHIBITIVE-2ND.AGENT.IRREALIS-See
'don't look at it'

1.4 Obligations.

The expression of an obligation with the prepronominal prefix kas- 'should,

be supposed to' also requires the Irrealis form:
(11) kassánáy?aw
kas-sa-náy=?aw
OBLIGATIVE-3RD.AGENT.IRREALIS-SÍng
'he should/is supposed to sing'

1.5 Conditions.

Conditions are also Irrealis. There are two prepronominal prefixes that ex­
press conditionality. One of them is the Particular Conditional prefix hí- 'if',
indicating the possibility of a particular event:
(12) hít' áybah
hí-t' a-yi=bahw
CONDITIONAL-1ST. AGENT. IRREALIS-See
'if I see it'
The other is the Generic Conditional prefix nas- 'if, whenever', used with
generic events:
(13) nast' áybah
nas-t' a-yi=bahw
GENERIC.CONDITIONAL-1ST. AGENT.IRREALIS-see
'if, whenever I see it'
The Realis-Irrealis Distinction 357

Such events are often predicted for the future:

(14) nast 'áybáw?a?
nas-t' a-yi—bahw-?a?
GENERIC CONDITIONAL-1ST.AGENT. IRREALIS- see-FUTURE

'if, whenever I see it (future)'

Negative Future conditions are expressed with nadús- 'if not':
(15) nadúst 'áybáw?a?
nadús-t' a-yi=bahw-?a?
NEGATIVE.CONDITIONAL-lST.AGENT.IRREALIS-see-FUTURE
'if I don't see it (future)'

1.6 Other Irrealis Options.

There are several other prepronominal prefixes that condition Irrealis pro­
nominal prefixes. One of them is the Simulative prefix dúy- 'as if:
(16) dúyt'áybah
dúy-t'a-yi=bahw
SIMULATIVE-1ST. AGENT.IRREALIS-See

'as if I saw it'

Another is the Infrequentative prefix wás- 'seldom':
(17) wást'áybah
wás-t'a-yi-bahw
INFREQUENTATIVE-1ST. AGENT. IRREALIS-see

'I seldom see it'

Of special interest is the Admirative prefix hús-, expressing surprise:
(18) húsbaasáy?k 'awihsa?
hús-ba-?a=sa-yi=k'awih-sa?
ADMIRATIVE-IST.BENEFICIARY.IRREALIS-name-know-PROGRESSIVE
'my goodness he knows my name!'
This last usage is perhaps itself surprising. It would seem that, although the
event or state itself is real enough (in this example he does know my name),
the fact that it is contrary to the speaker's expectation — a negation of
 358 Wallace Chafe

normality — is what is responsible for the use of the Irrealis. It is as if the

speaker were saying, "It's unreal that he knows my name!" (an interpretation
suggested to me by Alan Taylor).
Although the types of expressions listed above — yes-no questions,
negations, obligations, conditions, simulatives, infrequentatives, and admira-
tives — can be seen as functionally motivated members of the Irrealis cat­
egory, the occurrence of Realis pronominal prefixes with imperatives might
be seen as an inconsistency:
(19) dáy?bah
yah?-yi-bakw
IST. AGENT.REALIS-see
'look at it!'
More problematic, perhaps, is the use of Realis prefixes with futures, espe­
cially in view of the fact that future is so fundamental to the expression of
irreality in Papuan languages (Roberts ms.):
(20) cííbáw?a?
ci-yi=bahw-?a?
1ST. AGENT.REALIS-see-FUTURE
'I'll look at it'
(21) cííbáwčah
ci-yi-bahw-čah
IST.AGENT.REALIS-See-FUTURE.INTENTION
'I'm going to look at it'
Both imperatives and futures convey ideas of events that have not yet hap­
pened, having entered the speaker's consciousness through imagination.
Why, then, should Caddo treat them as realis? Two quite different answers
can be suggested. One is that reality is not a binary but a gradient dimension in
which imperatives and futures express ideas that are judged to be relatively
more in accord with reality than, say, yes-no questions or negations. Speakers
may have a relatively stronger expectation that commands will be obeyed or
that predicted events will take place. When we turn to the Northern Iroquoian
languages we will find, in fact, that futures (and their use as imperatives) have
been grammaticized with a status intermediate between the extremes of realis
and irrealis.
The Realis-Irrealis Distinction 359

But the exemption of Caddo imperatives and futures from the irrealis
category could have a different explanation. It is likely that ways of express­
ing imperatives and futures were established in the language well before the
negative and other particles came to be attached to what were originally
interrogative verbs, as suggested above. Imperatives and futures, that is, may
stem from a more ancient layer of Caddo morphology. Already entrenched in
the language, they would then have failed to participate in the more recent
grammaticization of irreality in the pronominal prefixes. It is difficult to
decide which of these explanations is correct (perhaps both are), but either is
plausible, and either serves to demonstrate how irreality (like other semantic
functions) may come through time to be represented in ways that may appear
synchronically to be inconsistent.

2. The Northern Iroquoian languages

Spoken in New York, Ontario, and Quebec, the languages that constitute the
northern branch of the Iroquoian family include Mohawk, Oneida, Onondaga,
Cayuga, Seneca, and Tuscarora, as well as other languages now extinct. All of
these languages behave in a similar fashion with respect to the matters to be
discussed here, and I will discuss them as a single entity.
When Northern Iroquoian verbs are inflected for the Punctual (perfec­
tive) aspect, they obligatorily contain one of a set of three prefixes that
segment a continuum of reality. The occurrence of these prefixes with other
aspects — the Stative (perfect) and Habitual (both imperfective and generic)
— is optional, and is marked with a special added suffix. At one end of this
reality continuum is the Factual prefix (reconstructable in most environments
as *wa?-), expressing the speaker's belief in the reality of the event or state. At
the other end is the Optative prefix (reconstructable in most environments as
*aa-), expressing either obligation or possibility. In some contexts the Opta­
tive is best translated 'should', in others 'might'. Intermediate between these
extremes lies the Future prefix (reconstructable as *ę-), used for a state or
event the speaker expects to materialize but is not certain of. Verbs with the
Future prefix are also often used as imperatives, a fact which suggests that the
exemption of both imperatives and futures from irrealis marking in Caddo, as
discussed above, may indeed have a functional basis. In short, whereas the
 360 Wallace Chafe

Caddo system divides this dimension simply into Realis and Irrealis, the
Northern Iroquoian system shows a three-way division. The examples below
are taken from the Seneca language:
(22) o?geegę?
wa?-ke-kę-?
FACTUAL-1ST. AGENT-See-PUNCTUAL
'I see/saw it' (direct perception or memory)
(23) ęgeegę?
ę-ke-kę-?
FUTURE-1ST. AGENT-See-PUNCTUAL
'I'll see it' (prediction)
(24) aageegę?
aa-ke-kę-?
OPTATIVE-1ST. AGENT-See-PUNCTUAL
'I should/might see it' (obligation or possibility)
This Northern Iroquoian continuum interacts with yes-no questions,
negations, and other irrealis contexts in complex ways. I will limit the discus­
sion here to questions and negations. Yes-no questions are expressed with
both an intonation contour and a phrase-final particle in most of the lan­
guages, but in Seneca with intonation alone. Such questions are expressed in
ways that are independent of the reality dimension. That is, they may be
marked as Factual, Future, or Optative, or (with aspects other than the Punc­
tual) they may have no reality marking at all.
The difference between Caddo, where all yes-no questions are Irrealis,
and Northern Iroquoian, where such questions may show any degree of reality
or none at all, also invites two possible explanations. One might suppose that
interrogation lies outside the scope of reality, so that, for example:

(25) ęhseegę?
ę-hse-kę-?
FUTURE-2ND.AGENT-see-PUNCTUAL
'will you see it?'
with a yes-no question intonation has a meaning that could be paraphrased "Is
it or is it not the case that you will see it?" whereas the corresponding Caddo
word:
The Realis-Irrealis Distinction 361

(26) sàyybáw?a?
sah?-yi=bakw-?a?
2ND.AGENT.IRREALIS-see-FUTURE
'will you see it?'
might be paraphrased "Will you see it or won't you see it?"
An alternative possibility, not inconsistent with the first, is that Northern
Iroquoian questions are actually requests for confirmation. Thus, (25) might
correspond more closely to English "You'll see it?" with a question intona­
tion, a translation that mirrors the Seneca surface form (regarding confirma­
tive questions in English see Chafe 1970a:333-337 and, in Onondaga, Chafe
1970b:24). A confirmative question allows any degree of reality in the propo­
sition for which confirmation is requested. This explanation can be combined
with that in terms of scope, in the sense that confirmation lies outside the
degree of reality. For present purposes it is enough to point out that the
different treatment of yes-no questions by Caddo and the Northern Iroquoian
languages can have a functional motivation.
Negation presents a different picture. A detailed discussion of negative
forms and their cooccurrences in the Northern Iroquoian languages would
take us too far afield, but we can note that the Negative prefix in these
languages (allowing for a few special exceptions) does not cooccur with any
of the reality prefixes. Negative past meanings are expressed with a Stative
(perfect) form of the verb, which does not require a reality prefix:
(27) da?ageegęęh
te?-wake-kę-ęh
NEGATIVE-1ST.PATIENT-see-STATIVE
'I haven't seen it'
Negative Futures are expressed with the so-called contrastive prefix (which
involves an element of surprise), together with the Optative prefix and a verb
that ends in the same way as an imperative (Chafe 1967: 32):
(28) thaageegęh
th-aa-ke-kç-h
CONTRASTIVE-OPTATIVE-1ST. AGENT-see-IMPERATIVE
'I won't see it'
362 Wallace Chafe

This unusual morphologized construction combines the three irrealis features

of negation, surprise, and the imperative to express idiomatically the Negative
Future meaning.
It was noted above that the reality prefixes are obligatory with the
Punctual aspect. They may occur with other aspects, but when they do, the
special nature of their occurrence is marked with an additional suffix. In
effect, reality is an obligatory category with verbs inflected as perfective, but
is optional and marked with verbs inflected as perfect, imperfective, and
generic. These languages thus suggest a special attachment between degrees
of reality and events that are regarded as particular. It would be interesting to
know whether this affinity appears in other languages.

3. English

In English the realis-irrealis dimension fails to receive the unmistakable overt

marking we find in Caddo and, distributed differently, in the Northern
Iroquoian languages. It nevertheless shows itself in a more subtle but no less
valid manner in its determination of the distinction between specific and
nonspecific referents (e.g. Chafe 1970a: 199; Jackendoff 1972, Chapter 7;
Givôn 1989, Chapter 5). This distinction is recognizable in the context of
certain verbs such as look for:
(29) I looked for a book.
where there is ambiguity as to whether or not the speaker has a specific book
in mind, as opposed to:
(30) / bought a book.
where only the specific meaning is at home. Of interest is the fact that the
nonspecific meaning is correlated with contexts that are typically irrealis:
(31) yes-no questions Did you buy a book?
negations I didn't buy a book.
futures I'm going to buy a book.
obligations I need to buy a book.
possibilities I might buy a book.
imperatives Buy a book.
prohibitions Don't buy a book.
conditions If I buy a book...
The Realis-Irrealis Distinction 363

Imagine, for example, that you and I are walking past an airport bookstore and
I say, "I need to buy a book." With the specific meaning, I might continue "It's
one I read a review of yesterday". With the nonspecific meaning, I might go
on to say "So I'll have something to read on the plane." It appears that in
irrealis contexts English allows and even encourages the interpretation of a
direct object referent as nonspecific. The contexts in (31) are just those that
are overtly marked as Irrealis in Caddo, except for the fact that they include
Future, thus confirming the natural inclusion of future in the Irrealis category.
It is possible that this relation between irreality and nonspecificity of refer­
ence is widespread if not universal, thus confirming the widespread or univer­
sal relevance of the irrealis category. But that remains to be seen.

4. Conclusion

The realis-irrealis distinction reflects judgments that certain ideas stem from
direct perception, memory, or expectations of what is normal, while others
have their source in imagining. This distinction can be thought of as a covert
semantic pressure that emerges in different languages in different ways. It
emerges in Caddo in a clear and ubiquitous formal distinction, every verb
being marked for either reality or irreality. The Caddo assignment of specific
functions to these two categories is motivated and coherent, with yes-no
questions, negations of various sorts, prohibitions, obligations, conditions,
simulations, and surprises in the irrealis category. Neither futures nor impera­
tives are marked as Irrealis, however, either because Caddo draws the line at
this point in the realis-irrealis continuum, and/or because ways of expressing
futures and imperatives were morphologized in the language prior to the
marking of irreality.
The distinction emerges differently in the Northern Iroquoian languages,
which show a three-way rather than a binary contrast with perceived and
remembered reality at one extreme, obligation and possibility at the other, and
between them the future (with an imperative use as well). Unlike Caddo, the
Northern Iroquoian languages do not treat yes-no questions as irrealis, per­
haps because the scope of questioning is external to reality, perhaps also
because Iroquoian questions are more confirmative in nature than the more
clearly disjunctive questions of Caddo. The Negative and Factual prefixes are,
as might be expected, incompatible in the Northern Iroquoian languages, and
negative futurity is expressed with a grammaticized combination of surprise
and possibility, along with an imperative ending. The three-way reality dis-
364 Wallace Chafe

tinction is obligatory with perfective verbs, but it is optional and marked with
perfects, imperfectives, and generics.
Reality fails to be overtly marked in English, but its force is nevertheless
felt through the irrealis creation of an environment in which referents may be
interpreted as nonspecific. The contexts that produce this effect line up well
with the Caddo Irrealis contexts, except for the inclusion of future.
I hope to have shown (1) that the realis-irrealis dimension has a consist­
ent functional basis in people's judgments concerning the degree to which
their ideas accord with what they believe to be objective reality; (2) that the
apparent inconsistency with which such judgments are manifested in different
languages can be explained sometimes in terms of a gradient rather than a
binary dimension of reality, sometimes in terms of layers of morphologiza-
tion, and sometimes in terms of scope; and (3) that recognition of the pres­
sures exerted by the reality dimension is ultimately essential to an
understanding of certain aspects of language change.
We are left with the question of the ontological status of the reality
dimension as it relates to the several moods and modalities. The extent to
which ideas are judged to accord with objective reality leads to a categoriza­
tion of specific moods and modalities, in combination with specific tenses and
aspects, into two or three overarching categories. The categories themselves
may surface as realis and irrealis semantic components of morphemes as in
Caddo, as complete morphemes as in the Northern Iroquoian languages, or as
indirect influences on the specific-nonspecific referential distinction as in
English. Realis and irrealis can thus be seen as comparable to broader catego­
ries such as tense or aspect, or for that matter mood and modality, within each
of which several values are available. They are most closely related to moods
and modalities in the sense that they reflect a speaker's attitude toward the
idea being expressed, but their effect is to combine the more specific moods
and modalities into larger groupings.

FOOTNOTES

1 Curiously, a search of Palmer's book does not reveal any such recommendation.
2 In the first, and sometimes the second lines the acute accent mark indicates a high
pitched vowel, the grave accent a pitch that falls from high to low. Long consonants and
vowels are shown with doubled letters. Caddo stems usually consist of a lexicalized
combination of two or more historically distinct morphemes, which are joined in the
The Realis-Irrealis Distinction 365

second line with an equals sign. The separate meaning of each of these stem-internal
morphemes is often obscure, and here I have simply given an English gloss for the stem
as a whole.

REFERENCES

Bybee, Joan, Revere Perkins & William Pagliuca. 1994. The Evolution of Grammar:
Tense, Aspect, and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Chafe, Wallace. 1967. Seneca Morphology and Dictionary. (Smithsonian Contributions
to Anthropology, Volume 4.) Washington, DC: Smithsonian Press.
Chafe, Wallace. 1970a. Meaning and the Structure of Language. Chicago: University of
Chicago Press.
Chafe, Wallace. 1970b. A Semantically Based Sketch of Onondaga. Indiana University
Publications in Anthropology and Linguistics, Memoir 25.
Chafe, Wallace. 1992. "Uses of the Defocusing Pronominal Prefixes in Caddo." Anthro­
pological Linguistics 32:57-68.
Chafe, Wallace. 1994. Discourse, Consciousness, and Time: The Flow and Displacement
of Conscious Experience in Speaking and Writing. Chicago: University of Chicago
Press.
Givón, Talmy. 1989. Mind, Code and Context: Essays in Pragmatics. Hillsdale, NJ:
Lawrence Erlbaum.
Givón, Talmy. 1994. "Irrealis and the Subjunctive." Studies in Language 18:265-337.
Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Cambridge,
MA: The MIT Press.
Mithun, Marianne. "On the Relativity of Irreality." This volume.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Roberts, John R. Ms. The Category 'Irrealis' in Papuan Medial Verbs.
 On the Relativity of Irreality

Marianne Mithun
University of California, Santa Barbara

0. Introduction

A modality distinction reported in languages in a number of parts of the world

is that between Irrealis and Realis. The terminology has been applied to
constructions in languages of Australia (Comrie 1985; Palmer 1986),
Austronesia (Dempwolff 1939, Chung & Timberlake 1985), Papua New
Guinea (Foley 1986; Roberts 1990), Africa (Givón 1994), Europe (Chung &
Timberlake 1985; Givón 1994), South America (Wise 1986, Payne & Payne
1990), Mesoamerica (Craig 1977; Givón 1994), North America (Buckley
1988; Miller 1990) and to various creole languages (Romaine this volume) to
mention only a few. It has been used widely in both grammars of individual
languages and general discussions of modality.
As pointed out by Bybee, Perkins & Pagliuca (1994), however, the
grammatical categories referred to by the terms are far from uniform cross-
linguistically. Construction types marked as Irrealis in one language may be
marked as Realis in the next. In some languages Imperatives are classified as
Irrealis, in others as Realis; futures, questions, and negatives also show some
variation. The formal expression of the distinction varies cross-linguistically
as well. In some languages, only the Irrealis category is expressed overtly, in
others both Irrealis and Realis are expressed, and in still others one or the
other is expressed by multiple markers. The distinction may be indicated in
various areas of the grammar: by particles, by clausal clitics, by verbal
inflection, and perhaps even by verbal derivation (Eatough ms on Nisenan). In
many languages it is carried as a feature of markers that also express other
functions, such as tense or aspect, clause linking, or pronouns, as in Caddo
368 Marianne Mithun

(Chafe this volume) and Amele (Roberts 1990). Irrealis/Realis distinctions

may even appear in several different parts of the grammar of a single lan­
guage. Buckley (1988) describes pairs of Completive and Inchoative suffixes
in Alsea; one member of each pair is Irrealis, the other Realis. In addition, an
Irrealis/Realis distinction is carried in this language by complementizers: sís
introduces Irrealis clauses, mís Realis clauses. Jamul Diegueño contains both
an inflectional Irrealis suffix -x and an Irrealis/Realis distinction carried on
the switch reference suffixes marking dependent clauses (Miller 1990).
Given this variation, the utility of the labels 'Irrealis' and 'Realis' for
cross-linguistic comparison is open to question. If there is no common seman­
tic or pragmatic basis for the terminology, its application to such diverse
phenomena could lead to unnecessary confusion. In what follows, the Irrealis
and Realis categories of several languages will be compared. It will be
suggested that despite the apparent heterogeneity of the categories to which
the distinction is applied, indeed perhaps because of it, the Irrealis/Realis
terminology can prompt fruitful cross-linguistic comparisons with potential
for furthering our understanding of certain processes of grammaticization.

1. Central Porno

Languages of the Pomoan family, spoken in northern California, contain

several pairs of verb-final markers that link clauses. The markers distinguish
Irrealis and Realis moods. Their forms are listed in (1).
(1) Central Porno clause linkers
SAME DIFFERENT
1. Irrealis 'while', 'and', 'then' -hi -hla
2. Realis
- Simultaneous 'and', 'while', 'as' -in -da
- Sequential 'and', 'then' -ha -li
The Irrealis markers in the top row (-hi, =hid) are used to link clauses
expressing actions or states that the speaker portrays as purely within the
realm of thought. Such constructions convey ideas known through imagina­
tion rather than direct perception (Chafe, this volume). The Realis markers in
the bottom two rows (-in, -da, -ha, =li) link clauses expressing events
portrayed as actualized, having occurred or actually occurring. The choice of
markers within each of these categories reflects further distinctions: whether
On the Relativity of Irreality 369

the linked actions or states are presented as components of a single event

(SAME) or as distinct events (DIFFERENT), and, if they are Realis, whether the
linked activities or states are simultaneous (overlapping in time) or consecu­
tive (occurring in sequence).
The uses of each marker are described below. Except where otherwise
specified, examples cited here are drawn from spontaneous speech, generally
conversation or narrative embedded within conversation. Speakers quoted
here are Mrs. Frances Jack and Mrs. Kate Daniels of the Hopland ranchería,
Mrs. Salome Alcantra and Mrs. Florence Paoli of the Yokaya ranchería, and
Mrs. Eileen Oropeza of the Point Arena ranchería. All material was tran­
scribed and translated with the help of Mrs. Jack.*

1.1 The Central Porno Irrealis Clause Linkers

The markers identifying Irrealis dependent clauses are -hi and =hla. The first,
-hi, links clauses expressing what are portrayed as components of the same
event.
(2) Irrealis same event linker: FJ
Mé-n=?ti mí= lálil čá-n-ma-hi kay
SO=but that=at b a c k run-IMPFV-COOPERATION-SAME.IRREALIS tOO
'But we could drive back there and
ya mu-l da ?=mí= =li hlá-qač=?le.
1.AGENT that road cop=that=at=with go.PL-up-=coNDITIONAL
go up that road.'
The second Irrealis marker, =hla, links clauses expressing what are portrayed
as different events.
(3) Irrealis different event linker: FP
Ma čalél qów=?-ne=hla
2.AGENT just out=by.gravity-set=DIFF.IRREALIS
'If you just threw it out
't- mu-l bá-č-i?le.
feel that big-INCHOATIVE.PFV=CONDmONAL
I guess it would grow.'
The clauses bearing the linkers are grammatically dependent, since they
cannot stand alone as independent sentences, but they are not necessarily
subordinate semantically. In (2), for example, 'We could go back up there and
370 Marianne Mithun

drive up that road', the Irrealis marker links two semantically and pragmati­
cally equivalent constituents of a larger sentence expressing root possibility.
(The free translations were provided by Mrs. Jack, who was present at all
conversations and aware of both the linguistic and extra-linguistic context.)
Irrealis linkers appear in a wide variety of constructions that express events
portrayed as non-actualized.
(4) Irrealis linker in counterfactual construction: KD
Me-n mí-hla mu.l
so say=DIFF.IRREALIS that
'If she'd said that

?a- ?čhá-=ne ?
a.
1 .AGENT Sit=CONDITIONAL 1 .AGENT
I'd stay longer.' [But she said ...]
(5) Irrealis linker in conditional construction: FJ
Wá-q-hi ?e
gO-level-SAME.IRREALIS COP
'If I go
ló'-h-du-w-?khe.
help-IMPFV-IMPFV-PFV=FUTURE
I'll be helping out.'
(6) Irrealis linker in deontic construction: FJ
?
čá-w-htow é y=yo-hi
house-LOC=from away=go-SAME.iRREALis
'He should leave home and
táwhal da-čé = ?le.
work handling-catch=CONDITIONAL
get a job.'
(7) Irrealis linker in future construction: FJ
Té-nta=lil wá-n-hi
town=to go-IMPFV-SAME.IRREALIS
'I'll bring it back
?
a qó=be-w=?khe.
1.AGENT toward=carry-PFV=FUTURE
on my way to town.' ('I'll go to town and bring it back.')
On the Relativity of Irreality 371

(8) Irrealis linker in hortative construction: FJ

Ya-ka háy š-dí-č-ma-hi
1.PL.AGT=INF wood drag-carry-INCH-co-SAME.IRREALIs
'Let's get some wood and
qhá=kay s-dí-c-ma-w=?khe.
water=too dragging-carry-INCHOATIVE-co-PFV=FUTURE
haul in some water.'
(9) Irrealis linker in imperative construction: EO
Qhá č h ní= ? el dó-č-hi,
water bread=the make-SEMELFACTIVE-SAME.IRREALIs
'Make the water bread and
mú'tu-ya-l qa-wá-c-ka-m.
3-PL-PATIENT biting-gO-IMPFV.PL-CAUSATIVE-IMPERATIVE
invite them to eat it.'

1.2. Central Porno Realis clause linkers

There are two pair of Realis dependent markers. They link past or present
events. (There are no past or present tense markers.) One pair, -in and -da
'while', link simultaneous actions or states. The actions or states need not be
simultaneous for their full duration, but simply overlapping at some point.
The first marker, -(i)n, links what are portrayed as components of the same
event.
(10) Simultaneous Realis, same event: EO
?a- Edna-to cá=l yó-h-du-n.
1 .AGENT =PATIENT house=to go-PFV-IMPFV-SAME.SIM-REALIS
T go to Edna's house and
híntil=?el ča-nó-'d-an-ya mú-tu.
Indian=the talk-IMPFV-IMPFV=EVID 3.PATIENT
talk Indian to her.'
The other, -da, links what are portrayed as distinct events.
(11) Simultaneous Realis, different events: SA
?
Mu-l oč qášóy-da
that still alive=DIFF.SIM.REALIS
'While he was still alive,
372 Marianne Mithun

me-n=da ?=mu-l to- qha?á-n ?e mu-l

such=at cop=that 1.PATIENT dream-IMPFV COP that
I dreamed that
?
to- -tá-w he šaní-c.
1.PATIENT with.fingers-sense-PFV and/or talk-SEMELFAcrrvE.PFV
he touched me and said something.'
The second Realis pair, -ba/=li 'and then', link clauses expressing consecu­
tive events.
(12) Sequential Realis, same event: SA
Mu-l só-č-ba qamát iá-?-du-w.
that hear-SML-SAME.SEQ.REALIS angry sense-RFL-IMPFV-PFV
'I heard that and got mad.' or 'When I heard that, I got mad.'
(13) Sequential Realis, different events: FP
?
to- met -né-ya-w=li
1 .PATIENT SUCh gravity-set-DEFOCUS-PFV=DIFF.SEQ.REALIS
'I was nominated and
míya- mé dá-?-du-w čhó-w.
3.POSS father like-RFL-iMPFV-PFV not-PFV
his father didn't like it.'
(or 'When I was nominated, his father didn't like it.')
Clauses expressing habitual and past habitual events are also linked with
Realis markers.
(14) Central Porno past habitual: FP
Behé-m...
pepperwood
'Pepperwood,
mét=li mu-l qhá- qóm-ad-an
that.kind=with that water-LOC bathe-IMPFV-IMPFV
'he would bathe with that
qasít=da šé-ma?wi.
be.cold=DIFF.SIM.REALIS early.morning
early in the morning when it was cold.'
On the Relativity of Irreality 373

Multiple markers can appear within a sentence. In (15), 'grab' and 'pull
out' are joined by -ba, both of which are in turn linked by -da to the verb 'cut'.
(15) Central Porno linked dependent clauses: SA
thaná=?el da-cé-ba
hand=the handling-seize-SAME.SEQ.REALis
'When they grabbed his hand and
s-thí-c-na-w=da,
drawing-Open-SEMELFACTIVE-AWAY-PFV=DIFF.SEQ.REALIS
pulled it (a knife) out,
Pun mu-l ča-qhá-č-ka-m=ma.
self that Sawing-CUt-RFL-CAUS-PFV=FACTUAL
he cut himself.'

1.3 The classification of questions

Questions show the same Irrealis/realis classification as their declarative

counterparts. The future question in (16) contains an Irrealis linker.
(16) Irrealis question: FP
?í=wa ma sa-có-t=?khe
be=Q 2.AGENT SWinging-whip-MULTIPLE.EVENT=FUTURE
'Are you gonna whip us
ya-l ča=-l dé-m-ma-hi?
1 .PL-PATIENT h0USe=to lead.PL-M.E.-CO-SAME.IRREALIS
when you take us home?'
The question about the past in (17) contains a Realis marker.
(17) Realis question: FJ
Thaná dasé-č-ba-wa
hand pulling-wash-RFL-SAME.SEQ.REALIs=Q
'Did you wash your hands and
?
ma é- čh?ól-či-w?
2.AGENT hair comb-RFL-PFV
comb your hair?'
374 Marianne Mithun

1.4 The classification of negatives

Similarly, negatives show the same Irrealis/Realis categorization as their

positive counterparts. The negative counterfactual construction in (18) con­
tains an Irrealis linker, as would its positive counterpart.
(18) Irrealis negative counterfactual: FJ
Mu-l ?a- mwl yhé-n č h ó-w=hla
that 1.AGENT that do-IMPFV not-PFV=DIFF.IRREALIS
'If I hadn't done that
sí-n phta- ya stó, bel mací, naphó=?le.
how wonder I.PL now this day sit.PL=coNDITIONAL
I wonder where we'd be today.'
The negative conditional construction in (19) contains an Irrealis linker, as
would its positive counterpart.
(19) Irrealis negative conditional: SA
Ma me-n ?í-w č h ó-w-hla,
2.AGENT SUCh do-PFV not-PFV=DIFF.IRREALIS
'If you don't do that,
ma bé-da ma- basét thabá-?či-w ph-wí-w=?k e.
2.AGT this=at things bad lie-INCH-PFV vis-perceive-PFV-FUT
you're going to see some bad things happening around here.'
The negative imperative in (20) contains an Irrealis linker, just like
positive imperatives.
(20) Irrealis negative command: FJ
? h
Dá-wi č á' č-hi
road=on stop-s AME. IRREALIS
'Don't stop and
khyá swé'l-an=?khe thín ?e ma.
game play-IMPFV=FUTURE not-IMPV COP 2.AGENT
play on the way home.'
On the Relativity of Irreality 375

The negative statement about the past in (21) contains a Realis linker, just like
positive statements about past events.
(21) Realis negative past: FP
Ranch =?el qdí yhé-t-ač čhó-w ?
í-n
= t h e g o o d do-M.E-IMPFV.PL not-PFV be-SAME.SIM.REALIS
'Because they didn't keep up the ranchería,
ya-l qo=-l mčá-w dá- ?-či-w.
1 .PL.PATIENT out=to throw.PL-PFV WANT-RFL-IMPFV.PL-PFV
they wanted to throw us out.'
The Central Porno Irrealis/Realis distinction is strikingly similar to those
in a large number of other unrelated languages. In the Papuan language
Amele, for example, described in Roberts (1990), dependent clauses are also
identified by markers that carry an Irrealis/Realis distinction. Counterfactual,
conditional, obligatory, future, hortatory, and imperative constructions appear
with Irrealis markers, essentially the same construction types as in Central
Porno. Past, present, and habitual actions are linked with Realis markers. The
Realis markers further distinguish simultaneous and sequential events. Ques­
tions and negatives show the same Irrealis/Realis classification as would their
positive declarative counterparts. It is not difficult to discern a motivation
behind the choice of terms 'Irrealis' and 'Realis' for this distinction. State­
ments categorized as Irrealis portray events as still within the realm of thought
alone, while those categorized as Realis are portrayed as actualized, actually
occurring or having occurred.
Categories identified as Trrealis' and 'Realis' are not uniform across all
languages, however. While many languages show patterns strikingly similar
to those in Central Porno and Amele, some show minor differences, and a few
show substantial ones. The variation suggests that either the distinction under­
lying the categories is not in fact comparable across languages, or that the
distinction is constant, but its application differs. If the basic distinction
varies, then the common terminology could be misleading. If the distinction is
the same, and differences are confined to its application, the variation might
be explicable in terms of the diachronic developments of grammatical catego­
ries in individual languages.
376 Marianne Mithun

2. Some cross-linguistic differences

Some kinds of constructions, such as counterfactuals and conditionals, are

generally classified as Irrealis in all languages with a grammaticized Irrealis/
Realis distinction. Several constructions show somewhat more variation
across languages than these, however, in particular imperatives, futures,
questions, and negatives. In what follows, suggestions will be made for kinds
of diachronic explanations that might underlie the variation.

2.1 Imperatives

In Central Porno and Amele, commands are classified as Irrealis, as in (9)

above. The classification seems motivated, in that commands can easily be
conceived of as expressing thoughts of actions rather than the realization or
actualization of them. In some languages, however, imperatives are classified
as Realis. In Maricopa, for example, a language of the Yuman family in
Arizona, reality is indicated by verbal suffixes. Gordon reports that the Irrealis
suffix -h(a) "is used typically in cases in which no part of the action or state
expressed by the verb is realized [...] It can be used to mark simple future,
though it is far more usual for it to be used to signal an exhortive or contrary-
to-fact sense [...] -h(a) is also used in Irrealis (future, possible, or obligation)
nominalizations" (1986:109).

(22) Maricopa Irrealis: (Gordon 1986:27, 109)

a. ny-aay-ha
1/2-give-IRREALIS
'I will give it to you.'
b. haat nyi-ttpooy-nt-ha
dog.PL PL.OBJ-kill-PL.ACTION-tOO-IRREALIS
'It might kill dogs too.'
?
c. aanylyviim m-vaa-kis nym-yuu-ha
yesterday 2-come-coND 2/1 -see-IRREALIS
'If you had come yesterday, you would have seen me.'
Realis suffixes indicate that 'the speaker is presenting the information as fact,
not as possibility, inference, or preference. A verb marked with [a Realis
suffix] expresses an action, event, or state which is taking place in the present
or has taken place in the past' (1986:24).
On the Relativity of Irreality 311

(23) Maricopa Realis: (Gordon 1986:24)

a. aham-m
hit-REALIS
'He hit him.'
b. hot-hot-m
good-good-REALIS
'It is very good.'
Imperatives appear with Realis suffixes.
(24) Maricopa imperative: (Gordon 1968:25)
k-tpuy-m
IMP-kill-REALIS
'Kill it!'
Several kinds of diachronic developments could lead to a categorization
of imperatives as Realis. One possible explanation could come from the order
in which an emerging Irrealis form might be applied to new contexts over
time. The distinction might first be made only in indicative clauses, then
spread only later into other contexts, among them the imperative. Systems in
which imperatives are not classified as Irrealis could represent the intermedi­
ate stage in such a development. A second possible explanation could come
from differing expectations of actuation. While counterfactual propositions
are expected not to occur or have occurred at all, imperatives could be seen to
imply reasonable expectations of compliance: in some societies, for example
the utterance of a command might be taken conventionally to entail immedi­
ate and certain performance of an action. A third possible explanation could
come from ways in which speakers might exploit the option of overtly
marking expectation. Speakers might intentionally mark commands as Realis
to imply strong certainty of their immediate actualization. In fact, many
languages contain two options: a polite imperative, classified as Irrealis, and a
strong imperative, classified as Realis.
Jamul Diegueño, a Yuman language of southern California, contains an
Irrealis suffix -x that is cognate with the Maricopa -h(a). Miller (1990:119)
reports that the suffix appears with counterfactuals, conditionals, potentials,
obligations, warnings, desideratives, and futures, as well as in adversative,
purpose, and hypothetical relative clauses and nominalizations. It marks
action that is 'desired, potential, or otherwise unrealized'. Realis construc­
tions show no -x. Jamul contains two kinds of Imperative constructions: a
378 Marianne Mithun

Polite Imperative, containing the basic second person Subject Prefix and an
Irrealis Suffix, and a Basic Imperative, containing a special Second Person
imperative Prefix and no Irrealis Suffix.
(25) Jamul Diegueño Imperatives (Miller 1990:119),
a. nya-m-mápa-pu m-rar-x-s
INDEF-2-NOM.Want-DEM 2-do-IRREALIS-EMPH
'Do whatever you want (polite imperative)'
b. k-naw
2-run
'Run! (basic imperative)'
Roberts (1990:390) notes that Alamblak, an East Sepik language of Papua
New Guinea, shows a similar choice. Polite Imperatives are categorized as
Irrealis, and strong imperatives as Realis. Roberts points out the parallel with
English commands: Would you be seated versus Sit down!. The presence of
both Irrealis and Realis imperatives in these languages shows that both
options are semantically possible.

2.2 Futures

Future events are often classified as Irrealis cross-linguistically, in languages

that have such a catogory, since they have not (yet) occurred (Comrie
1985:45; Givón 1994). In a few languages, however, including Caddo (Chafe
this volume), future events are classified as Realis.
One possible explanation for the existence of Realis futures is parallel to
one of those proposed for Realis imperatives. Futures can vary in their
probability of occurrence. Speakers could exploit the Irrealis/Realis distinc­
tion to mark their expectation of actuation. The Central Porno Irrealis/Realis
distinction is in fact used for this very purpose. While most Futures are
categorized by speakers as Irrealis, as in (5), (7), and (16) above, some are
categorized as Realis. Discussing how they would spend the Fourth of July,
Mrs. Paoli remarked that they would go to the annual community picnic, then
continued with (26).
(26) Central Porno Realis Future: FP
Ma?á qa-wá-č-in
food biting-gO-IMPFV.PL-SAME.SIM.REALIS
'We'll go around
On the Relativity of Irreality 379

hlá-?w-ač=khe.
walk.PL-around-IMPFV.PL=FUTURE
eating.'
On another occasion, Mrs. Jack related what her mother had told her as a
child.
(27) Central Porno Realis Future: FJ
?
sé- ul ma,
long.time already 2.AGENT
Tn the future,
yém-aq-da,
old-INCHOATIVE.PFV=DIFF.SIM.REALIS

when you are older,

?
a- čhó-w=da,
1 .AGENT not-PFV=DIFF.SIM.REALIS
when I am no longer here,
?
ma yá-q-an-ka-w=?khe.
2.AGENT kn0W-INCH0ATIVE-IMPFV-CAUS-PFV=FUTURE

you will remember. ...

?
a- č h ó-w=da,
1.agent not-PFV=DIFF.SIM.REALIS
After I am gone,
?
má bá-n-či-w= ? k h e....
2.AGENT Suffer-DUR-REFL-PFV=FUTURE
you will suffer.'
The use of the Realis linkers with futures was not a mistake. The choice was
significant. Mrs. Jack produced the sentence in (28) as an elicited translation.
(28) Central Pomo: FJ
Ma-báya cá-l yó-w-da
POSS-man h0USe=t0 gO-PFV=DIFF.SIM.REALIS
'When her husband gets home,
?
e mu-I ma?á čhú-w=?khe.
COP that food eat-PFV=FUT
she'll eat.'
380 Marianne Mithun

When asked about the use of =da in this context, she replied, 'With =da he'll
get home for sure.' The Irrealis counterpart =hla would suggest some uncer­
tainty. The Realis futures cited in (26) and (27) above were used for events the
speaker portrayed as certain to occur: eating at the picnic, getting older, and
dying.
The Central Porno option of categorizing futures as either Irrealis (fre­
quent) or Realis (rarer) shows that both are semantically coherent combina­
tions. A system in which futures were systematically categorized as Realis
could conceivably arise because the Irrealis form was so highly marked
semantically that speakers used it sparingly with futures. Such a scenario is
certainly not the only way in which such a pattern could develop. Chafe (this
volume) describes a possible path of development of a different sort for
Caddo, in which structural characteristics of the markers shape the system.

2.3 Questions and negatives

In many languages, including Central Porno and Amele, interrogation and

negation have no effect on the reality marking of clauses. Questions about
hypothetical situations are generally categorized as Irrealis, while questions
about actualized events are generally categorized as Realis, just like their
declarative counterparts. Similarly, negative constructions show the same
Irrealis or Realis categorization as their positive counterparts. The situation is
quite different in Caddo, however, where questions and negatives are system­
atically categorized as Irrealis.
In fact the differences between the two systems are not as chaotic as they
may first appear. They reflect a contrast in the semantic scope of the reality
markers on the one side and the interrogative and negative markers on the
other. In Central Porno, basic propositions are categorized as Irrealis or
Realis, independently of their status as statements or questions.

(29) Central Porno Questions

QUESTION ( REALIS ( PROPOSITION ))

In (30), cited earlier as (16), the basic proposition is categorized as Irrealis on

the basis of its futurity. The Irrealis proposition is then questioned.
On the Relativity of Irreality 381

(30) Central Porno questioned Irrealis: FP

?i=wa ma šačót=?khe yal ča-l démma-hi?
be=Q you whip=FUT us home take-when.IRREALIS
'Are you gonna whip us when you take us home?'
QUESTION ( - REALIS ( you will whip us ))
In (31), cited earlier as (17), the basic proposition is categorized as Realis on
the basis of its past perfective status. The complete Realis past perfective
event is then questioned.
(31) Central Porno questioned Realis: FJ
Thaná da-sé č-ba-wa ma ?
é- ch?ólčiw?
hand wash-SAME.SEQ.REALIS=Q you hair comb.PFV
'Did you wash your hands and comb your hair?'
QUESTION ( + REALIS ( you washed your hands and combed your hair))
Caddo questions show the opposite scope relations. Reality marking is sensi­
tive to the status of sentences as questions.
(32) Caddo Questions
- REALIS ( QUESTION ( PROPOSITION ))

All Caddo questions are categorized as Irrealis: questioned propositions are

not asserted to have occurred.
(33) Caddo Irrealis question: Chafe this volume
süyybáwnah
sah?-yibahw-nah
2.AGENT.IRREALIS-See-PERFECT
'Have you seen him?'
- REALIS ( QUESTION ( you saw him ))

Similar semantic scope differences can be seen by comparing the nega­

tive constructions in Central Porno and Caddo. In Central Porno, propositions
are categorized as Irrealis or Realis independently of negation.
(34) Central Porno Negatives
NEGATION ( REALIS ( PROPOSITION ))
382 Marianne Mithun

In (35), cited earlier in (19), the proposition is categorized as Irrealis on the

basis of its conditional status. The Irrealis conditional clause is then negated.
(35) Central Porno negated Irrealis: SA
?
Ma me-n íw č h ów=hla
you SUCh do.PFV nOt.PFV=if.IRREALIS
'If you don't do that, [you're going to see some bad things happening].'
NEGATIVE ( - REALIS ( you will do that )
In (36), cited earlier in (21), the proposition is categorized as Realis because it
expresses a past perfective event. This Realis construction is then negated.

(36) Central Porno negated Realis: FP

Ranch- ?el 'qdí yhétač čhów ?
í-n
the good do not-PFV not be.REALIS
'Because they didn't keep up the ranchería, [they wanted to throw
us out].'
NEGATIVE ( + REALIS ( they kept up the ranchería ))
The opposite scope relations can be seen in Caddo. Reality categorization is
sensitive to negation.
(37) Caddo negation
- REALIS ( NEGATIVE ( PROPOSITION ))

All negative statements are categorized as Irrealis in Caddo, since they report
non-actualized events or states.
(38) Caddo Irrealis negative (Chafe this volume)
kút' áybah
kúy-t'a-yibahw
NEG-1 .AGENT.IRREALIS-see
T don't see him.'
- REALIS ( NEGATIVE ( I see him ))'

Such cross-linguistic differences in semantic scope are not unusual. English

and German show a similar contrast in must not ('it is necessary that one not
...') versus muss nicht ('it is not necessary that one ...').
On the Relativity of Irreality 383

The scope relations described here for Central Porno and Caddo correctly
predict the categorization of negative commands, or prohibitives in each
language. In Central Porno, prohibitives are systematically categorized as
Irrealis on the basis of their imperative status.
(39) Central Porno scope relations: imperatives
commands:- REALIS ( IMPERATIVE )
prohibitions: NEGATIVE ( -REALIS ( IMPERATIVE )).
(40) Central Porno prohibitive: FJ
Dá-wi ?čhá-č-hi khyá swé-lar?khe thín ?e ma.
road.on stop-SAME.IRREALIS game play not it.is you
'Don't stop and play on the way home!'
NEGATION (- REALIS ( IMPERATIVE ( you stop and play on the way
home )))
In Caddo, prohibitions are systematically categorized as Irrealis as well but on
the basis of their negativity.
(41) Caddo scope relations: Imperatives
commands: + REALIS ( IMPERATIVE )
prohibitions: - REALIS ( NEGATIVE ( IMPERATIVE ))
(42) Caddo Prohibitive: (Chafe this volume)
kaššáy?bah
kaš-sah?-yibahw
NEG-2.AGT.IRREALIS-see
'Don't look at it!'
- REALIS ( NEGATIVE ( IMPERATIVE ( you look at it )
Such scope differences could arise in a number of ways. A simple
hypothesis would be that scope differences might result from differences in
the order in which the distinctions are grammaticized. Their paths of develop­
ment can be much more complex, however. Scope relations may even shift
within the history of a single language.
Mesa Grande Diegueño, a Yuman language of southern California
closely related to Jamul, contains an Irrealis suffix -x cognate with the Jamul
Irrealis suffix of the same form (Langdon 1970:158-9). The suffixes appear in
all of the same contexts except for one: in Mesa Grande, negated clauses
obligatorily carry the Irrealis suffix, but in Jamul they do not.
 384 Marianne Mithun

In Mesa Grande, negation is indicated by an inflected negative auxiliary

verb. In the sentence in (43), the verb 'for me to go' carries the Irrealis suffix
as the hypothetical complement of want, but the verb 'they want it' carries the
Irrealis suffix only because the sentence is negated.
(43) Mesa Grande negative (Langdon 1970:159)
?
ənya- puy ?əxap-x-vu dwa-p-x uma-w.
I there I.go.in-IRREALIS-SPEC they.want.it-IRREALIS they.not
'They didn't want me to go there.'
In Jamul, negation is indicated by a clause final negative word, such as
xemaaw in (44) below. The negative sentence contains no Irrealis suffix
unless required for some other reason, such as a conditional context.
(44) Jamul negative (Miller 1990:113)
nya 'wach yu 'ip xemaaw.
we.SUBJ hear.PL NEG
'We didn't hear.'
An examination of the forms of the Jamul Negative words indicates that there
has been a shift within Jamul. All three negatives, xemaaw 'not', xchan
'almost, but not', and xmir 'not yet', begin with x. The Irrealis suffix which
originally appeared on the verb immediately preceding the negative has been
reanalyzed as part of the negative marker itself. Miller (1990:203-5) reports
that the negative words, which she terms 'expiring auxiliaries', have lost
characteristics of auxiliaries, and no longer bear inflectional prefixes for
person of subject.

3. Conclusion
The comparison of the Irrealis/Realis distinctions described here suggests
some fruitful directions for the investigation of grammaticization patterns. In
Central Porno, as in many other languages, counterfactual, conditional,
deontic, future, hortative, and imperative constructions all contain Irrealis
markers. Some of these types, such as counterfactual and conditional con­
structions, are widely categorized as Irrealis cross-linguistically. They are in
fact often cited as the best evidence for the appropriateness of the label
'Irrealis' for a marker. Counterfactual and conditional constructions convey
ideas that are most clearly within the realm of thought (imagination) rather
than actualized reality (perception).
 On the Relativity of Irreality 385

Imperatives are classified as Irrealis in many languages, but not all.

Several kinds of diachronic explanations might underlie the discrepancy. An
emerging Irrealis distinction might first be applied within a small grammatical
context in a language, such as statements, before expanding to other construc­
tion types such as commands. An alternative explanation might come from
the fact that imperatives could be viewed as closer to potential actuation than
counterfactuals or conditionals. The utterance of a command might imply an
expectation of immediate compliance. A third kind of explanation could
involve speakers' exploitation of the distinction for expressive purposes.
Speakers might choose to use Realis imperatives to reinforce the implied
expectation that the act is certain to be carried out, soon to be actualized. The
possibility of such a scenario is strengthened by the existence in a number of
languages of two options: a polite imperative classified as Irrealis, and a
strong imperative classified as Realis.
Futures are classified as Irrealis in most languages; they represent events
that have not yet occurred, events still within the realm of thought. On
occasion, however, they are marked as Realis. Again, several possible paths
of diachronic development might lead to this cross-linguistic variation. Chafe
(this volume) shows how the presence of a future suffix in Caddo may have
obviated the need for the development of a new construction based on newly
emerging Irrealis forms. Alternatively, Realis futures might result from con­
ditions similar to those hypothesized to underlie Realis imperatives. Futures
might be seen to assert imminent actuation. Speakers might choose to exploit
the Irrealis/Realis distinction to express varying expectations of actuation. In
fact in Central Porno, futures may be classified as either Irrealis or Realis,
depending on the speaker's portrayal of the event in question as merely
probable or certain to occur.
Interrogation and negation have no effect on reality classification in
many languages. In Central Porno, for example, questioned or negated futures
are Irrealis, just like their positive declarative counterparts, while questioned
or negated past perfectives are Realis. In other languages, however, such as
Caddo, questions and/or negatives are systematically classified as Irrealis.
These differences are not as obviously attributable to varying expectations of
actuation. They may instead reflect a difference in the relative semantic scope
of the markers within the languages. In languages like Central Porno, reality
categorization has narrower scope than interrogation and negation. Proposi­
tions are categorized as Irrealis or Realis independently of their status as
questions or negatives. In languages like Caddo, the scope relations are
386 Marianne Mithun

reversed. Question formation and negation have narrower scope than reality
categorization. The Irrealis/Realis classification is thus sensitive to the status
of constructions as questions or negatives. A number of diachronic paths
could lead to such differences in semantic scope. One might be a difference in
the order of grammaticization of the individual constructions within a lan­
guage. Others might be more complex, such as the Jamul formal reanalysis of
the negative construction.
The comparison of these seemingly disparate systems shows complete
accord in the nature of the basic Irrealis/realis distinction that underlies them.
In all of the languages, events and states classified as nonactualized, those that
remain within the realm of thought and imagination, are overtly distinguished
from those portrayed as actualized, having occurred or currently occurring.
The differences among the systems lie in the application of the distinction to
various contexts, the categorization of individual grammatical constructions
within each language. Some differences may result from special uses of the
distinction for expressive purposes, such as the classification of Imperatives
as Realis to indicate certainty of immediate actualization. Some may stem
from differences in the degree of probability deemed necessary for Irrealis
marking, as in the case of futures. Some may mirror differences in the relative
semantic scope of the Irrealis/Realis distinction and other grammatical dis­
tinctions, such as interrogative or negative. Such scope differences could arise
in many ways, only a few of which have been explored here.
If the 'Irrealis/Realis' terminology were not used, the cross-linguistic
convergences in the semantic nature of the distinction, and the contrasts in its
application, might go unnoticed. Of course special care should be taken in
cross-linguistic comparisons of these categories, as with any modal catego­
ries, to ensure that the distinctions on which they are based are indeed
comparable. Grammarians describing particular languages can help to prevent
misunderstandings if they provide sufficient evidence that a particular marker
does indeed represent the nonactualization of Irrealis mood in a variety of
contexts, for example, rather than the temporal deixis of future tense. With
such care, this area of modality will give us much to discover about processes
of grammaticization that lead to the convergence and divergence of gram­
matical systems.
On the Relativity of Irreality 387

ABBREVIATIONS

CO COOPERATIVE SEQ SEQUENTIAL

DIFF DIFFERENT EVENT SIM SIMULTANEOUS
INF INFERENTIAL SPEC SPECIFIC
M.E. MULTIPLE.EVENT

NOTE

* Edith Bavin, Joan Bybee, and Suzanne Fleischman provided helpful comments on an
earlier draft of this paper.

REFERENCES

Buckley, Eugene 1988. "Temporal Boundaries in Alsea." Proceedings of the Berkeley

Linguistics Society, 14.10-22.
Bybee, Joan, William Pagliuca & Revere Perkins. 1994. The Evolution of Grammar:
Tense, Aspect and Modality in the Languages of the World. Chicago: University of
Chicago Press.
Capell, Arthur. 1971. "The Austronesian Languages of Australian New Guinea." Current
Trends in Linguistics 8: Oceania ed. by Thomas A. Sebeok, 240-340. The Hague:
Mouton.
Chafe, Wallace 1993. "Real and Unreal in Caddo." This volume.
Chung, Sandra & Alan Timberlake. 1985. "Tense, Aspect and Mood." Language Typol­
ogy and Syntactic Descrition III, ed. by Timothy Shopen, 202-258. Cambridge:
Cambridge University Press.
Comrie, Bernard. 1985. Tense. Cambridge: Cambridge University Press.
Craig, Colette G. 1977. The Structure of Jacaltec. Austin: University of Texas Press.
Dempwolff, O. 1939. Grammatik der Jaben Sprache auf Neuguinea. Hamburg:
Friederichsen de Gruyter.
Eatough, Andrew ms. Central Hill Nisenan: Grammar, Wordlist, and Texts. Santa Cruz:
University of California.
Foley, William A. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge
University Press.
Givón, Talmy 1994. "Irrealis and the Subjunctive." Studies in Language 18.2.
Gordon, Lynn 1986. Maricopa Morphology and Syntax. (University of California Publi­
cations in Linguistics 108.) Berkeley: University of California Press.
Langdon, Margaret 1970. A Grammar of Diegueño: The Mesa Grande Dialect. University
of California Publications in Linguistics 66. Berkeley: University of California Press.
388 Marianne Mithun

Miller, Amy 1990. A Grammar of Jamul Diegueño. Ph.D. dissertation, University of

California, San Diego.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Payne, Doris L. & Thomas E. Payne. 1990. Yagua. Handbook of Amazonian Languages,
ed. by Desmond C. Derbyshire & Geoffrey K. Pullum, 2.249-474. Berlin: Mouton de
Gruyter.
Roberts, John R. 1990. "Modality in Amele and other Papuan Languages." Journal of
Linguistics 26: 363-401.
Romaine, Suzanne 1993. "The Grammaticalization of Irrealis in Tok Pisin." This volume.
 The Grammaticalization of Irrealis in
Tok Pisin

Suzanne Romaine
Merton College, University of Oxford

0. Introduction*

In this paper I will address a case of semantic and syntactic change involving
the grammaticalization of irrealis marking in Tok Pisin, an English-based
pidgin/creole spoken in Papua New Guinea. Modality, and in particular, the
marking of so-called irrealis, has figured prominently in discussions of tense-
mood-aspect (hereafter TMA) systems in pidgins and creoles since
Bickerton's claims about the universality of a creole prototype consisting of a
three member inventory of preverbal particles marking the same semantic
functions. These elements all contrast with the unmarked form of the verb. In
most creoles the irrealis marker is rather transparently derived from verbs or
auxiliaries such as go or sa (<shall) in English-based creoles. In others such as
Tok Pisin, however, a temporal adverbial has been recruited for this function
(compare Portuguese-based creoles lo from logo 'next' with Tok Pisin
baimbai from 'by and by').
Bickerton (1981:80) identifies three stages in a gradual process which
leads to the incorporation of these particles from a clause-external syntactic
position to the status of preverbal auxiliaries. Some examples are given in 1 to
3 to illustrate these proposed stages with reference to the grammaticalization
of baimbai in Tok Pisin.
390 Suzanne Romaine

(1) Baimbai mi go
'By and by I'll go.'
(2) Bai mi go
T U go.'
(3) Mi bai go
'I'll go.'
In the first stage baimbai and other similar elements appear optionally
clause-finally or clause-initially. In the second, they are incorporated into the
auxiliary but cannot combine with any Aux constituents, while in the third
they display combinatorial capacities. Tok Pisin is not quite in step with
Bickerton's stages since in (2) we see a phonologically reduced and un­
stressed form of bai still in clause-external position rather than incorporated
within Aux. The third example shows bai in preverbal position, but does not
specify whether there are any restrictions on its co-occurrence with other Aux
constituents. Nevertheless, in many respects Tok Pisin represents an excellent
case for examining this alleged sequence of grammaticalization since all three
stages currently co-exist. In other ways, however, Tok Pisin's development is
atypical, both with respect to what has happened in other pidgins and creoles,
as well as with respect to the grammaticalization paths chosen by other
languages.
This general sequence of grammaticalization of TMA markers is also
taken by many others to be a significant hallmark of creolization. Labov
(1977:29), for example, says:
It is not at all obvious that a pidgin will develop obligatory tense markers
when it becomes a native language. Yet this has happened in case after case.
When pidgins become creoles, the system of optional adverbs gives way to
an obligatory tense marker next to the verb.

More recently, Markey (1982) and Mühlhäusler (1986:156-7) have cited the
use of sentence-external propositional qualifiers as a major typological char­
acteristic of pidgins (see also Kay and Sankoff 1974:64). Sankoff and Laberge
(1980) argued on the basis of research done in the late 60s and early 70s that
the Tok Pisin temporal adverb baimbai is becoming a marker of future tense.
Moreover, they link this change to "the passage of Tok Pisin from a second
language lingua franca to the first language of a generation of urban New
Guineans". I will say more about the extent to which such developments are
triggered by creolization and exclusive to creoles in Sections 3., 4.3 and 5.
The Grammaticalization of Irrealis in Tok Pisin 391

In 2.I situate my use of the term 'irrealis' within the context of modality and
explain its relevance to the Tok Pisin case.
Bickerton's claims have been based largely on his analysis of data from
Hawai'i Creole English (HCE), partly in the conviction that "data of any kind
on the antecedent pidgins of any creole but HCE are simply non-existent"
(1981:79). The latter assertion is unwarranted. In this paper I will make use of
new historical data on Pidgin English in the Pacific together with synchronic
Tok Pisin data I collected during 1986 and 1987 in Papua New Guinea.

1. Brief description of corpus

The spoken data were recorded over a two year period in 1986 and 1987 as
part of a study of rural and urban Tok Pisin in two provinces in Papua New
Guinea (see Romaine 1992 for further details). The two provinces, Madang
and Morobe, lie on the New Guinea Coast. The number of Tok Pisin speakers
has been steadily increasing there, though it is difficult to give precise figures.
In the 1966 and 1971 Censuses, which covered only 10% of the village
population, respondents were asked to say which members of the household
over the age of 10 were able to speak Tok Pisin. In Madang Province there has
been an increase in the number of reported speakers of Tok Pisin between the
years 1966 and 1971 from 62% to 69%, while in Morobe Province it has
increased from 47% to 59%.

1.1 Spoken Data

The spoken corpus to be dealt with here consists of tape-recorded material

comprising some 700,000 words from nearly 500 children and adults. The
majority of the speakers fall between the ages of 5 and 17 and were drawn
from 11 areas. I have indicated in Table 1 the nine areas which I will discuss
here; in those marked urban, the children are largely first language speakers,
while in rural areas the children are mainly second language speakers. Adults
are not included in the numbers given here, although I will discuss some data
collected from adults in one of the rural areas, Indagen. The Unitech sample is
a longitudinal mini-sample within the larger survey. These children were first
contacted and interviewed at a school in Lae. Their parents live on the campus
of the Papua New Guinea University of Technology and they took part in
 392 Suzanne Romaine

regular after-school recording sessions over a period of two years. The data
for these children consist of narratives, free play and video story retellings. In
other areas such as Indagen and Bulolo there are some longitudinal data too
since some óf the same speakers were interviewed in 1986 and 1987.

1.2 The Written Corpus

At the same time as the analysis of the spoken material is being carried out, I
have been compiling a corpus of written texts (both contemporary and histori­
cal) for comparison. Table 2 gives brief details of that part of the corpus I will
be referring to here and further information about the texts can be found in
Romaine (1992). Nupela Testamen (the 1966 translation of the New Testa­
ment) and the weekly newspaper in pidgin, Wantok, played important roles in
the standardization of Tok Pisin. The other two main sources which were used
to make up the written corpus are Save Na Mekim (Understanding and Doing;

Table 1. Details of Sample of Spoken Data

Province Date Location Setting N. children No. words
Morobe 1986 Lae/Taraka urban 74 70782
1986 Lae/Unitech 12 30619
1986 Bulolo urban 18 16092
1986/7 Indagen rural 79 108572
1987 Geraoun rural 33 14768
1987 Waritsian rural 32 51113
Madang 1987 Kusbau urban 104 178472
1987 Rempi rural 42 67815
1987 Erima rural 40 60497
1987 Bahor rural 48 106391
TOTAL 482 705121

Table 2. The written corpus

Text Date No. words
Nupela Testamen 1966 3720
Wantok 1982 9838
Save Na Mekim 1982 4825
Konstitusen 1986 3453
personal letters 1986/7 1294
TOTAL 23130
The Grammaticalization of Irrealis in Tok Pisin 393

Bergman 1982) and Konstitusen Bilong Independen Kantri Papua Niugini

(The Constitution of the Independent Country Papua New Guinea; Mihalic
1986a). In each case samples were taken from these texts for analysis. The
total corpus amounts to some 23,000 words.
The written corpus also includes a small sample of 7 personal letters sent
to me by various speakers. These were written by a houseboy in Lae (N = 2);
two women in Indagen (N= 3); 2 women in Madang who run a mobile
pre-school unit (N = 1); the headman of Musep village, near Indagen (N = 1).
There is also one official letter written by the assistant secretary to Morobe
Province advising various community governments about the research I was
doing.

2. Baimbai, bai and the marking of irrealis

It is not entirely clear from the historical record when baimbai began giving
way to bai. My guess is that this happened sometime in the 1950s and 1960s,
but I will have more to say about that in 3. All three stages are represented in
the data from Bulolo, Unitech (Lae), Erima and Indagen, as can be seen in
examples (4) to (8). The same is true incidentally for the speech of adults.
(4) Baimbai yu kaikai. [BL1F4]. 1
'Then [after you've cooked], you'll eat.'
(5) Mi droim maunten, na bihain bai mi wokim wara wantaim.
[BL1M5].
'I'll draw the mountain, and then I'll do the water that goes with it.'
(6) Em i go insait nau, bai ol kaikai [INDIFl]
'He went inside and then they ate.'
(7) Yu stretim ston insait long hul ia orait ston bai stap nau bai yu
stretim hul ok yu putim disla kaikai ia [BL M5]
'You arrange the stones inside the pit, and once the stones are
there, you arrange the pit and you put the food in.'
(8) Ol man bai bihainim em ol bai kisim em. [IND1F1]
'The men then followed him and they caught him.'
Example (4) is one of ten examples of baimbai I have found so far in the
children's corpus. The speaker is a fourth grade girl in Bulolo, who is a second
language speaker of Tok Pisin from Irian Jaya. She also uses bai in both
394 Suzanne Romaine

clause-initial and preverbal position, as can be seen in (9). The only other
instances of baimbai are from one of the Unitech girls and the children in
Erima.
(9) Nau wanpla longlong man ia draipla kanu blem, nau disla
longlong ia em sa kaikai ol man nau. Longlong man ia kam luk ol
manki na em tok, "Yupla sa waswas long ia, mi kisim yupla go, bai
mi lusim yupla na yu bai waswas long disla ". Na em kisim ol go.
[BL1F4].
'There was a crazy man who had a canoe, and he ate people. This
crazy man saw the boys and said, "You boys wash over here and
I'll catch you, now I'll leave and you'll wash over there." Then he
got them.'
Thus, even today all three stages of this process of grammaticalization
can be represented synchronically in the speech of one speaker. It is evident,
however, that even at the time when Sankoff and Laberge obtained their data,
baimbai was a recessive feature. In a corpus of 395 examples of bai in the
speech of 18 people (9 children and adolescents between the ages of 5 and 17,
and 9 adults between the ages of 25-45) recorded near Lae, they found only
five instances of baimbai. These were phonetically already somewhat re­
duced, i.e. [bəm'bai] or [bə'bai] and were found in the speech of three of the
nine adults. There were no instances of baimbai in the speech of the children
they interviewed (see Romaine and Wright (1987) on morphophonological
condensation in children's speech). Nevertheless, it is clear from my data that
the use of baimbai is not entirely exclusive to rural and/or second language
speakers, or adults.
As far as the meaning of baimbai and bai is concerned, it will be obvious
even from the few examples I have given here that bai is a highly redundant
marker which signals more than time relations. It often occurs with other
adverbs indicating relations of time or the sequence of events. For example, in
5, the connectives na bihain 'and then/afterwards' make it clear that the
activity of the second clause takes place after that reported in the first. In this
context bai is clearly redundant, and functions not so much as a tense marker
but merely as a sequence marker in discourse. This use of bai is in line with its
historical origin as a temporal adverb. In some texts it appears almost with
every clause. This can be seen in (10).
The Grammaticalization of Irrealis in Tok Pisin 395

(10) Krokodail tok olsem, "Nau bai mi go long warn na bai mi painim ol
dok na ol dok bai sindaun, bai mi kisim em go long aus bilong em.
Em bai wokabaut bilong mi. ", em tok, krokodail olsem. Krokodail
tok olsem, "Bai mi go long wara na painim dok". Na em go long
wara na em lukim bikpla ston na dok sindaun i stap. [IND1F6].
T h e crocodile said, "I'll go down to the river and look for dogs.
Dogs will be sitting there and I'll get one and take him back to his
house. Then he'll come to mine", the crocodile said. The crocodile
said, "I'll go down to the river and look for the dog." He went to the
river and saw a big stone and the dog was sitting there.'
Generally speaking, there is no obligatory marking of tense in Tok Pisin.
In this extract from a narrative in (10), the context makes it clear that the
events referred to take place in the past. The crocodile is of course making
statements here about what he will do in the future and uses bai to do so. In
this respect bai marks an intention, but it can also be understood in some
respects as a prediction about what will happen at some later point in time. It
is this meaning of prediction which is central to the notion of future according
to Bybee et al. (1991). I will return to this point in 4.3. Once the direct
discourse of the crocodile is finished, we are back into the frame of narrative
time, i.e. past events. That is why we find the unmarked verb form lukim 'he
saw', which is the usual form for reporting past events. The latter is cited by
Bickerton (1981) as a general characteristic of creole tense systems; namely,
the zero form of the verb encodes past tense for verbs of action. As might be
expected from the nature of reported speech, a great many occurrences of bai
are found in conjunction with it. At this stage I want to introduce the notion of
'irrealis' into the discussion and to claim that bai could be regarded either as
an irrealis or as a future marker. The paths leading to the grammaticalization
of futures and modality intersect at certain points because events in the future
are by definition hypothetical and there is some uncertainty about them. In
this sense we might say that the common demoninator of irrealis modality is
futurity or uncertainty (see Givón 1984:318).
Many linguists have noted the confusing and often overlapping ways in
which the terms 'aspect', 'tense', 'mood' and 'modality' have been used. In
their discussion of mood and modality Bybee et al. (1994: Ch 6, section 12)
comment, in particular, on some of the problems surrounding the use of the
term 'irrealis,' especially in discussions of non-Indo-European languages.
396 Suzanne Romaine

The concept of irrealis is clearly central to a consideration of futurity, which is

often treated as a matter of tense. Givón (1984:285), for instance, writes that
"the future is a clear irrealis tense, dealing with hypothetical, possible, uncer­
tain states or events that have not yet occurred. Roberts (1990:363) includes
within the category of irrealis future tense, imperative mood and counterfac-
tual mood, while for Bickerton (1981) irrealis consists of futures and condi­
tionals. These labels refer to meanings which traditionally have been
associated with and discussed under the heading of modality. Synchronically,
Bybee (1985:156-9) has cited many instances of languages whose future
markers have primarily a modal function. Moreover, the diachronic connec­
tion between irrealis modality and futurity is clearly indicated in Bybee et al.
(1991).
Some such as Roberts (1990:370) have also observed that the English
future tense is really a modality because it is expressed with a form which has
a modal function. Thus, in Old English, for example, will indicated modality,
but it has since come to be used as a future auxiliary. He (1990:369) also
points out that in many Papuan languages future tense is a modality.
Indeed, in many languages the distinction between realis and irrealis is
alleged to be central. Givón (1984:309), for instance, says that many Austro-
nesian languages make their major morphological distinction between realis
and irrealis in TMA systems. Roberts (1990) and Foley (1986) have claimed
that for many of the Papuan languages the basic distinction is between realis
and irrealis. The centrality of the realis/irrealis opposition in both these
language families is not without relevance to a discussion of Tok Pisin since
they form its substratum in its most comprehensive sense. All of Roberts'
principal examples from the Papuan family are spoken in Madang Province,
where some of the varieties of Tok Pisin to be discussed here were investi­
gated, but the general patterns he observes have a wider areal distribution,
extending into Morobe Province, where the other spoken varieties of Tok
Pisin I discuss here were recorded.
In attempting to dismiss the realis/irrealis distinction as ill-founded,
Bybee et al. (1994) query whether it is possible for the past habitual to be
categorized as irrealis in some languages when it is more generally taken to be
a prototypical case of realis in others. Personally, I do no find this problematic
(if indeed it is correct) since I believe there are cultural prerequisites to
grammatical analysis. In any event, I would suggest instead that it is perhaps
more often the case for languages to use the same morpheme to mark both past
The Grammaticalization of Irrealis in Tok Pisin 397

tense and unreal or hypothetical events and states. As Steele (1975) and others
have noted, the basic connection between the past tense and irrealis is their
sharing of the characteristic of being distant from present reality. Consider,
for instance, the use of the past in hypotheticals such as I wish I had a new car
(see the discussion in James 1982 and 1991). Verbs in the protases of con­
trary-to-fact or unreal conditionals referring to the future are also inflected
with -ed in English, e.g. If you travelled to Hawai'i, you'd like it there.
Not surprisingly, bai is also used in similar kinds of conditionals in Tok
Pisin. This can be seen particularly when bai occurs in construction with
sapos (from 'suppose') in a kind of if/then construction as in (11). It can also
be used on its own, as in (12), where both the antecedent or cause clause and
the result clause are introduced by bai. It is semantically equivalent to an if/
then construction, even though there is no overt marking as such. The exten­
sion from temporality to causality in this kind of construction is a natural one
in the sense that what is caused is later in time than the event or state which
precedes it. As I will show in 4., an understanding of the semantics of bai is
crucial to interpreting the significance of the syntactic developments I will
look at next.

(11) Sapos fesbon em fesbon meri, bai ol ol kolim Asan. [BL1F6]

'If the first born child is a girl, then they will call her Asan.'
(12) Bifo long mipela em ol sa wokim olsem. Man i gat laik long yu na
yu no laik. Orait, em man. Em bai makim yu, bai yu no laik long en,
bai em bai bagarapim yu disla kain olsem ... Bifo taim bilong
tumbuna em kilim men indai. Em tu bai indai bifo. Nau lo i strong
tru. Orait bihainim lo tasol [INDIEM].
'Before our custom used to be like this. If a man liked you and you
didn't like him. Ok, he's a man. He would mark you. If you didn't
like him, then he'd rape you like this. In the old days he would kill
a woman and she would die then. Now, however, our laws are
powerful and they just have to follow law.'

2.1 The syntactic distribution of bai in nine areas

I will now look at the syntactic position of bai in the nine areas in more detail.
My analysis revealed that 56% of the occurrences of bai are preverbal, while
44% are clause-initial. The data also show that although bai can either precede
398 Suzanne Romaine

or follow noun phrases, there is a tendency for noun phrase subjects to occur
with preverbal bai (see further in Sankoff and Laberge 1980:207). The third
person singular pronoun em strongly favors the occurrence of preverbal bai.
The shift to preverbal position appears to begin in the third person singular
and then spreads throughout the pronominal paradigm.
My findings on the position of bai in the nine individual areas can be
compared with those of Sankoff and Laberge shown in Table 3. If we are
dealing with on-going change, as they claim, then their data appear to have
intersected the change at a point just before bai shifts to preverbal position
since most of their occurrences of bai are still clause-initial. In my data the
balance has shifted in favor of preverbal positioning of bai. Interestingly, they
found no significant differences between the adults and children with respect
to the ordering of bai.
More specifically, it can be seen that the situation in Bulolo and Rempi
more closely parallels that reported by Sankoff and Laberge (1980), who
worked in Lae. That is, the shift from clause-initial to preverbal positioning of
bai is underway, but clause-initial position is at least marginally still the most
frequent. It is really only in Indagen that the shift is well advanced, and
possibly nearing completion, whereas it is only just beginning in Bulolo and
Rempi. The other areas, both rural and urban, represent an intermediate stage.
In general, apart from Bulolo and Waritsian, the areas in Morobe Province are
ahead of those in Madang. The Taraka children in Lae are most directly
comparable to the urban children interviewed by Sankoff and Laberge nearly
20 years ago and it would appear that change has progressed remarkably
quickly in that time span in this area.
Judging from past sociolinguistic studies of the role of social factors in
linguistic change, we might reasonably expect both sex differentiation and
significant age-grading, if change is taking place. Recall that Sankoff and
Laberge found no differences between children and adults. However, their
data base is too small to show the effect of age grading either between the
adult and younger population, or within these groups. Moreover, they do not
take into account the possibility of regional differences. My study found no
significant gender distinctions in the use of bai.
If it is the case that preverbal bai represents a change in progress, we
would probably find differences between both children and adults as well as
within the younger group of speakers. More specifically, we should find that
adults and older children showed a predominantly clause-initial pattern and
The Grammaticalization of Irrealis in Tok Pisin 399

Table 3. The distribution of bai in nine areas (compared with Sankojf and Laberge;
figures are percent)
% Clause-initial % Preverbal
[Sankoff/Laberge 59 41]
Rempi 58 42
Bulolo 56 44
Kusbau 50 50
Bahor 50 50
Waritsian 47 53
Erima 45 55
Geraoun 39 61
Lae (Taraka) 36 64
(Unitech) 32 68
Indagen 25 75

younger children showed increasing use of preverbal bai. There are some
indications that this is true, though there is not always a regular age-by-age
increment in certain areas. Age grading is particularly apparent for some of
the individual areas more so than others. For instance, there is no significant
age grading in Bahor, Erima or Geraoun. However, the oldest children in
Geraoun use mainly clause-initial bai, while the younger ones all use more
preverbal bai. In Unitech, there is significant age grading between younger
and older speakers, which happens to coincide with sex differentiation be­
cause the girls are all younger than the boys. All the Indagen children use
more preverbal bai, though the oldest age group, namely, 16 year-olds use
predominantly clause-initial bai. The Indagen adults who use 44% clause-
initial bai are thus behind the younger generation with respect to this change.
Age-grading is, however, particularly apparent in the two areas where the
shift appears not to be implemented yet, namely Rempi and Bulolo.
Thus, the change appears rather sporadic in its transition and is not
clearly progressing in some areas like Erima, Kusbau and Bahor in a regular
way which is tied solely to increasing age, but there is still more to be said
about the differences between children and adults in my corpus. It is obvious
that whatever changes were taking place in the status of bai were well
underway prior to the existence of a large number of native speakers. If
stabilization and nativization are prerequisites for this kind of grammaticali­
zation, then it is surprising that Sankoff and Laberge did not find the change
further along in Lae 20 years ago. In fact, they claim (1980:208-9) that not
only do native speakers appear to be carrying further tendencies which were
400 Suzanne Romaine

Table 4. Incidence of bai among first (i.e. urban) and second (i.e. rural) language
speakers (figures are normalized to distribution per 1000 words)
Normalized N bai N words
children: urban 8.0 2374 295965
rural 7.0 2878 409156
5252 705121
adults rural 11.5 498 43221

already present in the language, but also that fluent speakers show no differ­
ence from native speakers with respect to use of bai. Sankoff and Laberge
(1980:205) say that 'bai seems to be redundant and obligatory for both adults
and children', with no observable differences between them. They record
almost equal numbers of bai for both groups (children = 192; adults = 203).
Certainly at the moment in other areas of the grammar and lexis, those
innovations which are associated with town speakers are the ones most likely
to be picked up by others (see Romaine, 1989b and 1993). It may be that the
reason why Indagen and Geraoun show such a high rate of pre verbal usage is
that influence from Lae has spread out first to these areas and they have now
overgeneralized or accelerated the change, though if that is the case, it is
surprising that Bulolo does not follow Lae too. Bulolo is connected to Lae by
a relatively good road while Indagen, though closer to Lae (some 25 miles in
the mountains to the northeast), is connected by air only. Geraoun is con­
nected only by bush track to Indagen and surrounding areas.
However, by comparing Table 3 with Table 4, we can see statistically
significant differences between both rural and urban children as well as
between children and adults. Although first language speakers of Tok Pisin
clearly use bai more frequently overall than second language or rural speak­
ers, it is the second language speakers in rural areas who lead in the introduc­
tion of preverbal bai. While it would appear that children rather than adults
are ahead in introducing preverbal bai as the canonical way of marking
irrealis, it is the rural adults who show the highest rate of bai usage.
To interpret these findings more fully, I turn now to a consideration of the
syntactic developments in comparative and historical perspective, and then I
will look at the semantic aspects of this change. At that stage I will also return
to the question of age grading.
The Grammaticalization of Irrealis in Tok Pisin 401

3. Discussion of syntactic developments in historical and comparative

perspective

There are at least three issues which need to be discussed in order to clarify the
sequence of steps identified by Sankoff and Laberge as part of the grammati­
calization process. Firstly, do these stages reflect true diachronic ordering?
Secondly, what is the connection between syntactic positioning and phono­
logical reduction? Thirdly, is this sequence coincident with creolization? To
take the last question first, I have already indicated that it does not appear to
be, despite the fact that Sankoff and Laberge (1980:195) link this change in
bai to 'the passage of Tok Pisin from a second language lingua franca to the
first language of a generation of urban New Guineans'. In order to illuminate
all three questions, however, I will need to consider further diachronic and
comparative evidence.
Evidence from the formative stages of Tok Pisin's development is very
sketchy. Earlier sources mention baimbai but not bai. This is true even of
Mihalic's (1957) first edition of the Tok Pisin grammar and dictionary.
Murphy (1966:59), however, lists it in his dictionary and it can be found in
print in Tok Pisin newspapers, e.g. Nius bilong Yumi published fortnightly by
the colonial administration, from the 1960s onwards. In the 1971 edition of
his dictionary and grammar, Mihalic (1971:63) lists bai with baimbai and
says that it is derived from English 'by and by', and has the same range of
meanings.
It seems likely that bai is a result of phonological reduction of baimbai,
as Sankoff and Laberge claim, even though the two forms have apparently
co-existed for some time. Further support for this can be found in Solomon
Islands Pijin, a near relative of Tok Pisin. The form bambae and its reduced
variants babae and bae are all currently in use and have been for some time.
Older speakers (36+) tend to be conservative and use mainly bambae and
babae. The younger generation tend to use bae (Jourdan 1985). In Bislama, a
variety of Melanesian pidgin spoken in Vanuatu also closely related to Tok
Pisin, similar variants are in use, though Crowley (1990:262) reports that bae
is now more widespread than bambae in the modern language. Whatever the
precise historical details are, it seems clear that baimbai has become very
marginal in ordinary conversational use in Tok Pisin, and indeed, much more
so than in Bislama and Solomon Islands Pijin.
402 Suzanne Romaine

With regard to the positioning of bai, the historical sources which have
been discussed by other scholars such as Sankoff and Laberge indicate that
clause-initial positioning is the earlier. All of the 11 instances of baimbai cited
by Churchill (1911:37) are clause- or sentence-initial. Four of these occur
before a noun phrase, and the others before pronouns. Mihalic (1957:43) lists
baimbai with a group of adverbs which he claims are found only at the
beginning of a sentence. In his later edition he says only that it is one of a set
of adverbs found usually at the beginning of a sentence.
It is striking that none of the earlier historical sources mentions the
possibility of preverbal baimbai, and Sankoff and Laberge have overlooked it
too. I have, however, collected a number of attestations of it both diachron-
ically and synchronically in speech and writing, as can be seen in examples
(13) through (42). I will now use them to argue that the question of syntactic
positioning of bai and phonological reduction must be seen as separate issues.
Their conflation by Sankoff and Laberge (and Labov) fails to accommodate a
number of competing developments which have made the grammaticalization
process less straightforward than it appears. The existence of these examples
indicates that the incorporation of the full form baimbai within the verb
phrase probably existed as a grammatical option long before phonological
reduction or creolization, or indeed, the existence of a community of fluent
second language speakers.
Examples (13), (14) and (15) are from the earliest textual records from
the jargon stage. The first two attestations for pidgin English in New South
Wales and Queensland predate the labor trade. Crowley (1990:252) notes
there is evidence that the clause-initial future marker also began to appear
occasionally in post-subject position in the early 1880s. It is also attested in
Samoan Plantation Pidgin, as in (16). Examples (17) through (21) are re­
corded from the German colonial period.2
(13) me bye and by come back. [New South Wales, 1844, Troy 1985]
'I'll come back'.
(14) Suppose you no kill piccaninnies, that fellow by and by jump up
kipper. [Queensland, 1858, Praed 1885]
'If you don't kill the children, they will grow up into adolescents
and kill you.'
(15) Brother belong-a-me by and by he dead, [reported in Beche-la-mar
by E.L. Lazard 1880 in Noumea, New Caledonia to Schuchardt 1883]
'My brother will die soon.' 3
The Grammaticalization of Irrealis in Tok Pisin 403

(16) Brata bilong mi baimbai dai. [Samoan Plantation Pidgin,

Mühlhäusler 1978:103)
'My brother will die.'
(17) me think German by and by come and raus em altogether English.
[Detzner 1920:226]
'Perhaps the Germans will come and get rid of the English com­
pletely.'
(18) German by and by strong belong English. [Detzner 1921:131].
'The Germans will be strong against the English.'
(19) Master, I think fight by and by finished, Germans now strong
belong English. [Detzner 1921:177]
'Master, I think the fight will be finished, the Germans are strong
against the English.' 4
(20) Tripela de baembai ipinis. [Borchardt 1930]
'Three days will have passed.' 5
(21) Now bymby dry [Vogel 1911:212, Pak Island, Bismarck Archi­
pelago]
'Now the coat will become dry'.
Examples (22), (23) and (24) are all from adult speakers of contemporary
varieties of Melanesian Pidgin, namely Solomon Islands Pijin, and Vanuatu
Bislama.
(22) Solomon bombai pinis kaikai. [PM, Solomon Islander in Samoa,
1975]
'Solomon will finish eating.'
(23) ou, lotu ia baebae kam i spoel-em devol blong iumi.
[DA, Talise, Guadalcanal, SI, Keesing 1988:186]
'If this church comes, it will destroy our ancestral spirits'.
(24) Olsem we yumi save finis, sua nao se gavman bambae i no save
tekem tufala lanwis blong Inglis mo Franis i kam wanples.
[Ligo 1987:83, Bislama]
'As we already know, it is certain that the government will not
bring together the two languages English and French.'
Examples (25) through (30) are contemporary spoken instances from
adult Tok Pisin speakers. Examples (26) and (27) are from a 71 year-old man,
one of the oldest men in Indagen village, while (28) through (30) are from a 52
year-old man, who speaks Boiken, a Papuan language spoken near Wewak in
404 Suzanne Romaine

East Sepik Province. He is a former schoolteacher and government inter­

preter, who was partly educated in Australia. The Indagen adults produced 8
instances of baimbai, half of which were preverbal. Examples (31) through
(34) are from children in Erima, who produced a total of 4 instances of
baimbai, all of which were preverbal.
(25) Na Taña i tok, em baimbai kilim yu. [J. Z'Graggen. Daria A Nar­
ratives, Madang Province]
'And Zaria said, he would kill you.' 6
(26) Em yet baimbai olim em. [IND87MISC1BW]
'He himself will hold him.'
(27) Ol i baimbai i go. [IND87MISC1BW]
'They will go.'
(28-30) Mi tok olsem pisin baimbai i karamapim tasol long bihain Inglis
bai i karamapim na dispela kantri baimbai trupela tok bilong
mampapa bilongen long ples bilong em tru baimbai i pinis.
[Recording made by J. Fingelton, legal consultant to East Sepik
government in 1987].
'I say that pidgin will replace [them i.e. local languages], but later
English will replace it and the real language of the older generation
of the village will be finished.'
(31) Em baimbai kam lo kaikai yu. [ERM1FKW]
'He'll come to eat you.'
(32) Planti papa empla baimbai koros. [ERM4MWA]
'Many men will be angry.'
(33) Moning yumi go, baimbai dikim weisan. [ERM5FMM]
'In the morning we'll go and dig some sand.'
(34) Strongpela wot baimbai kisim disla meri. [ERM5FMM]
'Strong words will get this woman.'
The rest of the examples in (35) to (42) are written, most of them from
Bible translations, especially Nupela Testamen, but occurrences of preverbal
baimbai go back to some of the earliest mission texts such as in (35-36).
Example (41) is from the newspaper Nius bilong Yumi and (42) is from
modern written Tok Pisin.
(35-36) Em baimbai i bringim tok bilong Ridimer, em baimbai i mekim
stret ol man.
[Hall 1955:141, citation from Liklik Katolik Baibel 1934]
The Grammaticalization of Irrealis in Tok Pisin 405

'He will bring the word of the redeemer, and he will make all men
straight.'
(37) Long las de bilong dispela graun bambai i olsem tasol.
[Nupela Testamen, 1966. Mt. 13:49].
'And on the last day of the world this is the way it will be.'
(38) Em bambai i kam bek gen wankain.
[Nupela Testamen, 1966. Ap.l:ll.]
'He will come back again.'
(39) Em bambai i save as bilong dispela tok mi autim.
[Nupela Testamen, John 7:17]
'He will know the reason for what I am saying.'
(40) Skai wantaim graun baimbai i pinis.
[Nupela Testamen, Mk 13:31.]
'Heaven and earth will pass away.'
(41) Ol dispela man baimbai i stap long Australia inap olsem wan mun.
Oli baimbai i stap wantaim arapela kain kain taun kaunsel long
Victoria na Tasmania. As bilong limlimbur bilong ol em long
baimbai long lukim Australia taun kaunsel wok olsem wanem.
[Nius bilong Yumi 15 April 1966. Vol. 8 no.6., p.2].
'These men will stay in Australia for a month. They will visit other
kinds of town councils in Victoria and Tasmania. The reason for
their extended stay is so that they will be able to see how Australian
town councils work.'
(42) Dispela buk long Tok Pisin i bambai helpim planti pipol long
Papua Niugini save long nupela na guípela we long mekim gut
sindaun long ples bilong ol. [Preface to Save na Mekim, 1982.
Michael Somare, former Prime Minister]
'This book in Tok Pisin will help plenty of people in PNG to know
about new and good ways of making a good life in their villages.'
My examples are important for a number of reasons. They demonstrate
the option of using preverbal baimbai has apparently been available in the
language for at least a century, if we accept examples like (13) through (21) as
accurate renditions, and it still exists today in the speech and writing of fluent
users. They show that phonological reduction is neither a necessary nor
sufficient precondition for preverbal placement of the marker. Example (43)
may also provide evidence for the possibility of yet another variant, namely, a
clause-initial reduced form of baimbai.
406 Suzanne Romaine

(43) bam yu go stap? [ERM5FMM]

'Will you go and stay there?'
This means that there is good reason to question the strict chronicity and
implicational relationships among the three stages proposed by Sankoff and
Laberge. The existence of preverbal bai today may derive from reduction of
preverbal baimbai in situ rather than by movement, and therefore, their stages
do not correspond to discrete points in time. Thus, the use of clause-initial bai
need not represent an intermediate stage in the grammaticalization. It is
possible of course that the synchronic examples are not survivals in any direct
sense of the earlier attestations, and are therefore not genuine reflexes of this
construction. They may be simply analogical reformations patterned on pre-
verbal bai. In fact, Sankoff and Laberge (1980:201) note a personal communi­
cation from Anne Chowning, who claims that "in areas of New Britain in the
1950s, bai was the exclusively used form, with baimbai appearing later as a
novel introduction". It is dangerous to assume, although it is commonly done,
that older speakers preserve an earlier stage of the language and do not change
their speech over the course of their lifetimes.
It would likewise be a mistake to assume too much continuity in the life
cycle of a pidgin. It should not be surprising that the same structural innova­
tions arise at different stages in the development of a language and either
spread or fade away. The possiblity of convergent etymology as an important
factor in determining lexicalization in pidgin and creole languages (see Ro­
maine 1988) is now widely acknowledged, and it seems plausible to assume
that structures compete for grammaticalization too. The more potential
sources for grammaticalization of a construction, the more likely that con­
struction is to be incorporated, though different speakers may pull the lan­
guage in different directions.
There is also another structural possibility which could have paved the
way for the grammaticalization of bai in preverbal position. A number of the
children and adults use the form em bai, as in (44) and (45). Here the third
person pronoun em is not syntactically integrated as a clause argument. It is
easy to see how speakers might have regarded this as a short form of baimbai
if we return to example (12), where we have a case in which bai appears on
both sides of the third person singular pronoun em, i.e. bai em bai bagarapim
yu disla kain olsem. In rapid speech the sequence of bai em bai is almost
identical with baimbai. It is easy to see how the full form baimbai might have
been first reanalyzed in this position to the sequence bai em bai, which would
The Grammaticalization of Irrealis in Tok Pisin 407

then have set the precedent for the reduced form bai to occur both clause-
initially and preverbally. This also fits in with the finding that it is the third
person pronoun which provides the point of departure for the diffusion of
preverbal bai throughout the pronominal paradigm. If this is true, then phono­
logical reduction of baimbai is not the only source for the short form bai.
(44) em bai tupela sindaun. [NDMISC87FA]
T h e two of them will sit down'.
(45) i go pas turnas, em bai skin bilong en kamap hatpela turnas.
[IND86MISCSB]
'If it [airplane] goes very fast, the outside becomes very hot.'
There are also many examples where bai appears both preverbally and
clause-initially with a repetition of the same verb, as in the cases in (46-48).
There are also cases where preverbal and clause-initial uses are juxtaposed
within the same utterance, which suggests that for some speakers the two are
optional variants, possibly with some stylistic or pragmatic significance.
(46) Nau bai kau bai go.
'Now the cow will go.'[UTlF5]
(47) Bai bed ia bai bruk. Bed ia bai bruk.
'The bed will break. [UT1F1]
(48) Ol bai katim ol karot ia bai ol katim ol karot.
'They will cut the carrots.'[UTIF]

As far as developments in the written language are concerned, most of

the contemporary written texts I have looked at show predominantly clause-
initial bai. For example, Nupela Testamen (1966) has 80% and Save na Mekim
(1982), 81%, as can be seen in Table 5. The Tok Pisin version of the New
Testament is now noticeably archaic in aspects of its syntax, such as predicate
marking and the use of bai, as well as in lexicon, but it has set an important
standard for the written language (see Romaine 1989a). The weekly newspa­
per Wantok, however, has predominantly (63%) preverbal bai and is thus
more in line with what appears to be the long term trend in the spoken
language, however laggard the actual transition appears to be in some areas.
We can see that all of the written texts, except Wantok and the personal
letters, are characterized by more frequent use of clause-initial bai. This is
especially true for Nupela Testamen and Save na Mekim, as opposed to the
Konstitusen, which shows a pattern very similar to Bulolo and Rempi.
408 Suzanne Romaine

Table 5. Distribution of bai in the written corpus

Text % Clause-initial % Preverbal
Save Na Mekim 81 19
Nupela Testamen 80 20
Konstitusen 55 45
Wantok 37 63
personal letters 27 73

Keesing (1988:184) argues for substratum influence in the form of a

common Oceanic pattern for incorporation of the future marker within the
verb phrase. The early attestations in New South Wales Pidgin English (see
examples (13) and (14)) and Chinese Pidgin English (see note 4) make
superstrate influence more likely since the Australian and Chinese substratum
would have been different from each other and each would have been differ­
ent from the Oceanic substrate. However, since his primary aim is to argue for
early stabilization of Melenesian Pidgin, i.e. prior to separation into regional
dialects, he (1988:187) also claims that by the late nineteenth century bambae
seems to be a grammaticalized form and not simply an adverb temporally
framing the clause. He (1988:182) notes that what happened to bambae is
"theoretically important because [...] the transformation of what was until
recent decades a temporal adverb in sentence- (or clause-initial) position to a
grammaticalized preverbal particle is supposed to reflect a late phase in
Melanesian pidgin development, particularly associated with incipient
creolization." This indicates that for him syntactic position of the marker is
the criterion for deciding whether we are dealing with a grammaticalized
form. Sankoff and Laberge and Labov, on the other hand, rely on both
position and reduction. If we separate the issues of position and reduction
from semantic development for the moment, what is crucial is to determine at
what stage we are dealing with a form which is no longer a temporal adverb
(see Romaine 1990). This is of course not an easy matter. A number of
scholars of grammaticalization have cited category shift and semantic bleach­
ing as concomitant processes of grammaticalization. Some comments made
by an Indagen speaker in (49) when questioned about the variation between
clause-initial and preverbal bai are interesting because he mentions preverbal
baimbai as a possible variant. However, he does not attach any important
difference in meaning to the variants.
The Grammaticalization of Irrealis in Tok Pisin 409

(49) Baimbai, nogat, em nau liklik. Baimbai em i go, baimbai i kam. Em

baimbai i kam. Baimbai em. klostu bai. I gat kainkain mining i
stap. Bai baimbai ating wankain olsem. [IND87MISC9P].
'You only hear baimbai a little bit now, as in baimbai em i go,
baimbai i kam, em baimbai i kam. That's baimbai. It's just about
the same as bai. They have a similar sort of meaning. Bai and
baimbai are almost the same thing.'

Only a careful semantic analysis of the meanings encoded by baimbai

and bai in clause-initial and preverbal position from the earliest Pacific Jargon
English records to the present day will shed light on what has happened. I will
do this with reference to more general claims made by Bybee et al. (1991)
about the grammaticalization of future markers in language to see to what
extent the changes postulated for Tok Pisin fit their predictions.

4. Semantic development of bai

Bybee et al. (1991) make two strong claims relating to the development of
futures:
(1) Futures in all languages develop from a small set of lexical sources
and all future morphemes from a given source go through similar stages of
development.
(2) The semantic change is accompanied by formal reduction.
With regard to the first point, they make a more general distinction
between aspectual and non-aspectual futures because they say that these two
groups develop in entirely different ways. Tok Pisin bai is an example of a
non-aspectual future. 7 However, they identify four routes of development,
one of which can be applied to the case of bai in Tok Pisin; namely, cases
where temporal adverbs indicating a time after the moment of speech or a
reference to time come to encode the future. Since the other three routes are
more commonly attested in the sample which Bybee et al. (1991) looked at,
they do not exemplify the fourth one or discuss it in detail. Thus, the extent to
which Tok Pisin matches their claims in all the precise details will have to
await investigation of other languages in which temporal adverbs have devel­
oped into tense markers, such as the case of the Kru languages studied by
Márchese (1986).
410 Suzanne Romaine

Whether the second claim is substantiated by the developments in Tok

Pisin depends on the extent to which it is possible to determine what meanings
are encoded by baimbai, both synchronically and diachronically, particularly
when it appears in preverbal position. It is clear from the data that the reduced
form bai has become primarily an irrealis marker in preverbal position. To
demonstrate this, I will first look at the semantic development of baimbai and
bai in more detail.

4.1 Degree of semantic grammaticalization

Bybee et al. (1991:20) identify prediction as the protypical and defining use
for future grams, although more than half of the future grams in their data base
also had other uses. Examples (10) through (12) in Tok Pisin have shown a
similar range of uses other than prediction as components of the meaning of
bai. Bybee et al. (1991) take these as clues to the semantic development of
futures, telling us where the grams come from and how far they have ad­
vanced in their development. Other meanings such as imperative are typical
of later stages. The original meaning of a gram also determines the range of
use it will have at later stages.
There is evidence to suggest that in Pacific Jargon English and its
immediate descendents that baimbai already encoded an irrealis distinction. A
small corpus of 48 examples of baimbai collected from an on-going survey of
19th century textual records on Pacific Pidgin English was examined to see
what meanings were expressed. Out of a total of 48, only 8 were pre verbal and
these all encoded the meaning of later, or remote future, or prediction in the
sense of Bybee et al. (1991), as in (13). Of the cases of clause-initial baimbai,
most of these (N = 24) too encoded remote futures, but others encoded results
or later sequences of events, some of which can be understood as warnings
and threats, as they still do in modern Tok Pisin, as in (50).

(50) Bai i go antap tru tumach, baimbai em i blek, olsem blek.

[IND87MISC1]
'If it goes too far on top, it gets black, like black.
The existence of preverbal baimbai with future meaning indicates that
the preverbal slot was already becoming specialized for grammatical markers
some time ago. Eventually, the preferred option was for bai to be used in
The Grammaticalization of Irrealis in Tok Pisin 411

preverbal position with phonological reduction when it expressed certain

particular meanings. I will show now in more detail how this was accom­
plished.
Bybee et al. (1991:19) note that future marking is found primarily in
main rather than subordinate clauses because prediction is a type of assertion.
The same is true in Tok Pisin, as can be seen in examples like (51) and (52).
Sankoff and Laberge (1980:204-5) found that independent clauses in condi­
tional constructions appear to take bai obligatorily, which agrees with the
claim made by Bybee et al. (1991). Only after a long period of development
do futures come to be used obligatorily in subordinate clauses. There are,
however, some examples where bai does occur in subordinate clauses, as in
(53). In (54) both clauses take bai, although there is no explicit marking of
subordination.

(51) Bed ia bai bruk na em bai kam gen. [UTIFI]

T h e bed will break and he will come again',
(52) Sapos neks yia yutupla kam bek, mipla bai kam lukim yutupla.
[IND86MISCBW]
If you both return next year, we will come to see you.'
(53) Sapos draipla wulf ia bai kam brukim haus blol em no inap bruk.
[UT1M3]
'If the big wolf comes to break down their house, it won't break.'
(54) Nais tru, b'ol singim, b'ol man krai. [KUS6FMK]
'It's very nice. Whenever they sing it [the Australian national
anthem], everyone cries.'
There are always more cases where bai is preverbal in main clauses than
it is clause-initial (with the exception of Rempi), as can be seen in the data in
Table 6. This adds support to the idea that preverbal bai is a marker of future.
The implication of these calculations is that (bam)bai was not originally a
'future', or rather, a tense/aspect marker, but developed its future meaning in

Table 6. Number of preverbal and clause-initial bai in subordinate vs. main clauses
Bahor Kusbau Rempi Erima Lae Indagen
PV CI PV CI PV CI PV CI PV CI PV CI
main 66 34 55 45 47 53 62 38 63 37 79 21
subord 47 53 40 60 40 60 60 40 66 34 76 24
412 Suzanne Romaine

main clauses. The data in Table 7, which I will discuss shortly, support this
too, in that the most frequent meanings in the preverbal slot in main clauses
are the tense/aspect ones of future, habitual and sequence.
Before looking at Table 7 in more detail, however, we need to look more
closely at the meanings encoded by bai in preverbal and clause-initial posi­
tions. The relevant semantic categories appear to be the following: habitual
(aspect), sequence, result, imperative/suggestion and intention/purpose. I noted
earlier that some linguists such as Roberts (1990) categorize imperatives as
irrealis. Habituais also fall within the scope of irrealis for Givón (1984) since
they make reference to generics rather than specifics and are in this sense non-
referential. The merger between, or use of the same marker, for habituais and
irrealis apparently occurs in a number of creoles (see Taylor 1971). Bybee et
al. (1991) observe a connection between futurity and imperatives which arises
from situations in which a speaker makes a prediction about the addressee,
e.g. you're going to wash to hands before dinner.
Bybee et al. (1991:26) also say that developing grams often retain traces
of their earlier lexical meaning. Certainly, bai has retained traces of the earlier
meanings of baimbai. It has other uses too, such as imperative. This supports
their proposal that the use of future for imperative develops out of the
prediction sense and is therefore a late use (Bybee et al. 1991:28). I have
already given some examples of the other modal and aspectual meanings of
bai in (10) to (12). Some additional examples of bai used to encode result/
sequence/habitual in the past are in (55) to (56). Notice that both preverbal
and clause-initial bai can encode the past habitual. Example (57) shows the
use of bai in sequencing (cf. also (5)), while (58) shows its use in suggestions or
imperatives. Example (59) illustrates the intention/prediction meaning of bai.

Table 7. Meanings encoded by preverbal and clause-initial bai in main and subordinate
clauses (F/H/S = future, habitual, sequence; R = result; I/It/S =imperative/
intention/suggestion)
Kusbau Rempi Erima Lae Indagen
PV CI PV CI PV CI PV CI PV CI
M F/H/S 88 12 77 23 89 11 91 9 94 6
A R 33 67 33 67 25 75 25 75 100 0
I I/It/S 18 82 1 99 23 77 27 73 55 45
N
S F/H/S 38 62 47 53 65 35 79 21 84 16
U R 53 45 33 67 32 68 51 49 69 31
B I/It/S 19 81 20 80 34 66 18 82 48 52
The Grammaticalization of Irrealis in Tok Pisin 413

(55) Tasol man em putim kain hat olsem ol men bai respectim ol moa
yet. Em pasin bilong tumbuna. [IND86MISC1]
'However, if men wore this kind of hat, the women respected them
even more. That was the custom of our ancestors'
(56) Ol mipla save ron shutim ol sampla fis ol bai slip antap lo rif
[REM4MAM]
'We habitually catch the fish that sleep on the reef.'
(57) Bihain bai pepa bai i kam insait long ... komuniti
gavman na bai mipela lukim wanem tok bai i kam insait.
[IND87MISC1]
Then the document will come to the community and we will look
to see what it says.'
(58) Desla mangi ia tok: "Bai yumi go long bus". [REM2FBA]
'This boy said: "Let's go to the bush'.
(59) Ating bai mi go long maket nau. [REM1MFM]
T think I will go to the market now.'
Clauses which express imperatives, suggestions, hypothetical results and
intentions/purposes show the greatest clause-initial use of bai, while clauses
with future, habitual and sequence show the strongest tendency toward pre-
verbal bai. Thus, preverbal bai tends to be used where there is a good chance
of realization of the activity of the main verb modified by bai, while clause-
initial bai is preferred for cases where there is a dependency on some condi­
tion or state, e.g. in clauses introduced by sapos. Contrast, for instance, the
hypothetical or irrealis condition in (60), where bai is clause-initial, with (12),
where the events have a greater likelihood of occurrence and bai expresses
predictions and intentions.

(60) I no olsem Lae na bai yutupla tingting planti [raskol]. Em peles, i

no gat raskol, bai sidaun isi. Sapos long taun, yes, bai yutupla
wari. [IND87MISC2]
'It isn't like Lae where you both would think there would be a lot of
rascals [gangs of thieves]. This is the village, there aren't rascals,
you can rest easy. If you were in town, yes, you'd have to worry.'
It is by now evident that semantic factors are more important than
syntactic ones in determining the position of bai. Different varieties are at
different stages of development with regard to the shift towards preverbal bai
being used to encode the future. It is instructive to look at variation in one
414 Suzanne Romaine

syntactic environment, namely, long clauses. In example (61) from Bahor bai
is clause-initial in a subordinate clause introduced by long, whereas in (62)
from Indagen both instances of bai are preverbal. In (62) the first use of bai
encodes a prediction or future result, whereas the second one expresses
purpose or intention, as does (61). Both of these meanings now tend to be
expressed increasingly by preverbal bai in Indagen. In the other areas clause-
initial bai is still used. In (63), also from Bahor, we see a case of preverbal bai
used in a long clause because here the meaning is habitual result.
(61) Em tokim pikini bilong en lo [a reduced form of long] tumoro b 'ol
go chekimpig. [BAH5FSK]
'He told his son that tomorrow they would go to check on the pig.'
(62) Sapos sios i karnap bikpela, em bai gim graun long ol bai sanapim
aus lotu. [IND86MISCSN]
'If the church [membership] gets big, it will give land for them to
build a church.'
(63) Olgeta taim em sa singaut long mani long ol man bai helpim em.
[BAH1FBS]
'He is always crying out for money for people to help him'.
The children from Kusbau, Bahor, Rempi, Erima, Lae and the Indagen
adults all generally use the same system. The Indagen children, who are the
most advanced in their use of preverbal bai, behave differently. In cases
where the other children have primarily clause-initial bai, Indagen usage is
split roughly 50/50 between clause-initial and preverbal bai. Thus, there are
no environments where clause-initial bai is favored. Clauses expressing inten­
tion are on the point of becoming environments for preverbal bai. This
indicates the existence of a hierarchy of environments for change, as in Figure
1. The change begins in main clauses and is less advanced in subordinate
ones.
Table 8 shows the differences among four areas in more detail with
reference to the age grading dimension. The results show that Indagen clearly
leads the other areas in its use of preverbal bai in marking futures as well as
other modalities such as intentions, suggestions, etc. There is also age-grading
such that younger speakers are introducing more preverbal bai. The difference
between the Indagen younger (Grades 1-4) and older (Grade 5+) speakers is
considerable with respect to the use of preverbal bai in the marking of
suggestions, intentions, etc. The adults and older children (i.e. from Grade 5
The Grammaticalization of Irrealis in Tok Pisin 415

Direction of change towards more preverbal

Area: Bahor Kusbau Rempi Erima Lae Indagen
PREVERBAL F/H/S F/H/S F/H/S F/H/S F/H/S/R F/H/S/R/I
CLAUSE-INITIAL I/It I/It I/It I/It I/It It

Figure 1. Hierarchy of change from clause-initial to preverbal bai in main clauses

(F = future; H = habitual; S = sequence; R = result; I = imperative; It = intention)

Table 8. Regional differences in the use of bai

Percent of cases where bai encodes future/habitual/result
Grade/Age 1-2 3-4 5-6 30 yrs 70 yrs
PV CI PV CI PV CI PV CI PV CI
Indagen 91 9 89 11 84 16 94 6 72 28
Lae 75 25 91 9 81 19
Erima 72 28 71 29 77 23
Rempi 65 35 51 49 53 47
Percent of cases where bai encodes intention, suggestion, etc.
PV CI PV CI PV CI PV CI PV CI
Indagen 62 38 54 46 44 56 5 95 20 80
Lae 29 71 18 82 16 84
Erima 23 77 42 58 15 85
Rempi 19 81 8 92 6 94

onward) in Indagen have the older system which is more like Lae in that the
meanings of suggestion, imperative and intention are still encoded primarily
by clause-initial bai. It is the youngest children in Grades 1-4 whose usage is
most advanced. 8
I now return to the predictions Bybee et al. make about the development
of futures. They (1991:29) classify future grams into four semantic ages based
on the uses they have in addition to their future use, as shown in Table 9. They
classify grams which develop from temporal adverbs in the same schema as
movement-derived futures. They put all the non-aspectual future grams into
this schema since they illustrate a different path of development from the
aspectual futures. It is also clear from Table 9 that some of the postulated
meanings for FUTAGE 1 and 4 overlap with those of modality, i.e. possibility,
obligation, desire, probability, etc. Table 9 can also be compared with Figures
1, 2 and 3 in Bybee et al. (1994: Ch. 6), which implicate the future in
grammaticalization paths towards various modalities.
416 Suzanne Romaine

Table 9. Meaning components for each FUTAGE (from Bybee et al. 1991:33, Table 2)
FUTAGE 1 FUTAGE 2 FUTAGE 3 FUTAGE 4
obligation intention future probability
desire root poss possibility
ability andative imperative
venitive use in complements
immediate future use in protases

Bybee et al. (1991:57) explicitly list baimbai in their sample of future

grams and categorize it as a FUTAGE 3 (relying on Hall's 1943 texts). Although
they do not mention bai, presumbly it would belong to FUTAGE 4, on the
assumption that it is a later development from baimbai, and it encodes, as I
have shown, the meanings of possibility, imperative, etc. However, it still
preserves the meaning of intention and immediate future, so it has not com­
pletely lost these meanings, which Bybee et al. (1991) classify as belonging to
Stage 2, and therefore younger (see further in 4.3.).

4.2 Degree of grammaticalization and position

Bybee et al. (1991) also make some predictions about the form and position­
ing of grams in relation to the degree of grammaticalization they have under­
gone. They (1991:33) note that positioning of affixes is governed by the
source constructions from which the affixes arose diachronically, which in
turn is governed by general typological features of the language. They predict
a form/meaning correlation reflecting degree of grammaticalization. Grams
that are older and have undergone more development tend to be closer to the
stem, more fused and shorter or more reduced in segmental material (Bybee et
al. 1991:33). Bickerton (1981) and others (e.g. Givón 1982) have made much
of the tendency for creoles to encode tense, mood and aspect by means of
preverbal particles. Bickerton's claim is that there is a creole prototype. The
use of particles rather than bound morphology clearly reflects the influence of
typology since pidgins and creoles generally dispense with morphology.
As far as Tok Pisin is concerned, the preverbal slot is not the only one
reserved for grammatical markers. The tense, mood and aspect markers, save,
laik, ken, bin, etc. all occur preverbally, but the completive marker pinis is
almost always postverbal. By comparison, in Torres Strait Creole, for in­
stance, all grammatical markers precede the verb.
The Grammaticalization of Irrealis in Tok Pisin All

Bybee et al. (1991:34) also propose three measures of the extent of

grammaticalization of form: fusion, dependence and shortness. If we assume
that preverbal bai is closer to the verb stem and thus more fused, then by these
criteria, preverbal bai is the more recent form, while clause-initial bai then
would be less so. Thus, preverbal bai should be more grammaticalized than
clause-initial bai. Without going into the details of their measures, I will note
briefly how Tok Pisin bai and baimbai compare on the three parameters. The
results are shown in Table 10.
Using five measures of the degree of fusion between a gram and the verb,
Bybee et al. (1991) calculate for each gram a value ranging from 0 to 9 by
assigning numerical values to these five variables. Higher scores indicate a
greater degree of grammaticalization. Only preverbal bai shows any degree of
grammaticalization. It is less grammaticalized than English -ed, for example,
as in walked, which receives a score of 8, but more than will, which receives 0.
The dependence scale, also ranging from 0 to 9, measures the extent to
which the gram is losing its autonomy independently of whether it is fusing
with the verb. They believe that lack of fusion is usually due to the gram's
position. This applies to bai since in clause-initial position, it is losing its
autonomy, but not fusing with the verb. The four criteria depend on the
assumption that as a gram becomes more dependent, it develops more phono­
logical and morphological allomorphy and loses stress.
As is evident from examples like (61), there are a number of allomorphs
of bai, e.g. bai, b' and ba, but these are phonetically conditioned. There is also
loss of stress, but it is not obligatory. Only the preverbal and clause-initial bai
show any degree of grammaticalization on this scale. By comparison, English
-ed has a score of 7, and will, a score of 6.
The shortness scale measures the extent of segmental reduction of
a gram by starting with a score of 10 and subtracting points for each segment
of the gram. A consonant has a value of -1, a vowel, -2 and a long vowel, -3.1
have counted diphthongs as long vowels on the assumption that [e:], for
instance is equivalent to [ee]. Thus, baimbai scores 3, and bai, 8. Again,

Table 10. Degree of grammaticalization of baimbai and bai

fusion dependence shortness total
FUTAGE 1 baimbai 0 0 3 3
2 bai CI 0 1 8 9
3 bai PV 1 1 8 10
418 Suzanne Romaine

comparing these scores to those for English grams, bai has the same score as
will for shortness, but both score less than -ed.
By correlating the scores achieved on these three measures of fusion,
dependence and shortness, Bybee et al. (1991:39) predict that forms with
higher FUTAGES will have significantly higher values and they argue for a
reliable trend in their data towards higher measures of formal grammaticaliza-
tion as one moves from FUTAGE 1 through FUTAGE 4. It is somewhat problem­
atic to apply these measures to a language with little morphology since the
measures are best suited to cases which display highly developed phonologi­
cal and morphological conditioning of allomorphs. Nevertheless, we can see
in Table 10 that preverbal bai has the greatest degree of grammaticalization.
This is consistent with the fact that preverbal bai is attested only in recent
decades.
Bybee et al. (1991:41) conclude that there is a significant correlation
between the semantic properties hypothesized to belong to older grams and
the formal properties which accompany grammaticalization. There are several
biases in their method, which I have no space to comment on here. One,
however, which they recognize and attempt to rectify, is that typology may
interfere with the correlations. This affects Tok Pisin since it is a young
language by comparison with languages like English. Older languages may be
expected to have a greater number of older and therefore more highly fused
grams than languages with shorter time depth. Inflecting languages on the
whole may be expected to show more fusion, but this fusion may not indicate
a greater age of grams.

4.3 Evaluation of the claims made by Bybee et al. (1991) in relation to

baimbai and bai

Having now traced both the syntactic and semantic development from
baimbai to bai in Tok Pisin, I can now evaluate the relevance of the claims
made by Bybee et al. (1991) about the evolution of futures. I have no reason to
doubt their statement that all futures develop from a limited set of lexical
sources and go through similar stages of development. However, in the case
of Tok Pisin, it is clear that formal reduction was not an absolute prerequisite
for the development of future meaning. In fact, Bybee et al. (1991:42) found a
deviation in FUTAGE 4 for both the measures of dependence and shortness.
Upon examining their correlations in Table 3 and 4 more closely, it appears
The Grammaticalization of Irrealis in Tok Pisin 419

that as far as shortness is concerned, FUTAGE 2 shows a slightly lower score

than FUTAGE 1, and FUTAGE 3, a lower score than FUTAGE 4 (see Bybee et al.
1991:40). Thus, shortness is the least reliable of their measures of degree of
formal grammaticalization. If the case of Tok Pisin is indicative of other
cases, where several forms overlap in terms of their range of uses, it is not
surprising to find this result. While there is evidence of a long term trend to
specialize preverbal bai for the meanings typical of FUTAGES 3 and 4, at the
moment grammaticalization is very much in progress and 'old' uses of
baimbai still survive. As far as Sankoff and Laberge's hypothesis of a three
stage development is concerned, it would seem that there is no evidence for
reduction in situ of preverbal baimbai. By dint of the methods used by Bybee
et al. (1991), in particular, their reliance on published sources rather than
detailed semantic analyses of a large corpus of data from the languages
concerned, it is inevitable that their results are static rather than dynamic and
reflect the products of grammaticalization rather than the on-going process.
Thus, as I indicated earlier, there is reason to believe that the FUTAGE
assigned to baimbai is inaccurate. Bybee et al. (1991:27) say that FUTAGE 3
contains "all those grams that have future as their only meaning. This stage
represents the period during which all agent-oriented uses have been eroded,
and no further uses have yet developed". They do note, however, that their
classification may be inaccurate since they relied on reference grammars
which may have been incomplete. Within their schema we cannot reconstruct
an earlier stage of meaning for baimbai. It is clear, however, from the
examples I have given and my discussion of the uses of baimbai in Pacific
Jargon English as well as in modern Tok Pisin that its meaning is not
restricted to the future understood in the narrow sense of a marker indicating
time relations.
I would like to propose the schema in Table 11 for classifying clause-
initial and preverbal baimbai and bai with reference to the FUTAGES proposed
by Bybee et al. (1991). The semantic scope of baimbai covers at least FUTAGES
2 and 3, while bai extends from FUTAGE 2 through 4. The differences among
these options are quantitative, with preverbal bai tending to be the more
grammaticalized.
We can now make some further observations about the connection be­
tween irrealis and futurity and their grammaticalization channels. To begin
with, it is clear that the view of grammatical categories adopted by Bybee et
al. (1991) is one in which notions such as tense, mood and aspect are viewed
 420 Suzanne Romaine

Table 11. FUTAGES for baimbai and bai

FUTAGE 1 FUTAGE 2 FUTAGE 3 FUTAGE 4

obligation intention future probability
desire root poss possibility
ability andative imperative
venitive use in complements
immediate future use in protases
←CI BAIMBAI PV→ ?
←CI BAI PV→→

as sets of diachronically related functions. I find this perspective insightful

and would add to it the principle that there will be some overlap between
functions. Bybee et al. (1991:26) believe that the use of modalities to state
intentions is an important link in the chain of developments leading to the
grammaticalization of futures. Once the intention use becomes established, a
further inference can be made to the effect that one can predict the subject will
do what he/she intends to do (compare example (10) discussed earlier). Some
of the more modern uses of preverbal baimbai in Tok Pisin are of this
character and they may have provided the transition to the use of reduced
forms in preverbal position. Compare some of the examples I cited earlier
from Nupela Testamen (37-40), and (42). Bybee et al. (1994: Ch. 6:54)
comment that while intention can develop into prediction, which is the
criterial use for futures, and from there come to be used for purpose clauses, it
is apparently not necessary for the future to serve as an intermediate step
between intention and purpose.
It is also clear from the examples I have given that bai has evolved
beyond FUTAGE 3 because it expresses imperatives, and can also be used in
subordinate clauses, e.g. in complement clauses (i.e. long complements) and
in the protases of conditional sentences. While formal reduction doesn't seem
to be a prerequisite for Stage 3, it isn't clear that it is essential for FUTAGE 4
either. Compare (41), where baimbai appears in a long complement clause.
I don't have any spoken examples of baimbai in long clauses, which is
why I have put a question mark under FUTAGE 4 for baimbai, and (41) is odd in
certain respects. Note, for instance, the repetition of long. It could be that this
development is restricted to media Tok Pisin. The newspaper also uses
preverbal bai in complement clauses, as can be seen in (64). While some
issues appear to use mainly bai, both preverbally and clause-initially, other
The Grammaticalization of Irrealis in Tok Pisin 421

issues use mainly baimbai. I am currently doing a diachronic study of bai and
other syntactic features in media Tok Pisin, which I hope will shed light on
this usage.
(64) Bihain oli mekim pinis general survey o lukluk raun oli tingting
bilong mekim ol rot bai olsem bilong bung wantaim long ol ples
balus na bris bilong mekim isis moa wakabaut long olgeta hap long
Territory. (Nius bilong Yumi. October 1962, Vol. 4. No. 12, p. 4).
'After they complete the general survey they are thinking of mak­
ing the roads connect with the airports and bridges in order to
facilitate travel within all parts of the Territory.'
Another problem which may be specific to the Tok Pisin data or have
more general relevance for the predictions made by Bybee et al. (1991) once
we have more data from other cases where temporal adverbs have become
future markers is the classification of imperatives under FUTAGE 4. A strict
chronological progression from FUTAGE 1 to 4 would imply that more cases of
preverbal than clause-initial bai should be used with the meaning of impera­
tive. As I showed above, however, clauses which express imperatives, sug­
gestions, hypothetical results and intentions/purposes show the greatest
clause-initial use of bai.

5. Conclusion

From a cross-linguistic perspective it is obvious that neither the process nor

specific chain of grammaticalization transforming a sentence-initial temporal
adverb into a preverbal tense particle is unique or necessary to pidgin and
creole languages. Marchese (1986:254-7), for instance, notes the develop­
ment of tense auxiliaries from time adverbs in Kru languages. Some Kru tense
markers are clearly reduced forms of time adverbs and now have the distribu­
tional properties of auxiliaries rather than adverbs and can even cooccur in the
same clause with the corresponding adverb, e.g. a general past tense is derived
from the corresponding adverb meaning 'yesterday' by semantic extension.
This case bears further examination within the framework developed by
Bybee et al.
Creoles, however, appear to be no different from 'natural' languages with
respect to the routes of development they follow in the grammaticalization of
422 Suzanne Romaine

futures. The most common source for futures in pidgins and creoles is to use a
verb of movement, e.g. go in most English-based pidgins and creoles. Notable
exceptions occur among some of the Portuguese and Spanish-based Atlantic
Creoles such as Papiamentu and Cape Verde Creole, which use lo [<Spanish/
Portuguese luegollogo 'later']. In Kristang (Malacca Creole Portuguese),
which has logu, the future marker can occur both within and outside the verb
phrase.
Although I have already demonstrated that creolization is not a trigger for
the grammaticalization of bai as a future or irrealis marker in Tok Pisin, I
return now to this issue in my concluding remarks as it was my point of
departure. I have already argued that creolization per se, in so far as it is
mainly an urban phenomenon, does not appear to be responsible for the shift
of bai to preverbal position.
There are some further relevant parallels when we examine comparative
evidence from other pidgins and creoles, which weaken the link between
creolization and the grammaticalization of preverbal tense particles. The most
interesting and relevant case is that of Solomon Islands Pidgin. There nativi-
zation does not seem to have affected the positioning of bae. Jourdan (1985)
found that rural adults had the highest incidence of preverbal bae; however,
these accounted for only 11% of the occurrences of bae. Thus, preverbal bae
is not by any means as frequent a variant for younger speakers of Solomons
Island Pijin as it is for those of Tok Pisin. Monolingual speakers hardly ever
use preverbal bae. Thus, for both younger and older speakers in all areas,
clause-initial bae is still the most frequent variant. Although Crowley
(1990:496;270) suggests that Bislama in Vanuatu was probably acquiring
native speakers in small numbers for most of its history, he says that the
placement of bae probably parallels more closely the pattern found in Solo­
mon Islands Pijin than that in Tok Pisin.
Another relevant case from the Pacific is Hawai'i English Creole. In the
pidgin stage, temporal adverbs occur sentence-externally, either initially or
finally. When the creole developed, neither bambai nor pau (compare Tok
Pisin pinis), the completive marker, underwent any change of meaning, nor
were they incorporated into the auxiliary. A closer comparison of Tok Pisin
with Hawai'i Pidgin/Creole English, however, reveals a difference in out­
come which is probably due to the longer time period for stabilization in Tok
Pisin and in the Melanesian Pidgins in general by comparison to creolization
in Hawai'i, which was more abrupt. Although Hawai'i Pidgin English and the
 The Grammaticalization of Irrealis in Tok Pisin 423

Melanesian Pidgins have both baimbai and go, in Hawai'i, bambai never
became regularized as a future tense marker.
Further evidence to support the lack of connection between creolization
and grammaticalization of preverbal tense markers is found in Sango. Despite
the fact that there is now a second generation of creole speakers, the future
marker fade (identical to the adverb fade 'right now') always occurs first or
early in the sentence and is never phonologically reduced (Samarin 1975).
Nor has its use become obligatory. An example is given in (65).
(65) fade tongana kete nginza ti mo ake, mo goe vo a-Nivaquine quoi
If small money of you is, you go buy Nivaquine.
'If you should have a little money, you ought to buy some Niva­
quine (anti-malarial medicine).'
In conclusion then, I have argued that there is no evidence to indicate that
creolization and/or urbanization is associated with the incorporation of tense
markers into the verb phrase in preverbal position in Tok Pisin. The existence
of preverbal baimbai with future meaning shows that the preverbal slot was
available for tense markers long before phonological reduction or the exist­
ence of a community of native speakers. I have also shown that the grammat­
icalization process follows in many respects the universal paths of
development for futures predicted by Bybee et al. (1991) at the same time as it
intersects with many of the key routes leading to modality.

FOOTNOTES:

* I would like to thank Joan Bybee and Gillian Sankoff for their helpful comments on an
earlier version of this article. I would also like to thank the directors of the Max-Planck-
Institut-für-Psycholinguistik, WJ.M. Levelt and Wolfgang Klein, for their support and
encouragement, and also Fiona Wright and Rod Mitchell, who assisted in the research at
various stages. I am also very grateful to the Papua New Guinea University of Technol­
ogy for providing a research base and to the Provincial Government of Morobe and
Madang Provinces for permission to conduct research. I am also grateful to Susan
Hockey of the Oxford University Computing Service for assistance with the Oxford
Concordance Programme used in the analysis of the data.
1. The coding system is to read as follows: BL1 = Bulolo 1986; INDI = Indagen 1986; UT1
= Unitech 1986, etc. F = female; M = male. The number indicates the grade. Where
initials appear and no grade is given, the speaker is an adult.
2. I am grateful to Philip Baker for providing me with some of the examples from German
New Guinea and Australia.
424 Suzanne Romaine

3. Schuchardt (1883) also notes the occurrence of by and by in Chinese Pidgin English and
quotes the following example, which interestingly contains a preverbal usage: my
by'mby catchee he 'I will get it.' Although Baker (1987:179) has attested this feature for
Chinese Pidgin English and other Pacific Pidgin Englishes he does not note any prever-
bal occurrences. He dismisses one instance recorded by Leland (1876:110) as
unauthentic because the author never visited the Far East and composed all his examples
in London, e.g. he fedders by'mby stlate all-same nother hin 'Its feathers will become
straight like those of any other hen.'
4. Detzner claims to have lived among bush speakers of Tok Pisin for several years. He
continued fighting the war in the remote bush withough surrendering for four years after
Rabaul was firmly in the hands of Britain and Australia. He quotes numerous examples
of Tok Pisin as used by the police force. However, the use of the copula be and several
other unexpected non-standard features suggests that we are dealing with a somewhat
suspect literary form of Tok Pisin. After Detzner was repatriated, he published a
disclaimer of his book, Vier Jarhen unter Kannibalen. He said (1932:307-8) said the
book contained a number of misrepresentations concerning his journeys in New Guinea.
His travels were scientific in part only and were to be taken as primarily a fictional
account (see Biskup 1968).
5. Borchardt was a Catholic missionary on Manus Island. His language is clearly influ­
enced by the Rabaul tradition. His examples are fabricated; most other examples have
clause-initial baimbai. His Guidance for learning the Tok Boi was translated from the
German original in 1930.
6. I owe this example to John Z'graggen.
7. Grams that have 'future' as one of their meanings may also more typically develop from
auxiliary constructions which have the meanings of 'desire', 'obligation' or 'movement
towards a goal'. These routes are also illustrated by other ways of expressing future in
Tok Pisin, e.g. laik 'want/desire', go 'to go', and ken 'to be able'. Bybee et al. (1991:55)
classify laik as a FUTAGE 1 gram expressing an agent-oriented modality.
8. There may also be a developmental aspect to the progression of this change from clause-
initial to preverbal bai. The data partly reflect the fact that older children and adults use
more clauses expressing the meanings of intention, etc. and thus, have more environ­
ments for clause-initial bai to occur than do the younger speakers. The higher rates of
preverbal bai usage could indicate that the temporal meanings of bai are learned earlier
than the others, at least in some areas. The youngest children in both Lae and Indagen in
fact use fewer modal clauses than the younger children in other areas. However, this
tendency appears at odds with claims made about the priority of aspect over tense in
language acquisition more generally (see Romaine 1988:Chapter 7 for an evaluation of
these claims in relation to creoles).
The Grammaticalization of Irrealis in Tok Pisin 425

REFERENCES

Baker, P. 1987. "The Historical Developments in Chinese Pidgin English and the Nature
of the Relationships between the Various Pidgin Englishes of the Pacific Region".
Journal of Pidgin and Creole Languages 2.149-62.
Bergmann, U., ed. 1982. Save Na Mekim. Lae: Liklik Buk Information Centre.
Bickerton, D. 1981. Roots of Language. Ann Arbor: Karoma.
Biskup, P. 1968. "Hermann Detzner: New Guinea's First Coast Watcher". Journal of the
Papua and New Guinea Society 2.5-21.
Borchardt, K. 1930. Anleitung zur Erlernung des Tok-Boi, mit Wörterbuch. Manus, mimeo.
Bybee, J., Pagliuca, W. and Perkins, R.D. 1991. "Back to the Future". Approaches to
Grammaticalization. ed. by E.C. Traugott and B. Heine. Vol. II, 17-58. Amsterdam:
John Benjamins.
Bybee, J.L., Pagliuca, W. and Perkins, R.D. 1994. The Evolution of Grammar: Tense,
Aspect and Modality in the Languages of the World. Ch. 6: Mood and Modality.
Chicago: University of Chicago.
Churchill, W. 1911. Beach-la-Mar, the Jargon or Trade Speech of the Western Pacific.
Washington,D.C: Carnegie Institution.
Crowley, T. 1990. Beach-La-Mar to Bislama: The Emergence of a National Language in
Vanuatu. Oxford: Oxford University Press.
Detzner, H. 1921. Vier Jahren unter Kannibalen. Berlin: A.Scherl.
Detzner, H. 1932. Zeitschrift der Gesellschaft für Erdkunde 1932.307-8.
Foley, W.A. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge
University Press.
Givón, T. 1982. "Tense-Aspect-Modality: The Creole Prototype and Beyond". Tense-
Aspect. Between Semantics and Pragmatics, ed. by P. Hopper, 114-63. Amsterdam:
John Benjamins.
Givón, T. 1984. Syntax. A Functional-Typological Approach. Introduction. Vol. 1. Amsterdam:
John Benjamins.
Hall, R.A. 1943. Melanesian Pidgin English: Grammar, Texts, Vocabulary. Baltimore:
Linguistic Society of America.
Hall, R.A. 1955. Hands Off Pidgin English. Sydney: Sydney and Melbourne Publishing
Company Pty, Ltd.
James, D. 1982. "Past Tense and the Hypothetical: a Cross-Linguistic Study". Studies in
Language 6. 375-403.
James, D. 1991. "Preterit Forms in Moose Cree as Markers of Tense, Aspect, and
Modality". International Journal of American Linguistics 57.281-297.
Jourdan, C. 1985. Sapos lumi Mitim lumi. Urbanization and Creolization in the Solomon
Islands. Ph.D. dissertation. Australian National University.
Kay, P. and Sankoff, G. 1974. "A Language Universals Approach to Pidgins and Creoles".
Pidgins and Creoles: Current Trends and Prospects, ed. by D. De Camp and I.
Hancock, 61-72. Washington,D.C: Georgetown University Press.
Keesing, R. 1988. Melanesian Pidgin and the Oceanic Substrate. Stanford: Stanford
University Press.
426 Suzanne Romaine

Labov, W. 1977. "On the Adequacy of Natural Languages: The Development of Tense".
Linguistic Agency University of Trier Paper No. 23. Series B.
Leland, C.G. 1876. Pidgin-English Sing Song. London: Trübner and Co.
Ligo, G. 1987. Lanwis long nyus. Introdaksen long Stadi blong Bislama: Buk Blong
Ridim. ed. by T. Crowley, 81-4. Suva, Fiji: Ekstensen Sevis, Yunivesiti blong Saot
Pasifik.
Liklik Katolik Baibel. 1934. Alexishafen: Catholic Mission Press.
Morchese, L. 1986. Tense/Aspect and the Development of Auxiliaries in Kru Languages.
SIL Publications in Linguistics, 78. Dallas: Summer Institute of Linguistics.
Markey, T.L. 1982. "Afrikaans: Creole or Non-Creole?" Zeitschrift für Dialektologie und
Linguistik IL. 169-207.
Mihalic, F. 1957. Grammar and Dictionary of Neo-Melanesian Pidgin. Westmead, NSW:
The Mission Press.
Mihalic, F. 1971. The Jacaranda Dictionary and Grammar of Melanesian Pidgin. Port
Moresby: The Jacaranda Press.
Mihalic, F. 1986a. Konstitusen Bilong Independen Kantri Papua Niugini. Boroko, PNG:
Word Publishing Company.
Mihalic, F. 1986b. Stail Buk bilong Wantok Niuspepa. [reprint].
Mühlhäusler, P. 1978. "Samoan Plantation Pidgin English and the Origin of New Guinea
Pidgin". Papers in Pidgin and Creole Linguistics No.1. Pacific Linguistics. Series A-
24.67-119. Canberra: Australian National University.
Mühlhäusler, P. 1986. Pidgin and Creole Linguistics. Oxford: Blackwell.
Murphy, J. 1966. The Book of Pidgin English. Brisbane: W.R. Smith and Paterson Pty.
Nupela Testamen bilong Bikpela Jisas Krais na Buk bilong Ol Sam. 1966. Port Moresby:
The Bible Society of Papua New Guinea.
Praed, Mrs. Campbell 1885. Australian Life: Black and White. London: Chapman and Hall.
Roberts, J.R. (1990). "Modality in Amele and other Papuan Languages". Journal of
Linguistics 26: 363-401.
Romaine, S. 1988. Pidgin and Creole Languages. London: Longman.
Romaine, S. 1989a. "Some Differences Between Spoken and Written Tok Pisin". English
World Wide 9.243-69.
Romaine, S. 1989b. "Change and Variation in the Lexicon of Tok Pisin in Papua New
Guinea". Synchronic and Diachronic Approaches to Linguistic Variation and Change.
ed. by T. Walsh, 268-80. Washington,DC: Georgetown University Press.
Romaine, S. 1990. "Substratum, Stabilization and Grammaticalization in Melanesian
Pidgin English". Pacific Studies 14: 79-85.
Romaine, S. 1993. "The Decline of Predicate Marking in Tok Pisin". The Atlantic Meets
the Pacific, ed. by F. Byrne and J. Holm, pp. 251-61. Amsterdam: John Benjamins.
Romaine, S. 1992. Language, Education and Development: Urban and Rural Tok Pisin in
Papua New Guinea. Oxford: Oxford University Press.
Romaine, S. and Wright, F. 1987. "Short Forms in Tok Pisin". Journal of Pidgin and
Creole Languages 2.64-7.
Samarin, W.J. 1975. "Historical, Ephemeral and Inevitable Verbal Categories". Paper
presented at the International Conference on Pidgins and Creoles, Honolulu, Hawai'i.
The Grammaticalization of Irrealis in Tok Pisin All

Sankoff, G. and Laberge, S. 1980. "On the Acquisition of Native Speakers by a Lan­
guage". The Social Life of Language. G. Sankoff. 195-211. Philadelphia: University of
Pennsylvania Press. [Reprinted from Kivung (1973) 6.32-47].
Schuchardt, H. 1883. "Kreolische Studien V: Über das Melaneso-englische". Sitzungs­
berichte der kaiserlichen Akademie der Wissenschaften zu Wien, Philosophisch-his­
torische Klasse, 105.151-61.
Steele, S. 1975. "Past and Irrealis: Just What Does it All Mean?" International Journal of
American Linguistics 41.200-217.
Taylor, D. 1971. "Grammatical and Lexical Affinities of Creoles". Pidginization and
Creolization of Languages. 293-6 ed. by D. Hymes. Cambridge: Cambridge University
Press.
Troy, J. 1985. Australian Aboriginal Contact with the English Language in New South
Wales: 1788-1845. B.A. thesis. University of Sydney.
Vogel, H. 1911. Forschungsreise im Bismarckarchipel. Hamburg.
 The Evaluative Function of the Spanish
Subjunctive

Patricia V. Lunn
Michigan State University

1. Introduction

Whether or not mood encodes realis/irrealis is surely an unanswerable ques­

tion in language universal terms (but see Chafe, Mithun, Romaine, this
volume and Roberts, ms, on language-specific cases), given the impossibility
of satisfactorily defining 'reality' and the interference of internally motivated
diachronic change. But the fact remains that speaker judgment and mood are
more overtly linked than are speaker judgment and tense, aspect or voice
(though such links can, of course, be inferred). Fleischman (1989) suggests
that evaluation strategies in epic narration carried out by tense-aspect can
provide a window onto world view and cultural presuppositions of narrators,
and I think that the same can be said for the window that the Subjunctive
provides onto the world view of speakers of Spanish.
This paper argues that, in contemporary Spanish, the mood system is a
device through which speakers can evaluate the information value of clauses.
Further, because speakers may take an evaluative stance with respect to any
proposition, the identification of the Subjunctive with syntactic subordination
is breaking down. This process is fed by metalinguistic use of the Subjunctive
in certain registers, by usage that reflects the indicative history of the -ra form
of the Past Subjunctive, and by the identification of the highly non-assertive-se
form of the Past Subjunctive with the auxiliary verb haber, an identification
which allows the -se form to appear wherever haber is used.
The Spanish Subjunctive is primarily expressive of epistemic modality.1
Numerous studies, following Bolinger (1968) and Terrell and Hooper (1974),
430 Patricia V. Lunn

have shown that the best synchronic description of the Spanish mood system
links choice of the Indicative (I) to assertion and choice of the Subjunctive (S)
to non-assertion. A proposition may be unworthy of assertion because the
speaker has doubts about its veracity:
(1) Dudo que sea (S) buena idea.
'I doubt that's a good idea'
or because the proposition is unrealized:
(2) Necesito que me devuelvas (S) ese libro.
'I need you to return that book to me.'
or because the proposition is presupposed:
(3) Me alegra que sepas (S) la verdad.
'I'm glad that you know the truth.'
Assertion and non-assertion are not the only options, however. Klein
(1980) has shown that the territory between assertion and non-assertion is
occupied by a cline of speaker attitudes, with choice of the Subjunctive
becoming more likely as speakers report diminished degrees of certainty with
respect to propositions.
Lunn (1989) has shown that Subjunctive-marked propositions, whether
true or false, share the semantic quality of low-information value. Low-
information value is an instantiation of what Sperber and Wilson (1986) term
low relevance. Sperber and Wilson argue that the expectation that communi­
cated propositions will modify and improve existing information is the motive
force in discourse comprehension: "Every act of ostensive communication
communicates the presumption of its own optimal relevance" (1986:158). In
Spanish, propositions that are less than optimally relevant, either because they
add nothing new to existing information or because they add nothing true to
existing information, are marked with the Subjunctive.
Subjunctivizable information can be characterized as not possessing
certain qualities, a suggestive characterization in view of the language-univer­
sal link between negation and irrealis modalities. 2 Potentially assertable
information must have two qualities: in the speaker's opinion, it must be both
reliable as to truth value and informative as to news value. Information that is
believed to be flawed in truth value or flawed in news value is unlikely to be
asserted.
Evaluative Function of the Spanish Subjunctive 431

This description accounts both for grammaticized uses of the morphol­

ogy, like those exemplified above, and for ungrammaticized uses. Clearly,
speakers of Spanish are making judgments about information quality when
they use the Subjunctive, and this means that they have broad power to use the
Subjunctive electively. This, in turn, results in an increasingly confusing
syntactic pattern of mood usage.

2. Complementary theme/rheme analyses

Gsell and Wandruszka (1986:84-87) use the theme/rheme framework to show

that presupposed information in comment sentences (like (3) above, in which
the main clause is a comment on a presupposed proposition) is thematic.
Thematicity is related not only to mood marking, but also to word order.
Variation in mood choice after el hecho de que 'the fact that' is illustra­
tive of the complexity of mood choice in Spanish. It is well known that either
Subjunctive or Indicative can be used after el hecho de que, though discus­
sions of this structure are often based on decontextualized or invented sen­
tences (Lipski 1978; Woehr 1975). In an effort to monitor the context from
which data were taken, Krakusin and Cedeño (1992), studied mood choice
after el hecho de que in a ten-year series of magazine columns written by the
political analyst and columnist Mariano Grondona. Their study confirmed
that mood choice was directly linked to information value. When information
had already been introduced in the column or when it was of such moment that
it could be assumed to be common knowledge, it was marked with the
Subjunctive. In contrast, propositions neither previously discussed nor uni­
versally known were marked with the Indicative.
Moreover, a word-order generalization that can be linked to a theme/
rheme analysis emerged from the data. Subjunctive-marked clauses appeared
in initial position, i.e., before the main verb. The use of the Past Subjunctive
(PS) in (4) is an example.
(4) El hecho de que en los cuatro el régimen democrático se impusiera
(PS) facilitaba la tarea.
T h e fact that in all four the democratic regime was imposed made
the task easier.' (Krakusin and Cedeño 1992:1290)
 432 Patricia V. Lunn

Clauses marked with the Indicative, in contrast, appeared after the main verb.
(5) Lang soslaya el hecho mayor de que las ciencias físicas están (I)
penetradas de visiones filosóficas del universo y de todo lo que se
encuentra en él.
'Lang overlooks the major fact that the physical sciences are shot
through with philosophical visions of the universe and of all that is
in it.' (Krakusin and Cedeño 1992:1291)
With respect to the linear distribution of theme and rheme, Contreras
(1976:26) established that "the highest ranking rheme occurs in final posi­
tion". Clearly, the information value of these factive clauses determines both
their mood marking and their position in the sentence. Guitart (1991) has
suggested a similar approach to the Spanish Subjunctive by distinguishing
between pragmatic presupposition, based on speaker assumptions about
hearer knowledge, and semantic presupposition, based on sentence-level truth
conditions.

3. Journalistic uses of the Subjunctive

In journalistic discourse, there are subgenres that are characterized by very

heavy Subjunctive use. In particular, the specialized press which addresses a
knowledgeable readership (e.g. gossip magazines, sports pages and bullfight­
ing news) uses the Past Subjunctive to mark information which can be
assumed to be known to assiduous readers. These publications have an
enormous readership, which means that these uses of the Subjunctive are
widely disseminated. Many native speakers, in response to questions about
such usage, say that it "sounds like something you'd hear on the news", which
reveals that this kind of Subjunctive marking of known information forms part
of the input for native speakers' grammars. The use of the Subjunctive
discussed below serves the discourse function of backgrounding old informa­
tion and the metalinguistic function of identifying certain styles of writing as
journalistic.

(6) Casi once años después de que el Bayern de Munich alcanzara

(PS) su tercera y última Copa de Europa, los vigentes campeones
de la RFA y España van a volver a enfrentarse en un encuentro
oficial. (La Vanguardia, 8.4.87)
Evaluative Function of the Spanish Subjunctive 433

'Almost eleven years after Munich Bayern won its third and last
European Cup, the current champions of the FRG and Spain are
going to meet one another in an official match.'
(7) La pareja, que se hiciera (PS) famosa por interpretar el papel de
marido y mujer en "El pájaro espino", es en la vida real un
matrimonio feliz. {Hola, 6.7.85)
'The couple, who became famous for their role as husband and
wife in "The Thorn Birds", is happily married in real life.'
In this kind of writing, judgments about information value are based on a
very broad definition of discourse context. The discourse contexts in which the
subjunctivized propositions of (6) and (7) are judged to constitute old infor­
mation include previous editions and may span a fairly long period of time.
The Subjunctive in (7) appears in a non-restrictive relative clause, where
the grammars say that it should not appear at all (Solé and Solé 1977:187-94).
Given that non-restrictive relative clauses facilitate the introduction of addi­
tional information, it is surprising that the non-assertive Subjunctive should
appear in them. This anomaly can be explained in relevance terms: journalis­
tic information is marked for low relevance if it is considered to be known to
a faithful readership. The flagging of known information serves to communi­
cate the additional message that writers assume readers to be knowledgeable;
this use of the Subjunctive serves not just to classify information but to create
solidarity between a publication and its readers.
In some cases, all that is necessary in order for a proposition to be
presupposed is prior appearance in a headline. Then, when reiterated in the
following text, the information is marked with the Subjunctive.

(8) a. Headline: La bandera que besó es la que, en su día, también

besó (I) el Rey don Juan Carlos, y bordó (I) su tatarabuela la
Reina doña María Cristina. {Hola, 26.10.85)
T h e flag that he kissed is the one that one day King Juan
Carlos also kissed and his great-great-grandmother Queen
María Cristina embroidered.'
b. Text: Y, al final, besó la bandera roja y gualda que hace
treinta años besara (PS) su padre el Rey y que un día bordara
(PS) su tatarabuela la Reina doña María Cristina.
' And, at the end, he kissed the red and gold flag that his father
the King had kissed thirty years ago, and that his great-great-
grandmother Queen María Cristina had once embroidered.'
434 Patricia V. Lunn

Another example of the link between mood choice and old/new informa­
tion appeared in a recent newspaper report on a notorious murder. Under the
headline Una maleta con un niño muerto 'Suitcase with dead child inside'
appeared the following sentence.
(9) Nadie en la ciudad, ni la policía, ni los vecinos, encontraron los
motivos por los que una mujer ... pudo (I) matar al menor de los
hijos de su vecina ...y enviara (PS) después el cuerpo a Madrid
dentro de una maleta. (El País, 22.5.92)
'No one in the city, on the police force, or in the neighborhood
understood the reasons why a woman could kill her neighbor's
youngest child and then send his body to Madrid in a suitcase.'
The first subordinate verb, pudo, is in the Indicative mood, while the verb
enviara, in the clause which repeats the headline information, is in the
Subjunctive. The uniquely gruesome disposal of the body was talked about all
over Spain, so it is impossible—and unnecessary—to decide whether the
factor controlling mood choice here is the global discourse context or the
format of the article in which the verb appeared. In the terms of category
theory, this is a case of multiple motivation for the explanation of the exten­
sion of a category. As Lakoff (1987:86) says, "where motivation is concerned,
the more kinds of motivation, the better."
The discourse context that governs mood choice in the journalistic data
cited in this section and in Section 2 can be as narrow as one article, or as
broad as a chronologically extended series of references to one topic. The use
of the Subjunctive to mark known information serves as acknowledgment of
the shared context of propositions, and in this sense it serves an interpersonal
discourse function.

4. Dialect variation

In many Latin American dialects, when a conditional si clause is introduced

by no sé T don't know', the following verb appears in the Subjunctive, e.g. No
sé si pueda (S) 'I don't know if I can'. Continental speakers find this usage
odd, but it is widespread in America. Denial of knowledge in combination
with the conditional conjunction tips the semantic balance in favor of the non-
assertive Subjunctive.
Evaluative Function of the Spanish Subjunctive 435

For many Latin American speakers, starting a sentence with no sé appar­

ently licenses use of the Subjunctive even in clauses in which it would not
normally appear. There are prosodic and syntactic differences in Spanish
between the complementizer que 'that' and the interrogative pronoun qué
'what', yet even embedded questions can contain a Subjunctive verb if they
are introduced by no sé. I recently heard a Latin American speaker say (10).
(10) No sé qué tal sea (S).
'I don't know what it's like.'
The standard form of this sentence is No sé qué tal es (I). In sentences like
(10), a semantic factor, speaker uncertainty about a proposition, determines
use of the Subjunctive.
The use of the Past Subjunctive to make a polite request for hearer
participation is current in Chile, to judge from dialogue in the short story
"Santelices" by the contemporary writer José Donoso.
(11) El departamento es lindo, de lujo, viera (PS) qué moderno.
(Donoso 1965:190)
'The apartment is pretty, deluxe, you should see how modern.'
I have also heard, in Ecuador, the verb conocer 'to get to know' used in this way:
(12) Conocieras la quinta de mis suegros
'You should visit my in-laws' farm.'
Note that viera and conocieras are not in subordinate clauses, nor are
they modal verbs (which are conventionally used in the Past Subjunctive to
form polite commands, as in ex. (20)).
In Guatemala, I have heard the tags ¿cómo te dijera/explicara (PS)?
'how can I tell you/explain to you', which are uttered when the speaker is
searching for words. In most dialects, the Present Indicative form of these
verbs would be used. Again, these Subjunctives appear outside of the ex­
pected subordinate-clause context.
In Mexico, speakers telescope the meaning of parecer 'to seem' and the
low assertiveness of the Past Subjunctive to yield pareciera, which is used in
main clauses in the following way.
(13) Pareciera (PS) agotado, destruido físicamente.
'It seemed as if he were worn out, physically destroyed.'
(Taibo 1976:142)
436 Patricia V. Lunn

There are many other Indicative/Subjunctive contrasts cited in Kany's

American-Spanish Syntax (1951:170-185) which attest to a general pattern of
the form Politeness:Subjunctive::Assertion:Indicative, and also to the height­
ened politeness or non-assertiveness of the Past versus the Present Subjunc­
tive. This pattern is the semantic source of the main-clause uses of the
Subjunctive discussed in this section.

5. Other sources of instability in the mood system

The Subjunctive is associated with modesty because of its non-assertive

meaning, and this meaning sometimes overrides the syntactic parameters of
Subjunctive usage. In adverb clauses, the Subjunctive can have the force of
downplaying the importance of a clause. For example, Camilo José Cela made
the following remark after being awarded the Nobel Prize for literature.
(14) El mundo no va a dejar de girar porque me hayan (S) dado el
Nobel.
'The world's not going to stop turning because they've given me
the Nobel Prize.'
Subjunctive adjective clauses usually modify unknown or nonexistent
nouns, as in No conozco a nadie que hable catalán T don't know anybody
who speaks Catalan'. In (15), though, the modified noun is real and known.
(15) Sólo he conocido a dos personas que tuvieran (PS) tanto o más
miedo que yo a los automóviles. (El País, 18.12.86)
T v e only known two people who were as afraid or more afraid
than I am of automobiles.'
The force of the Subjunctive here is to minimize the importance of these
few people. The semantically minimizing use of the Subjunctive has no clear
syntactic limits.
Another source of confusion is the Indicative Pluperfect origin of the -ra
Past Subjunctive form. Via a process discussed by Klein-Andreu (1991) and
Lunn and Cravens (1991), the functions of the Latin Pluperfect Indicative
were taken over by the Romance analytic pluperfect, leaving to the Latin form
the backgrounding functions that are associated with the subjunctive mood.
This is the source of the modern Spanish Past Subjunctive in -ra. (A similar
Evaluative Function of the Spanish Subjunctive 437

process now going on in Mandarin is described by Guo, this volume.) Parallel

to the dominant -ra form is a form in -se, which comes from a Latin Pluperfect
(Romance Imperfect) Subjunctive.
Evidence of the original usage of the -ra form was still extant in the
Romantic period in Spain (mid-nineteenth century), and intentional usage of
the form with pluperfect meaning is one of the precious characteristics of
Romantic prose (Wright 1932). In modern Spanish, such usage continues to
characterize highly formal or archaizing prose, such as the following clauses
from a much longer sentence in the short story "Syllabus" by the Spanish
author Juan Benet.
(16) Aquellas herméticas sentencias, cuyos secretos sentidos tantas
veces escaparan (PS) a la concurrencia, volverían a aclararse por
obra de su propia ironía,... (Benet 1981:285)
'Those hermetic aphorisms, whose secret meanings had so many
times escaped the audience, would once again become clear by dint
of his own irony,...'
The -ra form is so thoroughly identified with the subjunctive mood that
its marginally productive use as an indicative pluperfect has been generalized
to the unquestionably Subjunctive -se form.
(17) El conductor se encuentra atrapado en el interior de su coche
después de que éste se precipitase (PS) a un riachuelo. (Hola,
21.5.87)
'The driver is seen trapped inside his car after it had run into a
stream.'
The Subjunctive must be used after antes de que 'before', and it is
possible that the appearance of the Subjunctive in cases like (17) is partly due
to the conceptual pairing of the opposites antes de que and después de que, as
happens also in French. The use of the Past Subjunctive after desde que
'since' raises the same issues.
(18) Varios meses han transcurrido desde que Carmen Harto y Gonzalo
de Borbón decidiesen (PS) separarse tras un fugaz y sorprendente
matrimonio. (Hola, 9.4.85)
'Several months have passed since Carmen Harto and Gonzalo de
Borbón decided to separate after a brief and surprising marriage.'
438 Patricia V. Lunn

In some cases, as in the one above, the Subjunctive may be triggered as

well by the journalistic assumption that a previous notorious event is well
known to the readership.
Sometimes, confirmation of the known status of a Subjunctive-marked
proposition is present in the text, as in (19).
(19) Al día siguiente de que Isabel Preysler iniciase (PS) su veraneo,
según informamos en la página 44 de este número,... (Hola, 17.8.85)
T h e day after Isabel Preysler had begun her summer vacation, as
we reported on page 44 of this issue...'

Clearly, it is not possible to separate the individual strands of motivation

for the use of Subjunctive morphology. Pragmatic and discourse motivation
exist alongside semantic motivation and lexical analogy.

6. The two forms of the Past Subjunctive in Spanish

Because of its indicative history, the -ra Past Subjunctive is used in ways that
the -se form usually is not. It is quite conventional to use the -ra form of
certain modal verbs in main clauses in order to make polite requests and
suggestions.
(20) Quisiera hacerle una pregunta.
T would like to ask you a question, if I may.'
The -se form sounds obsequious and pompous in this formula, and few
speakers report having heard it. It is crucial, however, that it is the exagger­
ated politeness of the -se form that is objectionable here. The -ra form is fully
identified with the Subjunctive mood, and the alternate Past Subjunctive -se
form is, at least in theory, substitutable for it.
The -ra form is also likely to be chosen for the 'old information' journal­
istic uses discussed in Section 5 above. The -se form is occasionally used in
this way, however, which shows that the criterion for substitutability is
morphological: speakers identify both the -ra and the -se forms as Subjunc­
tives. Examples (18) and (19) above illustrate the old information use of the
-se form.
If main-clause and pluperfect uses of the -ra form are classified as non-
Subjunctive, as most traditional grammars would have it, the difference
Evaluative Function of the Spanish Subjunctive 439

between the -ra and the -se forms is hard to describe. Gili y Gaya, for
example, says that "both forms are interchangeable as long as they are
Subjunctive" (1964:179). If, however, all of the -ra uses are considered
Subjunctive, then it is possible to make the generalization that the -se form
encodes less assertion than the -ra form both because it rarely appears in the
main clause and quasi-pluperfect environments of the -ra form, and because
of contrasts like that between (21) and (22).

(21) Quisiera que usted me aclarase (PS) algunas cosas. (Vázquez

Montalbán 1977:41)
'I'd like you to clear up a few things for me.'
(22) Quisiera que recordaras (PS) primero si ha sido cliente tuyo o si
puedes preguntárselo a alguna colega. (Vázquez Montalbán
1977:23)
T d like you to remember first if he's been a client of yours or if
you can ask a colleague about him.'
In these sentences (from dialogue in the novel La soledad del manager),
the -se form appears with the formal second-person pronoun usted, while the
-ra form appears with the familiar second-person pronoun tú.
The Past Subjunctive can be used in place of the Conditional (especially
with the verb haber, to be discussed in Section 7 below). It is usually the -ra
form that is so used, as exemplified in hubiera sido 'would have been' in this
advertisement.
(23) Esos sí que fueron tiempos difíciles. Noé necesitó un vehículo de
gran capacidad, aguantador, fuerte y sobre todo confiable. Y se lo
tuvo que fabricar él. Ahora hubiera (PS) sido diferente. ¡Pobre
Noé! En realidad nunca tuvo otra alternativa. Seguramente si
hubiese (PS) podido elegir, Volkswagen habría sido su otra opción.
'Those were sure hard times. Noah needed a large-capacity vehicle
that was durable, strong and above all reliable. And he had to make
it himself. Now things would have been different. Poor Noah! He
really didn't have any alternative. Surely, if he had been able to
choose, Volkswagen would have been his other option.'

In this text, the -se form is used in the protasis of a conditional sentence:
si hubiese podido 'if he had been able'. The contrastive value of the -se form
is thus exploited by having it appear in a subordinate clause with past time
440 Patricia V. Lunn

reference, while the -ra form appears in a main clause with present time
reference. This chain effect is often to be observed when, in a given context,
there are a number of syntactic slots for which degrees of modality may be
selected. Klein-Andreu (1986) explains the various possibilities in detail.
Lunn (1988) discusses other data which show the -se form to be a less
assertive variant of the -ra form.

7. Dialect variation in the use of the -se Past Subjunctive

The -se form is the minority form in all dialects, though its frequency varies a
great deal from dialect to dialect. In this section, -se usage in modern novels
from Peru, Argentina and Spain is examined. The novels were chosen because
their authors are famous for their ability to write prose that evokes the spoken
language.
Listening to Peruvian Spanish, one gets the impression that the -se form
is virtually unused in that dialect. This impression is borne out by the novel
Un mundo para Julius, written in 1970 by the Peruvian novelist Alfredo
Bryce Echenique. In this book, the -se form appears only 21 times in nearly
500 pages, which amounts to slightly less that 2% of all Past Subjunctives.
None of these -se forms appears in the Lima speech of the characters. The
verb hubiese, from the auxiliary haber 'to have', accounts for 10 of the 21
cases, or 48% of all -se verbs. Only one of the verbs in the -se form is a
transitive verb. Also, 8 of the -se forms, 38%, appear in negative clauses. In
Un mundo para Julius, residual -se usage can be described as linked with—
but not triggered by—negation and haber.
In the novel, the -se form is often used in the description of highly
unlikely or fanciful situations, which may or may not be marked with an
introductory irrealis conjunction.
(24) Muchas veces tropezó la chola con los mayordomos o con el
jardinero que yacían muertos alrededor de la carroza, para que
Julius, Jesse James o Gary Cooper según el día, pudiese (PS)
partir tranquilo a bañarse. (Bryce Echenique 1991:13)
'Often the Indian girl tripped over the servants or the gardener who
were lying dead around the coach, so that Julius, Jesse James or Gary
Cooper, whoever he was that day, could go off quietly to his bath.'
Evaluative Function of the Spanish Subjunctive 441

(25) Ni más ni menos que si estuviese (PS) reinventando el mundo.

(Bryce Echenique 1991:327)
'As if he were reinventing the world, no less.'
It is interesting to note that although as many as 75 pages may go by
without a -se form, on five occasions the form is used once and then again on
a directly or closely following page, giving the impression that use of the form
activates some schema which licenses its reuse, but which once forgotten may
lie dormant.
For Argentine Spanish, DeMello (1993:240) reports that -se usage is
relatively high in uneducated speech. In Manuel Puig's 1976 novel, El beso de
la mujer araña, 27% of all Past Subjunctive verbs are in the -se form. Most of
the novel takes the form of dialogue, from which it may be inferred that the
form appears in the popular Buenos Aires dialect of the characters. Appar­
ently, it appears even in highly spontaneous language.
(26) Ay, ayyy...miráy son unas puntadas tan fuertes...como si me aguje­
reasen (PS). (Puig 1976:118)
'Oh, ohhh...see, they're such strong pains...as if they were making
holes in me.'
What motivates the use of the -se form here is the counterfactual como si
'as if'; 38% of -se Subjunctives are found in contrary-to-fact clauses.
As in the Peruvian dialect, auxiliary haber accounts for a high percentage
of the verbs appearing in the -se form: 50%. The -se form is used in main
clauses several times, each time as a form of haber.
(27) Me hubieses (PS) despertado antes, Molina. (Puig 1976:96)
'You should have woken me up sooner, Molina.'
The modality expressed here is agent-oriented; once Past Subjunctives
appear in main clauses, the low assertiveness that they convey and the
assertiveness inherent in the main clause combine to license the inference
'should have happened but didn't'. There is even a case of main-clause -se in
the Peruvian data, also in the form of hubiese. Clearly, this verb licenses the
widest usage of the form.
-Se usage in the Spanish novel Si te dicen que caí'by Juan Marsé (1973)
falls between the poles of the virtual non-use of the form in the Peruvian text
and its frequent use in the Argentine text. The -se form appears in dialogue in
 442 Patricia V. Lunn

the Spanish text, which suggests its continued usage in the Madrid dialect of
the characters. However, the -se form accounts for only 13% of all Past
Subjunctive verb forms in the book, about half of the Argentine figure. As in
the Peruvian and the Argentine texts, the -se form appears most often with the
auxiliary haber, here in 55% of all cases.
In this text, as in the Peruvian text, there is an alternation of extended
non-use of the -se form and closely spaced occurrences of it. Such evidence
shows how a form is sporadically used even as it falls out of use. As speakers
produce longer and longer periods of non-use, the form disappears from
learners' input.
As in the Argentine text, a significant percentage (39%) of the -se forms
in Si te dicen que caí are introduced by a counterfactual conjunction. How­
ever, it is not the case in any of the texts examined for this paper that the -se
form must appear in counterfactual clauses; the -ra form may appear there as
well. That a high percentage of the minority -se forms should appear in
contrary-to-fact clauses is logical; in this highly non-assertive environment, a
form that encodes a high degree of non-assertion should be used.
Klein-Andreu (1991) traces the genesis of the Subjunctive use of the
originally indicative -ra form to its early backgrounding functions. In her
data, use of the -ra form in a fourteenth-century text correlated significantly
with the 'low focus' factors of negation and the use of the copular and
auxiliary verbs ser, estar and haber. It is striking that, in the modern language,
use of the most non-assertive of the Subjunctive forms continues to be linked
with these factors.
In Table 1, the first number shows the percentage of Past Subjunctive
forms that carry -se morphology. Note that, despite the differences in absolute
-se usage, the co-occurrence of -se morphology and auxiliary haber is remark-

Table 1. Percentage of use of-sc Subjunctive forms in these dialects.

% of PS HUBIESE w/NEG C to F in MC
Peru 2% 48 38 33 5
Spain 13 55 23 39 13
Argentina 27 50 20 38 15
NEG = negative clauses
C to F = contrary to fact
MC = main clauses
Evaluative Function of the Spanish Subjunctive 443

ably similar. The number on the right shows the percentage of -se forms that
appear in main clauses. Fully 100% of these main-clause uses, in all three
dialects, are realized as hubiese', -se morphology can travel with haber to the
main clause, even though the semantic value of the morphology would seem
to exclude it from this locus of assertion.
I have done the same analysis on other novels by the same authors, and
found that the percentage of Past Subjunctives realized as -se varies from
book to book, as authors adopt the point of view of very different narrators,
and as the authors themselves live for long periods away from their native
countries. The ranking, however, from least to greatest use of -se, does not
change. 3

8. Textual patterns of Subjunctive use

Even grammaticized uses of the Subjunctive can be avoided simply by pro­

viding each proposition with its own main clause. Sentences (l)-(3), for
example, can be recast as (28)-(30), in which modality in the sense of speaker
attitude towards a proposition is expressed, but the Subjunctive mood is
absent.
(28) Será una buena idea, pero lo dudo. 4
'It might be a good idea, but I doubt it.'
(29) ¿Sabes ese libro? Lo necesito.
'You know that book? I need it.'
(30) Sabes la verdad. Me alegro.
'You know the truth. I'm glad.'
The model of the active speaker of Spanish who manipulates mood
choice in order to structure discourse is particularly well described by
Lavandera (1983). Such use of mood marking suggests Traugott's identifica­
tion of a intermediate tendency in semantic change for meanings based in
extralinguistic reality to develop into meanings grounded in text-making
(1989:35).
If indeed non-assertion is a chosen stance, one might expect that fre­
quency of Subjunctive use should correlate with certain narratively non-
assertive stages in conventionally structured texts. In order to find out if this
444 Patricia V. Lunn

was so, I looked at Subjunctive use in a recent mystery novel, El balneario

(1986) by the Spanish author Manuel Vázquez Montalbán.
Writers of mystery novels play a conventionalized game with their
readers in which clues to the ultimate solution are hidden in narrative digres­
sions; these digressions serve both to entertain and to mislead the reader. In
order to see how Subjunctive use correlated with the structure of the plot, the
average number of Subjunctives per page was figured for each chapter, and
this crude measure was used to identify high-use and low-use chapters. What
was revealed by this analysis is that the Subjunctive is most used in those
chapters where information is withheld or implied, and is least used in those
chapters in which the dénouement is foretold or explained.
In El balneario, which is about a series of murders which take place in a
Spanish spa, use of the Subjunctive reaches its high point in a segment in
which the spa's patrons attempt to convince the investigating officer that they
each should be allowed to leave because they are important people with
important connections. The term 'veiled threat' comes to have a new iconic
meaning here. The other high-Subjunctive chapters contain theorizing about
the meaning of clues gathered up to that point, and a description of a hunger-
motivated attack on the spa's kitchen. In typical mystery novel style, these
chapters in which nothing new seems to be added to the reader's stock of
knowledge about the crime end with the introduction of information which
moves the plot along. Just as the reader is being lulled by hypothesis or
amused by clowning—and by the non-assertive message of the high-Subjunc­
tive text—something happens.
The most limited use of the Subjunctive comes in an early chapter in
which the detective visits the site where, as it turns out, the key to the mystery
lies. The high-Indicative text telegraphs the conclusion. The other low-Sub­
junctive sections come at the end of the novel, in which the dark history of the
spa is revealed. In these sections, the reader is modally invited to attend to
highly informative text material.
"The syntax of the text," in which plots are regarded as sentence-like
structures, is a European structuralist concept. It is not necessary, however, to
subscribe to all of the assumptions of such analyses in order to acknowledge
that highly conventionalized narrative forms—like mystery novels—are likely
to evidence patterning at the text level. One kind of macro-pattern may be
evidenced in frequency of occurrence of certain morphological forms. It is
this kind of pattern of Subjunctive usage that can be seen in El balneario.
Evaluative Function of the Spanish Subjunctive 445

9. Conclusions

It has been shown that choice of the Subjunctive in Spanish is sensitive to

semantic, textual and analogic factors. Though Subjunctive morphology con­
tinues to appear primarily in subordinate clauses, the Subjunctive can appear
in non-restrictive relative clauses (as in (7)), in embedded questions (as in
(10)), and in main clauses (as in (11), (12),(13), (23) and (27)) as the expres­
sion of a wide range of speaker attitudes and intentions. Some main-clause
uses of the Subjunctive are the reflex of the indicative history of the -ra form,
while the identification of the -se form with the auxiliary verb haber has
allowed that form—the most non-assertive—to migrate to main clauses as well.
Silva-Corvalán, following Bolinger (1956), has argued that, of a series of
syntactically and semantically related forms, the one that will be lost is "the
form which is farthest away from the speaker, in the sense that it refers to
objects or events which are farthest from him in his objective (e.g., actual
distance) or subjective (e.g., possibility of actualization) world" (1985:565).
The -se form is clearly the most remote modal form, both objectively and
subjectively; it is past tense and strongly non-assertive. Moreover, the -se
form of the Past Subjunctive is closely identified with haber, the auxiliary
used to form the temporally remote Pluperfect. This is so even in dialects
where the form is still frequently used.
In this discussion of the loss of Subjunctive forms, it is significant that in
the Quito dialect of Spanish, the Present Subjunctive is colloquially used in
the subordinate clause of past tense sentences. I have heard many sentences
like this one: Tenía vocación de pintor, pero sus padres nunca permitieron
que pinte Tainting was his vocation, but his parents never let him paint'. In
most dialects, the "sequence of tenses" would require the Past Subjunctive
{pintara) rather than the present Subjunctive {pinte). In quiteño, in which the
-se Subjunctive is little used, the -ra Subjunctive is the "outermost" form and
therefore the one, on this analysis, most vulnerable to loss. Indeed, some of its
uses are disappearing in this dialect.
The -se form of the Past Subjunctive is going out of the language via the
same path on which the -ra form came in: that of the temporally remote
pluperfect. That -se morphology is identified with a specific verb (rather than
with a specific syntax) means that the form can be used in main clauses as well
as subordinate clauses (as in (27) above), thus further muddying the syntactic
identification of Subjunctive morphology and subordination.
446 Patricia V. Lunn

Heine (this volume) has discovered that, for the German modal verbs,
"the highest values of agent-oriented and the lowest values of epistemic
modality are found with items associated with high verbality". This is sugges­
tive in view of the fact that, in Spanish, the copulas and auxiliaries—which
lack some of the characteristics of full verbs—have a special behavior with
respect to modality. In the texts examined, the low-assertiveness -se form
appears preferentially with copular and auxiliary verbs. The next form up in
assertiveness, the -ra form, appears freely in main clauses only as hubiera,
from haber. The verbs that appear in Participle form with this auxiliary are
usually non-action verbs. The list deseado 'desired,' esperado' hoped,'
pensado 'thought,' querido 'wanted' (all mental act verbs) and sido 'been,'
covers the majority of the possibilities. The one case of a second-person
subject has already been discussed as (27); here, significantly, the modality is
agent-oriented.
Finally, there is a hard-to-define but very real performance factor at work
in all of this. In modern Spanish, the mood system is a tremendously sophisti­
cated rhetorical device, and individual speakers are more or less competent
with respect to using it in a sophisticated way. There are native speakers of
Spanish who are masterful manipulators of the mood system, and there are
native speakers who are clumsy at using mood to produce rhetorical effects.
These varying degrees of rhetorical—as opposed to grammatical—compe­
tence constitute input for learners of the language, and this loop serves to
further subjectify the system.
Recent work in categorization has shown that the structure of categories
varies depending on what the goal of categorization is. According to Barsalou,
"it appears that people use a variety of differences between exemplars when
judging typicality" (1987:104); with respect to mood in Spanish, speakers are
presented with syntactic, semantic, discourse and metalinguistic differences
from which to judge what constitutes a typical Subjunctive. Concepts are
temporary constructs in working memory which typically "vary widely as a
function of goals, current context, and recent experience" (1987:135). The
data presented in this paper exemplify some of the goals and contexts which
influence choice of the Spanish Subjunctive. The grammar of mood in Span­
ish "emerges" (Hopper 1988) from the complex interaction of many factors.
Evaluative Function of the Spanish Subjunctive 447

NOTES

1. The use of the Subjunctive to express agent-oriented modality is limited in Spanish. In

addition to cases like (27), to be discussed below, there is the archaic and formulaic
quién +PS, e.g. quién pudiera escribir 'would that/if only I could write.' Agent-oriented
modality is usually expressed by lexical morphemes, as predicted by Bybee (1985:166).
2. Except for the affirmative singular and plural forms, the Imperative shows Subjunctive
morphology. The use of the unmarked verb form (third person, singular, present tense)
for the affirmative tú command follows a common pattern (Bybee 1985:172). The use of
the Subjunctive forms with all negative commands again suggests a correlation between
negation and irrealis. And the use of the Subjunctive for all formal commands is surely
due to the non-assertive (and therefore polite) value of the morphology.
3. The figures cited here for overall -se usage in Lima and Madrid are strikingly similar to
those found by DeMello (1993) in a study based on a corpus of educated oral speech.
The percentage for Lima was exactly the same (2%), and that for Madrid only slightly
higher (16%). Educated Buenos Aires speech, however, showed significantly lower -se
usage at 6%. The use of a popular dialect in the novel accounts for this disparity, as -se
usage is reportedly heavy in uneducated Argentine speech. Among the dialects studied
by DeMello, the highest -se usage was found in San Juan, Puerto Rico.
4. Example (28) shows how the Future tense may carry modal meanings; será is the Future
form of the verb ser 'to be'. This use of the Future has been widely commented on (see
e.g. Bybee, Pagliuca and Perkins (1994), and Klein (1980)). If the Synthetic Future (as
opposed to the analytic 'going to' future) develops into a kind of morphological modal,
as suggested in Fleischman (1982), it will be one more case in which mood marking has
moved into the main clause in Spanish. Niño-Murcia (1992) documents a development
along these lines in the Spanish of Andean Ecuador, where the Synthetic Future is being
used as a kind of polite command.

REFERENCES

Barsalou, L. W. 1987. "The Instability of Graded Structure". Concepts and Conceptual

Development: Ecological and Intellectual Factors in Categorization, ed. by Ulrich
Neisser, 101-140. Cambridge: University Press.
Benet, Juan. 1981. Una tumba y otros relatos. Madrid: Taurus Ediciones.
Bolinger, Dwight L. 1956. "Subjunctive -ra and -se: 'Free Variation'?" Hispania 39.345-49.
Bolinger, Dwight L. 1968. "Postposed Main Phrases: An English Rule for the Romance
Subjunctive". Canadian Journal of Linguistics 14.3-30.
Bryce Echenique, Alfredo. 1991. Un mundo para Julius. Barcelona: Plaza & Janés.
Bybee, Joan L. 1985. Morphology: A Study of the Relation between Form and Meaning
(=Typological Studies in Language, 9) Amsterdam & Philadelphia: John Benjamins.
Bybee, Joan L., William Pagliuca, & Revere D. Perkins. 1994. The Evolution of Gram­
mar: Tense, Aspect and Modality in the Languages of the World. Chicago: University
of Chicago Press.
 448 Patricia V. Lunn

Chafe, Wallace. This volume. "The Realis-Irrealis Distinction in Caddo, the Northern
Iroquoian Languages, and English".
Contreras, Heles. 1976. A Theory of Word Order with Special Reference to Spanish.
Amsterdam: North-Holland.
DeMello, George. 1993. "-Ra Vs. -Se Subjunctive: A New Look at an Old Topic".
Hispania 76.235-44.
Donoso, José. 1965. Los mejores cuentos de José Donoso. Santiago de Chile: Editora Zig-Zag.
Fleischman, Suzanne. 1982. The Future in Thought and Language: Diachronic Evidence
from Romance. Cambridge: University Press.
Fleischman, Suzanne. 1989. "Temporal Distance: A Basic Linguistic Metaphor". Studies
in Language 13.1-50.
Gili y Gaya, Samuel. 1964. Curso superior de sintaxis española (9a ed.). Barcelona:
Bibliograf.
Gsell, Otto & Ulrich Wandruszka. 1986. Der Romanische Konjunctiv. Tübingen: Nie­
meyer.
Guitart, Jorge M. 1991. "Aspectos pragmáticos del modo en los complementos de predi­
cados de conocimiento y de adquisición de conocimiento en español". Indicative y
subjuntivo ed. by Ignacio Bosque, 315-29. Madrid: Taurus Universitaria.
Guo, Jiansheng. Tis volume. "The International Structuring of Meaning: Children's Use
and Development of the Manderin Model Neng (Can)".
Harris, Martin. B. 1986. "The Historical Development of si-clauses in Romance". On
Conditionals ed. by E. C. Traugott, A. ter Meulen, J. S. Reilly & C. A. Ferguson, 265-
84. Cambridge: University Press.
Heine, Bernd. This volume. "Agent-Oriented vs. Epistemic Modality: Some Observations
on German Modals".
Hopper, Paul. 1988. "Emergent Grammar and the A Priori Grammar Postulate". Linguis­
tics in Context: Connecting Observation and Understanding ed. by Deborah Tannen,
117-34. Norwood, NJ: Ablex.
James, Deborah. 1982. "Past Tense and the Hypothetical: A Cross-Linguistic Study".
Studies in Language VI. 375-403.
Kany, C. E. 1951. American-Spanish Syntax (2nd ed.). Chicago:University Press.
Klein, Flora. 1980. "Experimental Verification of Semantic Hypotheses Applied to Mood
in Spanish". Georgetown University Papers in Language and Linguistics 17.15-34.
Klein-Andreu, Flora. 1986. "Speaker-Based and Reference-Based Factors in Language:
Non-Past Conditional Sentences in Spanish". Studies in Romance Linguistics ed. by
Carmen Silva-Corvalán & Osvaldo Jaeggli, 99-119. Dordrecht: Foris.
Klein-Andreu, Flora. 1991. "Losing Ground". Discourse Pragmatics and the Verb ed. by
Suzanne Fleischman & Linda Waugh, 164-178. London: Croom Helm.
Krakusin, Margarita and Cedeño, Aristófanes. 1992. "Selección del modo después de
el hecho de qué". Hispania 75.1289-1293.
Lakoff, George, 1987. "Cognitive Models and Prototype Theory". Concepts and Concep­
tual Development: Ecological and Intellectual Factors in Categorization ed. by Ulrich
Neisser, 63-100. Cambridge: University Press.
Lavandera, Beatriz R. 1983. "Shifting Moods in Spanish Discourse". Discourse Perspec­
tives on Syntax ed. by Flora Klein- Andreu, 209-31. New York: Academic Press.
Evaluative Function of the Spanish Subjunctive 449

Lipski, John. 1978. "Subjunctive as Fact?". Hispania 61.931-934.

Lunn, Patricia V. 1989. "The Spanish Subjunctive and 'Relevance'". Studies in Romance
Linguistics ed. by Carl Kirschner & Janet A. DeCesaris, 249-60. Amsterdam &
Philadelphia: John Benjamins.
Lunn, Patricia V. 1988. "Some Stops on the Modality Line". New Analyses in Romance
Linguistics ed. by Dieter Wanner and Douglas A. Kibbee, 221-33. Amsterdam &
Philadelphia: John Benjamins.
Lunn, Patricia V. 1989. "Spanish Mood and the Prototype of Assertability". Linguistics
27.687-702.
Lunn, Patricia V., and Thomas D. Cravens. 1991. "A Contextual Reconsideration of the
Spanish -ra 'Indicative'". Discourse Pragmatics and the Verb ed. by Suzanne
Fleischman & Linda Waugh, 147-63. London: Croom Helm.
Marsé, Juan. 1978. Si te dicen que caí. Barcelona: Seix Barrai.
Mithun, Marianne. This volume. "On the Relativity of Irreality".
Niño-Murcia, Mercedes. 1992. "El futuro sintético en el español norandino: Caso de
mandato atenuado". Hispania 75.705-13.
Puig, Manuel. 1976. El beso de la mujer araña. Barcelona: Seix Barrai.
Roberts, John R. ms. "The Category 'Irrealis' in Papuan Medical Verbs".
Romaine, Suzanne. This volume. "The Modal Function of bai in Tok Pisin".
Silva-Corvalán, Carmen. 1985. "Modality and Semantic Change". Historical Semantics,
Historical Word Formation, ed. by Jacek Fisiak, 547-542. Berlin: Mouton.
Solé, Carlos A. and Solé, Yolanda R. 1977. Modern Spanish Syntax. Lexington MA: D. C.
Heath.
Sperber, Dan and Wilson, Deirdre. 1986. Relevance: Communication and Cognition.
Cambridge MA: Harvard University Press.
Taibo, Paco Ignacio II. 1976. Días de combate. México: Editorial Patria.
Taylor, John. 1989. Linguistic Categorization: Prototypes in Linguistic Theory. Oxford:
Clarendon Press.
Terrell, Tracy & Hooper, Joan B. 1974. "A Semantically Based Analysis of Mood in
Spanish". Hispania 57.484-94.
Traugott, Elizabeth C. 1989. "On the Rise of Epistemic Meanings in English: An Example
of Subjectification in Semantic Change". Language 65.31-55.
Vázquez Montalbán, Manuel. 1977. La soledad del manager. Barcelona: Editorial Planeta.
Vázquez Montalbán, Manuel. 1986. El balneario. Barcelona: Editorial Planeta.
Woehr, Richard. 1975. "Grammar of the Factive Nominal in Spanish". Language Sciences
36.13-19.
Wright, L. O. 1932. The -ra Verb Form in Spain. (=University of California Publications
in Modern Philology 15.1). Berkeley: University of California Press.
 IV
Modality and Other Categories of Grammar
 Negation and the Modals of Possibility
and Necessity

F.R. Palmer
University of Reading

Introduction

Even a brief look at the forms used for modality accompanied by negation in a
few languages will show that there is a great deal of irregularity, in the sense
of lack of regular correspondence between form and meaning. This paper
investigates the negative modal forms of a number of diverse languages
within a simple typological paradigm, and shows that several kinds of supple-
tion are used and that there is a strong tendency for some terms in this
paradigm to be regular and other forms to be irregular.

1. The nature of the irregularity

The irregularity takes two forms:

(i) There is a simple lack of one-to-one correlation, between form and
meaning, so that it is not possible to predict for a given form what will be its
meaning, or, vice versa, what will be the form to express a particular modal
meaning. This is simply illustrated from both French and Italian:
(1) French It faut partir
it is.necessary to.go
'(We) must go'
Ilne faut pas partir
it NEG is.necessary NEG to.go
'(We) mustn't go'
454 F.R. Palmer

Italian Deve venire

must+3sG to.come
'You must come' (polite form)
Non deve venire
NEG must+3sG to.come
'You mustn't come'
Since in both cases the negative marker is formally placed with the
modal, it might be expected that it would be the modality, the meaning of
'must', that of obligation, that is negated, with the resultant meaning 'there is
no obligation' (English: 'needn't'); in fact, the meaning is 'there is obligation
not to' (English 'mustn't'). (The Italian sentence can also be interpreted in
terms of 'no obligation', but that does not invalidate the point being made; for
examples from other languages see Section 5).
Since form and meaning are to be compared and contrasted, the choice of
terminology is important. The terms 'modal' and 'full verb' will be used for
the formal, grammatical, elements, while for meaning, the terms used will be
'modality' and 'proposition;' provided these distinctions are clearly made, the
terms 'negated', 'negation' etc., can be used for both form and meaning. Any
lack of correspondence between negation of the modal and negation of the
modality or between negation of the full verb and negation of the proposition
is considered to be an irregularity.
In the examples above, then, while the modal is formally negated, seman-
tically it is the proposition, not the modality, that is negated. This irregularity
may be described in terms of the negation being 'misplaced.' It may be added,
however, that this kind of misplacement of the negative is not confined to
modals: it is also to be found in e.g. I don't think he'll come (= T think he
won't come').
(ii) A second type of irregularity is found where there are gaps in the
paradigm, especially where these gaps appear to be filled by suppletion from
other modal forms in the language. This can be illustrated from English:

(2) You must go

You musn 't go
You needn't go
Obligation to act and obligation not to act are expressed by forms of must,
but for no obligation to act the negative form of another verb, need, is required.
Negation and the Modals of Possibility and Necessity 455

No less important, though possibly less immediately obvious, is the kind

of suppletion found in:
(3) He must be in his office
He can't be in his office
He may not be in his office
The first may be roughly paraphrased 'I firmly conclude that he is in his
office', but the notions 'I firmly conclude that he is not in his office' and 'I do
not firmly conclude that he is in his office' are not usually expressed by a form
of must. Instead, the forms that may be used to express these meanings are
those of may and can, modals expressing a 'possible' conclusion ('It is
possible that...'); an explanation for this is discussed in Sections 2 and 7.

2. The modals of possibility and necessity

As the title of this paper indicates, it is concerned only with the modals of
possibility and necessity, and, moreover, only with epistemic and deontic
modality, deontic possibility and necessity being interpreted in terms of the
modals for permission (English may/can) and obligation (English must) re­
spectively. The can of ability (dynamic possibility) is not included for two
related reasons, that it is less modal and more like verbs such as wish, intend
etc., and it does not raise the same kind of problems with negation as the other
forms.
There are three reasons for this delimitation of the area of discussion:
(i) It is comparatively easy to identify such modals across languages,
though it must be allowed that there are some differences of meaning. In
particular, in some languages it would seem that there is no distinction
between 'mustn't come' and 'oughtn't to come', while a Hebrew speaker felt
that the form he gave me for 'He must be in his office', might be better
translated as 'Of course he is in his office.' But problems of this kind beset
any kind of cross-linguistic comparison.
(ii) It is possible to paraphrase them in terms of such expressions as 'It is
possible/necessary that ...', and thereby to indicate where the negation is
semantically located (its scope) — 'It is not possible/necessary that...' and 'It
is possible/necessary that... not....' This is important, because it is assumed
that in the ideal or regular situation the grammatical placement of the negative
456 F.R. Palmer

indicates the scope of the negation. If the modal is negated, the expected
paraphrase will be 'It is not possible/necessary that...', while if the full verb is
negated, the paraphrase will be 'It is possible/necessary that... not ....' One
concern of this paper is the way in which modal systems diverge from this
ideal and the forms are, in this sense, irregular.
(iii) Possibility and necessity are related logically in terms of negation,
and these relations are to some degree mirrored in language. This is clearly
seen from English. Consider:
(4) He may be in his office
He can't be in his office
He may not be in his office
These express 'possible that', 'not possible that' and 'possible that not'
respectively. (There is suppletion of can for may, but both are essentially
modals of possibility). However, as noted in Section 1 (ii), the second and
third examples can be seen as the negative forms of He must be in his office
('necessary that', indicating a firm conclusion). This can be explained in
terms of the logical equivalence Not Possible = Necessary Not and Possible
Not = Not Necessary.
Because of irregularity, the lack of correspondence between form and
meaning, it is essential to distinguish between the characterization (in terms of
negation and possibility/necessity) of the forms and the meanings of the
modals. This can be achieved by using quotation marks for the meanings, but
not for the forms; thus it can be said of the examples above that the not-
possible form (can 't) is used to express 'necessary-not' and the possible-not
form (may not) is used to express 'not-necessary' (but see Section 7 for the
problem with may not).
In my analysis there are twelve different categories (see Palmer 1990:
39). These are, with the relevant English forms:
(5) Epistemic 'possible' may
'not-possible' can't
'possible-not' may not

'necessary' must
'not-necessary' (may not)
'necessary-not' (can 't)
Negation and the Modals of Possibility and Necessity 457

Deontic 'possible' can/may

'not-possible' can't/may not
'possible-not' (needn't)
'necessary' must
'not-necessary' needn 't
'necessary-not' mustn't
(The forms in brackets indicate the suppletive forms used in English that
are dependent on the necessity/possibility equivalences).
It is not suggested that all the modal forms fit precisely into the logical
system, although this seems to be true of the epistemic modals in English,
where the negative necessity forms are provided by the logical equivalences.
With the deontic modals, however, there are forms that appear to be logically
equivalent, yet do not have identical functions in the language. Thus denial of
permission with can't can be interpreted as 'not-possible', and obligation not
to act with musn't as 'necessary-not', yet they are clearly not identical in
meaning, even though they share the feature of precluding action; here lan­
guage and logic do not wholly coincide. Yet, in contrast to this, there is only
one form to express both permission not to and no obligation — the not-
necessary form needn't', here language is consistent with logic.

3. The language material

Twenty-two languages were investigated. Most of the material was provided

by interviewing native speakers, many of them Ph.D students in the Depart­
ment of Linguistic Science in the University of Reading. They were asked to
provide the most natural equivalents of English sentences and no attempt was
made to check the material against published grammars; in any case, that
would probably have been a futile exercise, since most grammars contain
little or nothing on the topic. The Tigrinya material comes from personal
research. Only two written sources were consulted, Davidsen-Nielsen (1990)
and Thiagaran (1980), both written by scholars fully aware of the issues. This
sample may not be sufficient to draw any firm general conclusions, but it is
successful in pointing out features likely to be of typological interest.
There were two problems with this material:
458 F.R. Palmer

(i) Not surprisingly, several of the languages investigated provided few

relevant forms, and can be seen as having very little modal system. Thus for
Oromo (N. Kenya) the only apparent negative form was that for epistemic
'possible-not' (which was regular):
(6) inni afisa isa keesa jira fa
he office his inside be possible
'He may be in his office'
inni afisa isa keesa hinjiru fa
he office his inside NEG+be possible
'He may not be in his office'
For the 'not-possible' 'He can't be in his office' the speaker offered only
'He isn't', and for the epistemic necessity only forms with an emphatic
particle on 'office', while for the deontic forms he offered only imperatives.
Of course, there may be forms in the language that the speaker had not
thought of, but that does not necessarily invalidate my remarks, for my
interest lies primarily in establishing the most natural forms, rather than
attempting to establish a neat system. I am positive about the existence of gaps
in Tigrinya, on which I worked in detail many years ago, but have now been
able to check with both my original assistant and his daughter. There is no
modal expression for epistemic 'not-possible', the negative declarative being
used instead, while for deontic 'not-necessary' the only form they could
suggest was:

(7) bd-hayle 'aytəbəlla'

by-force eat+NEG+IMP
(Lit. 'Don't eat by force') 'You needn't (don't have to) eat'
There are, almost certainly, languages that have no system in terms of
negation and possibility/necessity, e.g., those with an evidential system (see
Palmer 1986: 66-76).
(ii) It was not always easy or even possible to decide which were the
relevant modal forms. Thus for French, it seemed at first that the forms to be
considered were the verbs pouvoir and devoir plus the infinitive, but, from a
consensus of a number of French scholars, it appears that, in practice, epis­
temic possibility is expressed by il se peut que, and the impersonal il faut is
more commonly used for deontic necessity. There was a different problem
with Welsh. One speaker (a professional linguist) provided me with an almost
perfectly regular set of forms using the verb gallu for possibility and the
Negation and the Modals of Possibility and Necessity 459

impersonal rhaid for necessity. A second speaker (with no knowledge of

linguistics) assured me that gallu could only be used to expressed ability, and
came up with non-modal forms of the type 'It is possible that ...'for epistemic
possibility and forms of the verb cael 'get' for deontic possibility; he offered a
regular form for epistemic 'necessary-not' and regular forms for deontic 'not-
possible' and 'not-necessary', but none of the other negatives (see Section 5).

4. Regular and irregular systems

I did not discover a language with a perfectly regular system. The closest to it
was found in Modern Greek, where only one form, that for deontic 'not-
necessary', was irregular. The regular forms were:
(8) bori na ine sto yrafio tus (epistemic possibility)
it.can that they.are in.the office theirs
'They must be in their office'
Sen bori na ine. (not-possible)
not can that they.are...
'They can't be ...'
bori na min ine... (possible-not)
it.can that not they.are...
'They may not be ...'
prepi na ine sto yrafio tus (epistemic necessity)
it.must that they.are in.the office theirs
'They must be in their office'
Sen prepi na ine... (not-necessary)
not it.must that they.are...
'They may not be ...'
prepi na min ine... (necessary-not)
it.must that not they.are...
'They can't not be ...'
boris na fivis (deontic possibility)
you.can that you.leave
'You can (may) leave'
460 F.R. Palmer

Sen boris na fiyis (not possible)

not you.can that you.leave
'You can't (may not) leave'
boris na mi fiyis (possible-not)
you.can that not you.leave
'You needn't leave'
prepi na fiyis (deontic necessity)
it.must that you.leave
'You can (may) leave'
prepi na mi fiyis (necessary-not)
it.must that not you.leave
'You mustn't leave'
For the irregular 'not-necessary', two forms were suggested:
(9) den ine anangi na fiyis
not is necessary that you.leave
'You needn't leave'
den xriazete na fiyis
not is.needed that you.leave
'You needn't leave'
A relevant observation about the Greek system is that there is a finite
subordinate clause introduced by a 'that' conjunction. It is this, undoubtedly,
that influences the regularity. Such regularity was found in several other
languages for epistemic modality where that type of modality alone employs a
finite subordinate clause, e.g., in one set of forms in French (see Section 3
above):
(10) Il se peut qu'il soit...
it REFL can that.he is ...
'He may be ...'
Il ne se peut pas qu'il soit...
It not REFL can not that.he is ...
'He can't be...'
Negation and the Modals of Possibility and Necessity 461

Il se peut qu'il ne soit pas...

it REFL can that.he not is not...
'He may not be ...'

Where the 'ideal' situation is not to be found, i.e., where the scope of the
negation is not reflected in the grammatical placement of the negative, there
are several possibilities:
(i) The simplest is that another verb is used. Two examples of this are
found in English: the use of can in place of epistemic may for 'not-possible'
{can't) and the use of need in place of deontic must for 'not-necessary'
{needn't).
(ii) The logically equivalent form is used. An example is the use of
('not- possible') can't in English as the form for epistemic 'necessary-not'
corresponding to must (see (2) and, for other examples, Section 5).
(iii) The negative is 'misplaced', placed on the 'wrong' verb. This was
found commonly and almost exclusively with 'necessary-not' (mostly with
deontic modality); this was signalled grammatically by the not-necessary
form. This too was illustrated, from French and Italian, in Section 1, and there
are further examples in Section 5.
(iv) There is a rather different construction, but using the relevant modal.
Thus for Latvian, the form proposed for deontic 'possible-not' (permission
not to) was literally 'You can come in, or not':
(11) tu drīksti nākt ieksa vai ne
you can come in or not
'You needn't come in'
(v) The form used is not related grammatically to the modal system.

5. Logical suppletion

One of the most interesting points is the suppletion in terms of the logical
equivalences. For most of the languages, suppletive forms of this kind were
offered for at least one place in the paradigm, but there is a problem with some
languages, in that there were gaps in the paradigm and no suppletive forms
were offered. Yet it cannot be said with any certainty that suppletive forms
462 F.R. Palmer

could not have been used, for it is possible the native speaker simply did not
immediately think of them (and I carefully refrained from suggesting them).
There was only one language with no suppletive forms of this type —
Modern Greek, where all the forms except that for deontic 'not-necessary'
were regular (see Section 4).
Where suppletive forms were offered, they followed two main patterns:
(i) Suppletive forms are used for both types of negation of necessity (both
'not-necessary' and 'necessary-not'), possibility forms being used for both.
This is especially common with epistemic modality, as already illustrated for
English by (3) which is repeated here:
(3) He must be in his office
He can't be in his office
He may not be in his office
Similar forms were found for Assamese, Italian and one variety of
Welsh. Examples from the first two are:
(12) Assamese
ofis-ot thakibc lage
office-in to.be he.must
'He must be in the office'
ofis-ot na thakibc pare (possible-not → 'notnecessary')
office-in not to.be he.may
'He may not be in the office'
ofis-ot thakibo nware (not-possible →'necessary-not')
office-in to.be he.NEG+may
'He can't be in the office'
Italian
Deve essere nell ufficio
He.must be in.the office
'He must be in the office'
Pud non essere nelV ufficio (possible-not —»
He.can not be in.the office 'not-necessary')
'He may not be in the office'
Negation and the Modals of Possibility and Necessity 463

Non pud essere nell ufficio (not-possible →

Not he.can be in.the office 'necessary-not')
'He can't be in the office'
However, for Italian, the not-necessary form was suggested as an alterna­
tive for the last:
(13) Non deve essere nelV ufficio
Not he.must be in.the office
'He can't be in the office'
(Although this uses the 'must' verb, the negation is misplaced, the not-
necessary form being used for 'necessary-not', as it is also used with deontic
necessity in Italian — see Sections 1 and 5).
Japanese, alone of the languages investigated, appeared to use suppletion
for both types of deontic necessity. Surprisingly, perhaps, even the positive
necessity form was suppleted, using a double negative ('can't not');
(14) tabe naku te wa ikenai (not-possible-not → 'necessary')
eat not may.not
?
'You must eat
tabe te wa ikenai (not-possible → 'necessary -not')
eat may.not
'You mustn't eat'
tabe naku temo ii (possible-not →'not-necessary')
eat not may
'You needn't eat'
(ii) Suppletive forms were used for negation of the proposition alone
('possible-not' and 'necessary-not'), the forms negating the modality being
regular and used to provide the suppletive forms (the not-necessary and not-
possible forms respectively).
This pattern is found in Danish for the epistemic modals only. Thus
Davidsen Nielsen (1990: 78-87) recognizes only two forms that negate epis­
temic modality:
(15) Det kan ikke vœre sandt (not-possible)
that can not be true
'That can't be true'
464 F.R. Palmer

Det behøver ikke vœre sandt (not-necessary)

that mustnot be true
That may not be true'
These are treated as the two negatives of both:
(16) Det kan vœre sandt
that can be true
'That may be true'
Det må vœre sandt
that must be true
'That must be true'
If in Danish the negative formally negates the modal (see Section 7),
these forms are regular when used to negate the modality (to express 'not-
possible' and 'not-necessary'), but are used suppletively for the forms that
negate the proposition (as 'necessary-not' and 'possible-not' respectively).
There was a similar situation with deontic modals for one version of
Welsh: only the modals are negated, and these negate the modality, so that the
forms negating the proposition are suppletive:
(17) Mi gewch chi ddod yfary
You get you to.come tomorrow
'You may come tomorrow'
(Ni) chewch chi ddim ddod yfary ('not-possible/
(not) get you not to.come tomorrow 'necessary-not')
'You can't come tomorrow' (or 'You mustn't...')
(Mae'n) rhaid i chi ddod yfary
(It.is) necessary to you to.come tomorrow
'You must come tomorrow'
('does) ddim rhaid i chi
ddod yfary
('not necessary /'possible-not')
(it.is.not) not necessary to you to.come tomorrow
'You needn't come tomorrow'
However, Welsh has another means of expressing 'necessary-not' — see
Section 6.
Negation and the Modals of Possibility and Necessity 465

6. Deontic 'not-possible' and 'necessary-not'

The most striking result of this exercise lies in the comparison of the forms for
deontic 'not-possible' (denial of permission) and 'necessary-not' (obligation
not to). In every one of the languages except Oromo, which, as was noted
earlier, barely has any system, and Tamil, the 'not-possible' form was regular
in that negation was formally marked on the modal. Some examples are:
(18) Tigrinya
Iətə 'atto 'aytəkə'ələn 'ika
that.you.come.in you.not.being.able you.are
'You cannot/may not come in'
Latvian
tu ne drīksti nākt ieksa
you not can come in
'You cannot/may not come in'
By contrast, in all the languages except two, the form used for 'neces­
sary-not' was irregular. The possibilities were:
(i) In ten of them, Arabic, Assamese, Danish, English, Farsi, French,
Italian, Kinyarwanda, Tamil and Tigrinya, it is the modal that was formally
negated, i.e., there was misplacement of the negative (although in Italian and
Tamil the same form was used for 'not-necessary', while in Farsi the regular
form with negation on the full verb was also possible — and see Section 6 for
a discussion of Danish and English). Examples are (see also (1) for Italian):

(19) Arabic
la:zim jizi
must you.come.in
'You must come in'
ma:/mu laizim jizi
not must you.come.in
'You mustn't come in'
Kinyarwanda
agomba kwinjira
he.must to.come.in
'He must come in'
466 F.R. Palmer

ntagomba kwinjira
NEG+he.must come.in
'He mustn't come in'
The form used for expressing 'not-necessary' in these languages varied.
While Italian and Tamil (see above) used the same form, Kinyarwanda and
Assamese used rather different constructions, while Arabic, Danish, English
and Farsi employed a different verb and the others had no form that was
clearly modal.
(ii) The form given in five languages, Bahasa Malaysia, Chinese, He­
brew, Kikuyu and Japanese, was the suppletive not-possible form, and this
was also given as an alternative to the form with misplaced negative for Italian
and Tamil. In all of these languages except Chinese and Japanese, negation on
the necessity modal had the regular 'not-necessary' sense (Chinese and Japa­
nese had no necessity forms with negation, and used the suppletive 'possible-
not' for 'not-necessary'). Examples are:
(20) Bahasa Malaysia
dia mesti/perlu masuk
he/she must come.in
'He/she must come in'
dia tak mesti/perlu masuk
he/she not must come.in
'He/she needn't come in'
dia tak boleh masuk
he/she not can come.in
'He/she mustn't (can't/may not) come in'
Chinese
ta mín tjen bi ∫y lé
he tomorrow must come
'He must come tomorrow'
ta mín tjen kŏe i bu Jy Jé
he tomorrow can/may not come
'He needn't come tomorrow'
ta mín tjen bu kŏe i Jy Jé
he tomorrow not can/may come
'He mustn't come tomorrow'
Negation and the Modals of Possibility and Necessity 467

There is probably also similar suppletion in German (but see Section 7):
(21) Du musst herein kommen
You must in.here come
' You must come in'
Du musst nicht herein kommen
You must not in.here come
'You needn't come in'
Du darfst nicht herein kommen
You can/may not in.here come
'You mustn't come in'
The point here is that the verb dürfen (darfst) is probably best seen as a
possibility verb (though translated 'need' as well as 'can' by dictionaries),
used both in positive forms and as the lexically suppletive verb for mögen
(although können is also used):
(22) Du magst herein kommen
You can/may in.here come
'You can (may) come in'
Du darfst nicht herein kommen
Du kannst nicht herein kommen
You can/may not in.here come
'You can't (may not) come in'
This shows that the possibility verb dürfen is used suppletively to express
necessity, in place of müssen (not-possible to express 'necessary-not').
Welsh has another alternative. Instead of the suppletive 'not-possible'
form given in (17), the 'necessary' form may be used if followed by the
negative verb peidio, which is, slightly misleadingly perhaps, glossed as
'refrain' below:
(23) Rhaid i chi beidio â ddod yfory
Must to you refrain from to.come tomorrow'
'You mustn't come tomorrow'
The two languages in which negation is regular, with the negation of the
full verb giving the 'necessary-not' interpretation were Modern Greek and
one version of Welsh. The Greek example has already been given and dis-
 468 F.R. Palmer

cussed. What is surprising is that it is the form used for 'not-necessary', rather
than the form for 'necessary-not', that is irregular in Greek. This is expressed
not by negation on the modal, but by a different, probably non-modal, verb, as
illustrated in (9), which is repeated here:
(9) den ine anangi na fiyis
not is necessary that you.leave
'You needn't leave'
den xriazete na fiyis
not is.needed that you.leave
'You needn't leave'

7. Function of the negative

There is a problem with some of the Germanic languages in that it is not at all
certain that it can be determined whether it is the modal or the full verb that is
negated. In English, it is to be assumed that the negative is formally associated
with the modal, since it is generally cliticised. In that case, there is misplace­
ment of the negative in You mustn't' come. If this were regular, it would have
the 'non-necessity' meaning, which is, in fact, expressed by needn't. The
position in Danish and German is less obvious. In Danish skulle is commonly
used for obligation (as well as mätte), but the negative forms, with beh0ve
used suppletively, are (Davidsen-Nielsen 1990: 200-1):

(24) Du skal ikke svare

you must not reply
'You mustn't reply'
Du beh0ver ikke svare
you must not come
'You needn't come'
The first of these should be compared with the German Du musst nicht
herein kommen in (21). The difference in meaning is surprising. If it is
assumed that, as in English, the negatives ikke and nicht are formally associ­
ated with the modal, Danish is irregular, but German is regular. Conversely,
of course, if the negatives were assumed to negate the full verb, Danish would
be regular and German irregular. Since, however, these negative markers are
Negation and the Modals of Possibility and Necessity 469

placed between the modal and the full verb, it may be that there is no way of
determining which of these two they formally negate.
In Danish, however, there is some evidence to suggest that the negative
belongs formally with the modal: in certain subordinate clauses it precedes the
modal. Examples are (Davidsen-Nielsen personal communication):
(25) Jeg mener at man ikke skal/behøver dømme
]I think that one not must pass judgment on
nogen så hurtigt
someone so quickly
'I think one mustn't/needn't pass judgment on someone that quickly'
I know of no similar evidence to decide the issue in German. Since,
however, negatives normally follow other verbs in German, it is not unreason­
able to associate the negative formally with the modal.
There is, perhaps, a problem with English epistemic may not. this is not
normally cliticized, so that the negative is not so clearly associated formally
with the modal. Consistency with the other modals, however, would suggest
that it is the modal that is negated. In that case, however, the form is irregular,
since it is a not-possible form expressing 'possible-not;' moreover, it is the
only example from all the languages considered in which there is displace­
ment of the negation with this meaning. It is tempting to suggest that it is
precisely because it has this meaning that the form is not cliticized, and so
suggests negation of the full verb (and regular negation of the proposition).
However, if that is so, one might expect (falsely) that deontic must + not
should also not permit cliticization, and so similarly indicate negation of the
full verb and of the proposition.
One solution is to say that in these languages it is not possible to signal
the contrast of negation of the modality and negation of the proposition by
differential placement of the negative (negating the modal and negating the
full verb): the negative formally belongs to the whole verbal complex, as in
English with other auxiliaries — with isn't coming and hasn't come. Moreo­
ver, the contrast is almost always made by the use of suppletion (e.g., English
must/need, German müssen/dürfen, Danish skulle/beh0ve etc.; it may be
precisely because negative placement cannot signal the contrast that supple­
tion is used. However, these languages can still be included within the general
typological framework, because, if the modality is taken to negate the whole
verbal phrase, the most natural interpretation would be 'It is not the case that it
 470 F.R. Palmer

is possible/necessary ...;' this would be the same as with the regular marking
of negation of the modal. Judgments about regularity (and about suppletion)
are, then, still possible: it can still be said that the German must nicht,
expressing 'not-necessary', is regular, while the corresponding forms in Eng­
lish and Danish (mustn't, skal ikke), expressing 'necessary-not' are irregular.

8. Possible explanations

Is there any explanation for the features discussed in Sections 6 and 7? The
issue of suppletion is, perhaps, easy to account for: economy is achieved by
using the logical equivalences. They are particularly appropriate with epis-
temic modality, where logic and language appear to coincide, but less appro­
priate with deontic modality, where, for instance, 'not-possible' (denial of
permission) and 'necessary-not' (obligation not to) are not identical in mean­
ing, even though both preclude action.
The comparative irregularity of the 'necessary-not' form is probably due
to two factors. The first is that this is a very commonly used form, and, in
particular, much more common than 'not-necessary.' The second is that
elsewhere in language the scope of negation is not accurately indicated by the
grammar, as shown by the example noted in Section 1,I don't think he 'll come
(= 'I think he won't come'). It is, perhaps, not wholly surprising if the form
that negates the modal is used for this purpose; it would emphasize the
essentially negative aspect of the prohibition. This is especially so if there is
only one formal way of associating the modal verb with negation, as in
English. By contrast, the 'not-possible' form is usually regular, but this too
may be accounted for by its frequent use, in that there would be no reason not
to use the most obviously available form.

REFERENCES

Davidsen-Nielsen, N. 1990. Tense and Mood in English; A comparison with Danish.

Berlin & New York: Mouton de Gruyter.
Palmer, F.R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Palmer, F.R. 1990. Modality and the English Modals. London: Longman.
Thiagaran, K. 1980. Modal systems of English and Tamil. Ph.D. thesis, University of Madras.
Negation and the Modals of Possibility and Necessity 471

APPENDIX

The table below indicates the type of form offered for each modal type in the languages
investigated. The symbols used are:
■4 Regular
s Suppletive in terms of the possible/necessary equivalences
M Negation misplaced
D Different modal form
c Construction different, but modal form used
No form offered or form not in the modal system

Epistemic Deontic
Not-Pos Pos-Not Not-Nec Nec-Not Not-Pos Pos-Not Not-Nee Nec-Not
Arabic (Hijazi) - √ √ √ √ √ D M
Assamee √ √ s s √ - - M
Bahasa Malaysia √ √ √ √ √ s √ S
Chinese √ √ s s √ √ s S

Danish V s √ s √ s √ M
English D R s s √ s? D M
Farsi √ √ - √/R √ - D √/M
French √ √ s s √ √ - M
German √ s √/D s √ - √ S

Greek

Hebrew √ √ - √ √ √ √ s
Italian √ √ s R/S √ √ √ M/S
Japanes D s s/√ s S
Kikuyu √ s s s S
Kinyarwanda √ √ c R
Latvian √ c √
Luo √ D - - √ - √ -
Oromo - √ - - - - - -

Serbo-Croatian D √ √ s √ s √ D
Tamil √ s D D s D √ M/S
Tigrinya √ √ M
Welsh (i) √ √ s S √ S √
(ii) - - - √ √ S √ s
 A Functional Theory of Complementizers*

Zygmunt Frajzyngier
Dept of Linguistics, University of Colorado

1. Introduction

My main goal in this paper is to present a functional theory of complementiz­

ers. The category of complementizers, an important component of complex
and simple sentence structure, remains undefined, as documented further in
the paper. A secondary goal is to show the interconnectedness of elements of
the same functional domain in a grammatical system. More specifically, I
show that presence of one or more complementizers, treated as a parametric
variation within the GB framework, is closely linked to the synchronic or
diachronic presence or absence of devices encoding various kinds of modali­
ties, and to the place these markers occupy in the clause. Applying the
proposed theory I describe (1) the motivations and conditions contributing to
the absence of complementizers in the matrix clause, and (2) the occurrence of
two complementizers in a sequence. The results of the present study fully
support postulating mood as a component of a sentence. Since this work
addresses substantial issues in the structure of language, its results should be
of interest to functional theories and to formal theories such as GB and GPSG,
which postulate the existence of COMP as a component of a sentence. Some of
the properties that have been ascribed to the category COMP in various works
can be explained by its semantic function of encoding modality, while other
properties may turn out to be artifacts of analyses.
474 Zygmunt Frajzyngier

2. The notion of complementizer

The non-lexical category COMP, first introduced in Bresnan (1970) plays an

important role in several contemporary theories. Within GB theory COMP
interacts with the theory of Government, Case theory, and Binding theory,
and is an important element with respect to movement rules (cf. Chomsky
1986a; 1986b; 1992; Rizzi 1990). COMP may be empty or may be filled by such
elements as the English 'that', 'for', etc. It may also be filled by WH-words as
a result of WH-movement. In English there cannot be both a complementizer
and a WH-word (i.e., the COMP position cannot be filled by two words). There
are, however, languages in which there may be two or more complementizers
in a sequence, as is the case in Dutch (cf. Koster 1987:206ff). Within GB
theory such cases are treated as resulting from movement rules usually
involving WH-movement, (e.g., Bouchard 1984; Wachowicz 1974; Cichocki
1983; Comorovski 1986; Rudin 1986; and Rizzi 1990). In other theoretical
approaches two complementizers in a sequence are either explicitly allowed
(e.g., Kac 1978:76) or implicitly allowed (e.g., Givón 1980, 1990).
While there is some consensus among linguists when it comes to identi­
fying a morpheme as a complementizer, there is considerable divergence as to
which morphemes should or should not be included in the set of complemen­
tizers. This divergence is due to the fact that the term itself is rather poorly
defined. The term 'complementizer' includes the old term 'subordinating
particle' that 'marks' sentential subjects and objects, but also includes some
morphemes, such as interrogative markers occurring in the matrix clause.
Often the identity of this category is taken for granted, (e.g., Lefevbre 1980;
Bickerton 1981; Dasgupta 1983; Borsley 1985; Joseph 1985; Lawal 1986;
Dimmendaal 1989; Heine et al. 1991). In many works, an example or a list of
examples is given in lieu of a precise definition. Chomsky (1986:161), for
example, states that in English COMP may be 'that, for or null'. The comple­
mentizer may also be defined by analogy, e.g., as the morpheme that corre­
sponds to the English 'that' (Lefevbre 1980). It is through the examples of
complementizers that we may be able to detect what the underlying and
largely tacit assumptions are with respect to this category.
A random selection from some recent attempts to define complementiz­
ers indicates that they are most often treated as morphemes signaling that the
following clause is a complement. Thus Noonan (1985:44-45) states that
'Complement types often have associated with them a word, particle, clitic, or
A Functional Theory of Complementizers 475

affix whose function is to identify the entity as a complement. Such forms are
known as complementizers'. He further (p.47) lists among English comple­
mentizers 'that', 'if', and 'to'. Palmer (1986) in an extensive discussion of
types of complement clauses, uses the term 'conjunction' for what other
linguists call 'complementizers'. He does not define the term, but in discuss­
ing conjunctions in various languages he ascribes to them modal functions.
Ransom (1986:87-88) defines complementizers as 'semi-lexical forms that
occur within the complement, setting it off from the main clause and, at the
same time, signaling its modality'. She lists among English complementizers
'that', 'whether', 'for-to', and 'whether-to'. Givón (1990:552ff) refers to the
morphemes in question as 'subordinating —"separating"—morphemes', but
also uses the term 'complementizer' (p. 554) and describes their function as
participating 'in the coding of the event-integration scale'. The underlying
principle in these studies appears to be that for a morpheme to be classified as
complementizer, it has to occur in a specific syntactic position, viz. before the
embedded clause (cf. Bickerton 1981:109), regardless of the relative order of
the two clauses.
In the present paper I discuss morphemes that are usually analyzed, or
likely to be analyzed, as complementizers. I show that complementizers are
lexically separate modality markers that happen to occur in a specific syntac­
tic position. As such they are just one of the means by which languages
encode modality.
Descriptions that take separating the main and the embedded clauses to
be the main function of complementizers cannot explain the occurrence of the
same morphemes in simple sentences or in the main clauses of complex
sentences, i.e., in syntactic environments that do not require or allow the
function of 'separation'. The analyses that postulate separation as the main
function of complementizers face the additional problem of motivating the
existence of a morpheme whose function is to 'separate' clauses, since there is
no possibility of clauses ever overlapping in time. It is not therefore clear what
the metaphor of 'separation' stands for. One could argue that the function of
complementizers is to mark a clause as a complement, but even that function
is not well justified in view of the fact that that there are many types of
complement clauses that do not require or even allow a complementizer, even
in languages that have complementizers. A third problem with such descrip­
tions is that they do not account for morphemes that occur in positions other
than at the beginning of the embedded clause. These morphemes are, as will
 476 Zygmunt Frajzyngier

be shown, in complementary distribution with complementizers, and hence

belong to the same set. And finally, if the separation of the main and the
embedded clauses, or the marking of a clause as a complement, are the
functions of complementizers, why should a language have two or three
complementizers in a sequence when one morpheme should be enough?
While early transformational grammar was concerned with the choice of
particular complementizers (cf. Rosenbaum 1967), GB theory so far has not
shared the same concern. Functionalist approaches are very much interested
in the function of specific complementizers. The works dealing with particu­
lar complementizers in English and other languages are much too numerous to
be listed in this article. For a selection of such works see (Bolinger 1968,
Kirsner and Thompson 1976, Ransom 1985, Frajzyngier 1991a, Frajzyngier
and Jasperson 1991) and the references there. None of these works answers
the following questions: Why are there complementizers in some main
clauses but not in others? Why is it that in some languages there may be only
one complementizer in the embedded clause, but in other languages there may
be two and possibly more complementizers in the embedded clause? I claim
that the answer to these questions lies in the function of complementizers. I
also discuss phenomena linked with the presence of complementizers that
affect other components of a sentence.

3. Hypothesis

The main hypothesis of the present paper is that complementizers constitute a

part of the system of modality markers, more specifically, that they encode
deontic, epistemic, and other types of modality. As such, they may mark
either the main or the embedded clause. The term modality is understood in
the present work as it is in Palmer (1986), viz. as the speaker's attitude toward
the proposition, but it also includes the agent-oriented and the speaker-
oriented distinction of modalities (see Bybee et al., 1994). I consider inter­
rogative also to be a part of epistemic modality, as it is either an inquiry about
the truth of the proposition ('yes/no questions'), an inquiry about a specific
element of the proposition ('WH-questions'), or an inquiry about the truth of
an assumption ('tag questions'). The hypothesis affirms the function of com­
plementizers envisioned to a certain degree in Palmer (1986) and covers an
area left untouched in Bybee et al. (1994). It is possible that in some languages
A Functional Theory of Complementizers 477

complementizers eventually assumed the role of 'separating' two clauses, or

of 'introducing' an embedded clause. In the present paper, this last stage in the
development of complementizers, even if it is true, will not be discussed.
There are several ways in which this hypothesis can be supported. One is
by examining the meaning of complementizers, more specifically by contrast­
ing sentences with a complementizer with those without, or by contrasting
sentences with different complementizers. There are many studies of indi­
vidual languages whose results support the hypothesis, though these studies
were published before the hypothesis was postulated (see Bolinger (1968),
Kirsner and Thompson (1976), Frajzyngier and Jasperson (1991) for English,
and Bickerton (1981) for various Creole languages).
But the hypothesis may also be supported in another way which I will
pursue in this paper. If the hypothesis is true, one should expect complemen­
tary distribution between complementizers and other modality markers essen­
tially along the following lines: If there is a complementizer marking an a
modality there should not be other a modality markers in the clause, and if
there is an a modality marker already present in a clause there should not be a
complementizer encoding the a modality. If an epistemic or a deontic modal­
ity is not marked in the embedded clause it may be marked by a complemen­
tizer. If an a modality is marked within the embedded clause it will not be
marked by a complementizer.
Another important support for the main hypothesis comes from its ability
to explain a number of cross-language syntactic phenomena. These phenom­
ena are: (1) Lack of complementizers in one major type of simple or main
clause; (2) presence of two complementizers associated with one embedded
clause; (3) presence of one complementizer in the embedded clause (this falls
out naturally from the discussion of (2)); and (4) relationship between com­
plementizers and inversion in English.

4. The 'null' COMP in the main clause

Even in theories which postulate that every sentence has a COMP, it is recog­
nized that the COMP position may remain empty. Bresnan (1970), who first
proposed the rule S' —> COMP S, attempted to explain the empty COMP by
postulating that complementizers are also a part of the underlying structure of
simple sentences but that they are deleted in the surface structure. She associ-
478 Zygmunt Frajzyngier

ated declaratives with the deleted 'that' and interrogatives with the deleted
'WH' (Bresnan 1970:301). For an analysis that does not postulate COMP
deletion in the simple sentence see Emonds (1985:318), who identifies S' and
P' as COMP and P. In the main clause the underlying P is empty.
I explain the absence of complementizers in simple sentences by postu­
lating two factors, both of which agree with the proposed hypothesis. One is
that indicative sentences convey what the speaker intends to be taken as his
belief. In other words, indicative sentences have an inherently defined epis-
temic modality. I do not defend this hypothesis here as I have argued for it
extensively elsewhere (Frajzyngier 1985 and 1987; but cf. Palmer 1987).
With respect to the deontic modality, I propose, that the indicative clause is
the unmarked case, i.e., it indicates neither obligation nor wish. The main
support for this analysis comes from the fact that obligation, wish, and all
other types of deontic modalities have to be marked by inflectional markers,
modal verbs, modal adverbs, or other devices that a language may have.
The reason why even non-indicative main clauses sometimes lack com­
plementizers is that their modalities are coded by syntactic devices such as
word order, by morphological changes on the verb, or by phonological
devices such as intonation, vowel or consonant length. Thus the Gothic verb-
final -u, which marks the interrogative modality, would most probably not be
considered a complementizer, e.g.:
(1) skuld-u ist kaisaragild giban kaisara
'is it lawful to give tribute money to Caesar?' (Visser 1969, 3:1547)
The following examples illustrate the complementary distribution of comple­
mentizers with other devices marking modality in the main clause. As indi­
cated earlier, the interrogative 'whether' is considered by most authors a
complementizer in English:
(2) I asked whether she came
(3) I don't know whether she came
Polish czy is also a complementizer, as shown by the following examples:
(4) Spytalem czy przyszta
ask: 1SG:PERF:PAST COMP come:3F:sG:PERF:PAST
T asked whether she came'
A Functional Theory of Complementizers 479

(5) Nie wiem czy przyszła

NEG know COMP come:3F:SG:PERF:PAST
'I don't know whether she came'
Unlike contemporary English 'whether', but similar to its Old English equiva­
lent, Polish czy may also occur in main interrogative clauses inquiring about
the truth, e.g.:
(6) Czy przyszla
COMP come:3F:SG:PERF
'Did she come?
The complementizer czy has an interesting distribution with respect to the
interrogative intonation. The sentence (6) may, but does not have to, have the
interrogative intonation marked by heavy stress on the penultimate vowel of
the verb przyszla. But an interrogative sentence may also be produced without
the complementizer czy. In such a case, however, the verb przyszla must have
the interrogative intonation, viz.:
(6) a. Przyszla
come-3F-SG-PERF
'Did she come?'
The absence of deontic complementizers in the simple/main clause in
many languages can be explained in a way similar to the absence of epistemic
complementizers, viz. by the complementary distribution of various means to
encode the same modality. I will illustrate the problem using examples from
Polish and French.
In Polish the wish, a component of mood of obligation, for first and third
person is marked by a complementizer, as in the following examples:
(7) a-by tylko przyszła
COMP-COMP only come:3F:PERF
'I wish she would come'
(8) niech by (tylko) przyszla
COMP COMP (only) come:3F:PERF
'I wish she would come' (chances are she won't )
'I dare her to come'
480 Zygmunt Frajzyngier

In Old French obligation or wish was marked by an inflectional device, the

subjunctive mood:
(9) Dex te desfande, chevaliers, de mesaventure
'may God protect you, knight, against bad luck' (Jensen 1990:356)
The subjunctive in French was also used, however, for functions other than
encoding the mood of obligation. Bourciez (1967:709) states that from the end
of the 17th century que, which is identical with the complementizer, becomes
an obligatory marker before the subjunctive, indicating a wish or an order.
(10) que les dieus nous aient en lor garde
'may the gods keep watch over us' (Medieval French, Jensen 1990:356)
Que son nom soit béni!
'May his name be blessed'
Qu 'il périsse!
'May he perish!' (Bourciez 1967:709)
Therefore what on the surface may appear to contradict my earlier claim about
the complementary distribution of devices coding the same modality is in fact
a confirmation of this principle. The subjunctive in contemporary French is no
longer primarily a marker of the mood of obligation.
The facts with respect to the main clause may be summarized as follows:
The main clause does not have a complementizer if one of the following two
conditions is met: (1) The clause is indicative, and hence is inherently marked
for epistemic value expressing the speaker's belief in the truth of the proposi­
tion and is unmarked for the deontic value, i.e., it does not express a wish, an
order, or an obligation; or (2) there exist clause-internal, mainly inflectional,
devices to encode the modality of the clause.

5. Two complementizers in a sequence

Theoretically, a situation could arise in which the simple clause is marked for
more than one modality, and each of the modalities is encoded by a separate
morpheme. The previously cited examples (7) and (8) from Polish could be
considered instances of two complementizers occurring in a simple clause. In
example (7) the sequence of complementizers is a-by and in example (8) niech by.
The first complementizer encodes obligation and the second a hypothetical
mood. Compare also the following example where the sequence is o-by:1
A Functional Theory of Complementizers 481

(11) o-by przyszla

COMP-COMP come:3F:PERF:PAST
T wish she would come'
As for the occurrence of two complementizers in a complex sentence, in
addition to the markers of deontic and epistemic modalities already outlined
for the simple clause, there may be a complementizer marking the embedded
clause as being in the de dicto or in the de re domain (both domains being
types of epistemic modality). This is especially true for sentential comple­
ments of verbs of saying and verbs of perception (see Frajzyngier 1991a;
Frajzyngier and Jasperson 1991). In many languages de dicto complementiz­
ers have been extended in function to serve as markers of indirect evidence, as
described in Frajzyngier (1991a). The potential composition of the embedded
clause modality markers, whether clause-internal or -external (these are usu­
ally referred to as 'complementizers'), is therefore as follows: (1) De dicto/de
re - (2) epistemic - (3) deontic (I make no claims with respect to the order of
these elements). In this representation 'epistemic' indicates any epistemic
distinction other than the de dicto/de re. The possibilities are as follows: One,
two, or all three types of modalities may be overtly coded. It is also possible
that the clause may be characterized by one modality only, and that that
modality is an inherent property of the clause, i.e., there is no overt modality
marker. The evidence for the hypothesis that complementizers are just one of
the means of coding modality consists in showing that the presence of a
particular complementizer is linked with the absence of clause-internal mark­
ers of modality, and that the absence of a particular type of complementizer is
linked with the presence of clause-internal markers of modality.

6. Two disjoint complementizers

Whether two complementizers occur in a sequence or are separated by some

other material depends crucially on the position of the second complementizer
within the embedded clause (and of course on the presence of other devices
coding the same modality as would have been coded by a complementizer).
This dependence will be illustrated by deontic complementizers in three
languages: Polish, which does not have inflectional devices to mark obliga­
tion with respect to first and third person but has two complementizers;
French, which has the subjunctive mood and a complementizer; and Guider
482 Zygmunt Frajzyngier

(Central Chadic), which does not have inflectional devices and whose marker
of obligation does not occur at the beginning of the embedded clause. In all
three languages, complements of volitional verbs and volitional expressions
do not have markers of obligation if the subjects of the main and the embed­
ded clauses are the same, e.g.:
(12) chce-my jesc
want-1PL eat:INF
'we want to eat' (Polish)
(13) chce jesc
want:3sG eat-INF
'he/she wants to eat'(Polish)
(14) Je veux le savoir
lSG want it know
'I want to know it' (French)
In Guider, one of the equivalents of the verb 'to want' is a periphrastic
construction with the literal meaning 'x's mouth is on . . . '. Such an
expression may be followed by an NP to realize the equivalent of 'X wants an
NP',e.g.:
(15) má-w á kà mbáy lá
mouth-lSG PREP on drink
'I want a drink'
Such an expression may be also followed by a clause, to realize the equivalent
o f ' X wants S', e.g.:
(16) má w á kà zəmá
mouth-lSG PREP on eat
'I want to eat'
(17) má-w á kà zəm hlúá
mouth-lSG PREP on eat meat
'I want to eat meat'
In Polish, if the subjects of the main and the embedded clauses are different,
then the embedded clause is marked for the deontic modality of obligation.
The marker is composed of the free complementizer ze and the enclitic by,
which has a hypothetical function, e.g.:
A Functional Theory of Complementizers 483

(18) chcemy że-by-ś jad-ł owoc-e

want: 1 PL COMP-COMP-2SG eat-PERF fruit-PL:ACC
'we want you to eat fruit'

(19) chce że-by-m jadl owoc-e

want:3sG COMP-COMP-1SG eat-PERF fruit-PL:ACC
'He wants me to eat fruit'

In Guider the marker of obligation g_ ni (g n in non-pausal position) occurs

after the verb of the simple sentence, e.g.:

(20) m l g ní
1PL go OBLIG
'let us go'
(21) m mbát g ni
lPL depart OBLIG
'let us leave'

(22) ki d s w g, n wàyá
2SG cook:-OPT BEN 1SG OBLIG food
'you should cook for me'

The marker of obligation g ni occurs in the embedded clause when the

subjects of the main and the embedded clauses are different. Moreover, the
marker remains in clause-final position and hence there is no complementizer
as it is traditionally understood, i.e. there is no marker at the beginning of the
embedded clause, e.g.:

(23) mà-w à kà d zó- g nî

mouth- ISG LOC on 3 come-PL OBLIG
'I want them to come'
In French when the subjects of the main clause and the embedded clause are
different, the embedded clause is in the subjunctive mood and is introduced by the
complementizer que, an automatic consequence of the mood of obligation, e.g.:

(24) Je veux qu'il le sache

1SG want COMP-3M it know:suBJ
'I want him to know it'
 484 Zygmunt Frajzyngier

7. Two complementizers in Lele

7.1 The system

Lele (East Chadic) codes obligation for first and third person and the impera­
tive for second person through a suffix on the verb. Consequently, as pre­
dicted by the hypotheses presented in this paper, the language should not have
complementizers marking these specific deontic functions. The prediction is
borne out by the facts. In the following sentences the modality of obligation in
the embedded clause is coded by the imperative form of the verb ír 'go'. Note
that the sentences have a de dicto complementizer ná since the main verb is a
verb of saying:
(25) ń-yá bé-gè ná gé ir-à
lSG-tell BEN-3PL COMP 3PL gO-IMPER
T told them to go'
(26) yà-di bú-dú ná ir-à
Say-3M BEN-3F COMP gO-IMPER
'he told her to go'
The doubt in truth marker sàn, a marker of epistemic modality, occurs in
clause-final position and therefore there is no sequence of two modality
markers, or complementizers, e.g.:
(27) ná-i né báy gō ségrè sàn
COMP-3M COP person DEM hunter DOUBT
'hej said that hej is a hunter' (it may be that he only pretends to be one)2
cf.
(28) cánígè ná-í nè báy gō ségrè
Canige COMP-3M COP person DEM hunter
'Canige j said that hej is a hunter'
Lele does not, however, have an inflectional form of the verb that indicates
hypothetical or irrealis mood, which in Indo-European languages often con­
veys hearsay information. According to our prediction, in the absence of
inflectional markers, if such a mood were to be marked, a complementizer
would be one of the ways to realize it. In what follows I show that Lele indeed
has a complementizer whose function is to mark less than direct evidence.
A Functional Theory of Complementizers 485

7.2 Complementizer go

This complementizer is used with verbs of perception. It has the epistemic

function of indicating direct perception on the part of the subject of the main
clause. The complementizer occurs at the beginning of the embedded clause,
hence the structure of a sentence with this complementizer is as follows:
Subject V (Object) gō S, e.g.:

(29) n-dénlí kúlbá lám-do gō gè jè wál-dù

lSG-hear cow noise-3F COMP INDEF PROGR slaughter-3F
'I heard the cow being slaughtered'

(30) rj-dérjlí lám-di gō jè èjé

lSG-hear noise-3M COMP PROGR come
T heard him coming'

(31) n-gòl-í gō jè wàl-di kúlbá

lSG-see-3M COMP PROGR slaughter-3M cow
T saw him slaughter a cow'

(32) n-gôl-dù gō jè wàl-dú kúlbà

lSG-see-3F COMP PROGR slaughter-3F cow
T saw her slaughter a cow'

The importance, for the proposed hypothesis, of the complementizer gō as the

marker of direct perception is that it codes de re modality. English, for the
same kind of modality, uses either infinitive clauses or '-ing' complements, as
illustrated by translations of the examples given above and as described in
Frajzyngier and Jasperson (1991), and to a certain extent in Bolinger (1968).
The complementizer gö is identical with the relative clause marker and a
subordinator introducing purpose clauses, e.g.:
(33) báíndí gō éjè ná ín-ín kán nè bügán-ín
man REL come CONJ ASSOC-lsG here COP friend-1SG
'the man who brought me here is my friend'

(34) jènè-dù kányà lé gō kílé cáanì kùsùgù-ni

prepare 3 F food PURP trade away market-AT
'she prepares food to sell it on the market'
486 Zygmunt Frajzyngier

(35) n-jè né kïré gö ódé

lsG-PROG look means PURP go
'I am looking for the means to go'
Given the fact that relative clause markers and complementizers in many
Chadic and other languages often derive from demonstratives, it is more than
likely that the de re complementizer gō and the relative clause marker derive
from the same source, probably a demonstrative whose traces can be found in
the pronominal system as well.

7.3 Complementizer ná

The complementizer ná (sometimes realized with mid rather than high tone,
cf. Palayer 1981) occurs after verbs of saying and thinking and indicates that
what follows is in the domain de dicto, e.g.:
(36) cánígè ná dá-ì kólôŋ nè báy gō ségrè
canige COMP 3M there COP man DEM hunt
'Canige said that he [one over there] is a hunter'
(37) kìrbí-dí ná-ì nè báy gō ségrè
think-3M COMP-3M COP man DEM hunt
'he i thinks that hei is a hunter'
Embedded interrogative clauses that follow the main verb tón 'ask' are also
introduced by the complementizer ná, e.g.:
(38) ŋ-tón ná é-jé gè wán gà
lSG-ask COMP come-PERF 3PL nevertheless INTERR
'I asked whether they nevertheless came'
(39) rj-tón ná bálè dí gō ôje'-ŋ gà
lSG-ask COMP be able 3M COMP help-lSG INTERR
'I asked whether he could help me'
(40) rj-tón gí ná gí bál gō oje'-rj gà
lSG-ask 2M COMP 2M be able COMP help-lsG INTERR
'I asked you whether you could help me'
(41) ŋ-tôn ná wéî bà bál gō oje'-rj gà
lSG-ask COMP who be able COMP help-lSG INTERR
'I asked who could help me'
A Functional Theory of Complementizers 487

As is the case with de dicto complementizers in other languages (cf. Kirsner

and Thompson 1976; Frajzyngier 1993,1991a, and Frajzyngier and Jasperson
1991), ná indicates indirect evidence when it occurs after verbs of perception:
(42) ŋ-gol ná wàl-dí kúlbà
lSG-see COMP slaughter-3M cow
T saw that he slaughtered a cow'
(43) cáníg dérŋí ná gírbí-dú jè kôjô kô-rô
Canige hear COMP forget-3F PERF hoe POSS-3F
'Canige heard that she forgot her hoe'
Another piece of evidence that ná is a de dicto complementizer is provided by
causative constructions. In one construction, the main clause has the verb jíbé
'push', 'cause', followed by a nominal or a pronominal object, followed by a
clause with an infinitival verb, i.e., a verb without any subject affixes. The
construction has the meaning: 'make X do Y', e.g.:
(44) jíb-íŋ-dí jé dà kán
cause-1SG-3M come LOC here
'he made me come here'
(45) jíb-íŋ-dí jè e jè nà
cause-1SG-3M come there
'he made me go there (close to him)'
If the VP with the verb jíbé is followed by the complementizer ná and then by
a clause with an overt subject whose verb is in the imperative form, the
construction has the meaning 'tell X to do Y', e.g.:
(46) ŋ-jîb-dù ná nà kányà lé
lsG-cause-3F COMP cook:IMPER thing eating
'I told her to cook some food'
(47) jíb-íŋ-dí ná ŋ-'údá
pUSh-lSG-3M COMP lSG-go:lMPER
'he told me to go'
(48) ŋ-jîbí-î ná dí údá
lsG-push-3M COMP 3M go
T told him to go'
488 Zygmunt Frajzyngier

(49) dàdú jíb-gè ná kùjà-gé kûnî kùlèn-di

3F cause-3PL COMP clean:IMPER-3PL house belly-3M
'she told them to clean the inside of the house'
As indicated by the translations, the sentences above do not entail that the
action of the embedded clause has taken place. But the crucial evidence comes
from the fact that the above sentences may be followed by a negative proposi­
tion denying the accomplishment of the action of the embedded clause, e.g.:
(50) jìb-íŋ-dí ná ŋ-'údà bá ŋ'é dé béi
pUsh-lSG-3M COMP lSG-go:IMPER CONJ lSG-go NEG ?
'he told me to go but I didn't '
Hence the function of the complementizer ná is to transfer the potential reality
into the de dicto domain.
An additional piece of evidence for the de dicto/de re distinction between
gō and ná is provided by complements after the verb narí 'refuse' and the
phrase nàrí.. . kîré 'forbid', where the element kîré is an independent lexical
item with the meaning 'way', 'road', 'means'. The structure of the sentence
with an embedded clause is as follows: (Nominal Subject) nàrí Object (Pro­
nominal Subject) kîré gō S. Note that the pronoun occurring after the verb
nàrí is an object pronoun rather than a benefactive. Since kîré must be
considered an object, the main clause in fact has two objects, e.g.:
(51) nàrî-î-dú kírè gō kílé bú-dù gúrbálò
refuse-3M-3F way COMP buy BEN-3F cloth
'she forbade him to buy her a cloth'
If the verb nàrí is not followed by the noun kîré then it is treated as a verbum
dicendi and has the complementizer ná, and the verb of the embedded clause
occurs in the imperative form, e.g.:
(52) rj-nàrí-î ná ír-à
lSG -refuse-3M COMP go-IMPER
'I forbade him to go'

7.4 Complementizers after the verb sèn 'to know'

The verb sèn 'to know' may be followed by the single complementizer gō, e.g.:
A Functional Theory of Complementizers 489

(53) ŋ-sèn gō ngú gírbí köjd kú-ŋgù

lSG-know COMP 2PL forget hoe POSS-2PL
'I know that you(pl) forgot your hoes'
It may also be followed by two complementizers in a sequence whose order
must be gō ná, e.g.:
(54) ŋ-sèn gō ná gírbí-gè köjd kè-gè
lSG-know COMP COMP forget-3PL hoe POSS -3PL
'I know that they forgot their hoes'

When gō is the only complementizer it indicates that the source of knowledge

is the speaker's personal experience. The evidence for this hypothesis is
provided by the fact that a sentence with gō as the only complementizer may
be followed by a clause corresponding to 'because I have seen it' and may not
be followed by a clause corresponding to 'because I was told so', e.g.:
(55) ŋ-sèn gō gl gírbí jè kèjè kèm bè kōlō ŋ-gól
lSG-knowCOMP2M forget PERF hoe 2M:POSS give cause lSG-see
'I know that you forgot your hoe, because I have seen [you forget­
ting the hoe]'

(56) ŋ-sèn gō mé gírbí kôjd kèrè *bè kōlō

lSG-know COMP 2 F forget hoe 2 F POSS give cause
gé yá-rj
INDEF tell-lSG
for 'I know that you forgot your hoe, because I was told so'
If the verb sèn is followed by the two complementizers gö and ná then the
sentence means that the source of knowledge is hearsay. The evidence for this
function is provided by the fact that a sentence with the two complementizers
may be followed by a clause corresponding to 'because I have been told so' and
may not be followed by a clause corresponding to 'because I have seen it', e.g.:

(57) rj-sèn gö ná gí gírbí jè kôjô kôm bè

lSG-know COMP COMP 2M forget DIST hoe 2M.POSS give
kölö gé yá-rj
cause INDEF tell-lsG
'I know that you forgot your hoe, because I was told so'
490 Zygmunt Frajzyngier

(58) *ŋ-sèn gō ná gî gírbí jè kôjô körn

lSG-know COMP COMP 2M forget DIST hoe 2M:POSS
bè kōlō ŋ-gôl
give cause lSG-see
'I know that you forgot your hoe, because I saw it'
Another piece of evidence that the function of ná is to indicate hearsay as
the source of knowledge is provided by the fact that in an imperative sentence
with the verb 'know', i.e., in a sentence where the complement is obligatorily
a fragment of speech rather than a fragment of reality, the complementizer is
ná rather than gō e.g.:

(59) sínà ná . . .
know COMP
'know that
The last piece of evidence I present here for the functions of ná and gō is the
fact that one cannot use both complementizers after a verb of perception. Thus
the following sentences are ungrammatical:
(60) *ŋ-gôl-í gō ná wàl-dí kûlbà
lSG-see-3M COMP COMP slaughter-3M cow
for 'Ï saw him slaughter a cow'
(61) *ŋ-gôl-i ná gō wàl-dí kúlbà
lSG-see-3M COMP COMP slaughter-3M cow
for T saw him slaughter a cow'
The ungrammaticality stems from the fact that the main verb of perception is
followed by the direct perception marker gō and also by the indirect percep­
tion marker ná, creating an internal contradiction within the clause.

8. Two complementizers in Mupun

The importance of the data from Mupun lies in the fact that the language has a
clause-internal epistemic marker coding doubt in the truth of the proposition,
and inflectional markers coding imperative and obligation. According to the
proposed hypothesis, neither doubt in truth nor obligation should be realized
by complementizers. The prediction is fully borne out by the data. The
A Functional Theory of Complementizers 491

language also has two other modality markers, both occurring at the begin­
ning of the clause: a prohibitive marker and an interrogative marker. Accord­
ing to our predictions, such a marker may be preceded by a de dicto
complementizer, resulting in a sequence of two complementizers. The data
support this prediction as well. A discussion of the four cases follows.

8.1 Epistemic marker

The epistemic modality marker páa, which encodes a hypothetical mood,

occurs after the subject of the embedded clause. This marker and its theoreti­
cal implications have been described in Frajzyngier (1985). As predicted by
the proposed hypothesis, there is no other complementizer in Mupun coding
the hypothetical function.
The verba dicendi complementizer is né (described in Frajzyngier 1991b,
1993). There is a productive construction with a verb of saying followed by an
embedded clause with the doubt in truth marker páa (in the examples to
follow, L stands for 'logophoric'):
(62) a sat nd wu paa mbә yo muan
2M say COMP 3M DOUBT FUT go trip
'you said that he will go on a trip (but he may or may not go)'
cf.
(63) a sat nd wu mbd yo muan
2M say COMP 3M FUT go trip
'you said that he will go on a trip'
(64) wu sat nd din paa cß cin hankuri
3M say COMP 3M.L DOUBT 3M.L make patience
'he said that he was patient (but I have my doubts)'
There is a shift in the function of páa from doubt in truth marker to hypotheti­
cal when the subject of the embedded clause is first or second person, as
illustrated by the following examples, where the embedded clause expresses a
state that may occur only if the subject consents to the request:
(65) wu sat nә ha paa a cin hankuri
3M say COMP 2M DOUBT 2M do patience
'he asked you to be patient (if you please)'
 492 Zygmunt Frajzyngier

(66) wu sat nd wa paa a ein hankuri

3M say COMP 2M:EMPHATIC DOUBT 2M do patience
'he asked you to be patient (if you please)'

8.2 Deontic modality

The optative and imperative in Mupun are marked by low tone on subject
pronouns in both main and embedded clauses, e.g. (only tones on pronouns
are marked):
(67) mó grdp pupwap
3PL cut:PL fish
'they cut a fish into pieces'
(68) mo grdp
3PL cut:PL
'they should cut it into pieces'
Since the mood of obligation is marked by inflectional devices, there is, in
accord with our prediction, no complementizer marking the mood of obliga­
tion. The embedded clause may occur without a complementizer or be pre­
ceded by a de dicto complementizer nә, e.g.:
(69) dәm à sat mo grdp
go 2M tell 3PL cut:PL
'go and tell them to cut it into pieces'
(70) mo sat nd yì cin dik
3PL say COMP 2 F do marriage
'they said that you should marry'
cf.
(71) mo sat nd yì ein dik
3PL say COMP 2 F do marriage
'they said that you got married'
(72) mo sat nd à la mat
3PL say COMP 2M take wife
'they said that you should marry'
A Functional Theory of Complementizers 493

cf.
(73) mo sat nә á la mat
3PL say COMP 2M take wife
they said that you married'

8.3 Prohibitive marker

Unlike the mood of obligation, which is coded by inflectional devices, the

prohibitive mood is marked by a separate lexical item táají, occurring at the
beginning of the clause, e.g.:
(74) taaji yidam pee n-an
COMP 2Fbother place PREP-1SG
'don't bother me'
In a complex sentence the de dicto complementizer nә may be followed by the
complementizer táajì. When this combination occurs, the distribution of
functions between the two complementizers is as follows: The first
complementizer indicates that what follows is in the de dicto domain, which is
an epistemic function. The second complementizer functions as a marker of
the deontic modality of the embedded clause. Thus each of the complementiz­
ers encodes a different modal function, e.g.:
(75) n-sat n-wur nә taaji gwar dam pee n-an
lsG-tell PREP-3M COMP COMP 3M:ADR bother place PREP-1SG
'I told him not to bother me'
(76) mo sat puo n-an nd taaji n-ddm n-jos kas
3PL say word PREP-1SG COMP COMP lSG-go PREP-Jos NEG
'they warned me not to go to Jos'

8.4 Interrogatives

Mupun has a rich system of clause-final interrogative markers for yes/no

questions. Some of these are -a, -wo, a ri , -on, and -e. Presumably these
would not be considered complementizers because they do not occur between
the main and the embedded clauses, e.g.:
494 Zygmunt Frajzyngier

(77) n-kak nd get mu sin makaranta ndaŋ

ISG-think COMP PAST IPL make school follow
together-INTERR
siak-e
T think we went to school following each other, didn't we?'
(78) n-tal pә wur a nә a gwar a miskoom-e
lSG-ask PREP 3 M COP 3PL COP 3M:ADR COP chief-INTERR
T asked him whether he was a chief
The difference between embedded interrogatives and main clause interroga­
tives can be seen in yes/no questions. Embedded yes/no questions, which may
end with any of the interrogative markers listed above, also have an interroga­
tive complementizer két. Since the verb of the main clause is a verb of saying,
the embedded clause may also be preceded by the verba dicendi com­
plementizer nә or by ko, giving the double complementizer construction nә
két or ko két. There is even a triple complementizer construction when ko két
occurs after the complementizer nә, e.g.:
(79) mo tal pd an nd ko ket n-man war-a
3PL ask PREP lSG COMP COMP COMP 1SG-knOW 3F-INTERR
'they asked me if I knew her'
(80) mo tal pd an nd ko ket n-man war-wo
3PL ask PREP ISG COMP COMP COMP lSG-know 3F-INTERR
'they asked me if I knew her'
(81) mo tal pd war nd ket pa man an-a
3PL ask PREP 3 F COMP COMP 3F.L know ISG-INTERR
'they asked heri whether shei knew me'
The verba dicendi complementizer nd is optional even if the main clause verb
is a verb of saying, e.g.:
(82) n-tal pd wur (nә) ko ket wurkә kes akaranta
lSG-ask PREP 3 M (COMP) COMP COMP 3M PERF finish school
'I asked him i if hei has finished school'
If the embedded clause is fronted, then két is the only complementizer before
the embedded clause, e.g.:
A Functional Theory of Complementizers 495

(83) ket mo wul n-tulu be ba n-man kas

COMP 3PL arrive PREP-home CONS NEG lSG-know NEG
'I don't know if they arrived home'
It may be argued that the presence of the interrogative complementizer
két and a clause-final interrogative marker contradicts the principle stated
earlier that there should not be two markers coding the same modality. In fact
there is no contradiction here, as the function the marker két is to delimit the
scope of the question, from the whole sentence to just the embedded clause.

9. Two complementizers in Polish

The importance of the discussion of the Polish data is as follows: There are no
clause-internal markers of epistemic modality nor of deontic modality for the
first and third person. In accordance with the hypothesis of the present paper
one can expect that such a modality may be expressed through modal verbs,
adverbs, or a complementizer. If the modality is expressed through a comple­
mentizer one can expect that after verba dicendi, and possibly after verba
sentiendi, a sequence of at least two complementizers may occur, of which
one is a complementizer of verbs of saying and the other codes another
epistemic or deontic modality.
The Polish enclitic by, mentioned earlier in the paper, indicates a hypo­
thetical or irrealis mood. It is clearly a modality marker and can be attached to
verbs, nouns, pronouns, adjectives, complementizers, adverbial clause mark­
ers, and exclamations, e.g.:
(84) byt by-m w Warszawie gdy-by. . .
be-PERF HYP-ISG in Warsaw when-HYP
T would have been in Warsaw if. . .'
The enclitic by can be added to the morpheme ni to express the epistemic
modality of doubt in truth of the proposition; while by added to the morpheme
o encodes the deontic modality expressing wish. The resulting structures ni-by
and o-by occur in clause-initial position, i.e., in the position of complementiz­
ers such as the interrogative czy, e.g.:
(85) Ni-by byt w Warszawie
COMP be:3M:sG:PERF PREP Warsaw:DAT
'Apparently he was in Warsaw'
496 Zygmunt Frajzyngier

(86) O-by byt w Warszawie w lipcu

COMP-COMP be:3M:SG:PERF PREP Warsaw:DAT PREP July
T wish he were in Warsaw in July'
The de dicto complementizer in Polish is ze, e.g.:
(87) Powiedzialem ze byłem w Warszawie
say:PERF:PAST: 1SG COMP be:PAST: 1SG PREP Warsaw:DAT
T said that I was in Warsaw'
As in many other languages, the de dicto complementizer in Polish can be
used after verbs of perception to indicate indirect evidence, e.g.:

(88) widzialem ze byl w dom-u

see:PERF:PAST-1SG COMP be:PAST:3sG.M PREP home-GEN
'I saw that he was at home'
Given the fact that hypothetical markers in Polish occur at the beginning of
the clause, one can expect that after verba dicendi there will be two comple­
mentizers between the main and the embedded clause. This is indeed the case.
The following example illustrates the occurrence of an epistemic complemen­
tizer after the de dicto complementizer:

(89) Powiedzial ze ni-by byl w Warszawie

say:PERF:PAST:3sG COMP COMP be:PAST:3sG PREP Warsaw:DAT
'He said that he was in Warsaw' (I have my doubts about it)
The sequence ze oby is ungrammatical. Instead the particle by can be added
directly to the de dicto complementizer że with the resulting meaning of the
mood of obligation, e.g.:
(90) Powiedzialem ze-by przyjechal
say:PERF:PAST: 1SG COMP-COMP come:PAST:3sG
do Warszawy
PREP Warsaw:GEN
'I told him to come to Warsaw'
(91) Powiedzial ze-by-m spalil papier-y
say:PERRPAST:3sG COMP-COMP-1SG burn:PERF paper-PL
'He said that I should burn the papers'
Thus it is possible to have in Polish a sequence of two or three complementizers,
each performing a different modal function.
A Functional Theory of Complementizers 497

10. Complementizers and inversion in English

I am not aware of any attempts to explain the lack of subject-auxiliary

inversion in embedded interrogative clauses, such as:
(92) *I asked whether should he go
(93) I asked whether he should go
and the presence of inversion in the simple clause, e.g.:
(94) Should he go?
(95) He should go
The hypothesis proposed in the present paper explains the lack of inversion in
a principled way. If we assume that subject-auxiliary inversion encodes the
interrogative modality (cf. Quirk and Greenbaum 1975), and that the comple­
mentizer 'whether' encodes the same interrogative modality, then the two
devices should not occur together in the same clause. And this is indeed the
case. In Old English, the inversion was a much more widespread device than it
is today because it included not only subject-auxiliary but also subject-verb
inversion. Unlike in contemporary English, the interrogative main clause in
Old English could be marked by the particle hwaeðer, corresponding to
contemporary 'whether' (cf. Closs Traugott 1972:73). The interesting fact
about this particle is that when 'whether' was used in the main clause there
was no subject-auxiliary or subject-verb inversion. Similarly in contemporary
English, there is no inversion in embedded interrogative clauses, e.g.:

(96) Hwaeðer ge nu secan gold on treowum ?

'Whether you now seek gold in trees?
'Do you now seek gold in trees?' (Closs Traugott 1972:73)
This fact is fully predictable from the proposed hypothesis. 'Whether' marks
interrogative modality, the same modality that is marked by subject-auxiliary
or subject-verb inversion. Therefore, since the use of the two markers would
be redundant, only one device coding interrogative function is used in the
interrogative clause.
What remains to be explained is why there is inversion in simple 'WH'
interrogative clauses, but no inversion when these clauses are embedded in
another clause.
498 Zygmunt Frajzyngier

(97) Which books have you lent him?

When will you come back?
How did you mend it? (Quirk and Greenbaum 1975:197)
(98) I asked which books *have you lent him
I asked when *will you come back
I asked how *did you mend it
cf.
(99) I asked which books you have lent him
I asked when you will come back
I asked how you mend it
Here is an explanation for the presence of inversion in specific interroga­
tives in the main clause and the lack of inversion when these clauses are
embedded. I have proposed elsewhere for Chadic languages (Frajzyngier
1985) that the Chadic equivalents of 'WH' interrogatives were actually not
interrogative markers but rather indicators of what the question is about. The
interrogative modality in Chadic is usually coded by clause-final interrogative
particles and by intonation (cf. Leben 1989). I propose that the situation in
English reflects to some degree an older system which in many features
resembled the system postulated for Chadic. In Indo-European languages
interrogative pronouns are identical with the indefinite pronouns roughly
corresponding to 'someone' and 'something'. The contemporary 'what',
'who' and 'which' also derive most probably from Old English indefinite
pronouns hwaet 'whatever', hwa 'whoever' and hwilc 'any'. Closs Traugott
illustrates this function with a contemporary example:
(100) Who pays the piper calls the tune (Closs Traugott 1972:153)
I propose that the WH-words in main clauses in contemporary spoken English
are not markers of interrogative modality but rather indicate what component
of the clause the question is about. The interrogative modality is provided by
at least two other devices: subject-auxiliary inversion (with auxiliary 'do'
provided according to familiar rules) and by interrogative intonation. Without
the two devices the clause does not carry the interrogative meaning. I propose
that in the complex sentence the WH-word of the embedded clause becomes a
complementizer, making the subject-auxiliary inversion a second, and hence,
superfluous, device to mark interrogative modality.
A Functional Theory of Complementizers 499

11 Conclusions

I have shown in this paper that morphemes labeled complementizers consti­

tute a device to code the modality of a clause, and as such they are related to
other markers of modality such as modal verbs, modal adverbs, inflectional
modality markers, devices to code interrogative function, etc. The absence of
complementizers in indicative main clauses in some languages is explained
by the fact that such clauses are inherently marked as conveying a speaker's
belief in the truth of the proposition. Such clauses are also unmarked with
respect to deontic modality. It is possible that in some languages complemen­
tizers may acquire a purely formal function and became obligatory markers of
embedded clauses with different modal characteristics. It is also possible that
complementizers may lose their function of coding modality completely.
The explanation provided in this paper may be useful also to those formal
theories that postulate the existence of COMP as a component of some types of
clauses, on the condition that COMP is understood as representing modality of
the clause. Whether these theories will accept this reinterpretation of COMP is
an entirely different matter.

ABBREVIATIONS

BEN Benefactive LOC Locative

COMP Complementizer M Masculine
CONJ Conjuction NEG Negation
COP Copula OBLIG Marker of obligation
DAT Dative PAST Past tense
DEM Demonstrative PERF Perfective
DOUBT Doubt in truth marker PL Plural
F Femine POSS Possessive marker
FUT Future Possessive pronoun
GEN Genitive PROGR Progressive
HYP Hypothetical PURP Purpose marker
IMPER Imperative REL Relative marker
INDEF Indefinite SG Singular
INTERR Interrogative SUBJ Subjunctive
 500 Zygmunt Frajzyngier

NOTES

* Work on this paper was supported by a grant from the National Endowment for the
Humanities, No. RT-21151-90. I would like to acknowledge the generous help of
colleagues, through conversations and correspondence. For the French data I am grateful
to Anne Abeille, Philippe Alcouffe, and Sylvie Bruno; for information on Quebecois
French and some bibliographic pointers I am grateful to Gary Coen, and for Dutch data
(not used in the present paper) and references I am grateful to Jan Odijk, J. Zwart, and
Gerrit Dimmendaal. I would like to thank Eric Atwell, University of Leeds, for checking
the LOB corpus for me for the occurrence of two complementizers in a sequence in
English (none were found, but cf. Caroll 1983). I gathered data on Lele, an East Chadic
language spoken in Chad, during fieldwork in North Cameroon in the summer of 1990.
I gathered data on Guider, a central Chadic language spoken in North Cameroon, in the
summers of 1990 and 1991. Data on Mupun, collected over a period of many years, are
taken from Frajzyngier 1991b and 1993.
I would like to thank Immanuel Barshi, Karen Ebert, Bernd Heine, Elizabeth
O'Dowd, David Pesetsky, Erin Shay and participants of the Symposium on Mood and
Modality, especially John Haiman, Gillian Sankoff, Eve Sweetser, and Werner Abraham,
for questions raised and comments on various aspects of the present paper. None of them
is, however, responsible for any shortcomings of this paper.
1 For a discussion of by in Russian see Brecht 1977.
2 In Lele, as in many Chadic languages, the verb 'to say' is often omitted. The only trace
of its presence is the complementizer.

REFERENCES

Bickerton, Derek. 1981. Roots of Language. Ann Arbor: Karoma.

Bolinger, Dwight. 1968. "Entailment and the Meaning of Structures." Glossa 2.2.119-127.
Bolinger, Dwight. 1972. That's That. The Hague: Mouton.
Bouchard, Dennis. 1984. On the Content of Empty Categories. Dordrecht: Foris.
Bourciez, Edouard. 1967. Eléments de Linguistique Romane. Paris: Klincksieck.
Borsley, Robert D. 1986. "Prepositional Complementizers in Welsh". Journal of Linguistics
22.67-84.
Brecht, Richard. 1977. "Ctoby or cto and by". Folia Slavica 1.1.33-41.
Bresnan, Joan W. 1970. "On Complementizers: Toward a Syntactic Theory of Comple­
ment Types". Foundations of Language 6.3.297-321.
Bybee, Joan with Revere Perkins and William Pagliuca. (1994) The Grammaticization of
Tense, Aspect and Modality in the Languages of the World. Chigago: University of
Chigago Press
Caroll, Susanne. 1983. "Remarks on For-To Infinitives". Linguistic Analysis 12.4.415-451.
Chomsky 1986a. Barriers. Cambridge, Mass.:MIT Press.
Chomsky 1986b. Knowledge of Language. New York: Praeger.
Chomsky, Noam. 1992. A Minimalist Program for Linguistic Theory. {MIT Occasional
Papers in Linguistics I.) MIT: Cambridge, MA.
A Functional Theory of Complementizers 501

Cichocki, W. 1983. "Multiple Wh-Questions in Polish: A Two COMP Analysis". Toronto

Working Papers in Linguistics, 4.53-71
Comorovski, Ileana. 1986. "Multiple Wh-Movement in Rumanian". Linguistic Inquiry
17.1 171-177.
Dasgupta, Probal. 1983. "Bangla Equatives, Complementizers, Final Foci, and Root".
Linguistic Analysis 11.2.103-137.
Dimmendaal, Gerrit F. 1989. "Complementizers in Hausa". Current Progress in Chadic
Linguistics ed. by Z. Frajzyngier. 87-110. Amsterdam: Benjamins.
Emonds, Joseph E. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris.
Frajzyngier, Zygmunt. 1985. "Truth and the Indicative Sentence". Studies in Language
9.243-254.
Frajzyngier, Zygmunt. 1987. "Truth and the Compositionality Principle: A Reply to
Palmer". Studies in Language 11.211-217.
Frajzyngier, Zygmunt. 1991a. "De Dicto Domain in Language". Approaches to Grammat-
icalization. Ed. by Traugott, Elizabeth C. and Bernd Heine. Volume 1. Amsterdam:
Benjamins, 219-252.
Frajzyngier, Zygmunt. 1991b. "Typology of Complex Sentence in Mupun". Phrase
complexe dans les langues tchadique. ed. by Herrmann Jungraithmayr and Henry
Tourneux, 27-72. Paris: Geuthner.
Frajzyngier, Zygmunt. 1993. A Grammar of Mupun. Berlin: Reimer.
Frajzyngier, Zygmunt. in press. "Two Complementizers of Lele". Studia Chadica and
Hamito-Semitica. ed. by Rudolf Leger and Dymitr Ibriszimow.
Frajzyngier, Zygmunt and Robert Jasperson. 1991. "That Clauses and Other Comple­
ments". Lingua 83.133-153.
Givón, Talmy. 1980. "The Binding Hierarchy and the Typology of Complements".
Studies in Language 4.333-377.
Givón, Talmy. 1990. Syntax. A Functional Typological Introduction. Vol. 2. Amsterdam:
Benjamins.
Heine, Bernd; Ulrike Claudi and Friederike Hünnemeyer. 1991. Grammaticalization. A
Conceptual Framework. Chicago: University of Chicago Press.
Jensen, Frede. 1990. Old French and Comparative Gallo-Romance Syntax. Tübingen:
Niemeyer.
Joseph, Brian D. 1985. "Complementizers, Particles, and Finiteness in Greek and the
Balkans". Folia Slavica 7.3.390-411.
Kac, Michael. 1978. Corepresentation of Grammatical Structures. University of Minne­
sota Press.
Kirsner, Robert, and Sandra A. Thompson. 1976. "The Role of Pragmatic Inference in
Semantics: A Study of Sensory Verb Complements in English". Glossa 10:2, 200-240.
Koster, Jan. 1987. Domains and Dynasties. Dordrecht:Foris.
Lawal, Nike S. 1986. "A Note on Yoruba Subordinator 'ti'". Journal of West African
Languages 16.2. 27-33.
Leben, William. 1989. "Intonation in Chadic: An Overview". Current Progress in Chadic
linguistics ed. by Z. Frajzyngier. 199-218. Amsterdam: Benjamins.
Lefevbre, Claire. 1980. "Cases of Lexical Complementizers in Cuzco Quechua and the
Theory of COMP". Journal of Linguistic Research 1.2.91-112.
502 Zygmunt Frajzyngier

Noonan, Michael. 1985. "Complementation". Shopen 1985:42-140.

Palayer, Pierre. 1981. "Notes de grammaire." Livre de lecture lele. Ed. by Martine
Garrigues Cresswell avec la participation de Christophe Weibegue. Sarh, Tchad:
Centre d'Études Linguistiques, Collège Charles Lwanga.
Palmer, F. R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Palmer, F.R. 1987. Truth Indicative? Studies in Language 11.1-2, 206-210.
Quirk, Randolph and Sidney Greenbaum. 1975. A Concise Grammar of Contemporary
English. New York: Harcourt, Brace, Jovanovich.
Ransom, Evelyn N. 1986. Complementation: Its Meaning and Forms. Amsterdam: John
Benjamins.
Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press.
Rosenbaum, P. 1967. The Grammar of English Predicate Complement Constructions.
Cambridge, Mass.: MIT press.
Rudin, Catherine. 1986. Aspects of Bulgarian Syntax: Complementizers and WH-con-
structions. Columbus, Ohio: Slavica Publishers.
Shopen, Timothy, (ed). 1985. Language Typology and Syntactic Description. Cambridge:
Cambridge University Press.
Traugott, Elizabeth Closs . 1972. A history of English Syntax. New York: Holt, Rinehart
and Winston.
Visser, F. Th. 1969. An historical Syntax of the English Language. Leiden: Brill.
Wachowicz, Krystyna A. 1974. "Against the Universality of a Single WH-Question
movement". Foundations of Language 11.155-166.
 The Semantic Development of Past Tense
Modals in English

Joan L. Bybee
University of New Mexico

The English modal auxiliaries, would, should, might, and could are histori­
cally the Past Tense forms of will, shall, may and can respectively.1 However,
their meaning and usage in Modern English are not derivable from the
combination of past meaning with the meaning of the Present modals. Rather
the uses of the Past Tense modals may be divided into three types: (i)
hypothetical uses; (ii) present tense uses; and (iii) past tense uses.
In Modern English the hypothetical uses are the most common according
to the analysis of spoken and written texts done by Coates (1983). They are
most obvious in the then-clauses of hypothetical conditional sentences. In
some cases the hypothetical meaning occurs along with what is sometimes
called the 'root' meaning of the modal. For instance in (1) would indicates
'hypothetical willingness' (Coates 1983: 211), while in (2) could indicates
'hypothetical ability':
(1) If you had that job lined up, would Fulbright then pay up?
(2) If you helped me, I could finish this in an hour.
In the example in (3), no additional root meaning is present, and the modal
conveys only the hypothetical conditional sense (Coates 1983:214):
(3) / mean we all want be to millionaires, but if we were of course
money wouldn't be worth anything.
The present tense uses of originally Past modals are illustrated in examples (4)
and (5) (from Coates 1983: 58, 152).
(4) You should walk round the ramparts of the old city too.
504 Joan L. Bybee

(5) / think it unlikely actually, but he might do it today.

Much less common are uses of the Past modals in past contexts, although they
do occur, as the examples in (6) and (7) show (Coates 1983: 211, 111):
(6) He believed Mr. Weaver would perform "outstanding service " in
the post.
(7) "Ijust cannot remember a time when I couldn 't swim, " she told me.
The ambiguity of could can be seen in the following bit of wisdom, expressed
on a bumber sticker:
(8) George Bush couldn't run a laundromat.
This statement could be taken to mean that he tried and failed (the past tense
reading), or that even if he tried, he would fail (the hypothetical reading).
Even though these modals are Past in form, the past uses of most of them
are the least frequent of their uses, and for should, one could argue that there
are no past uses at all.
This situation, especially with regard to the hypothetical uses, is paral­
leled in other languages, for it is common for the Conditional Mood to consist
formally of the Past tense of the Future, as in Spanish and French, or the Past
of another modal form, as for instance in Nahuatl (Andrews 1975) or Sierra
Miwok (Freeland 1951). Such facts suggest that there may be something
predictable about the diachronic development of the Past tense forms of
modal verbs and that it may be fruitful to investigate the question of how and
why past tense modals develop hypothetical and present tense uses, and
whether there is any relation between these two developments.
In the following I will trace the semantic development of would and should
only, but I expect the general findings to apply to might and could as well.

1. Should and would in Old and Middle English.

That should and would are originally past tense in meaning is clear from their
uses in Old English texts, such as Beowulf. There we find many clear in­
stances of should signalling destiny, duty or obligation of the subject in the
past, corresponding to sceal, which has the same meaning in the present.
Consider the examples (9) through (11) and their translations, based on
Gordon 1926.
The Semantic Development of Past Tense Modals in English 505

(9) Nœs ßcet forma sid /þœt hit ellen-weorc œfnan scolde. (1464)
'That was not the first time that it (a sword) had to perform a deed
of valor.'
(10) ße he wid ßam wyrme gewegan sceolde. (2400)
'when he had to fight the dragon.'
(11) ßcet ßcet ðeodnes beam geßeon scolde. (910)
'that the king's son was destined to prosper.'
Similarly, wolde is used almost exclusively to signal volition of the subject in
the past, just as wylle was used to express volition in the present. Consider the
examples in (12) through (14). Example (13) refers to a time that is past with
reference to the time of the narrative, which is also past.
(12) wolde self cyning symbel ßicgan. (1010)
'The king himself wished to join in the banquet.'
(13) donne sweorda gelac sunu Healfdenes efnan wolde; (1040-41)
'when the son of Healfdene wanted to practice sword-play.'
(14) woldewig-fruma Wealhßeo secan, cwen to gebeddan. (664-5)
'the war-leader wanted to look for the queen to bed down with.'
By the Middle English period, however, both should and would had made
their way into present contexts, especially with first and second person.
Consider first wolde, which is especially prominent in present usage in Sir
Gawain and the Green Knight, where it can express present volition or
willingness. Consider the following examples:
(15) "I wolde yowre wylnyng worche at my myzt..." (1546)
'"I am willing to do your desire as far as I can...'"
(16) "'Wher is', he sayd, "The gouernour of this gyng? Gladly I wolde
Se that segg in syzt, and with hymself speke raysoun." (224-6)
"Where is', he said, "The lord of this company? I would gladly
See that knight and speak reason with him.'"
To explain the present uses of Past forms, we must refer to the special
properties of modal verbs. Modal verbs, whether they express desire, obliga­
tion, necessity, intention or ability, have in common the semantic property
that they do not imply the completion of the action or event expressed by the
infinitive with which they occur. (This property is in contrast with other
complement-taking verbs such as finish and manage, which do imply comple­
tion [Givón 1973].) In the present tense, however, these modal verbs do imply
506 Joan L. Bybee

a close relation between the agent and the main predicate, which may imply or
predict a future completion of the event or activity of the main predicate.
Especially in first person, present tense modal verbs such as Modern English /
wanna, I'm gonna, I hafta or / can are used to state intentions, make offers and
promises which our interlocutors can expect to be carried out. The same verbs
in the third person are used to report intentions, offers and promises. Simi­
larly, I will and I shall in the Gawain text are used to make promises, state
intentions and resolutions, which one trusts will be carried out.
The situation with these same modal verbs in the past tense is quite
different, however. The relation between the agent and the main predicate is
much more tenuous. He/I wanted to for instance may be used both in contexts
in which the desire was carried out and in contexts in which the desire was not
carried out. As stative verbs, the past forms of modals assert that a state
existed before the moment of speech, but they do not say whether that state
still exists in the present or not. Thus past modals offer two areas of vague­
ness: (i) whether or not the predicate event was completed; and (ii) whether or
not the modality remains in effect.
A modal in past time, then, leaves open the possibility that some condi­
tions on the completion of the main event were not met, and therefore the
modality may still be in effect. That is why when we say, for example,
(17) / wanted to help you.
in a situation where I didn't, it means that there were some conditions that
were not met: something was standing in my way. Interestingly, we can also
say I wanted to help you in a situation in which you still need help, but I do not
intend to help you because something IS standing in my way. In other words,
the existence of PAST blocking conditions is generalized to include the exist­
ence of PRESENT blocking conditions.
Note that the use of a Past Tense in a conditional sentence results in a
hypothetical conditional. The hypothetical conditional is contrasted with the
reality conditional which uses Present tense or a Future auxiliary:
(18) Hypothetical conditional:
If she saw Judy, she would tell her the news.
(19) Reality conditional :
If I see Judy, I will tell her the news.
The difference between the two is that the hypothetical conditional implies
The Semantic Development of Past Tense Modals in English 507

that it is considerably less likely that the condition will still be met. The
difference in likelihood originates from the tense distinction in the following
way: The past in both clauses of the hypothetical conditional sentence signals
that the conditional relation held in past time. The modality in the then-clause
signals that the relation is still in effect. (Cf. If Mike saw her, he told her the
news.) On the other hand, the reality condition with Present tense means that
the conditional relation begins in present time and projects into the future. The
reason that the one framed in the past seems more hypothetical is that the
conditional relation has existed in the past but the condition has not yet been
met, suggesting that it may never be met.
This description of the origins of hypotheticality for past tense modals
predicts that the uses of hypothetical conditionals that span past and present
time are more basic and historically prior to those that are set in the future. As
Suzanne Fleischman (personal communication) has pointed out, a hypotheti­
cal conditional may also refer to the future: If in the next year the interests
rates went down again, I would definitely consider refinancing. No uses of
this sort are found in the texts examined in the study reported below, suggest­
ing that this use is a less common extension of the hypothetical use derived in
the way described above.
The if-clause of a hypothetical conditional always contains a past form,
and the past modals show up here as well. However, it is their use in the main
clause that is more important for explaining how they come to be used in
present contexts.
It is interesting that the then-clause of the hypothetical may take other
modal verbs besides would, such as those shown in (20), even though some of
them are not very highly grammaticized:

(20) I wanted to
I was gonna
If I saw Judy, I was supposed to tell her the news.
I intended to
I might
The Past tenses of all these modal verbs behave in a similar manner: They
allow the interpretation that the modality is still in effect, and the predicate
action will be carried out if the right conditions are met.
For this reason it seems appropriate to interpret Modern English wanted
to as an analogue to wolde in the Late Old English and Middle English period.
508 Joan L. Bybee

I am suggesting that the use of wanted to in present time in Modern English is

possible because it implies that certain conditions on carrying out the desired
predicate may not be met. The so-called polite or remote uses of past tense as in
(21) I wanted to ask you a question.
arise in the same way. That is, (21) implies that there might be conditions that
are unmet. Among these implied conditions is the question of whether the
addressee wants to be asked a question — thus the deferential use of the past
(see Fleischman 1989).
Note that under this analysis, the polite use of the Past tense depends
upon the presence of a modal verb; it is not possible to get a polite reading of
just any verb with the Past tense. Thus He asked you a question is no more
polite than He is asking you a question. Some of the verbs that commonly
have polite readings in past tense across languages are 'be obliged' (English
should, Spanish debía), 'want' (English would, Spanish quería), 'be able'
(English could, Spanish podía), and also mental or emotional state verbs, such
as 'think'. All of these verbs have in common the fact that they are stative and
that their past tense meaning leaves open the possibility that the past state
continues into the present.
The difference between I wyl and I wolde in Gawain illustrates the
conditional nature of the Past form nicely. / wyl is used to state intentions or
willingness and may be taken as making a promise or resolution. Compare the
use of wyl in (22) and (23) to that of wolde illustrated in (15) above.
(22) Quoth Gawayn, "I schunt onez,
And so wyl I no more" (2280-1)
'Gawain said, "I flinched once,
But I won't do it again.'"
(23) "For sothe," quoth that other freke, "so felly thou spekez,
I wyl no lenger on lyte lette thin ernde rizt nowe." (2302-3)
"Truly," said the other (the green knight), "so fiercely you speak,
I will no longer delay your errand (right now).'"
These statements with wyl are waiting on no conditions. The first is a knight's
resolution, which is apparently not conditional, and the second is followed by
action.
Wolde also occurs with the conditions made explicit, as in the examples
in (24) and (25): 2
 The Semantic Development of Past Tense Modals in English 509

(24) And I wolde loke on that lede, if God me let wolde. (1063)
'And I want to see that knight, if God would let me.'
(25) Bot wolde ze, lady louely, then leve me grante... (1218)
I wolde boze of this bed, and busk me better; (1220)
'But if you would, lovely lady, grant me leave...
I would get out of bed and dress myself better.'
The conclusion, then, is that the hypothetical and the present uses of Past
modals are basically the same in semantic content. With the present uses of
wolde, however, the conditions are not stated, they are only implied.
Turning now to schulde in Middle English we find a similar situation.
Shall refers to what is to be. The deontic source may vary: it may be divine
destiny, social obligation or mutual arrangement. Schulde refers to what was
to be and carries the same vagueness of implication as the other modalities:
the action may or may not have been completed; the modality may or may not
still be in effect. In Middle English we find many uses of schulde to refer to
what was to take place, without any implication that it did take place:
(26) ßere watz much derue doel driuen in the sale
That so worthé, as Wawan schulde wende on that ernde...(558-9)
There was great lamenting made in the hall
That so worthy as Gawain was to go on that errand.'
(27) And went on his way with his wyze one,
ßat schulde teche hym to tourne to pat tene place...(2074-5)
'And (he) went on his way with the man
That was to show him how to get to that perilous place.'
As with wolde, all the conditions necessary for the completion of the main
predicate may not be met, so the use of schulde is appropriate in a hypothetical
conditional.
(28) 'For were I worth all fie wone of wymmen alyue... (1269)
fier schulde no freke vpon folde bifore yow be chosen. ' (1275)
'"For if I were worth all the host of women alive...
No man upon earth would be chosen before you.'"
The fact that a hypothetical conditional relation still holds in the present
makes the few present uses of schulde found in Gawain possible. In (29) I
assume the condition to be 'if I looked for you.'
510 Joan L. Bybee

(29) "Where schulde I wale ße, " quoth Gauan, "where is ßy place?"
(398)
'"Where shall I find thee," said Gawain, "where is thy place?'"
(30) And quy ße pentangel apendez to ßat prynce noble
I am in tent yow to telle, ßof tary hyt me schulde: (623-4)
(Author speaking)
'And why the pentangle belongs to that noble prince
I intend to tell you, though it (should) slow my tale.'
(31) "paz I hade nozt of yourez,
zet schulde ze haue of myne." (1815-6)
'"Though I have nothing of yours,
Yet you should have something of mine.'"
One very interesting example, (32), refers to an obligation established in the
past, which is being met in the present. This example shows clearly that the
use of a Past modal may imply the continuation of the modality into present
time.
(32) "At pis tyme twelmonyth ßou toke ßat ße failed,
And I schulde at zis Nwe zere zeply ze quyte. " (2243-4)
'"At this time a year ago you took what fell your lot,
And I am/was obliged at this New Year to promptly repay you.'"
In an embedded or hypothetical context, the original lexical meaning of
obligation or destiny has weakened considerably even in Middle English, so
that it contributes little additional meaning:
(33) Quat! hit clatered in Pe clyff, as hit cleue schulde. (2201)
'Quat! It (a noise) clattered on the cliff as if it would cleave it.'
(34) And he asoyled hym surely and sette hym so clene
As domezday schulde haf been dizt on the morn. (1883-4)
'And he absolved him surely and made him so clean
As (if) doomsday should have been the following morn.'
A very interesting point to note about the distribution of schulde in the
Gawain text is that all of the past uses save one are in embedded clauses — in
complements, relative clauses, as if oriƒclauses — where the main clause is
marked for Past tense. The one exception is where schulde occurs in a main
clause in the main past narrative line, but in this case it has haf with it:
The Semantic Development of Past Tense Modals in English 511

(35) And he schunt for ße scharp, and schulde haf arered;

A rach rapes hym to, ryzt er he myzt,
And ryzt bifore ße hors fete ßay fel on hym alle, (1902-5).
'And he (the fox) swerved for the sharp blade, and should have retreated,
A hound hurries to him, just before he could,
And right before the horses' feet they all fell on him.'

In this Middle English text, schulde never has to convey a past sense by itself.
It was compatible with a past context, but to convey pastness, it had to be
bolstered with have.

2. Should and would in Shakespeare

If we continue to follow should and would into Early Modern English, and
consider a Shakespearean play such as The Merchant of Venice, we find that
the trend toward using should and would in present and hypothetical contexts
has continued to the point where both modals appear in these contexts almost
to the exclusion of past contexts. 3 Consider Table 1.
The present uses continue to be much the same for would although they
grow in frequency. They occur mainly in the first person singular and mean T
want' or T would like'. The present uses of should include the first singular
question, as in (36), the second singular statement of obligation, as in (37),
and a new development, the occurrence of should in complement clauses
where no past time is signalled or implied, as in (38).
(36) What should I say to you ? Should I not say
"Hath a dog money? ..." (1.2.115-6)

Table 1. Uses of should and would in Middle and Modern English.

hypothetical past present other total
should
Gawain 17%(6) 62%(21) 15%(5) 5%(2) (34)
Merchant 58%(30) 6%(3) 21%(11) 14%(7) (51)
Would
Gawain 35%(12) 38%(13) 17%(6) 8%(3) (34)4
Merchant 38%(20) 4%(2) 42%(22) 15%(8)5 (52)
512 Joan L. Bybee

(37) You should in all sense be much bound to him. (V.l. 136)
(38) You teach me how a beggar should be answer'd. (IV. 1.436)
It will be recalled that the hypothetical uses of modals may involve a
retention of some of their lexical meaning, such as hypothetical obligation and
hypothetical willingness. Thus in Gawain we find examples of would which
signal hypothetical or conditional willingness, as examples (15) and (24).
There is also an example in Gawain which is devoid of any volitional meaning
and might be characterized as a conditional prediction:

(39) Bot who-so knew the costes that knit ar therinne,

He wolde hit prayse at more prys, parauenture; (1849-50)
'But whoever knew the qualities that are knit into it,
He would value it more highly, perhaps.'
In such examples, would is a pure conditional, since it contributes no other
meaning to the clause.
In The Merchant of Venice such uses are common, but only for third
person. Would occurs rarely with first person in then-clauses, and when it
does, it signals volition. Compare (40) and (41) with third person, where
would is conditional, to (42) where it is volitional.
(40) Believe me sir, had I such venture forth,
The better part of my affections would
Be with my hopes abroad. (1.1.15-17)
(41) For if they could, Cupid himself would blush
To see me transformed to a boy.
(42) If every ducat in six thousand ducats
Were in six parts, and every part a ducat,
I would not draw them, I would have my bond.
Should complements would by occurring with a pure conditional sense in first
person, and rarely in second or third person then-clauses.
(43) If I could bid the fifth welcome with so good a heart as I can
bid the other four farewell, I should be glad of his approach.
(1.2.121-3)
Another difference between would and should is that should appears fre­
quently in if-clauses with only hypothetical meaning, while would appears
The Semantic Development of Past Tense Modals in English 513

rarely in if-clauses and then only with volitional meaning. Thus should has
come to signal pure conditional or hypothetical in conditional sentences,
while would maintains some of its volitional content in first person and if-
clauses.
The difference between Elizabethan and Modern English with regard to
these modals is that would is replacing should in the first person in then-
clauses, and has in fact already done so in American English, and would is
now appearing in //"-clauses with only hypothetical meaning, where formerly
it conveyed willingness. 6
As for the present uses, my proposal is that the Past modals come to be
used in the present via the hypothetical conditional: originally the present uses
are hypothetical then-clauses without the conditions stated explicitly. As
should illustrates, however, the conditional sense can be lost. The resulting
present tense modal expresses obligation, as shall and should originally did,
but in a weakened sense. The hypothetical meaning has had the effect of
weakening the force of the obligation, through the implication that there are
outstanding conditions. Thus modern should expresses weak obligation. This
may be seen by comparing it to the stronger have to, in examples (44).
(44) I should mail this today, vs. I have to mail this today.
The other English Past tense modals, could and might, when used in a present
context, also express a meaning that is weaker than their Present tense
counterparts, can and may.1 If Past modals evolve universally into present
uses in the way proposed here, they will always express a meaning that is
similar to their Present counterparts, but weakened by hypotheticality.

3. Implications

These facts about the gradual development of the hypothetical and present
senses of Past modals and their loss of past meaning have a number of
implications for the synchronic and diachronic analysis of Past tense modals.
In particular, we have seen that it is the combination of the modal sense and
the past sense that produces the hypothetical reading, and not past alone, nor
the modality alone. Moreover, the hypothetical sense replaces the past sense
over time. Therefore a synchronic analysis of Modern English such as the
classical Chomskyan one (Chomsky 1957), which derives would and should
 514 Joan L. Bybee

by combining Past Tense with will and shall cannot be justified. The case for
should is the clearest, since should no longer has any past readings at all and
has needed have added to it to attain a past reading ever since the Middle
English period. 8
A case can also be made for would not containing any past sense, except
for its past habitual use. 9 Note that the past uses of would in the corpus of
spoken Modern British English examined by Coates represent only 16% of
the total, with 82% of the uses being hypothetical. Furthermore, the small
percentage of past uses occur in past contexts in which would does not on its
own signal past time. I would argue, then, that would agrees with or is
compatible with a past context, but does not contain as part of its own
semantics a reference to past time. The gradual semantic changes that take
place among the English Past tense modals are similar, but do not affect all the
modals at the same time. Despite having some syntactic properties demon­
strating classhood, each modal verb follows its own path and timetable of
development (Bybee 1986).
There are also consequences for the analysis of the meaning of past tense
morphemes. It has been suggested, for instance by Steele (1975), Waugh
(1975), Langacker (1978) and James (1982) that in languages in which a past
tense is used as a hypothetical that the past tense actually means 'remote from
present reality' rather than 'preceding the moment of speech'. James points
out a problem with this approach, which is that in normal contexts, the past
tense, say in English or French, continues to mean 'before the moment of
speech' and cannot be interpreted as merely 'remote' or 'distal'. That is, there
is no justification for claiming that the normal sense of a past morpheme has
become more general and no longer refers to past time just because in some
contexts it means hypothetical.
What all of these authors have failed to stress is that it is not the past tense
alone that is contributing the hypothetical meaning, but rather the past in
combination with a modal verb, a subjunctive mood, a hypothetical marker
(such as if), or, in some cases, the imperfective aspect (Fleischman, this
volume). Moreover, I have argued that the diachronic progression shows that
it is not the 'remoteness' of the past that leads to the hypothetical sense, but
rather the fact that past combined with modality leaves open the possibility
that certain conditions on the completion of the predicate have not been met.
Thus there is no justification for claiming that the meaning of past tense in
some languages is 'removed from present reality'.
The Semantic Development of Past Tense Modals in English 515

Another issue concerns whether or not a 'Conditional Mood' should be

viewed as the 'future of the past'. Since in would and should, as well as in the
conditional mood in other languages, such as Spanish or French, we can see
that formally the 'conditional' consists of the Past tense of the Future marker,
it is tempting to claim that conditional meaning is the combination of past and
future meaning and is historically derived from the future-of-the-past.
Fleischman (1982) makes this claim for the Romance conditional, arguing
that sentences such as (45) show that would may be conditional and future of
the past at the same time.
(45) Peter said he would come (if he got the money).
For the Romance conditional she claims that "the conditional use was a
secondary development contingent upon the earlier use of this form as a future
for past time" (p. 66), though she presents no direct historical evidence for this
position.
There are a number of reasons for rejecting this particular explanation of
the relation between future and conditional, at least for English, and perhaps
generally. First, in the English data, we have seen that the conditional uses of
would and should developed while these two modals and their Present tense
counterparts still signalled their original lexical meaning of 'desire or voli­
tion' and 'obligation or destiny'. Not only did they not wait until will and shall
had become futures, they also did not wait until they had lost their original
modal semantics. That is, I have claimed that it is the particular combination
of these modalities with past time that creates the conditional, not the combi­
nation of past with future. If this hypothesis is correct, then we would expect
to find languages in which a past modal has developed into a conditional
without its present tense counterpart developing into a future. Indeed such
cases exist. For instance in Classical Nahuatl, the form used in 'then'-clauses
of conditional sentences (with the condition implied or explicit) is the Imper­
fect form of the verb meaning 'to want' (Andrews 1975). However, the future
is not formed from the Present of this verb, it is a completely independent
suffixal formation. Similarly, in the east central subdialect of Sierra Miwok,
the Andative (be going to) marker -y:i: modified by the Distant Past marker is
used to form a conditional. However, the Present Andative is not used as a
future in that dialect, although it is so used in a reduced form in the southern
dialect. The evidence that this is not a case of the 'future of the past' is that the
conditional meaning is expressed by the unreduced form of the Andative,
 516 Joan L. Bybee

while in the dialect in which the Andative is used as a future, it is reduced to yi

(Freeland 1951).
Finally, a more general conclusion is that we must expect grammaticizing
semantic elements to merge in some cases, especially if they are always
expressed together, and despite the fact that they may involve different
grammatical categories, such as tense and modality. Our overzealous ten­
dency as linguists to analyze every element down into minimal components
often obscures rather than illuminates the diachronic tendencies for some
types of grammatical meaning to merge.

NOTES

1. An earlier version of this paper was presented at the Eighth International Conference on
Historical Linguistics in Lille, France, August 1987, and appeared in the Buffalo
Working Papers in Linguistics, A special issue for Paul Garvin, 1990, 13-30. Special
thanks are due to Suzanne Fleischman, William Pagliuca and the participants in the
Mood and Modality Symposium for discussion of ideas contained herein. This work was
supported by a fellowship from the John Simon Guggenheim Memorial Foundation.
2. Wolde also occurs in the if-clauses in these examples. Here it means 'want' or 'be
willing' in the past.
3. The Merchant of Venice is written partly in poetry and partly in prose, making it more
appropriate than one of the plays written entirely in verse. Of course, we cannot assume
that Shakespeare's texts give a true representation of spoken Early Modern English. At
the time, however, they apparently sounded natural enough for audiences to accept them,
and moreover, they are the best evidence we have concerning spoken English of that
period.
4. For the Gawain text, the numbers for wolde include nolde (ne + wolde).
5. The other uses of would in The Merchant of Venice are three generic uses in phrases such
as 'as one would say' and four uses to mean 'I wish', in which the complement clause
has a different subject than the main clause.
6. In American English, examples such as the following, where would is purely hypotheti­
cal and conveys no sense of willingness, are frequently heard in colloquial speech:
(i) If I would see her, I would tell her.
1. Must is also considered Past tense in some analyses. However, its strong meaning
suggests that it might derive from the Second Singular Present form, moste.
8. In American English was supposed to conveys past obligation (I was supposed to go
yesterday) and should does not (*/ should go yesterday). Even should with have has
specialized to counter-factual obligation (/ should have gone yesterday).
9. The past habitual uses of would date from the Old English period and develop independ­
ently of the hypothetical uses.
The Semantic Development of Past Tense Modals in English 517

REFERENCES

Andrews, J. R. 1975. Introduction to Classical Nahuatl. Austin: University of Texas Press.

Bybee, Joan L. 1986. "On the Nature of Grammatical Categories: A Diachronic Perspec­
tive." Proceedings of the Second Eastern States Conference on Linguistics, ed. by
Soonja Choi 17-34.
Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.
Coates, J. 1983. The Semantics of the Modal Auxiliaries. London: Croom Helm.
Fleischman, S. 1982. The Future in Thought and Language. Cambridge: Cambridge
University Press.
Fleischman, S. 1989. "Temporal Distance: A Basic Linguistic Metaphor". Studies in
Language 13.1.
Freeland, L. S. 1951. The Language of the Sierra Miwok. (Memoirs of UAL or Indiana
Publications in Anthropology and Linguistics VI.) Bloomington: Indiana University.
Givón, T. 1973. "The Time-Axis Phenomenon". Language 49.800-825.
Gordon, R. K. 1926. Anglo-Saxon Poetry. London: Dent.
James, D. 1982. "Past Tense and the Hypothetical: A Cross-Linguistic Study". Studies in
Language 4.375-403.
Langacker, R. W. 1978. "The Form and Meaning of the English Auxiliary". Language
54.853-82.
Steele, S. 1975. "Past and Irrealis: Just What Does it All Mean?". IJAL41.
Waugh, L. 1975. "A Semantic Analysis of the French Tense System". Orbis 24.436-85.
 Imperfective and Irrealis*

Suzanne Fleischman
University of California, Berkely

1. Introduction

In many languages of the world we encounter a more than chance connection

between the aspectual category imperfective and irrealis modality. The con­
nection manifests itself both synchronically and to a degree diachronically,
and involves the use of verb forms marked for imperfective aspect, particu­
larly but not exclusively where combined with past tense, to express a spec­
trum of meanings and functions subsumable under the modal heading of
irrealis.
Synchronically, for example, we find imperfective pasts (including pro­
gressives) pressed into service in the negotiation of roles and settings that
typically serves as a prelude to children's make-believe games, as in (1):
(1) J: ... Pretend I was moving this up and down and up and down
[moves appendage on a doll]....
Z: And pretend somebody was iceskating on the rink1
Imperfectives appear likewise in the narration of dreams and other less-than-
fully-conscious states, as in the example in (2), from a novel about the
American west, in which the semi-conscious narrator ponders a statement
made by the character Winder:
(2) Now Winder was wanting to know what the hell the stage was
doing on the pass at night anyway. (Walter Van Tilburg Clark, The
Ox-Bow Incident, p. 126).2
 520 Suzanne Fleischman

A marked use of the imperfective past occurs in French — and conceivably

other languages as well — in a style of hypocoristic discourse used by adults
to address small children and pets, as in (3):
(3) Comme il était [IMP] sage! 'Now aren't you a good boy!'
And across languages, we find imperfective verbs used in styles of indirect
quotation, including so-called free indirect discourse.
Diachronically, imperfective indicative forms have come to be used in a
variety of hypothetical contexts, notably in conditionals. And in many lan­
guages, imperfective verbs and verbal periphrases have become grammatical-
ized, or further grammaticalized3, as futures, conditionals, and subjunctives.
The examples cited above will be taken up again in the sections that
follow, where the usages they illustrate will be analyzed in more detail. For the
moment they should suffice to give readers an idea of the thrust of this inquiry,
which grew out of some 'loose ends' left over from previous research.
In a 1989 paper on temporal distance as a metaphorical template for the
expression of other kinds of distance that languages see fit to express, I
pointed out that past-tense verb forms were often used to carry out a number
of non-temporal functions, including those illustrated in (1) and (3) above.
The time reference of these examples, it will be noted, is not past; thus the
past-tense verbs must be doing some alternative work in the discourse.4 The
focus of my 1989 paper, however, was on pastness, with merely an acknowl­
edgment that the verbs in question were also predominantly imperfective.
In my recent work on narrative, especially Fleischman (1990) and (1991),
I looked at the tense-aspect categories used by fiction writers to report the
speech and thought of their characters, notably in the quotation styles known
as interior monologue and free indirect discourse, which privilege, respec­
tively, the present tense and imperfective forms of the past. Yet here too, I
merely speculated on why the verb forms chosen as the grammatical vehicles
for these quotation styles are invariably imperfective. In the present inquiry
I take up these dangling threads, weaving them together into an investigation
focused specifically on the connection between imperfective aspect and
irrealis modality,5 and one that seeks to explain why imperfective forms so
frequently gravitate toward irrealis meanings and contexts. The data I cite
here are drawn primarily from Romance languages and from English, with
occasional references to parallel — or nonconforming — phenomena in other
languages.
Imperfective and Irrealis 521

2. Defining the categories

Before proceeding with my analysis, and in light of the fuzziness that comes
to surround grammatical categories once we start comparing them across
languages, let me propose some working definitions of the macro-categories
at issue here, imperfective and irrealis.

2.1. Experts on aspect will presumably agree that imperfective aspect

— which in some languages receives explicit coding only in past contexts 6 —
presents a situation7 as unbounded, i.e., without endpoints. Though the situa­
tion may well have endpoints in the world to which the utterance refers, by
choosing a verb with imperfective aspect, the speaker opts to leave these
endpoints out of consideration.
A metaphor often called upon to contrast perfective and imperfective
aspects is that of a camera lens focusing on a situation as if it were a
visualizable object. Thus Lunn (1985:52) remarks, apropos of this aspectual
opposition in Spanish:
When a speaker's perspective on a time span permits [... a] situation to be
viewed in its entirety and in focus, the speaker will refer to it in the preterit
[= perfective past]. When a speaker's perspective precludes such focus, the
situation will be referred to in the imperfect [= imperfective past], (my
emphasis)

As we shall see, the visual notion of focus provides an important key to

understanding the relationship between imperfective aspect and irrealis mo­
dality.
Of the two endpoints of a situation, initial and terminal, it is the latter that
tends to be of greater linguistic consequence in discourse. Accordingly,
imperfective aspect is also commonly said to present a situation as non-
completed. Thus Le Goffic (1986:57) observes, apropos of French, that
insofar as the Present and the Imparfait can be used to express non-temporal
meanings — unlike the Passé Simple, which cannot —, this is by virtue of
their shared property of aspectual non-completion.
One additional point concerning imperfective aspect: following Comrie
(1976) and Bybee & Pagliuca (1985), I include under the umbrella of 'imper­
fective' its aspectual subcategories continuous, habitual/iterative, and progres­
sive, even though certain of these subcategories, in certain languages, may be
subject to constraints not shared by the generalized imperfective (cf. n.7).
522 Suzanne Fleischman

2.2 As a cross-language category of grammar, irrealis constitutes a decidedly

untidy can of worms, given the extent to which languages differ over what
they classify as realis and what they classify as irrealis. In fact, several papers
in this volume that deal with irrealis (notably those by Chafe and Mithun)
move this theoretical problem right up to center stage.
In his Dictionary of Grammatical Terms in Linguistics, Trask (1993)
characterizes irrealis as "a label often applied in a somewhat ad hoc manner to
some distinctive grammatical form, most often a verbal inflection, occurring
in some particular language and having some kind of connection with unreal­
ity." This definition clearly won't get us very far. But precisely because of its
failure to sharpen our understanding of the modality in question, it under­
scores the extent to which defining irrealis as a cross-language category has
become a vexed question.8 Though some investigators, for this reason, favor
jettisoning irrealis altogether from the arsenal of linguistic metalanguage —
the position Trask ascribes to Palmer —, I find myself more in agreement with
Mithun's position (this volume, p. xx) that "despite the apparent heterogeneity
of the categories to which the distinction is applied, indeed perhaps because of
it, the irrealis/realis terminology can prompt fruitful cross-linguistic compari­
sons with potential for furthering our understanding of certain processes of
grammatic[al]ization." It is in the spirit of this statement — with a crucial
codocil to the effect that irrealis, like any linguistic category one seeks to
compare across languages, should be understood as having a prototype struc­
ture — that I undertake the present investigation into the widespread cross-
language tendency for imperfectively -marked verbs to be pressed into service
to express a range of modal meanings that in different ways signal:

(4) a speaker's lack of belief in or lack of commitment to (a) the reality,

realization, or referentiality of an event or sequence of events
predicated in an utterance; (b) the realization of an agent's wishes,
hopes, or intentions, as expressed in the proposition of an utter­
ance; (c) the authenticity of an utterance or chunk of discourse (i.e.
a sequence of utterances); or (d) what for lack of a better term I will
call the 'canonicity ' or normalcy of a discourse or of a communica­
tive situation.

These various avatars of irrealis will become clearer as they are illustrated in
the paragraphs that follow. But before proceeding to the analysis, a point of
clarification is in order concerning the basis for my classification of a context
or usage as irrealis.
Imperfectie e and Irrealis 523

As used here, irrealis refers solely to meaning, with no requirement that

this meaning be expressed formally by a 'dedicated' irrealis morpheme. This
caveat is necessary a fortiori given the limited degree to which irrealis
receives explicit morphological coding in the languages best represented here.

3. Imperfective aspect in conditional sentences

The imperfective-irrealis connection most frequently discussed in the litera­

ture involves bi-clausal conditionals of the type referred to as unreal, hypo­
thetical, or counterfactual. Without entering into the terminological debates
over the synonymy (or hyperonymy) of these labels, suffice it to say that
conditionals of this type are used to refer to events about which the speaker
expresses some kind of negative belief (Palmer 1986:189). Examples from
standard French and Spanish are given in (5) and (6). The examples in (5)
involve present and future (i.e. nonpast) unreality:
(5) Nonpast conditionals
Fr. Si j'avais le temps, je t'écrirais
[IMP] [COND] 9
Sp. Si tuviera tiempo, te escribiría
[IMP SUBJ] [COND] 10
Tf I had time, I would write to you'
For past unreality, the general rule is simply to add a past tense marker in
some way, retaining the imperfective aspect (Dahl Forthcoming). For the
Romance constructions illustrated in (6), this involves use of a conditional
form of the 'have' auxiliary plus a past participle:
(6) Past conditionals
Fr. Si j'avais eu le temps, je t'aurais écrit
[PLUPERF] [PAST COND]
Sp. Si hubiera tenido tiempo, te habría escrito
[PLUPERF SUBJ] [PAST COND]
Tf I had had time, I would have written to you'
Other languages using imperfective forms in unreal conditionals include
Latin, Greek, German, Old Irish, Cree, various Balkan languages (Palmer
1986:192; James 1982:399; Galton 1987), and doubtless others as well.
 524 Suzanne Fleischman

The standard French and Spanish conditional constructions given in (5)

and (6) use imperfective forms in the apodosis (main clause), and in French in
the protasis (subordinate clause) as well. In Spanish the forms used in the
protasis are subjunctive; albeit referred to as the Imperfect Subjunctive, 10 the
verb form used in the protasis of (5) is imperfective in name only, given that
the perfective/imperfective opposition is neutralized in Spanish in the past
subjunctive. In colloquial and/or regional varieties of Romance it is not
uncommon to find 'modal harmony' in unreal conditionals, involving either
double Imperfects — indicative (Portuguese, Italian) or subjunctive (Spanish)
— or double Conditionals (French, Spanish, Italian) (see Harris 1978:234-
246). But again, the crucial fact for our purpose is that these forms are
virtually all — nominally and/or historically — imperfective.
An alternative construction for past unreality, likewise documented in
several varieties of Romance, substitutes the imperfect indicative for the past
conditional, as exemplified in (7) and (8) from French and Italian:
(7) Fr. Un simple coup de téléphone, je venais [IMP] tout de suite
'All you had to do was call, I would have come right away'
It. Perché non me Vhai detto? Veniva [IMP] di sicuro
'Why didn't you tell me? Of course I would have come'
(8) Fr. Une minute de plus (et) le train déraillait [IMP]
It. Ancora un attimo, e il treno deragliava [IMP]
'One more second (and) the train would have derailled'
(examples from Le Goffic 1986, Bertinetto 1986)
The attraction of imperfective aspect for unreal conditionals is discussed
further in Section 10.

4. Motion-derived futures

A second context illustrating the migration of imperfectives into irrealis

territory involves grammaticalization of the motion verbs 'come' and 'go' as
auxiliaries in future and future-of-the-past constructions, as illustrated in the
(semantically equivalent) sentences in (9), again from English and Romance:
(9) Eng. I'm going [PRES PROG]/was going [PAST PROG] to do it myself
Sp. Voy [PRES]/iba [IMP] a hacerlo yo (mismo)
Imperfective and Irrealis 525

Fr. Je vais [PREs]/j'allais [IMP] le faire moi-même

Ptg. Vou [PRES] ia [IMP] fazé-lo eu mesmo
How is it, one might ask, that in so many languages deictic verbs of
spatial motion come to be used to locate situations in a timespan that is
unrealized at reference time? Bybee & Pagliuca (1987:116) observe that
'come' and 'go' are likely candidates for future meaning (including future-of-
the-past) when they occur in the imperfective aspect. For the languages
documented above — and there are many others one could cite with motion-
derived futures11 — the auxiliaries are either (a) in the simple present tense,
which in most languages is imperfective or neutral with respect to the perfec-
tive/imperfective distinction, but in any case not perfective (Comrie 1976:66);
(b) in the imperfective past, or, (c) in the progressive, an imperfective
subtype. Note that if perfective verbs are substituted for the imperfectives in
the examples in (9), only a 'motion' reading is possible, no longer 'future-of-
the-past' :

(10) Eng. I went to do it myself

Sp. Fui a hacerlo yo (mismo)
Fr. Je suis allé le faire moi-même
Ptg. Fui fazé-lo eu mesmo
Bybee & Pagliuca (1987:116-17) contend that the movement (in time or
space) conveyed by the construction in (9) should be understood as "already
in progress" at the temporal or spatial reference point; specifically, they trace
the futurate meaning back to an original meaning according to which "the
subject is on a path toward a goal (which may be an event, state, or activity)."
It is this 'in progress' meaning, then, that correlates with, and calls for, the
imperfective aspect.

5. The 'pre-ludic' imperfect

In Fleischman (1989) I discuss in some detail the imperfective usage illus­

trated in (1), repeated for convenience as (11a), which Warnant (1966) has
labeled the 'pre-ludic' imperfect, inasmuch as it occurs characteristically —
and exclusively — in the negotiation of roles and settings that serves as a
preface to children's make-believe games. Additional examples from Spanish
and French are given in (llb)-(llc):
526 Suzanne Fleischman

(11) a. J: ... Pretend I was moving this up and down and up and down
[moves appendage on a doll] [...]
Z: And pretend somebody was iceskating on the rink
a. Sp. Yo era [IMP] el jefe de la banda y tu eras [IMP] el hermano
'(Pretend) I was the leader of the gang and you were my brother'
c. Fr. Moi, j'étais [IMP] le gendarme, et tu me volais [IMP] mon vélo
'(Pretend) I was the policeman and you were stealing my bike'
The key word here is 'make-believe': for in effect, the imperfective verbs
serve to locate the situations they refer to in the realm of fantasy, of fictional
activity. 12 Interesting to note in this connection is a link various investigators
(Le Goffic 1986, Cappello 1986, Adam 1992) have posited between the
French Imparfait and fictionality, particularly as this form is used in the
protasis of unreal conditionals.
In a thought-provoking discussion, Adam (1992) proposes that the
Imparfait should be viewed not so much as a tense of the past but as an
"operator of fictionality." He bases this proposal on the modal uses of imper­
fective pasts that are of concern in the present inquiry. 13 Cappello (1986)
refers similarly to the Imparfait as "the fictional analogue of the Present
Indicative," 14 a tense which is likewise pressed into service to express non-
temporal meanings, including meanings that come under the heading of
conditional or imaginary. (The present, we recall, is commonly found in the
protasis of real, or open, conditions and is also the tense most frequently used
across languages for generic or gnomic statements.) Seeking out a common
property that will explain why the Present and the Imparfait in French can
express non-temporal (i.e., modal) meanings, whereas the Passé Simple can­
not, Le Goffic (1986) points to their shared imperfective — or as he calls it,
non-completive — aspect.15 It is this property, he argues, that explains,
moreover, why the imperfective past (the Imparfait) and not its perfective
counterpart (the Passé Simple) is chosen as the vehicle for free indirect
discourse (see Section 9).

6. Politeness forms

Another category of utterances discussed in Fleischman (1989), which I

subsume here under the irrealis rubric, includes politeness forms, in particular
attenuated requests of the type exemplified in (12)-(15):
Imperfective and Irrealis 527

(12) Sp. Quería [IMP] pedirle un pequeño favor

Querría [COND] pedirle un pequeño favor
Quisiera [IMP SUBJ] pedirle un pequeño favor
Fr. Je voulais [IMP] vous demander un petit service
Je voudrais [COND] vous demander un petit service
J'aurais voulu [PAST COND] vous demander un petit service
It. Volevo [IMP] chiederle un piccolo favore
Vorrei [COND] chiederle un piccolo favore
'I wanted to ask you a little favor'
(13) a. I was hoping [PAST PROG] we might get together for dinner.
What do you say?
b. ?? I hoped [SIMPLE PAST] we might get together for dinner.
What do you say?
(14) a. I was thinking about [PAST PROG] going to Italy in the summer.
Are you interested?
b. ?? I thought about [SIMPLE PAST] going to Italy in the summer.
Are you interested?
(15) a. I was wondering [PAST PROG] ifyou could tell me what time it is.
b. ?? I wondered [SIMPLE PAST] if you could tell me what time it is.
The Romance examples in (12) all translate rougly as 'I wanted to ask you a
little favor', with the different forms of the modal 'want' — all imperfective 16
— conveying different degrees of deference. (The fact that the English gloss
uses the Simple Past is due to a language-particular constraint on the use of
statives with progressive aspect.)
Bybee (this volume) emphasizes that utterances of this type typically
involve modal auxiliaries, which as a class "do not imply completion of the
action or event expressed by the infinitive with which they occur." 17 Inas­
much as Bybee includes under the modal umbrella mental-activity verbs (e.g.,
'hope', 'think about' and 'wonder', as in (13)-(15)), her claim strikes me as
well founded, provided these verbs are combined with imperfective aspect.
I.e., inasmuch as they do not predicate the completion of the mental situations
they refer to, by implicature they allow for the possibility that these mental
activities are still going on, thereby leaving the addressee an opportunity to
respond. This explains why the Past Progressive versions of the English
examples in (13)-(15) are more natural than their Simple Past counterparts,
which are all more or less pragmatically odd. But what remains to be ex­
plained is why I classify these polite request forms under the semantic rubric
of irrealis. 18
528 Suzanne Fleischman

Indirect speech-acts of this type operate by camouflaging the illocution-

ary force of assertive and potentially face-threatening speech acts, such as
requests, by clothing them in the form of simple declarative statements about
the speaker's intentions or desires in the past. Statements of desire or intention
in the past are generally used when the intended actions turn out to be
unrealized, 19 and as Hale (1969) has observed, unrealized intentions consti­
tute "a specific sort of irrealis." 20 What specifically marks these past inten­
tions as unrealized, I would argue, is the imperfective aspect of the
mental-activity verb. Whence the oddity of (13b) and (14b), where the closure
effected by the perfective Simple Past (I am no longer hoping that/thinking
about...) clashes with the present time reference of the questions that follow.
(15b) seems less awkward, for reasons not entirely clear to me; my intuition,
however, is that speakers would be less likely to produce (15b) spontaneously
than they would the corresponding utterances with Simple or Progressive
Present verbs, as in (15c) below, which are less assertive and involve less
potential loss of face than the present-tense counterparts of (13) and (14).

(15) c. / wonder/I'm wondering if you could tell me what time it is.

7. The 'hypocoristic' Imperfect

Several interpretations have been proposed for the so-called hypocoristic

Imperfect, illustrated by the French examples given in (3) (repeated here as
(16)) and in (17):
(16) Comme il était [IMP] sage! 'Now aren't you a good boy!'
(17) Il avait [IMP] fort mal à son petit doigt, mon bonhomme ?
'Does my little man have a big 'hurt' on his little finger?'
This usage occurs exclusively in an endearing form of address (whence the
label 'hypocoristic') used by adults — mainly women 21 — in speaking to
young children and pets. Though the temporal reference is to the present of
the speech situation, statements of this type are characterized by use of the
Imparfait together with reference to the addressee in the third person.
Of the various explanations proposed for this phenomenon, the most
compelling in my view is that of Gougenheim (1970). In using the
hypocoristic Imperfect, Gougenheim argues:
Imperfective and Irrealis 529

the speaker refuses to take seriously the words she utters. She says, in effect,
for the benefit of an adult who is or might be listening — or for her own
benefit, out of basic human self-respect: "Pay attention! What I'm saying is
not serious. This is a game; this is fiction."22

Through this usage, then, the adult speaker distances herself from the
words of her own utterance, acknowledging that they are not cast in the
conventional adult forms. She marks the utterance in question as an inauthentic
token of adult discourse. In line with Gougenheim's statement, Wilmet (1968-
69:105) points out that the hypocoristic Imperfect is always a 'fiduciary'
form: its use rests on an understood linguistic contract among speakers, from
which young children — and, obviously, animals — are excluded inasmuch
as they do not qualify as full-fledged interlocutors. In sum, the hypocoristic
imperfect operates as a kind of metalinguistic evidential signaling the non-
authenticity or non-canonicity of the speech situation itself — a usage I
include under the last category of irrealis meaning given in (4).
Commenting on the hypocoristic Imperfect, Martin (1971:102) describes
it as creating a world "centered around the child or pet addressee," "a reality in
which the [adult] speaker is not really immersed." This brings us to the use of
imperfective aspect to signal the irrealis status of represented worlds.

8. Imperfective aspect and irrealis worlds

As I have argued at various points in my work on narrative, notably in

Fleischman (1990), adult speakers, as part of their linguistic competence,
possess a typology of narrative genres, one of the most basic divisions of
which seems to be realis vs. irrealis, i.e., a distinction between reports of
events construed by the speaker and addressee(s) as real vs. reports of events
construed as hypothetical, or as fictional vs. non-fictional. Moreover, there
appear to be grammatical features, including tense-aspect categories, that
correlate predictably with these narrative primes. Narration in the perfective
past, which I take to be the unmarked tense of narration, correlates with stories
that are, or purport to be, realis — including conventional fiction, which
purports to be someone's experience; whereas genres which choose a basic
reporting tense other than the perfective past are often irrealis. Jokes, for
example, which are referentially irrealis — thus instantiating the first cat-
 530 Suzanne Fleischman

egory of the definition of irrealis given in (4) —, are routinely told in the
present tense. Any speaker of English immediately recognizes a story that
starts out "This guy walks into a bar and says to the bartender..." as a joke,
whereas the same information reported in the past tense might be construed,
initially at least, as a narration of realis events. Both the imperfect and the
present are selected across languages as the basic reporting tense of genres
that refer to a legendary or mythical past, e.g. the Sanskrit Rgvedas, the Tamil
Puranic stories, and epic universally (see Fleischman 1990:123-24).
The imperfective aspect of the present and the imperfect23enables these
tenses to represent past experience as if it were in the process of occurring,
unmediated by the post-hoc reflective consciousness of a reporting speaker. 24
Thus in many languages (e.g., Spanish, French, Italian, Rumanian, Dutch)
these two tenses are used to report the contents of dreams, hallucinations, and
other semi-conscious states, as seen in (2) above, repeated for convenience as
(18):
(18) Now Winder was wanting to know what the hell the stage was
doing on the pass at night anyway. (Walter Van Tilburg Clark,
The Ox-Bow Incident, p.126)
This statement occurs at a point in the story just after the (first-person)
narrator has been shot by a stagecoach guard. As Chafe (1994:242) observes,
Winder's statement would normally have been perceived as a punctual event,
but it entered the narrator's impaired consciousness "as if he were tuning in on
part of a speech event that was diffuse and out of focus." What is captured by
the imperfective aspect of the main-clause predicate is the semi-conscious
state of the narrator, specifically his inability of focus in on particular events.
In the best known of medieval French epics, The Song of Roland, the very
few Imperfects that occur at all are concentrated in the dream visions. 25 The
passages given in (19) and (20), though taken from literary fiction, are
representative of the reporting style used to narrate the contents of dreams and
fantasies. The passage in (19), from a Spanish novel of the 19th century,
reports a dream of one of the characters:

(19) Oía [IMP] el reloj de la catedral dando las nueve; veía [IMP] con
júbilo a la criada anciana durmiendo con beatífico sueño, salía
[IMP] del cuarto muy despacito para no hacer ruido; bajaba [IMP]
la escalera tan suavemente que no movía un pié [IMP] hasta no
estar segura de poder evitar el más ligero ruido. Salía [IMP] a la
Imperfective and Irrealis 531

huerta, deteniéndose un momento para mirar al cielo que estaba

tachonado [IMP] de estrellas. (Benito Pérez Galdos, Doña Perfecta,
p. 478)
'She heard [IMP] the cathedral clock striking nine; jubilant, she
saw [IMP] the old housemaid deep in a beatific sleep. She walked
out [IMP] of the room ever so slowly so as not to make a noise; she
went down [IMP] the stairs so quietly that she didn't lift her foot
[IMP] until she was sure not to make the slightest sound; she
walked out [IMP] to the garden, stopping for a moment to observe
the sky, which was studded [IMP] with stars.'
In the passage in (20), from a fin-de-siècle Italian novel, the protagonist
fantasizes about his father's dying (at the time of the fantasy the father is still
alive):
(20) Subitamente, gli si formo [PRET] nello spirito l'imagine del padre
agonizzante: strammazzava [IMP] come fulminato, a terra, di
schianto; sussultava [IMP], non anche morto, lívido, muto, contraf-
fatto, con l'occhio pieno dell' orrore di morire; rimaneva [IMP]
immobile, come sotto il secondo colpo d'un maglio invisibile,
carne inerte [...] La madre gli disse [PRET] [...] (Gabriele
D' Annunzio, Trionfo delia Morte; as cited in Bertinetto 1986:370)

'Suddenly, an image formed itself [PRET] in his mind of his father

in the throes of dying: collapsing [26] [IMP] suddenly to the ground,
as if struck by lightening; convulsing [IMP], as yet not dead but
deathly pale, speechless, deformed, eyes filled with the horror of
dying; he remained [IMP] immobile, as if struck by the second
blow of an invisible hammer, his flesh inert, [...] His [the protago­
nist's] mother said [PRET] to him [...]'
It will be observed from these passages that the contents of dreams and
fantasies tend to be packaged not in the form of punctual, bounded events,
which are normally encoded by perfective pasts, but as situations in the
process of being visualized, their endpoints out of focus; whence the choice of
imperfective pasts. Accordingly, what we often find in reports of this type is a
neutralization of the conventional discourse contrast between perfective and
imperfective aspect for reporting eventive vs. non-eventive material. If the
passage in (19) were a conventional narration, all predicates except estaba
 532 Suzanne Fleischman

tachonado ('was studded') in the last clause would be reported as narrative

events in the Preterit (= perfective past). As Lunn (1985:55) observes, "disori­
ented characters make strange aspectual choices: they find it impossible to
place themselves at a vantage point from which events could most conven­
tionally (i.e. coherently) be described." 27

9. Speech and thought quotation

The hierarchy given in (21) below, adapted from Haiman (1989), lists several
strategies for reporting speech, ordered according to their ostensible degree of
fidelity to the original speaker's utterance, 28 a parameter which lends itself to
interpretation along a realis/irrealis cline.
(21) Speech and thought reports: hierarchy of attenuation
a. DIRECT QUOTATION: She said/thought: "What the hell are you
doing here in my hotel room ?"
b. SEMI-INDIRECT QUOTATION (including free indirect discourse):
What the hell was he doing here in her hotel room ?
c. INDIRECT QUOTATION: She wondered what he was doing in her
hotel room.
d. EVIDENTIALS (including quotatives): He is said to have been/
was reportedly found in her hotel room (adapted from Haiman
1989:149; examples mine)
These are not the only categories of speech and thought representation, but
they will suffice for our purpose. 29

9.1. The 'framework' 30 of the original speaker is best represented (i.e., the
representation is maximally realis) in the case of direct speech, where, accord­
ing to the quotation convention (see n.32), the original speaker's words, at
least, are reproduced verbatim. As we move down the hierarchy, the quoted
speaker's framework becomes progressively more attenuated, i.e. more re­
moved from the reality of the original utterance, until finally, in the case of
quotative evidentials, the original speaker's very identity becomes immaterial
(Haiman 1989:149). The quotative evidential examples given in (22), from
newspaper journalism in French, use forms of the Conditional, which is
formally and historically imperfective:
Imperfective and Irrealis 533

(22) Paul VI envisagerait [COND] de faire le tour du monde (Figaro,

Jan. 11, 1966) '[Pope] Paul VI is reportedly contemplating making
a world tour'
L'association de l'Angleterre au Marché Commun serait proposée
[PASSIVE COND] par l'Allemagne {France Soir, Nov. 30, 1967)
'Germany is said to have proposed the entry of England into the
Common Market'
In journalistic discourse the source and accuracy of reported information are
crucial, given the legal consequences of libel laws. Whence this quotative has
acquired in French grammar the name 'conditionnel de presse' from the
variety of discourse in which it most frequently occurs.

9.2. If direct quotation purports to reproduce an utterance verbatim,31 indirect

quotation — in languages that allow this style of speech report (see n.36) —
provides only a semantic equivalent of the original utterance, i.e., it repro­
duces the propositional content of the original statement, integrated into the
framing utterance of the quoter. Several investigators (Haiman 1989,
Maingueneau 1990) have referred to indirect quotation as a type of 'transla­
tion' — from the original speaker's framework into that of the quoting
speaker. This can be seen by comparing the examples of direct and indirect
quotation given in (21a) and (21c) respectively. Chafe (1994: 222-23) sees
indirect speech as constituting an acknowledgment of "the inability of a
remembering consciousness to replicate distal language verbatim." 32

9.3. The quotation style illustrated in (21b) has been variously labeled 'style
indirect libre', 'erlebte Rede', 'represented speech and thought', 'narrated
monologue', 'semi-indirect quotation', or 'free indirect discourse'. Addi­
tional examples are given in (23) and (24), from French and Italian respec­
tively (sentences of free indirect discourse are bolded):
(23) Jamais Frédéric n'avait été [PLUPERF] plus loin du manage.
D'ailleurs, Mlle Roques lui semblait [IMP] une petite personne
assez ridicule. Quelle différence avec une femme comme Mme
Dambreuse! Un bien autre avenir lui était réservé! [IMP] Il en
avait [IMP] la certitude aujourd'hui. (Flaubert, Sentimental Educa­
tion, Part 3, Chapter 2, p. 350)
 534 Suzanne Fleischman

'Frédéric's thoughts had never been [PLUPERF] further from mar­

riage. Besides, Mademoiselle Roques struck [IMP] him as a
somewhat ridiculous little thing. What a difference with a
woman like Madame Dambreuse! A very different future was
awaiting [IMP] him! He was [IMP] certain of that now.'
(24) Domenico gli disse [PRET] di non poter accettare su due piedi
Vinvito a trasferirsi. Qui aveva [IMP] una rete di conoscenze,
frequentava [IMP] ambienti che lo interessavano [IMP], otteneva
[IMP] riconoscimenti; laggiù, chissà? Ma potevano [IMP] ripar-
larne Vindomani, quando veniva [IMP] a cena da loro, (cited by
Bertinetto 1986:392)
'Domenico told [PRET] him not to accept right away the offer to
move. Here he had [IMP] a network of acquaintences, he fre­
quented [IMP] milieus that interested [IMP] him, he received
[IMP] recognition; there, who knows? But they could [IMP] talk
about it again tomorrow when he came [IMP] to their place for
dinner.'

In a nutshell, what typically demarcates utterances of free indirect discourse is

the presence of features of direct quotation — direct questions, exclamations,
fragments, repetitions, deictics, emotive and conative words, colloquialisms
— reported in the style of indirect quotation, i.e., using the deictic system of
the narrative frame (notably with regard to tense and person) but normally
without the inquit formulas characteristic of indirect quotation (e.g. s/he said/
thought that..., wondered why..., etc.). 33 I hasten to add that this description
oversimplifies considerably a complex topic that has produced volumes of
scholarly debate, for the most part among literary narratologists, since it is
widely believed — and for epistemologically valid reasons — that this
particular style of quotation is confined to fiction.34
As noted above, the utterances bolded in (23) and (24) are utterances of
free indirect discourse, which is one linguistic strategy (though not the only
one) for representing the contents of another mind, the subjectivity of the
other. In the passage in (23) the bolded utterances correspond to the thoughts
of Flaubert's protagonist Frédéric Moreau, in (24) to the advice of the re­
ported speaker Domenico, 'translated' in both cases into the language of an
unpersonified narrator. The narrator's language imposes on these words and
thoughts, produced originally by a first-person subjectivity, the third-person
pronouns and imperfective past tenses (imperfect, pluperfect, and condi­
tional) of so-called 'speakerless' narration.
Imperfective and Irrealis 535

As Banfield (1982:108) notes, free indirect discourse is neither an inter­

pretation of the character's speech or thought, which implies an evaluating
speaker; nor a direct imitation of the quoted individual's voice; rather, the
words or thoughts of the represented self retain all their expressivity without
suggesting that their grammatical form was that originally uttered, aloud or
silently.

9.4. The point of this somewhat protracted discussion of quotation styles is

this: if we take direct quotation to be the the most realis representation of
speech and thought — the prototype quotation strategy, as it were, 35 with
respect to parameter (c) of the characterization of irrealis given in (4) —, then
the remaining styles represent varying degrees of irrealis. Thus it is not
coincidental, I submit, that in languages with a perfective/imperfective oppo­
sition in the past, irrealis quotation styles consistently privilege imperfective
past tenses.

9.5. Several investigators have claimed that in languages with established

written traditions of narrative fiction, free indirect discourse or analogous
forms of semi-indirect quotation are only possible in literary fiction,36
whereas in other languages this kind of mixed-perspective discourse that
"wears its hybrid status on its sleeve" (Haiman 1989:149) can be found in
natural narration. Let me suggest, however, that semi-indirect discourse can
occur in natural narration even in languages like English or the standard
Romance languages that have long established belle-lettristic traditions. And
one of the features that marks this mixed quotation style, I submit, is the
presence of inquit formulas in imperfective tenses.
Consider in this regard the natural-narrative examples given in (25)-(26),
which introduce quotation with Progressive inquit formulas (all verba dicendi
are bolded):
(25) [...] an' I was about two blocks from my house when a cop got be­
hind me and started following me
and I was going <Q Oh God, I know he's gonna, he's gonna pull
me over for the [license] plate. Q>37
(26) [...] an' I was saying <Q Well, you can't just try, You hafta, you
hafta — we can't hang up without you guaranteeing me that you or
somebody's gonna show up here within two hours Q>
and... so she said <Q OK Q>
an' we hung up [...]
536 Suzanne Fleischman

Direct quotations normally represent semelfactive speech events, and are

accordingly introduced by perfective verbs, e.g. said in the second-to-last line
of (26). But elsewhere in (25) and (26) what is 'marketed' as direct quotation
is not something the quoted speaker actually said, but something he was
thinking or something he, as narrator, imagines he was thinking at the time,
along the lines of the free indirect discourse of narrative fiction. (The fact that
this is a first-person narration, i.e. one in which the narrator is also a character
in the story and says T , makes the case slightly less compelling, since first-
person narrators are in a position to have had access to the consciousness of
their character-selves in the past (cf. n.35), which is not the case in non-
personal — what is traditionally referred to as 'third-person' — narration. 38 )
In any event, what matters here is that these forays into the consciousness of a
quoted speaker or thinker appear to be signalled by the imperfective aspect of
the introduction-to-discourse verbs.
I puzzled for some time over these imperfective verba dicendi — 'S/he
was going...', T was telling John...', T was thinking...' — and their analogues
in Romance (see Bertinetto 1986:399-401 for examples from Italian). From
the contexts it was clear that the speech-events they referred to were not
iterative/habitual, though they gave the impression of being so because of the
imperfective aspect of the verbum dicendi?9 Ultimately, I concluded that in a
manner similar to the colloquial English quotative like, so cherished by every
teenager in America (see Romaine & Lange 1991), imperfective verba dicendi
in narrative signal that what follows is not a direct-speech quote, though it
may have all the formal trappings of the direct-quotation style. Rather, it is a
verbalization of something the quoted speaker was turning over in his or her
mind. On this reading, it would be the 'ruminative' quality of the quoted
material in irrealis quotations of this type that motivates the imperfective —
and specifically iterative/habitual — aspect of the speech and thought tags.
Bybee (p.c.) has suggested an alternative interpretation of these imper­
fective inquit formulas which privileges not the iterative/habitual meaning
associated with imperfectivity, as proposed here, but rather the ongoing/
continuous meaning (on these two meanings as the major subcategories of
imperfective aspect, see Comrie 1976:25). According to the latter view, an
imperfective verbum dicendi gives the sense that a quoted utterance is 'in
progress' and perhaps also backgrounded, given that what it reports is an
ongoing mental attitude. An imperfective inquit formula will be chosen not
when the speaker is reporting a sequence of events (where perfective verbs are
Imperfective and Irrealis 537

called for), but when what is being reported on is rather how the speaker feels
about something. Imperfective verba dicendi formally release the quoting
speaker from a commitment to the exact replication of what the original
speaker said — this is in line with category (c) of the characterization of
irrealis given in (4) — and provides instead "a characterization that reflects
the quoted speaker's feelings" (Bybee, p.c.).
Both of the above interpretations of imperfective inquit formulas (ha­
bitual and continous) are plausible. And both motivate the use of verbs with
imperfective aspect to mark this style of quotation as irrealis, in the sense of
being an inexact representation of a quoted speaker's utterance(s).

10. Habitual and irrealis

Several investigators have looked cross-linguistically at the classification of

habitual as realis or irrealis (Givón 1984; Willett 1989; Dahl Forthcoming).
Givón regards habitual as mixed. 40 On the one hand, it represents "a clear
strong assertion of facts" (p. 285), which would account for the realis classifi­
cation of habituais in Sherpa (Woodbury 1986:190-93), Bikol (Givón
1984:309), and no doubt elsewhere as well. But on the other hand, habitual
does not refer to unique events — whence Banfield's (1985) reference to the
French Imparfait as a 'mass' tense (in opposition to the perfective Passé
Simple, which is a 'count' tense) 41 —, but is rather the aspect of generic, non-
referring expressions, which fall under our irrealis heading (4a). This would
explain why Bargain (a Papuan language), for example, treats past habitual as
Irrealis (Roberts 1990). In Givón's examples, reproduced in (27):
(27) a. Joe always cuts logs
(27) b. Every day Joe cuts a log, then...
'log(s)' does not refer to any individual log or unique group of logs. 42 Rather,
these are generic, non-referential statements. It is in this sense, Givón argues
(p.285), that habitual resembles irrealis.
The link between irrealis and habitual (or iterative) is underscored by the
fact that in many languages habitual — or the generalized imperfective, if the
language lacks a distinct habitual — and conditional are formally identical.
The best context for comparison is the apodosis (main clause) of unreal
conditionals of the type illustrated in (5).
538 Suzanne Fleischman

Commenting on Bulgarian forms that can mark either the modal Condi­
tional or the aspectual Iterative, Aronson (1977:15) sees the connection as so
close and so widespread as to constitute a potential universal. In support of
this position he offers the semantically equivalent pairs of sentences given in
(28), from English, Serbocroatian, and Hebrew respectively:
(28) a. Eng. He would play golf every day when [ITER] / if
[COND] he lived in Chicago
b. Serbobcr. Svakog bi dana igrao golf dokje [ITER] / kadbi
[COND] zivevo u Cikagu
c. Heb. Hu haya msoxek golf yom-yom kaašer [ITER] / ilu
[COND] gar bSikago

For each pair of examples, the main clause is identical; only the particle in the
subordinate clause tells us whether the sentence is iterative/habitual or condi­
tional (cf. also Galton 1987).
Lazard (1975) cites parallel evidence from a spectrum of ancient and
modern Indo-European languages in which a single morpheme expresses both
past habitual and irrealis/conditional (and in certain cases also future-of-the
past, analogous to Eng. would)43. He points out (pp.225f.) that where a
generalized imperfect is used to express certain irrealis meanings, it is the
habitual meaning, among the various possible meanings a durative past can be
used to express, that specifically correlates with irrealis. He sees this affinity
as particularly striking in that it is realized by a diverse range of morphologi­
cal structures (the data he cites have nothing in common from the standpoint
of form, their only commonality being their shared functions). Even in geneti­
cally related languages, the respective developments are apparently independ­
ent of one another — what Sapir referred to as 'drift'. Seeking to explain this
affinity, Chung & Timberlake (1985:221) observe that if habitual subevents
are indefinite both in number and in time (cf. n.42 above), they can also be
viewed as extending over possible worlds.
In the grammatical environment under consideration here, the apodosis
of unreal conditionals, a number of investigators (Lazard 1975, James 1982,
Restan 1989, Dahl Forthcoming) have noted that in languages that distinguish
perfective and imperfective aspect, what we invariably find is the imperfec-
tive past or one of its subtypes (progressive or, in particular, habitual), and not
the perfective form that the context would predict (given that the main-clause
Imperfective and Irrealis 539

verb most commonly denotes a single completed act). On the basis of broad
cross-language investigation, Dahl speculates about a general tendency to use
imperfective aspect in counterfactual constructions.

11. Imperfective and irrealis: why the attraction?

Seeking to account for why many languages use past tenses — specifically
imperfective pasts — to express the hypothetical, James (1982) points to the
shared semantic feature of non-completion. That is, imperfective and hypo­
thetical both present predicated situations as unrealized or not fully realized at
the time of reference. 44 An alternative explanation, not incompatible with the
non-completion hypothesis, involves the discourse notion of backgrounding.
The correlation between imperfective aspect and backgrounding, first
proposed by Weinrich (1964/1973) as an alternative to traditional interpreta­
tions of the Imparfait/Passé Simple opposition in French, became popularized
among linguists through several papers by Hopper from the late '70s and early
'80s (notably Hopper 1979, 1981). By now it has become a commonplace in
the discourse literature. Background information, Hopper (1981:238) ob­
serves, involves a reduced assertion of the finite reality of an event. It is not
involved in "the asserting of events in the story line, but makes statements that
are CONTINGENT [his emphasis] and dependent on story line events. Typically,
therefore, one finds in backgrounding those forms associated with a lower
degree of assertiveness and even forms designated as irrealis" (1981:215, my
emphasis). So if imperfective aspect is indeed associated in discourse with a
reduced degree of assertiveness, then, as James observes, this would provide a
natural explanation for why many languages choose it over perfective aspect
to encode the hypothetical, since hypothetical statements likewise do not
assert the truth of their propositions. Or, as Bybee, Perkins, and Pagliuca
(1994:239) argue, mood is not about truth values, but about the speaker's
choice between assertion and contrasting functions.
The foregrounding and backgrounding of information in discourse like­
wise correlates with the visual notion of focus, which, as I suggested in the
introduction, provides a well motivated link between imperfective and
irrealis. Simply put, information a speaker chooses to foreground is presented
as in focus, background information as defocused, or out of focus. 45 In a now
540 Suzanne Fleischman

classic paper on transitivity, Hopper & Thompson (1980) make the case for a
language-universal correlation between aspect and focus. 46 Their findings
have subsequently been confirmed by a number of studies on narrative show­
ing the focus and non-focus roles of perfective and imperfective aspects
respectively.
In an illuminating paper contrasting the Preterit and Imperfect in Span­
ish, Lunn (1985) distinguishes these two categories of the Spanish past system
in terms of the conferring vs. witholding of focus. There are two points of
view, she notes, from which things are not in focus. On the one hand, if an
observer is at a point from which foreground is in focus, then that observer is
too far away for background to be in focus. This would explain the conven­
tional connection between imperfective aspect and background. On the other
hand, imperfective forms are also used to describe situations that are too close
— physically and/or psychologically — to be sharply in focus. Thus Lunn
explains novelist Juan Rulfo's use of the Imperfect to render the obsessive
fantasy life of his protagonist Pedro Páramo, offering examples analogous to
those cited in (18)-(20) above. The same out-of-focus perspective is strikingly
conveyed in the passage in (29) below from Robbe-Grillet's novel Jealousy,
which relies on a series of Present tenses (bolded in the example) to report the
disjunctive perceptions of a jealous husband as he peers through the Venetian
blinds at his wife and her surroundings:

(29) C'est à une distance de moins d'un mètre seulement qu'apparais­

sent dans les intervalles successifs, en bandes parallèles que sépa­
rent les bandes plus larges de bois gris, les éléments d'un paysage
discontinu: les balustres en bois tourné, le fauteuil vide, la table
basse où un verre plein repose à côté d'un plateau portant les deux
bouteilles, enfin le haut de la chevelure noire, qui pivote à cet
instant vers la droite, où entre en scène au-dessus de la table un
avant-bras nu, de couleur brun foncé, terminé par une main plus
pâle tenant le seau à glace. La voix de A... remercie le boy. La main
brune disparaît. Le seau de métal étincelant, qui se couvre bientôt
de buée, reste posé sur le plateau à côté des deux bouteilles.
Le chignon de A... vu de si près, par derrière, semble d'une grande
complication. Il est très difficile d'y suivre dans leurs emmêle­
ments les différentes mèches: plusieurs solutions conviennent, par
endroit, et ailleurs aucune (pp. 51-2).
Imperfective and Irrealis 541

'It is only at a distance of less than a yard that the elements of a

discontinuous landscape appear in the successive intervals, paral­
lel chinks separated by the wider slats of grey wood: the turned
wood balusters, the empty chair, the low table where a full glass is
standing beside the tray holding the two bottles, and then the top
part of the head of black hair, which at this moment turns toward
the right, where above the table turns a bare forearm, dark brown
in color, and its paler hand holding the ice bucket. A...'s voice
thanks the boy. The brown hand disappears. The shiny metal
bucket, [which is] immediately frosted over, remains where it has
been set on the tray beside the two bottles.
The knot of A...'s hair, seen at such close range from behind,
seems to be extremely complicated. It is difficult to follow the
convolutions of different strands: several solutions seem possible
at some places, and in others none.'
Scrutinized at such close range, the figure of the wife can only appear
fragmented: a hand, a forearm, a knot of hair, its individual strands. A voice,
not a character, thanks the serving boy. Verbalizing his perceptions as he
watches, in the imperfective Present tense, the observer is unable to gain
perspective on focalized objects, his vision severely distorted by the lack of
distance betwen himself and what he observes and by the slats of the blinds.
As Lunn points out (p.59), the lack of definition which such characters'
aspectual choices impute to the situations they describe is precisely what an
observer sees who is 'too close' to a scene.
A point of view internal to a situation also precludes focus on the part of
an observer, and therefore results in the choice of imperfective forms when
the observer undertakes to describe that situation (Lunn, p.57). This occurs,
for example, when a narrator's consciousness rather than a situation external
to the narrator becomes the focus of a story — which is often the case in
modern fiction. The situations in which that narrator takes part are encoded in
a non-focus way. This, Lunn claims, accounts for the sense of reverie evoked
by the use of the Imperfect in Spanish to describe situations that could have
been described objectively using the Preterit. It also accounts, I have argued
(Fleischman 1990, 1991), for the use of the present tense in interior-mono­
logue novels and related forms of fiction which undertake to verbalize the
contents of a character's mind unmediated by a narrator's language.
542 Suzanne Fleischman

The concept of focus, then, provides us with a fruitful metaphor linking

aspect and perception. Generalizing this observation, Lunn argues that "the
existence and comprehensibility of metaphors linking aspect and perception
are meant to be taken as evidence that the aspect-perception link is real. That
is, metaphors are held to be verbalizations of legitimate comparisons" (52f.).
They capture truths about conceptualization, thus providing a solid cognitive
ground on which to chart observed pathways of grammatical change.

12. Conclusion

What I have undertaken to present here is a spectrum of irrealis contexts,

conforming to the typology set forth in (4), in which we find verb forms
marked for imperfective aspect. These include hypothetical conditionals,
motion-derived futures, politeness forms, the prefatory negotiation of chil­
dren's make-believe games, hypocoristic baby-talk, the narration of dreams
and other semi-conscious states, and various indirect forms of speech and
thought quotation. There are no doubt additional contexts which have not
come to my attention that likewise use imperfective forms to express irrealis
meanings 48 — readers are encouraged to take note of such contexts —, just as
there are languages in which we don't find imperfective aspect markers in
certain of the irrealis contexts discussed here, or where we instead find
perfectives.
One piece of apparent counterevidence involving aspect and evidentials
is Nichols' proposal, based on evidence from Chinese Pidgin Russian, Turk­
ish, Tibetan, and Sherpa, of a universal co-variance between perfective aspect
and inference vs. imperfective aspect and "immediateness" or directness of
evidence (Nichols 1986:254f.). Potential counterevidence is also available
from Slavic, depending on how one interprets it.
In Bulgarian, for example, it is Perfective not Imperfective verbs that are
acquiring a meaning more modal (i.e. conditional) than aspectual; however,
the Perfective verbs in question are those built on the present stem, i.e. the
Nonpast and, notably, the Imperfect. Moreover, the meanings of these cat­
egories, which Bulgarian classifies formally as Perfective, seem to correspond
to meanings which in other languages come under the heading of imperfective
(Aronson 1977). Macedonian presents a similar situation (Hacking 1993), 49
and references abound in the Slavic literature on the ability of the Russian
Imperfective and Irrealis 543

Perfective Nonpast to carry modal (specifically future) meaning, a characteri­

zation which Hacking extends to conditional contexts, "to make explicit the
connection between the perfective nonpast and hypothetical meaning" (194).
Though here and there we encounter developments that fail to support or
appear to contradict the macro-correlation proposed here between imperfec­
tive aspect and irrealis modality, such counterevidence seems neither suffi­
cient nor of such nature as to undermine the overall well-foundedness of the
correlation. The weight of evidence in support of the correlation, from a broad
spectrum of related and unrelated languages, seems sufficient to establish this
meaning shift as a widely documented pathway of linguistic development,
leading in some instances to the grammaticalization of new expressions of
evidentiality and/or epistemic modality. 410

NOTES

* Abbreviated sections of this paper were presented at a symposium on Tense-Aspect

(Cortona, Italy, October 10-13, 1993). The present version has benefitted from observa­
tions, data, and references offered by several colleagues present at that symposium,
notably Carl Bache, Jacques Boulle, Oesten Dahl, Ekkehard König, and Co Vet. Ac­
knowledgement is also due to Grace Fielder, and especially Jane Hacking, for guiding
my foray into the arcane regions of Slavic aspectuality and conditionals, and to Wally
Chafe, John Haiman, Sophie Marnette, Linda Waugh, and especially Joan Bybee for
comments and critique of an earlier version.
1 Data from an exchange between two 4-year-old girls recorded by Amy Kyratzis.
2 My thanks to Wally Chafe for this example, reported on in Chafe (1994).
3 This statement assumes a view of grammaticalization that is not limited to the develop­
ment of lexical items into grammatical morphemes, but includes also the "expansion"
(Heine & Reh 1984) of items already operative in the 'service sector' of the language
beyond their original function into other areas of grammar (see also Heine, Claudi &
Hünneneyer 1991, chapter 6).
4 This in accord with a principle of linguistic recycling which Lass (1990) has labeled
'exaptation'.
5 For the record, this connection has not passed unnoticed. It has been alluded to, in
diverse contexts, by Aronson (1977), James (1982), Le Goffic (1986), Cappello (1986),
Bertinetto (1986), Bazzanella (1987), Adam (1992), Bybee (this volume), and no doubt
elsewhere. However, none of the studies mentioned specifically targets the aspect-
modality connection, as I propose to do here, bringing together a diverse range of
phenomena under a single grammaticalization umbrella.
In a paper frequently cited as a paradigmatic case study of the interrelationship of
grammatical categories, Slobin & Aksu (1986) offer compelling evidence that although
544 Suzanne Fleischman

tense, aspect, and modality are clearly distinct categories at the theoretical level, in
practice (i.e., in actual usage) they cannot always be teased apart and studied in isolation
from one another.
6 In languages that contrast perfective and imperfective aspect, this opposition may not be
overtly expressed across the tense spectrum (in languages that have tense). Where this is
the case, the opposition is explicitly (i.e., morphologically) expressed most often in the
past. Thus Dahl (1985:92) observes that whereas progressives tend to appear in all tenses
(in a language that has tense), general imperfectives tend to be limited to past tenses.
7 'Situation' will be used here as a cover term for the predicate classes traditionally
referred to as events, processes, and states, or, according to the now familiar typology
originally proposed by Vendler (1967), as states, activities, achievements, and accom­
plishments.
8 Trask adds (p. 147) that Palmer, in Mood and Modality (1986), recommends avoiding
the term irrealis altogether in linguistic theory on grounds that it corresponds to "no
consistent linguistic content." Like Chafe (this volume), I have searched Mood and
Modality to corroborate this statement but was unable to locate it.
9 The French and Spanish Conditional forms (which double as Futures-of-the-Past)
evolved from a Latin periphrasis combining Past Imperfective forms of 'have' with an
infinitive: scribere habeba(m) (lit. to-write I-had [IMP]) > Fr. écrirais/Sp. escribiría.
Standard Italian scriverei departs from this common Romance pattern in having chosen
the Perfective form of the 'have' auxiliary, though both northern and especially southern
Italian dialects show forms traceable to the Imperfective periphrasis as well as hybrid
paradigms containing forms historically Imperfective alongside forms historically Per­
fective (see Rohlfs 1968:§§593-599).
10 Historically, this form began life as a Pluperfect Indicative. On this frequently analyzed
development, see Lunn & Cravens (1991), Klein-Andreu (1991).
11 Among many commentaries on these constructions, see Givón (1973), Fleischman
(1982), Bybee & Pagliuca (1987), Lichtenberk (1991), Bourdin (1992) and Forthcoming.
12 An alternative explanation, suggested by Oesten Dahl (p.c.), is that the verb form
children select in this context will be whatever form their language happens to use in the
protasis of unreal conditionals, e.g., the Conditional in Finnish, the Subjunctive in
German, etc. This usage, in other words, would represent a special type of truncated,
mono-clausal conditional. Undermining this explanation, however, is the situation in
Spanish. Standard Spanish uses the Imperfect Subjunctive in the protasis of unreal
conditionals, as in (5), yet Spanish children use the Imperfect Indicative in 'pre-ludic'
contexts.
13 A less radical statement of this position, formulated with respect to Italian, is that of
Bertinetto (1986:390), who argues that in the modal uses of the Imperfetto the meaning
of temporal pastness is, if not neutralized, at least attentuated to the point of conveying
'non-actuality' or simply 'virtuality'. Cf. also Bazzanella (1987).
14 In his words, "comme le conditionnel, qui prend souvent des valeurs de 'futur fictionnel'
(ou hypothéthique), l'imparfait de l'indicatif semble pouvoir assumer dans certains
emplois non temporels une valeur de présent fictionnel" (38f.).
15 Commenting on hypotheticality in Slavic, Restan (1989) similarly suggests 'incomplete­
ness' as the common feature linking this modality to imperfective past forms in many
Imperfective and Irrealis 545

languages. However, Restan does not systematically distinguish between imperfective

and imperfect, the latter of which can, in certain Slavic languages (Bulgarian, Macedo­
nian), be aspectually perfective (Jane Hacking, p.c.).
16 Save for the standard Italian Conditional, which, as explained in n.10, is historically
perfective.
17 Martin (1971:101) claims, similarly, that polite requests in French always involve an
auxiliary construction, e.g. with venir 'come' or vouloir 'want' (Je venais/voulais vous
proposer de venir dîner à la maison 'I wanted to ask you to come for dinner'). As for the
irrealis meaning, he refers to a paper by Valin (1964), who interprets the psychological
mechanism involved as a projection of the predicated situation back into a fictitious past
in which the speaker adopts a subordinate stance vis-à-vis the addressee. By shifting the
main verb (i.e., the auxiliary) back into this fictitious past, the speaker makes the
realization of the subordinate situation (expressed by the infinitive) appear to be contin­
gent upon the good will of the addressee. A similar explanation is proposed by Bertinetto
(1986:373) for this usage in Italian, though not all his examples involve a modal
auxiliary, e.g. Dove andavi [IMP] COSÌ di fretta? 'Where are [lit. were] you going in such
a hurry?' (uttered while the speaker is walking alongside the addressee).
18 Mithun (this volume) reports on several languages that, in a similar fashion, classify
polite imperatives as irrealis, while stronger imperatives are treated as realis (though
aspect is not involved in the contrast). She interprets the latter situation, i.e. classifica­
tion of an unrealized situation as realis, as reflecting a high degree of speaker confidence
— approaching certainty — that the situation will be realized.
19 If they had been realized, the utterance would presumably have been formulated differ­
ently, per Grice's maxim of quantity.
20 Hale (1969:22), as cited in Steele (1975).
21 Whether this usage represents a characteristic phenomenon of so-called women's lan­
guage, or alternatively, is simply an epiphenomenon of the fact that childcare has been
traditionally the responsibility of women, is a question which, albeit interesting, falls
outside the scope of the present inquiry.
22 In Gougenheim's words: "...on ne veut pas prendre au sérieux les paroles qu'on
prononce. On dit en somme, à l'adresse de l'adulte qui entend ou pourrait entendre ces
propos (où vis-à-vis de soi-même, par respect humain): Attention, je ne parle pas
sérieusement, je suis dans le jeu, dans le fictif' (p. 154).
23 Excluded from discussion here are languages in which the imperfect is not ipso eo
imperfective, e.g. Bulgarian and Macedonian, which operate with two aspectual opposi­
tions, Perfective/Imperfective and Aorist/Imperfect. The modal extensions of these and
other Slavic aspectual categories will be taken up briefly in §§10 and 12.
24 Argumentation supporting this claim is presented at various points in Fleischman (1990).
25 Verses 719-20, 726, 2556-60 of the Oxford version.
26 For a more idiomatic translation capturing the vividness of the character's vision, I have
rendered certain of the Italian Imperfects (stramazzava, sussultava) as nonfinite present
participles ('collapsing', 'convulsing').
27 For additional literary examples, see Lunn (1985:55f), Fleischman (1990:248f; 1991).
546 Suzanne Fleischman

28 I say 'ostensible' because fidelity to an original speaker's utterance turns out to be a

complex business involving more than simply a mechanical reproduction of the proposi-
tional content of that utterance and, in the case of direct quotation, also its form (cf.
n.32). The various pragmatic factors that come into play (e.g., intentionality, non-literal
meaning, opaque reference, point of view) are sufficient — and sufficiently complex
(messy?) — to have spawned a vast literature on the topic of speech-and-thought
representation that has, over the years, engaged numerous linguists, literary analysts,
and philosophers of language.
29 For a more finely grained typology, and one of the more illuminating presentations of the
topic, see Leech & Short (1981, chapter 10); also Chafe (1994, chapters 16-17).
30 Haiman's term 'speaker's framework' provides an umbrella for the range of meanings,
at various levels of linguistic functioning, that an utterance communicates.
31 Various investigators have argued and offered evidence for the 'constructed' nature of
directly quoted speech (Tannen 1986,1989:110-119; Mayes 1990:333-334; Chafe 1994,
chapter 16). However, the issue of whether a quote is in fact a verbatim report is often
not what matters most — except to journalists and their editors, at risk of costly libel
suits, and to the individuals they (mis)quote. What matters is rather the illocutionary and
perlocutionary force of direct quotation, i.e. what the speaker purports to — and
effectively does — accomplish by reporting what someone said or wrote in a form that is
conventionally understood to preserve maximum representational authenticity. Mayes
speculates — correctly, I believe — that the assumption that direct quotes are more
authentic than indirect quotes may have originated in literate cultures, where direct
quotation is by convention exact (359). The fact that many direct quotes are inventions
of the speaker suggests strongly that speakers use direct quotation for specific pragmatic
purposes (Mayes, p. 330; see also Chafe, chapter 16).

32 Similarly Tannen (1986). I agree with this statement in theory. Nonetheless, and as
Chafe acknowledges, speakers often use the direct-speech format to report language
they know was not uttered in precisely the form in which they quote it, as evidenced by
the following example, from a natural narrative, where the direct quote is preceded by an
explicitly hedged inquit formula: I was saying things like: "Well, could we stop by my
house so that I can tell my family that I have to go to jail haha an ' I can't make the
funeral?" That speakers act as if they can remember distal speech accurately and
reproduce it verbatim is also suggested by the following utterance, from the same natural
narrative (reference in n.38): I remember him saying: "Son, you're under arrest, you
have to go to jail".
33 For a more detailed analysis of the formal features and constraints of free indirect
discourse, see Banfield (1978/1993). Like many analysts, Banfield identifies utterances
of free indirect discourse on the basis of clause-level formal features. For a compelling
challenge to this position, relevant in particular to languages like English that lack a
distinctive imperfective aspect, see Ehrlich (1991), which establishes important dis­
course-based criteria for identifying this style of quotation.
34 A distinctive feature of fictional narrative that sets its apart from nonfiction is the
cognitive access the narrator has to the mental states of other individuals; only in
constructed narratives (i.e. fiction) which are true in constructed worlds can one indi­
vidual enter — and represent through language — the consciousness of another. When
conversational narrators appear to do this — and we often do — the thought quotation
must technically be viewed as an inference, not an assertion.
Imperfective and Irrealis 547

35 Echoing Kuno (1972), Wierzbicka (1974:271) insists that "direct speech is in some
important and intuitively clear way more basic than indirect speech. To see this it is
sufficient to think of all those languages which do not have indirect discourse at all
(whereas the existence of direct discourse seems to be universal), and also of the
historical priority of direct discourse in those languages in which the gradual develop­
ment of the indirect discourse is historically attested."
36 The most vehement advocate of this position is Banfield (1982, 1985, and elsewhere in
her writings), who insists on the chronological coincidence of free indirect discourse
with the emergence of the modern novel in the 19th century. For discussion, see
Fleischman (1990, 230ff.), (Haiman 1989).
37 From "Jeff goes to jail," a narrative tape-recorded and transcribed by Annie Jaisser. In
the transcribing of an oral narration editorial decisions inevitably have to be made, e.g.,
about punctuation, capitalization, and — most important for our purpose — conventions
for quotation (cf. Chafe 1988). The quotation conventions chosen will, of course, reflect
the transcriber's interpretation of the style(s) of quotation used. In (25) and (26) I have
changed the transcriber's double quote marks to the stylistically neutral convention <Q
... Q>.
38 In Fleischman (1990, Chap. 7) I discuss the influence of the grammatical person of a
narration (first vs. 'third') on what are here referred to as irrealis quotation styles.
39 An alternative reading of the imperfective aspect of these inquit formulas is presented
below.
40 This should occasion no surprise. As shown in the papers in this volume by Mithun and
Chafe, grammatical categories (e.g. future, imperative) or constructions (questions,
negation) are not uniformly classified as realis or irrealis across languages, nor even at
times within the same language (e.g. imperatives in Jamul Diegueño and Alamblak, as
reported by Mithun).
41 This explains why the Imparfait will be chosen when the past events in question are
construed collectively, whereas the Passé Simple (or Passé Composé in speech) will be
chosen when they are construed individually. The collective construal corresponds to
what is here referred to as habitual, the individual construal to what is here referred to as
iterative. Hacking (1993:158) offers a particularly nice formulation of this distinction
(with reference to Slavic, but easily generalizable), observing that imperfective aspect is
consistent with a "global, stative overview of a series of repeated events," i.e. habitual,
while perfective is called for if the focus is on "the individual quality of the subevents,"
i.e. iterative. Though there are languages that formally distinguish iterative from ha­
bitual (e.g. Old Persian, per Lazard 1975), for our purpose it is unnecessary to insist on
the distinction. For the record, in Old Persian it is the Habitual marker -i that is also used
to express irrealis meanings, notably in unreal conditionals of the type illustrated in (5).
42 Givón (1984:285) notes that the most common test for realis/irrealis involves the
referentiality of indefinite arguments under the scope of these modalities.
43 Regarding Eng. would, Bybee, Perkins, & Pagliuca (1994:239) point out that historically
the habitual and irrealis (conditional) uses are "entirely independent developments."
Whereas for Macedonian, Galton (1987:365) opines that "the two probably developed
alongside each other."
 548 Suzanne Fleischman

44 As James observes, "Imperfective aspect is usually held to have as its essential feature
the fact that it indicates that an action or state is not being viewed as a complete whole,
but rather as something going on in time and not yet completed. But likewise, if an action
or state is hypothetical, then it is unrealized and in that sense is not an actual complete
action or state. Thus, imperfective aspect and the hypothetical share a semantic feature in
that they both indicate something which is in some way not fully realized" (1982:p. 399).
45 One difficulty with the backgrounding hypothesis, pointed out to me by Ekkehard
König, is that background information tends to be presupposed, i.e., it is presented as
being not subject to challenge. It would thus seem more likely to correlate with realis
than with irrealis.
46 At issue here is 'focus' in the visual, perceptual sense, not as a category of information
structure contrasting with 'topic'.
47 Chafe (1994 and p.c.) suggests that these various contexts, or the irrealis meanings they
instantiate, have as a common denominator the marking of an event as imagined rather
than directly experienced. This distinction figures as the last of a hierarchy of opposi­
tions he sets up, according to which experiences are divided into those that are immedi­
ate and those that are displaced, and within the latter category, those that are remem­
bered and those that are imagined, though he acknowledges the fuzziness of this last
distinction.
48 Like Bulgarian, Macedonian has a dual aspectual system in the past involving two
oppositions: Perfective/Imperfective and Aorist/Imperfect, though the status of the latter
opposition — tense or aspect — remains controversial (see Fielder 1993, §§2.4.7-8).
According to Aronson (1977), the latter opposition has remained fundamentally aspec­
tual, while the former, superordinate opposition has become more modal, with the
(suffixally derived) Imperfectives giving rise to the new "Simple Conditional" forms (cf.
also Galton 1987, Hacking 1993, Fielder 1993).
49 Though these two categories of the linguistic metalanguage are differentiated at the
theoretical level, in practice (i.e. when applied to real-language data) the distinction
often blurs (see Willett 1988).

REFERENCES

Adam, Jean-Michel. 1992. "Si hypothétique et l'imparfait: une approche linguistique de la

fictionnalité." Etudes Littéraires 25,1/2.147-166.
Aronson, Howard I. 1977. "Interrelationships between Aspect and Mood in Bulgarian."
Folia Slavica 1,1.9-32.
Banfield, Ann. 1978. "Where Epistemology, Style, and Grammar Meet Literary History."
New Literary History 9,3.415-54. Abridged version repr. in Reflexive Language.
Reported Speech and Metapragmatics ed. by John Lucy, 339-364. Cambridge: Cam­
bridge University Press, 1993.
Banfield, Ann. 1982. Unspeakable Sentences. Narration and Representation in the Lan­
guage of Fiction. London: Routledge.
Banfield, Ann. 1985. "Grammar and Memory." Berkeley Linguistics Society 11.387-397.
Imperfective and Irrealis 549

Bazzanella, Carla. 1987. "I modi dell'imperfetto." Italiano e oltre 1.18-22.

Bertinetto, Pier-Marco. 1986. Tempo, aspetto e azione nel verbo italiano. Il sistema
dell'indicativo. Florence: Accademia della Crusca.
Bourdin, Philippe. 1992. "Constance et inconstances de la déicticité: la resémantisation
des marqueurs andatifs et ventifs." La Déixis: Colloque en Sorbonne, 8-9 juin 1990 ed.
by M-A. Morel & L. Danon-Boileau, 287-307. Paris: Presses Universitaires de France.
Bourdin, Philippe. Forthcoming. 'Aller' et 'venir' en linguistique générale: Pour une
typologie des filières de la resémantisation. Toronto: Editions du G.R.E.F.
Bybee, Joan L. This volume. The Semantic development of Past Tense Modals in English.
Bybee, Joan L. & William Pagliuca. 1985. "Cross-Linguistic Comparison and the Devel­
opment of Grammatical Meaning." Historical Semantics: Historical Word Formation
ed. by Jacek Fisiak, 59-83. Berlin: Mouton.
Bybee, Joan L. & William Pagliucal987. "The Evolution of Future Meaning." Papers
from the 7th International Conference on Historical Linguistics ed. by A. Giacalone
Ramat et al., 109-123. Amsterdam: Benjamins.
Bybee, Joan L., Revere D. Perkins, & William Pagliuca. 1994. The Evolution of Gram­
mar: Tense, Aspect, and Modality in the Languages of the World. Chicago: University
of Chicago Press.
Cappello, S. 1986. "L'imparfait de fiction." In: Le Goffic, ed., 31-41.
Chafe, Wallace. 1988. "Punctuation and the Prosody of Written Language." Written
Communication 5,4.396-426.
Chafe, Wallace. 1994. Discourse, Consciousness, and Time. Chicago: University of
Chicago Press.
Chafe, Wallace & Johanna Nichols, eds. 1986. Evidentiality: The Linguistic Coding of
Epistemology. (= Advances in Discourse Processes, 20). Norwood, NJ: Ablex.
Chung, Sandra & Alan Timberlake. 1985. "Tense, Aspect, and Mood." Language Typol­
ogy and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon ed.
by Tim Shopen, 202-258. Cambridge: Cambridge University Press.
Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press.
Comrie, Bernard. 1985. Tense. Cambridge: Cambridge University Press.
Dahl, Oesten. 1985. Tense and Aspect Systems. Oxford: Blackwell.
Dahl, Oesten. Forthcoming. "The relation between Past Time Reference and Counterfac-
tuality: A New Look." On Conditionals Again (Proceedings from the 1994 Workshop
"Conditionals and Co. ", Duisberg, Germany) ed. by A. Athanasiadou & R. Dirven.
Ehrlich, Susan. 1991. Point of View. A Linguistic Analysis of Literary Style. London &
New York: Routledge.
Fielder, Grace. 1993. The Semantics and Pragmatics of Verbal Categories in Bulgarian.
Lewiston, NY: Edwin Mellen Press.
Fleischman, Suzanne. 1982. "The Past and the Future: Are they 'Coming' or 'Going'?"
BLS 8.322-34.
Fleischman, Suzanne. 1989. "Temporal Distance: A Basic Linguistic Metaphor" Studies
in Language 13,1.1-51.
Fleischman, Suzanne. 1990. Tense and Narrativity. From Medieval Performance to
Modern Fiction. Austin: University of Texas Press/London & New York: Routledge.
Fleischman, Suzanne. 1991. "Verb Tense and Point of View in Narrative." Fleischman &
Waugh, 1991,26-54.
550 Suzanne Fleischman

Fleischman, Suzanne & Linda R. Waugh, eds. 1991. Discourse Pragmatics and the Verb:
The Evidence from Romance. London & New York: Routledge.
Galton, Herbert. 1987. "Unreal Hypothetical Periods in South Slavic and Greek and
Related Features." In Honor of Use Lehiste ed. by R. Channon & L. Shockey, 357-368.
Dordrecht: Foris.
Givón, T. 1973. "The Time-Axis Phenomenon." Language 49.890-925.
Givón, T. 1984. Syntax. A Functional-Typological Introduction. Vol. 1. Amsterdam:
Benjamins.
Gougenheim, Georges. 1970. "La valeur psychologique des temps dans le 'Monologue de
Figaro'." Etudes de grammaire et de vocabulaire français, 149-54. Paris: Picard.
Hacking, Jane F. 1993. A Comparative Typology of Conditionals in Russian and Macedo­
nian. Unpublished dissertation, University of Toronto.
Haiman, John. 1989. "Alienation in Grammar." Studies in Language 13,1.129-170.
Hale, Kenneth. 1969. "Papago/ci-m." Cambridge, MA: MIT (mimeo).
Harris, Martin B. 1978. The Evolution of French Syntax. A Comparative Approach.
London: Longman.
Heine, Bernd & Mechthild Reh. 1984. Grammaticalization and Reanalysis in African
Languages. Hamburg: Buske.
Heine, Bernd, Ulrike Claudi & Friederike Hünnemeyer. 1991. Grammaticalization. Chi­
cago: University of Chicago Press.
Hopper, Paul J. 1979. "Some Observations on the Typology of Focus and Aspect in
Narrative Language." Studies in Language 3,1.37-64.
Hopper, Paul J. 1981. "Aspect and Foregrounding in Discourse." Discourse and Syntax (=
Syntax and Semantics, 12) ed. by T, Givón, 213-41. New York: Academic Press.
Hopper, Paul J. & Sandra A. Thompson. 1980. "Transitivity in Grammar and Discourse."
Language 56,2.251-99.
James, Deborah. 1982. "Past Tense and the Hypothetical: A Cross-Linguistic Study."
Studies in Language 6,3.375-403.
Klein-Andreu, Flora. 1991. "Losing Ground: A Discourse-Pragmatic Solution to the
History of -ra in Spanish." Fleischamn & Waugh, 1991. 164-178.
Kuno, Susumo. 1972. "Pronominalization, Reflexivization, and Direct Discourse." Lin­
guistic Inquiry 3,2.161-195.
Lass, Roger. 1990. "How To Do Things With Junk: Exaptation in Language Evolution."
Journal of Linguistics 26,1:79-102.
Lazard, Gilbert. 1975. "La catégorie de l'éventuel." Mélanges linguistiques offerts à
Emile Benveniste, 221-232. Louvain: Peeters.
Leech, Geoffrey N. & Michael H. Short. 1981. Style in Fiction. London: Longman.
Le Goffic, Pierre. 1986. "Que l'imparfait n'est pas un temps du passé." Le Goffic, 1986.
55-69.
Le Goffic, Pierre, ed. 1986. Points de vue sur l'imparfait. Caen: Université de Caen.
Lichtenberk, Frantisek. 1991. "Semantic Change and Heterosemy in Grammaticaliza­
tion." Language 67.475-509.
Lunn, Patricia V. 1985. "The Aspectual Lens." Hispanic Linguistics 2.49-61.
Lunn, Patricia V. & Thomas Cravens. 1991. "A Contextual Reconsideration of the
Spanish -ra 'Indicative'." Fleischman & Waugh, 1991. 147-163.
Imperfective and Irrealis 551

Maingueneau, Dominique. 1990. Éléments de linguistique pour le texte littéraire, 2d ed.

Paris: Bordas.
Martin, Robert. 1971. Temps et aspect. Paris: Klincksieck.
Mayes, Patricia. 1990. "Quotation in English." Studies in Language 14:325-363.
Mithun, Marianne. This volume. "On the Relativity of Irreality."
Nichols, Johanna. 1986. "The Bottom Line: Chinese Pidgin Russian." Chafe & Nichols,
1986. 239-257.
Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Restan, Per. 1989. "Ne bylo, no moglo by but': O gipoteticheskoi modal'nosti." [There
Wasn't But There Could Have Been: Concerning Hypothetical Modality] Scando-
Slavica 35.203-210.
Roberts, John R. 1990. "Modality in Amele and Other Papuan Languages." Journal of
Linguistics 26.363-401.
Rohlfs, Gerhard. 1968. Grammatica Storica della Lingua Italiana e dei suoi Dialetti. Vol.
2: Morfología, trans, by Temistocle Franceschi. Turin: Einaudi.
Romaine, Suzanne & Deborah Lange. 1991. "The Use of Like as a Marker of Reported
Speech and Thought: A Case of Grammaticalization in Progress." American Speech
66,3.227-279.
Slobin, Dan I. & Ayhan A. Aksu-Koç. 1986. "A Psychological Account of the Develop­
ment and Use ofEvidentialsin Turkish." Chafe & Nichols, 1986. 159-167.
Steele, Susan. 1975. "Past and Irrealis: Just What Does It All Mean?" International
Journal of American Linguistics 41.200-217.
Tannen, Deborah. 1986. "Introducing Constructed Dialogue in Greek and American
Conversational and Literary Narrative." Direct and Indirect Speech ed. by F. Coulmas,
311-332. Berlin: Mouton de Gruyter.
Tannen, Deborah. 1989. Talking Voices. Cambridge: Cambridge University Press.
Trask, R.L. 1993. A Dictionary of Grammatical Terms in Linguistics. London & New
York: Routledge.
Valin, Roch. 1964. La méthode comparative en linguistique et psychomécanique du
langage. Québec: Presses de l'Université Laval.
Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca: Cornell University Press.
Warnant, Léon. 1966. "Moi, j'étais le papa... L'imparfait préludique et quelques remar­
ques relatives à la recherche grammaticale." Mélanges de grammaire française offerts
à M. Maurice Grevisse pour le trentième anniversaire du Bon usage, 343-66.
Gembloux: Duculot.
Weinrich, Harald. 1973. Le temps, trans, by Michèle Lacoste. Paris: Seuil (originally
published in German, 1964).
Wierzbicka, Anna. 1974. "The Semantics of Direct and Indirect Discourse." Papers in
Linguistics 7.267-307.
Willett, Thomas. 1988. "A Cross-Linguistic Survey of the Grammaticalization of Eviden-
tiality." Studies in Language 12.51-97.
Wilmet, Marc. 1968-69. "L'imparfait dit 'hypocoristique'." Français Moderne 36.298-312
& 37.54. Repr. in Etudes de morphosyntaxe verbale, 83-106. Paris: Klincksieck, 1976.
Woodbury, Anthony C. 1986. "Interactions of Tense and Evidentiality: A Study of Sherpa
and English." Chafe & Nichols, 1986. 188-202.
 Index of subjects

A analogy 406, 438,445

ability 5, 78, 80-81, 82, 84-85, 131, apodosis (of conditional statements),
141,207-208,210-218,459 281,282,524,537,538
acquisition, second-language 18,394, apprehensional-epistemic modality 9,
400, 402 294,295-303,306-311.See also fear
actualization 368,370,375,376,382, development of 294, 303-306,
413,430 311-315,318,319-323
expectation of 377, 378, 385, aspect 13,91,92,532,540,542,543,
386 544
admirative modality 333, 357 continuous 521,536
adult-child interaction 167,176,201, durative 125-126,538
202, 262, 289 habitual (see habitual)
adverbs (temporal), 11,389,390,394, imperfective (see imperfective
408,421,422 aspect)
into future tense 409, 415, 421 iterative (see iterative aspect)
age grading (pidgins/creoles), 398- perfect 35
400,401,414-415 perfective (see perfective aspect)
agent-oriented modality 5, 7, 12, 13, progressive (see progressive as­
17-48,131,165-66,424,441,446, pect)
447, 476 punctual 359, 362
crosslinguistic expression of 6 assertion 9-10, 146, 411, 506, 528,
defined 6, 102, 107, 165 539, 547
into epistemic modality 6,7,17, and irrealis 146,381,385,537
18,38,39,42,46,56,62-64, and subjunctive mood 11,429-
166, 293 430,433,434,435-436,439-
into speaker-oriented modality 6 446
agentivity 26,27,35,74 strength of 81, 89, 92, 98, 146
allomorphy 417,418 authority 120, 122, 130-131, 243,
ambiguity 40, 41, 72, 208, 209, 313, 267, 269, 273, 275
504 auxiliary reduction 19
contextual 72 avertive modality 299,300-301,302,
defined 39-40 303,306,307,309,311,313
pragmatic 299 defined 298
semantic 299, 302
554 Index of Languages

B facilitating factors in 166, 167,

baby talk. See hypocoristic 175,176-177,199,201-202
backgrounding 11, 169, 432, 436, and iconicity {see under
442, 536, 539-540, 548 iconicity)
basic meaning. See under meaning and imitation 201, 230-233
bleaching (semantic), 42, 64, 115, and input 167, 176, 196
229,321,408 and knowledge status (of child),
181-188, 197
and make-believe 519,525-526
canonicity (of discourse), 522, 529 of mood 166, 180, 181
causality 397 and old and new information 8,
causative 487 189-190, 194, 196-199
certainty 59-60, 90, 174, 293, 294- and order of acquisition 167,
295, 296, 318-319, 322, 395-396, 178-179, 197-198, 207, 209-
430, 435, 545 210
in child language 175,189-194, clauses, order of 431-432
198 cliticization 468,469
challenge, challenging 8, 205-206, cognition (link to grammar), 137-138,
207, 210-217, 218, 221-222, 225- 281,349,351,364
226, 227, 230, 231, 233, 235, 548 cognitive linguistics 137, 151, 206,
child language development 394,542 349
agented-oriented before epistemic complement clause 474-476
modality 8, 17-18, 64, 166, complementizers (as expressing mo­
175, 207, 209-210, 235 dality), 11, 12, 294, 368, 473-500
and certainty {see under cer­ absence of 473, 474, 477-480,
tainty) 481,499
and cognitive development 167, function of 475-477
176, 180-181, 186-187, 188, notion of 474-476
198-199 position of 12,475
and diachrony {see under and presence of other modal
diachronic, diachrony) markers 12, 477, 478, 479-
and discourse contingency 178, 480,481
187-188,197, 199,201 sequence of 473,474,476,480-
discourse interactive functions of 481,489,491,494,495,496
modals in 8, 205-207, 210- completion 505-506, 509, 514, 521,
235 527, 539, 544
of epistemic modality 8, 166- conceptual properties 7, 29-35, 41,
167, 175-202, 224-226 56-59, 62, 65
of evidentiality {see under conditionals 99, 356-357, 370, 396,
evidentiality) 397, 506-507, 512-513, 515, 520,
Index of Languages 555

523-524, 532, 543, 544, 545, 548 de dicto modality 484,486-488,

counterfactual {see counterfac- 491,492-493,496
tual) de re modality 485-486
hypothetical {see hypothetical) deference 43,267,269,318,325,508,
and iterative/habitual 537-538 527. See also mitigation; polite­
and subjunctive mood 434, 439 ness
unreal {see unreal conditionals) definite description 343, 344
context, influence of 7, 21, 72, 100- deixis 152
101,107-108,122,131,339 deontic modality 4-5, 13, 148, 209,
contextual frame 26-29, 35-36, 455, 457, 461, 464-468, 470, 478,
47 480,481,499
extralinguistic 68-70,72-74,76- defined 4
91, 120, 129-130, 250, 299, deontic necessity {see under ne­
302, 309 cessity)
and grammaticalization 38 deontic possibility {see under
linguistic 24-26, 27, 34-36, 47, possibility)
68-70, 72-74, 76-99, 114, deontic source 77, 79, 87, 88,
125, 250, 433 101,102,78-80,85,101,102,
continuity (of clauses), 281-283 208, 509
continuity (in diachrony), 18, 37-38, into epistemic modality 5, 6
40-41,42,47,406 grammatical categories express­
cooperation 282 ing 4-5
counterfactual 103, 125, 336, 370, deontic necessity. See under necessity
377, 396, 397, 516, 523, 539. See dependence 417,418
also hypothetical; unreal desire 5, 133, 258, 528
conditionals into future tense 6,177,243-244
and obligation 254, 257, 266, and necessity 128-130,131
269, 286 and obligation 128, 131
and subjunctive mood 441-442 diachronic, diachrony 108,120,130,
creoles 389,390,395,412,416,421, 284, 396,401,402,406,410, 416,
422, 424. See also pidgins 420. See also under specific devel­
creole prototype 389,416 opments
creolization 390-391,401,402, acceleration of 400
408, 422-423 and child language 400, 414,
culture, influence of 285, 377, 396, 424
429 and complementizers 473, 499
and future tense 239, 396, 243
D and imperfective-irrealis connec­
de dicto/de re distinction 331-332, tion 12,519,520
481,488 and obligation 240, 290
556 Index of Languages

and past tense modals 504,513- 48, 142-149, 165, 208, 209, 429,
514,516 446, 455-457, 462-464, 470, 478,
pathways of 3, 5, 349, 380,445 480,481,543
and pidgins/creoles 11, 401- crosslinguistic expression of 6
402,410,416,420 defined 4,6,55,165,293
and polysemy 5,325 development of 321-322
and realis/irrealis 10, 352, 375- epistemic necessity. See under neces­
377, 385-386,429 sity
and signed language 140, 141, epistemic possibility. See under pos­
144, 150, 151, 157 sibility
social factors in 398, 400 evaluation. See speaker evaluation
direct experience 168,173,485,489, evidentiality 4, 13, 77, 85, 101, 165,
490, 542, 548 458, 529, 532, 542, 543. See also
discourse quotative
discourse context 8,433, 434 and child language 194-196,
discourse interactive functions 197
3, 199-201, 212, 214, 216, expression of 168-169, 172,
219,221,225,227,229-230, 173, 175
234,432,434 {see also under indirect evidence 481,484,487,
child language development; 490
obligation) evolution (of language), 137
genres of 85 exaptation 543
interactive 3, 8, 9, 11, 12, 169, existential subject 26, 35-36, 58
173, 175 expected/unexpected 169, 357
doubt 333,430,484,490-491,495
dreams (narration of), 519, 530-531, F
542 facial expression. See non-manual
drift 538 signs under signed language
dubitative modality 2 factive 432
dummy subject 35-36 factual 29, 88, 92, 94, 359, 363
dynamic modality 209 fantasy (narration of), 530-531, 540
fear 296,297,302,304,305-306,307,
E 309, 310, 311, 314, 315, 319, 320,
emergent grammar 446 322. See also apprehensional-
emphasis 332, 337, 343 epistemic modality
enabling conditions 5, 77 fiction, fictionality 520, 526, 529,
endearment 339-340, 528. See also 530, 534, 535, 536, 541, 544, 546
hypocoristic focal sense (of a category). See under
epistemic downtoner 294-295, 301- meaning
302, 316-317, 318, 320, 321, 322 focus (contrastive), 280, 281, 282
epistemic modality 4,5,7,12,13,17- focus (visual), 521, 539-542, 548
Index of Languages 557

force dynamics 71, 159, 206-207, grammaticalization 2, 37-47, 108,

214, 221 115,130,228, 229, 239, 243, 278-
foregrounding 539-540 279, 323, 325, 383, 386, 408-409,
form-function relation 3,135,416 516, 520, 524, 543. See also un­
form-meaning correspondence 453, der specific developments
454, 456 crosslinguistic comparison of
formal context 57, 60, 62, 175, 437 130
formal theory 12,135,136,138,151, degree of 23,46,117,131,228,
157-158, 473, 476, 477-478, 499 284-285,290,410-419
free indirect discourse 520,526,532, early stages of 294, 321
533-536, 546, 547. See also quo­ and erosion of meaning 340-343
tation styles fusion 126,166,289,353,416,
full verb 454, 465, 467, 468-469 417,418
functionalist approach 1, 137, 151, lack of 9, 329, 344
205, 210, 234, 239, 241, 473, 476 lexical verb as source of 125-
future tense 20, 33-34, 98-99, 125, 127, 130
282, 320, 410, 447, 506-507, 515- and pidgins/creoles 11, 389-
516,543 390, 393-395, 399, 401-402,
development of 11,37,358-359, 406, 410-423
390,395,409,410-416,418- and realis/irrealis distinction
423, 520 349, 358, 359, 368, 384-386
expression of 149,395,396,414 and signed language 157, 158
and irrealis 363, 370, 378-380, grammaticalization chains 38, 45,
395-396,419 323, 420,421
into imperative mood 358,410, grammaticalization paths 383, 395,
412 415, 419,421-422, 423, 542, 543
lexical sources of 409,415,422, grammaticalization theory 18, 38-39
424 grammaticized use 431, 443
motion-derived {see under mo­ 'guiltive' modality 9, 334-337
tion verbs)
future-of-the-past 297,515,524-525, H
538,544 habitual 256,270,271,287,372,396,
fuzziness, fuzzy categories 64, 70, 412,414, 514,516,521, 536, 537-
71, 85, 91, 95, 100-101, 242, 298, 538
521,548 hearsay 168,173,194,484,489-490.
fuzzy set theory 71,72 See also quotative; reported speech
generic (reference), 256, 268, 356, hedges, hedging 64, 208, 268, 269
412, 526, 537. See also habitual homonymy 18
temporality 79 hope 324
gradience 39, 71 hortative 113
gram type 1, 2, 3, 10 hypocoristic 520, 528-529, 542
 558 Index of Languages

hypothetical 43, 44-45, 92, 96, 103, inference 74, 82, 168, 173, 546, 542
395-396, 397, 413, 480, 482, 491, information value 429,430-432,433.
495-496, 520, 523, 529, 539, 542, See also knowledge status; new in­
543,544,548. See also counterfac- formation; old information
tual; unreal conditionals intention 120,123-124,131,149-150,
and past tense 503-504, 506, 260, 262, 263-264, 395, 413-414,
509-514,516 420, 508, 528
interior monologue 520,541
interrogative mood. See questions
iconicity 136, 152,281,444 intonation. See suprasegmental fea­
vs. arbitrariness 135, 151, 152, tures
344 invariant meaning. See under mean­
and child language 151-152 ing
erosion of 145, 151, 152-153 inversion (subject-auxiliary), 477,
and signed language 7,138-139, 497-498
144, 151-158 irony 333
illocutionary force 43, 44, 336, 528 irrealis 11, 282, 369-370, 397, 410,
image schemata 137-138 440, 495. See also realis/irrealis
imagination 10, 349, 350, 358, 363, distinction
368, 384, 386, 548 categories classified as 396,412
imperative mood 3, 6, 147, 358-359, contexts conditioning 353-358,
363, 376-378, 396, 412, 447, 484, 362-363, 370-371, 373-375,
490, 492, 545 377,384-385,413
from future 410,412,420,421 crosslinguistic variability of 3
vs. obligation 275, 280-283 defined 9-10, 12-13, 350-351,
imperfective aspect 11, 12, 514. See 522-523
also perfective/imperfective dis­ and future tense 395-396,419
tinction grammaticalization of 353,354,
defined 521 359, 389-390, 393-394, 395,
and irrealis 12,519-548 400, 410,422
impersonal 107, 116, 117, 118, 120, and imperfective aspect 12,519-
126,127,130,458-459 548
impersonal subject/marker 35,36,57 and negation 430
implicature, conventionalization of redundancy in marking of 10
38, 39, 322 irregular, irregularity 11, 453, 456,
incorporation 402,408,423 465, 468-470
indefinite 10,436, 498, 538, 547 systems of 459-461
indicative mood 430,431,478,480 types of 453-455
indicative/subjunctive contrast iterative aspect 521, 536, 537-538,
11,430-431,434,436 547
Index of Languages 559

J metaphor (in diachrony), 37, 38, 39,

jokes, joking 255,529-530 40-41,42-43,45,47,48,71, 118,
145, 229, 322
K metonymy 305,306,319,322
knowledge status 168-169, 172-173, mitigation 78,81,84-85,89-90,250-
174, 175. See also information 252, 257, 262, 264. See also def­
value; new information; old infor­ erence; politeness
mation modal logic 4
modal verbs 67-68,70,170-171,464,
L 465, 470
logical equivalence 456, 457, 461, choice of 72
470 vs. full verbs, 454,456,468-465
inflectional marking of 19, 47
M iteration of 19
make-believe (and irrealis), 519, 525- order of 19
526, 542 modality 454
media language 420-421 agent-oriented (see agent-ori­
meaning 73 ented modality)
basic/core 35, 71, 72, 87, 206, crosslinguistic comparison of 3
227, 284 defined 2, 70-71, 74, 165, 476
contextualized 73-74 deontic (see deontic modality)
focal sense 21-22,25-26,27 dynamic 209
invariant 6,7,69,70,71,72-73, epistemic (see epistemic modal­
75-76,78,80,86,90,98,100- ity)
101 mixed 293-294, 295
literal interpretation 37, 288 modal force 29, 30, 41, 42, 47,
marginal sense 21-22 58
peripheral 78, 284 modal function 2
prototypical (see prototype pro­ redundancy in marking of 2
totypical meaning) root 55,75, 139-142
referential (see referential mean­ scope of 74,116,118,120,124,
ing) 330, 419
secondary 72 speaker-oriented (see speaker-
semantic (see semantic meaning) oriented modality)
merger 31, 39,41, 56, 61-62, 64, 65, social/interactive functions of
245,313,412 3,8
defined 40, 61 modesty 436
metalinguistic function 329,331 -332, module, modularity 135, 137
344, 429, 432,433, 529 monosemy 68, 69-70, 71, 100-101,
metamessages 335-337, 345 298-299
560 Index of Languages

mood 3, 13, 91, 108, 139, 329, 341, and subjunctive mood 440
429,431,473 neutralization (of distinctions), 40,
choice of 11, 443, 445, 446, 249,251,256,353,524,531
431-432,434 new information 174-175, 431. See
conditional {see conditionals) also information value; knowledge
defined 2,70-71,329 status; old information
imperative (see imperative mood) noteworthiness 172
indicative (see indicative mood)
irrealis (see irrealis) O
realis (see realis) obligation, obligative modality 4, 5,
subjunctive (see subjunctive 6,8,33,38,55,90,107-131,139-
mood) 140, 258, 356, 359, 363
motion verbs 48,389 and complementizers 479-480,
into future tenses 37, 38, 125, 481-484,490,492,496
415,422,524-525 crosslinguistic comparison of 9,
movement rules (formal theory), 474 240, 245-246, 269-271, 283,
284-285
N development of 7, 107-108,
narrative 8, 429, 443-444, 519, 520, 115-120,159,289
529-532, 534-536 degree of 140, 241-242, 247,
necessity 4, 107, 120, 121-123, 131, 272
140, 455,456-457, 459,462, 467 discourse interactive functions
and constraints 120, 121, 128, of 9,240,241,247-257,259-
129 271,273-277,283,284
epistemic 4, 55, 59, 60, 86, 87- into other categories (epistemic
89, 458-460 modality, future tense, infer­
root/deontic 55, 59, 60, 128- ence) 38,239,243-244,245,
130, 455, 458-460, 462, 463 278, 279, 284
negation 11-12, 63, 143, 150, 453- and negation 454, 455, 457,
470. 465-468,470
de dicto (metalinguistic)/de re and past tense 504, 510, 512,
102,329,331-332 513,515,516
and future tense 361-362 word order marking of 240,246,
(mis)placement of 12,332,454, 278, 279-283, 288
461,463,465,466,468,469 old information 431, 432-433, 434,
and realis/irrealis 430,447,354- 438. See also information value;
355,360,374-375,380,381- knowledge status; new information
384 optative modality 329, 359, 361,492
scope of 11,74,78, 102,455-
456,461,470
Index of Languages 561

and contextual factors 7

parametric variation 12,473 and diachrony 5
passive voice 287 of deontic and epistemic forms 5
past tense 11, 12, 43-45, 397, 503- of agent-oriented and epistemic
516, 519-520, 523, 525, 527-528, forms 6,7, 17, 18, 19
535,539,540, 544. See also under of root and epistemic forms 55-
subjunctive mood 56
perceptuomotor capacity 136-137 possibility 4, 7, 67, 302-303, 306,
perfective aspect 359, 364, 526, 529, 359, 455,456-457, 462, 467
536, 542, 544, 545 epistemic 4,5,55-65,71-72,74,
perfective/imperfective distinction 3, 78, 82-85, 86, 87-89, 124-
521, 524, 525, 531, 535,538, 539- 125, 142-144, 208, 458-460,
540, 544, 545, 547, 548 469
performance constraint 344-345 root/deontic 5,7,20,55-65,71-
permission 4, 5, 6, 8, 20, 33, 55, 78- 72, 74, 78,79, 82, 84-85, 94-
80, 85, 208, 455, 457, 218-221, 96, 142, 208, 370, 455, 459-
222-223 460
and negation 455, 457, 461, pragmatic factors 72, 73
465-468, 470 pragmatic strengthening 38,294,322-
philosophy of language 4 323
phonological change 140 pragmatic weakening 323
phonological reduction 115,126,166, precautionary situation 297-298,307-
394,407,418-419,420,422 308,311,322
and degree of grammaticaliza- precautioning function 302,305,306,
tion 416,417-418 307, 309, 313, 314, 315, 319-320
and syntactic position 401,402, prediction 177, 411, 412, 149-150,
405-406,408,409,410,411 395,413-414
pidgins 389,390,406,416,421,422 as prototype of future tense 410,
jargon stage of 402, 409, 410, 420
419 present tense 12, 78, 84, 87, 88, 503-
nativization of 399, 422 507, 509, 511, 513, 515, 520, 525,
stabilization of 399, 408,433 526, 528, 530, 540-541
standardization of 392 presupposed, presupposition 275,
politeness 43-45, 81, 267, 317-319, 430-431,432,433,548
323, 336, 377, 378, 385, 435-436, probability 5, 29-30, 32, 70, 124
438, 447, 508, 526-528, 542, 545. progressive aspect 25,26,34,35,519,
See also deference; mitigation 521, 525, 527, 528, 535, 538, 544
polysemy 70,71,101,102,108,207- prohibition 216-217, 222, 223-224,
208, 258-259, 298, 299, 302-304, 231,311,356,383,491,493
306, 325. See also synchronic proper name 343-344
variation
 562 Index of Languages

protasis (of conditional statements), crosslinguistic validity of 3,

282, 439, 524, 526 350-351,363,368,396,429
prototype, prototypical meaning 12, crosslinguistic variability of 10,
31,57,58,73,74,410,420,522, 360-361,362,363-364,367-
535 368, 375-386, 396, 522, 547
prototype model 78, 85 inconsistent marking of 349,
purpose 299-300,302,304,305,309- 350-351,359,364
310,311,312,420,485 grammaticalization of 349,352,
358-359, 375-377, 380, 383,
Q 384-386
questions 491 reality 10
confirmative 361, 363 defining 429
embedded 445, 486, 497-498 gradient nature of 358-364
rhetorical 208,225 scope of 360-361, 363, 364,
scope of 495 380-381,382-383,385-386
tag 146,435,476 reanalysis 116, 118, 126, 127, 384,
wh-, 148,159,354,476,497-498 386, 406-407
yes-no 63, 174, 353-354, 360- context-induced 38, 39, 40-41,
361,363,373,380-381,476, 45,47
478, 479, 493-495, 497 redundant marking 2, 10, 394, 400,
quotation styles 520, 532-537, 542, 497
547. See also evidential; free indi­ reduplication 337, 343
rect discourse referential meaning 206, 211, 218.
direct 532-536, 546 See also semantic meaning vs. in­
indirect 520, 532-534 terpersonal meaning 227
quotative 333, 334, 532-533. referentiality/non-referentiality 10,
See also evidential; hearsay; 537, 547
reported speech regular, regularity 453, 458, 462,
463-470
R systems of 459-461
realis 3, 9, 371-373, 535, 537. See reinterpretation. See reanalysis
also realis/irrealis distinction relative clauses (restrictive), 433,445
contexts conditioning 354,358, relative clause markers 485-486
372, 373, 375 relevance (of information), 430, 433
realis/irrealis distinction 3, 10, 351- replication (repetition), 337-343, 344
353, 359-360, 362, 368-369, 396, reported speech 395. See also hear­
529 say; quotative
cline of 532 root modality. See under modality
cognitive basis of 349,350,363,
364
Index of Languages 563

speaker beliefs 12, 478, 480, 499,

sarcasm 9, 330-334, 335, 343, 344 522, 523
scope of marking 330, 333 speaker commitment 4,12,125,165,
semantic change 107, 118, 229, 294, 174,341,344,522,537
321-323, 389, 408-409, 504-516 speaker evaluation 248,251 -252,254,
and repetition 340-343 271,284,296,429
degree of 410-416, 421 speaker-hearer relationship 228-229,
to expressive/interpersonal mean­ 275
ings 228-229 speaker involvement 228, 242-243,
to textual/cohesive meanings 256,286. See also subjectification;
321,322,325,443 subjectivity
semantic loss 13, 44, 127, 130, 322, speaker judgments 429,431,433
445 speaker-oriented modality 13,476
semantic meaning 207,211 -212,214- crosslinguistic expression of 6
215, 284. See also referential defined 6
meaning specific/non-specific reference 80,
semantic weakening 510, 513 83, 349, 362-363, 364
sequence of tenses 445 speech acts 6, 9, 321, 322, 325, 336
sequentiality 372, 375 indirect 528, 547
markers of 394,412 subjectification 17,38,321,322
vs. simultaneity 371, 375 subjectivity 59-60,64,108,242,286,
sex differentiation (language use), 398, 294, 296, 534. See also speaker
399 involvement
signed language subjunctive mood 44-45, 112-113,
iconicity in 7, 138-139, 144, 113,481,514,524. See also un­
151-158 der indicative mood
movement types 140, 143, 144, and assertion 10, 11, 430, 434
153-155 crosslinguistic variability of 10
non-manual signs 143, 144, development of 436-437, 442,
146-148 445, 520
phonological transitivity 154- dialectal variation in usage 434-
155, 156-157 436, 440-443
semantic phonology 138, 154, and irrealis 440
156 journalistic uses of 432-434,
vs. spoken language 137, 158 438
spatialization-of-form hypothesis and obligation 480, 483
137-138 past forms of 429, 432, 435,
speaker attitudes 9,74,101,120,123, 436, 438-443
165, 293, 364,443, 445,476 textual usage patterns 443-444
cline of 430 subordination 429, 439, 445, 460,
474,475,485
 564 Index of Languages

substrate influence 408

superstrate influence 40
u
suppletion 10-11, 453-457, 461-470 universals, universality 6, 240, 243,
suprasegmental features 9, 62, 63, 260, 282, 285, 344, 363, 389,429,
174, 330, 335, 342, 478, 479, 498 430,513,538
surprise 333, 357-358, 363 unreal conditionals 397, 523-524,
synchronic variation 108, 117, 131. 526, 537, 538, 544. See also
See also polysemy counterfactual; hypothetical
synonymy 70 V
syntactic change 107, 118, 389, 397- verb types
409,413-414,422 action/dynamic 25, 29, 34, 35,
and degree of grammaticaliza- 74
tion 416-419 durative 36
and phonological reduction (see mental activity 25, 486, 488-
under phonological reduc­ 490, 508, 527, 528
tion) motion (see motion verbs)
perception (verba sentiendi),
T 481,485,487,490,495,496
temporal adverbs. See adverbs (tem­ speech (verba dicendi), 81,481,
poral) 484,486,488,491,494-495,
temporal relations 394, 400 496, 535-537
tense 13,91,92-98,544,548 stative 25, 26, 34, 35, 83, 287,
future (see future tense) 506, 508
past (see past tense) telic 25
present (see present tense) 'verbiness', degree of 23,455
tense-mood-aspect 389, 396 volitive, volition 149-150,296,324,
grammaticalization of 389-390, 482,505,512,513,515
416,419-420
tentativeness 63 w
thematicity 431 warning 304,305,320,322,325,410
theme/rheme 431-432 wishes 324,479-480,495
transitivity word order 239,431-432,478
interpersonal 148-149 effect on modal interpretation
phonological 154-155,156-157 111,114,287
semantic 154 as obligation marker (see under
truth (of propositions), 4, 8, 55, 71, obligation)
124, 165, 174,430,476,480,490, world knowledge 28, 76
495, 499, 539 world view 285,429
typology 239, 453,457, 469 written language 407
and diachrony 416,418
 Index of languages

A
Acholi 107-117,119-127,130-132
c
Alamblak 378, 547 Caddo 10, 349, 351, 353-354, 358-
Albanian 333 365, 367, 378, 380-383, 385
Cape Verde Creole 422
Alsea 368
Catalan 436
Alur 108-111, 113, 117-120, 125-
Cayuga 359
126, 128-130, 132
Central Siberian Eskimo 320
Amele 375-376, 380
Chadic languages 486, 498, 500
American Sign Language 7,135,138-
Chinese 240-242,245-247,253,257-
140, 142, 144-146, 148, 150-159
258, 260, 264-274, 277-278, 285-
Antiguan Creole 228
286, 288, 466 (see also Mandarin)
Arabic 80-81,465-466
Chinese Pidgin English 408, 424
Assamese 462, 465-466
Chinese Pidgin Russian 169, 542
Austronesian languages 9, 294, 315,
Cree 523
325, 367, 396
Czech 302,309,319
B D
Bahasa Malaysia 466 Danish 463-466,468-470
Balkan languages 523 DhoLuo 108,110,113,118-120,129-
Bargam 537 131
Basque 304 Dhopadhola 108,109,111,113,115,
Bauan 315 119,126, 130,132
Betelgeuse 334 Diegueño
Bikol 537 Jamul 368, 377-378, 383-384,
Bislama 403, 422 547
Boiken 403 Mesa grande 383-384
Boumaa Fijian 315,317-318,320 Diyari 306
Brazilian Cities Sign Language (BCSL) Dutch 474, 500, 530
155
Bulgarian 542, 545, 548 E
Buriat 2 East Sepik 378
English 4, 5, 7-10, 12-13, 17-20, 26-
566 Index of Languages

27,33-35,37,39,40,43-44,47,55- F
61, 65, 68, 70-71, 78, 85, 91, 94, Farsi 465,466
96, 100, 102, 118, 120, 124-125, Fijian 315,322,325
129,166,170,176,201,204, 206- Boumaa 315,317-318,320
208, 212, 227-228, 232, 235-236, Nabukeleva 325
240-249, 252-253, 257-260, 262- Standard 315-318
266, 268-275, 277-281, 283, 285- Wayan 320
286, 288-291, 294, 302, 304, 313, Finnish 544
320, 325, 333-334, 342, 349, 350- French 12, 102, 330, 352, 403, 437,
351, 361-365, 378, 382, 389, 391, 453, 458, 460-461, 465, 479-483,
396-397, 401-404, 408-410, 417- 500, 504, 514-515, 520-521, 523-
419, 422, 424-427, 454-457, 461- 526, 528, 530, 532-533, 537, 539,
462, 465-466, 468-470, 474-479, 544-545, 550
485, 497, 498, 500, 503, 504-511, Old 480
513-516, 520, 524, 527, 530, 535- French Sign Language (FSL) 158-159
538,546-547 549,551 Old 140
American 61, 63-64, 330-331,
513,516 G
British 47, 60, 63-64, 244, 331, German 7, 17-20, 23-29, 34-36, 42,
541 46-48, 50, 56, 58, 76, 352, 446,
Early Modern 12,511,516 467-470, 523, 544
Elizabethan 513 High 27,48, 52
Late Old 507 Modern 48
Middle 12, 505, 507, 509-511, Gothic 478
514 Greek 166, 289, 311, 460, 467-468,
Modern 43, 315, 503, 506-508, 523
511,513 Classical 311,320
Old 314, 479, 497, 498, 504, Homeric 311
507,516 Modern 459, 462, 467
English-based pidgins and creoles Guider (Central Chadic) 481-483
389, 391, 407, 427 (see also
Chinese Pidgin English H
Hawai'i Creole English Hawai'i Creole English (HCE) 391,
Hawai'i Pidgin English 422
New South Wales Pidgin English Hawai'i Pidgin English 422
Pacific Jargon English Hebrew 455, 466, 538
Pacific Pidgin English) Biblical 9, 240-241, 246-247,
European languages 7,10-11,13,367 277-279, 281-283, 285-286,
Ewe 343 288
Index of Languages 567

Hopi 9, 240-241, 245-247, 271-278, M

285-286, 288-290 Macedonian 542, 545, 547-548
Hua (Papuan) 309,319 Malacca 422
Mandarin Chinese 8,9,64,205,207-
I 208, 212, 214, 224-227, 235-237
Indonesian 243 Beijing 236, 287
Irish, Old 523 Taiwan 236
Iroquoian 359 Maricopa 376, 377
Northern 349,351,358-364 Martuthunira 302-303,307-308,319
Italian 12, 453-454, 461-463, 465- Melanesian Pidgin 325, 401, 403,
466, 524, 527, 530-531, 533, 536, 408, 423 {see also Tok Pisin)
544-545,549,551 Mesoamerica 367
Mohawk 359
J Mupun 490, 492-493
Japanese 343, 463, 466
N
K Nabukelevu 325
Kashaya 334 Nahuatl, Classical 504, 515
Kikuyu 466 New South Wales Pidgin English 408
Kinyarwanda 465, 466 Ngiyambaa 305
Korean 8, 165-170, 172-173, 175- Nisenan 367
177, 180, 193, 197, 201, 203-204
Kru languages 409,421 O
Kwai 303 Oneida 359
Kwaio 303 Onondaga 359
Oromo 342,458,465
L
Lango 107-117, 119-121, 123, 125- P
132 Pacific Jargon English 410,419
Latin 436-437, 523 Pacific Pidgin English 410,424
Latvian 461,465 Papiamentu 422
Lau 303 Papuan languages 358,367,396,403,
South 303-304 537
Lele (East Chadic) 484, 500 Persian, Old 547
Lhasa Tibetan 168 Polish 478-482, 495-496
Luo 109 Porno 334
Central 368-369,371-373,375-
376, 378-385
Portuguese 422, 524-525
Portuguese-based creoles 389
 568 Index of Languages

R Torres Strait Creole 416

Romance languages 100, 520, 523- Turkish 168,333,342,542
524, 527, 535-536, 544, 550 {See Tuscarora 359
also under individual Romance
languages) U
Rumanian 334, 530 Ugandan languages 109
Russian 80-81, 236, 330, 342, 500,
542 w
Welsh 458, 462, 464, 467
S Western Nilotic languages 7, 107-
Samoan Plantation Pidgin 402 108,112,118,125,127,130-131
Sango 423
Sanskrit 530 Y
Seneca 359-360 Yiddish 343
Serbocroatian 538 Yuman languages 376-377, 383
Sherpa 169,537,542
Sierra Miwok 504,515 z
Solomon Islands Pijin 403, 422 Zulu 304
Spanish 2, 7, 11, 67-69, 87, 90, 93-
94, 98, 103, 239, 243, 245, 339,
422, 429-432, 435-440-447, 504,
508, 515, 521, 523-527, 530, 540-
541,544,550-551
Argentine 441,442,417
Chilean 96, 103
Madrid 96, 442
Latin American 96,435
Peruvian 440
Spanish-based Atlantic Creoles 422

T
Tamil 465, 466, 530
Tibetan 542
Tigrinya 457,458, 465
To'aba'ita 294, 299, 301-306, 315,
320, 325
Tok Boi 424
TokPisin 11,389-391,393,395-397,
400-401, 403-405, 407, 409-411,
416-424 {see also Melanesian
Pidgin)
 Author Index

A Biskup,R 424
Abraham, W. 18,19 Bloom, H. 289
Adachi,T. 343,334 Bloom, L. 176,187,201,289
Adam, J. 526, 543 Bolinger, D. 429,445-477, 485
Adams, D. Q. 334 Borchardt,K. 403,424
Aijmer, K. 38 Borges, J. 338-339
Akatsuka, N. 168, 201 Borsley, R. 474
Aksu-Koç, A. 166, 168, 333, 543 Bosch, P. 73
Albert, R. 288 Bouchard, D. 474
Alcantra, S. 369 Bourciez, E. 480
Alerotek 122 Bourdin,R 544
Andrews, J. 504,515 Boyes-Braem, R 140, 159
Ansre, G. 343 Brecht, R. 500
Aristotle 313 Bresnan,J. 474,477-478
Amstrong, D. 138, 154 Brito,F. 155
Aronson, H. 538, 542, 543, 548 Brody,G. 176,201
Austin, R 306,307 Browman, C. 137
Brown, P. 149
B Bryce,A. 440
Baker, C. 144 Buckley, E. 367, 368
Baker, P. 424 Bull,W. 72
Banfield, A. 535, 537, 546, 547 Buscha,J. 19
Barnes, A. 287,288 Bybee, J. 1,2,4-6,12-13,17,20,33-
Barsalou, L. 446 35, 37, 39, 42-47, 62, 64, 70, 73,
Bavin, E. 7, 109 75, 77, 93-94, 102, 107-108, 115-
Bazzanella, C. 543,544 116,118,120-121,124,126, 129,
Bello, A. 72 131, 133, 137, 156, 159, 165-166,
Bellugi, U. 136, 152, 153, 158, 159 177, 193, 229, 236, 239, 243-244,
Benet, J. 437 279, 284, 287, 293, 321, 351, 367,
Bergmann, U. 393 395, 396, 409-412, 415-421, 423-
Bertinetto, P. 524, 534, 536, 543- 424,447,476, 514, 521, 525, 527,
544, 545 536-537, 539, 543-544, 547
Bickerton, D. 389-391,396,416,474- Byrnes, J. 175
475, 477
 570 Index of Names

c Dasgupta, P. 474
Davidsen-Nielsen, N. 457,463,468-
Calbert,J. 34
469
Cao, Y. 258-259, 261-262, 266-268,
Deane,P. 137-138
270, 287
DeLancey, S. 168-169
Cappello, S. 526,543
DeMello,G. 441,447
Cedeño,A. 431-432
Demers, R. 19,47
Chafe, W. 4, 10, 13, 165, 168, 333,
Dempwolff, O. 367
349, 352, 361-362, 368, 378, 380-
Dench,A. 307-309
383, 385,429, 522, 530,533,543-
Detzner, H. 403,424
544, 546-548
Dixon, R. 295,317,318
Choi, S. 8,17,166-167,172,181,183
Donaldson, T. 305
Chomsky, N. 151,474,513
Donoso, J. 435
Chowning, A. 406
Dore, J. 203
Chung, S. 367,538
Driberg,J. 112,115,119
Churchill, W. 402
Duff,M. 175
Churchward, C. 316
Cichocki,W. 474
E
Clancy, P. 201
Eatough, A. 367
Clark, W. Van Tilburg 519,530
Edelman, G. 136
Claudi, U. 21, 37, 38, 40, 42,45, 48,
Ehrlich, S. 546
543
Emonds, J. 478
Coates, J. 7,18,31,27,33-35,39-41,
47,55-56,61,64,68,71-72,74,78,
F
94, 100, 103, 108, 124, 206, 236,
Faltz,L. 73
242-245, 284, 286-287, 313, 503-
Fielder, G. 548
504,514
Fischer, S. 139
Cokely,D. 144
Flaubert, G. 533,534
Collins 313,315
Fleischman, S. 11-12, 43-45, 92,
Comrie,B. 367,378,521,525,536
108, 158, 429, 447, 507-508, 514-
Conteras, H. 432
515, 525-526, 529-530, 541, 544-
Cooreman, A. 239
545, 547
Craig, C. 367
Foley, W. 367,396
Cravens, T. 436,544
Fowler, C. 136-137
Crazzolara, J. 112-113, 115
Fox, A. 239,280,303
Criper,C. 109
Frajzyngier, Z. 12,476,477-478,481,
Crowley, T. 401-402,422
485,487,491,498
D Freeland,L. 504,516
Dahl,Ö. 1,2,523,537-539,544 Friedman 333
Daniels, K. 369 Frishberg,N. 140,145,150-151,157
D'Annunzio, G. 531 Furrow, D. 175-176
Index of Names 571

G Helbig,G. 19
Galton, H. 523, 538, 547-548 Hernández, C. 67
García, E. 72-73 Herring, S. 239
Geertz,C. 273,276 Heschel,A. 289
Geraghty, P. 317-318 Higgins,D. 141, 145
Gerhardt, J. 8, 236, 228, 235 Hirst, W. 175
Gili y Gaya, S. 439 Hood,L. 176,187,201
Givón,T. 48,77, 135-137, 146, 149, Hooper, J. (see Bybee) 429
151-152, 157, 159, 228, 239, 243, Hopper, P. 1,154,236,239,320,325,
281-282, 318, 325, 351, 362, 367, 446, 539, 540
378, 395-396, 412, 416, 474-475, Horn,L. 299,331,343
505, 537, 544, 547 Humphries, T. 140,147, 158
Glick,R. 109 Hiinnemeyer, F. 21,37,38,40,42,45,
Goldstein, L. 137 48, 543
Goodwin, W. 311-312 Huntingford, G. 118-119
Gopnik,A. 167,181
Gordon, L. 376-377,504 J
Gordon, R. 504 Jack,F. 369
Gougenheim, G. 528, 529, 545 Jackendoff, R. 362
Gough, B. 139, 150 Jaisser, A. 547
Greenbaum, S. 497, 498 Jakobson, R. 13
Gregerson, E. 118 James. D. 397, 514, 523, 538-539,
Grice,R 545 543, 548
Grondona, M. 431 Janssen, T. 18-19
Gsell,0. 431 Jasperson 476-477,481, 485,487
Guitart,J. 432 Jelinek, E. 19,47
Guo,J. 8,64,437 Jenkins, L. 19
Jensen, E 480
H Jespersen, O. 29, 32,42, 56, 58,135,
Hacking, J. 542-543, 545, 547-548 314
Hackman, B. 303 Johnson, M. 137
Haegeman, L. 55,71, 102 Johnson, R. 159
Haiman,J. 9,151,159,309,330,332, Joseph, B. 474
343, 532, 533, 535, 546-547 Jourdan,C. 401,422
Hale,K. 528,545
Hall, J. 314,404 K
Halliday, M. 68,70,94,102,205,228 Kac,M. 474
Harris, M. 524 Kany,C. 436
Heine, B. 7,20-21,37-38,40,42,45, Kay, P. 390
47-48, 56-58, 65, 76-77, 158,446, Keesing, R. 303,403,408
543 Kemmer, S. 331
572 Index of Nantes

Kim, Y. 166,167 Leech, G. 61,64,546

Kimura,D. 136 Lefevbre,C 474
King,L. 72 Leland, C. 424
Kipling, R. 315 Levinson, S. 28,149
Kirsner, R. 476, 477,487 Li, C. 236, 258, 288
Kitagawa,C. 19,47 Lichtenberk, F. 9,544
Kitching,A. 115 Liddell,S. 159
Klaeber 314 Ligo, G. 403
Klein, F. 430,447 (see Klein-Andreu) Lipski, J. 431
Klein-Andreu, F. 92, 436, 440, 442, Liu,Y. 258
544 (see Klein) Lo,L. 287,288
Klima, E. 136,152-153,158-159 Lomatuway'ma, M. 273-277, 288
König, E. 38-39,236,321,323,548 Long,J. 139,140-142,145,149-150,
Kononov, A. 333 159
Koster, J. 474 Lü, S. 224
Krakusin,M. 431,432 Lunn,R 11,430,436,440,521,532,
Kratzer, A. 71-72, 102 540-541,544-545
Kuno, S. 547 Lyons, J. 4, 13, 59, 78, 86, 102, 206,
Kwilosz,D. 102 211-212,214,228,234,242

L M
Laberge, S. 390, 394, 398-402, 406, Maingueneau, D. 533
408,411,419 Mallinson, G. 334
Labov,W. 390,402,408 Malo, S. 119
Ladefoged,P. 109 Malotki,E. 273-277,288
Lakoff, G. 137 Marchese, L. 409,421
Lakoff,R. 137,434,236 Marcos Marín, F 67
Lambdin, T. 286 Markey,T. 390
Langacker, R. 17, 135, 137, 514 Marsé, J. 441
Langdon, M. 383-384 Martin, R. 529, 545
Lange, D. 536 Mayes, R 546
Lass,R. 543 McCay,W. 331
Lavandera, B. 443 McDonald, B. 157
Lawal,N. 474 Meier, R. 151-152
Lazard, G. 538,547 Meltzoff,A. 167,181
Lau,J. 287 Mihalic, F. 393,401,402
Le Goffic, R 521, 524, 526, 543 Miller, A. 367-368, 377-378, 384
Leben, W. 498 Miller, G. 102-103
Lee, H. 169, 172-175, 177, 187, 194 Mitchell, B. 314
Lee,K. 172 Mithun,M. 10,354,429,522,545,547
Index of Names 573

Moliner,M. 103 Paoli,F. 369

Moore, C. 175, 176 Paul G. 317
Moravcsik, E. 337 Pawley,A. 320,325
Mowrey, R. 137, 159 Payne, T. 361
Mühlhäusler, P. 390,403 Payne, D. 239,367
Murphy, J. 401 Pellegrini, A. 176,201
Myhill, J. 8, 239, 280, 282, 287 Pérez, A. 38
Pérez, B. G. 531
N Perkins, M. 68,71-72,75-76,86,102,
Narbona,A. 67, 102 206, 227
Nichols, J. 4, 13, 165, 168-169, 333, Perkins, R. 4-6,22,33-35,39,42,46-
542 47,71,73,75,77, 107, 156, 159,
Niño-Murcia,, M. 447 351, 367, 395-396, 409-412, 415-
Noonan, M. 474 421, 423-424, 447-476, 539, 547
Perlmutter, D. 34
O Piaget,J. 198
Oehrle,R. 19,47 Plato 312
Ogden, C 211 Plaut, W. 278,280
Omondi,L. 118-120 Poizner, H. 136, 158
Ong,W. 339 Pottier, B. 78, 102
Oropeza, E. 369 Puig,M. 441
O'Rourke, T. 140, 147,158 Pure, K. 175, 176
Oswalt, R. 334
Q
P Quirk, R. 497,498
Padden, C. 139, 140, 147, 158
Pagliuca, W. 4-6, 17, 20, 33-35, 37, R
39, 42, 46-47, 62, 73, 75, 77, 102, Ramat,A. 18
107,118,126,137,156,159,177, Rand,C. 287
239, 243-244, 279, 293, 321, 351, Ransom, E. 475-476
367, 395-396, 409-412, 415-421, Real Academia Española 67
423-424, 447-476, 521, 525, 539, Reh, M. 543
544, 547 Restan, P 538,544,545
Palayer,R 486 Reuse, de, W. 320
Palmer, F. 11, 13, 18, 29, 33, 55, 58, Richard, J. 341
63, 68, 70, 78, 165, 168-169, 206, Richards,I 211
219, 229, 236, 241-243, 275, 283, Ringe, P. 119-120
286, 293, 295, 324, 329, 351, 367, Rivero, M. 67
456, 458, 475-476, 478, 522-523, Rizzi,L. 474
544 Robbe-Grillet, A 540
 574 Index of Names

Roberts, J. 358, 367-368, 375, 378, Stoneman, Z. 176,201

396,412,429,537 Studdert-Kennedy, M. 136
Rocissano,L. 176, 187,201 Sun,C. 239
Rohlfs,G. 544 Sweetser, E. 18,37,39,42,71-72,77,
Romaine, S. 11, 367, 391-392, 394, 145, 158, 206-207, 229, 232, 236,
400, 406-408, 424, 429, 536 304, 325
Rosenbaum, P. 476
Rosenberg, D. 289 T
Roth, P. 334 Taibo, P. 435
Rudin,C. 474 Talmy,L. 71,159,206-207
Rulfo,J. 540 Tannen, D. 546
Taylor, A. 358
S Taylor, D. 412
Samarin,W. 423 Terrell, T 429
Sankoff, G. 390, 394, 398-402, 406, Thiagaran, K. 457
408,411,419 Thibault, P. 102
Sapir,E. 538 Thompson, C. 239
Saussure, F. 151 Thompson, S. 154, 236, 258, 288,
Sayaba,T. 320 476-477, 487, 540
Scheurweghs, G. 313 Timberlake, A. 367,538
Schick, B. 159 Tomasello,M. 167
Schuchardt, H. 424 Trask, R. 350, 522, 544
Schütz, A. 315-316 Traugott, E. 17,38-39,42,46-48,62,
Shatz,M. 176,201 64, 108, 120, 127, 228-229, 236,
Shaul,D. 288 294, 296, 321-323, 443, 497-498
Shearin,H. 314 Trudgill,P 62
Shepherd, S. 17,46,228 Tryon,D. 303
Short, M. 546 Tsang,C. 224,236
Silva-Corvalán, C. 7, 72, 92, 103,
235, 445 U
Sirbu,D. 87,90 Ultan,R. 243
Slobin, D. 168, 176, 181, 333, 543
Smith, A. L. 9 V
Solé,Y. 433 van Valin, R. 545
Sperber, D. 430 Vater, H. 20,23
Stafford, R. 118-120 Vázquez, M. 439,444
Steele, S. 19,47, 193, 397, 514, 545 Vendler, Z. 544
Stephany, U. 18, 64, 166, 175, 180 Visser, F. 478
Stokoe,W. 138,154 Vogel, H. 403
Stolz, T. 37
Index of Names 575

W Wilson, L. 247-248, 250-256, 286,

Wachowicz, K. 474 288
Wandruszka, U. 431 Wise, M 367
Wang,T. 287,288 Woehr,R. 431
Warnant, L. 525 Woodbury, A. 169,537
Wasow,T. 19,47 Wright, G. von 77, 293-394
Waugh,L. 514 Wright, L. 437
Weil, J. 175
Weinrich,H. 539 X
Wells, G. 166 Xing, Z. 280, 287
Wertheimer, R. 103
Wierzbicka, A. 343,547
Wilbur, R. 159 Yu, Cao 257, 287
Wilcox, P. 7, 145
Wilcox, S. 7, 136 Z
Willett,T. 13,537,548 Zadeh,L. 71
Wilmet,M. 529 Z'graggen, J. 424
Wilson, D. 430 Zoh,M. 166
Typological Studies in Language
A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com

71 Nedjalkov, Vladimir P. (ed.): Reciprocal Constructions. With the assistance of Emma Š. Geniušienė and
Zlatka Guentchéva. Expected June 2007
70 Zúñiga, Fernando: Deixis and Alignment. Inverse systems in indigenous languages of the Americas.
xii, 309 pp. Expected November 2006
69 Aranovich, Raúl (ed.): Split Auxiliary Systems. A cross-linguistic perspective. viii, 268 pp. + index.
Expected December 2006
68 Abraham, Werner and Larisa Leisiö (eds.): Passivization and Typology. Form and function. 2006.
x, 553 pp.
67 Veselinova, Ljuba N.: Suppletion in Verb Paradigms. Bits and pieces of the puzzle. 2006. xviii, 236 pp.
66 Hickmann, Maya and Stéphane Robert (eds.): Space in Languages. Linguistic Systems and Cognitive
Categories. 2006. x, 362 pp.
65 Tsunoda, Tasaku and Taro Kageyama (eds.): Voice and Grammatical Relations. In Honor of
Masayoshi Shibatani. 2006. xviii, 342 pp.
64 Voeltz, F. K. Erhard (ed.): Studies in African Linguistic Typology. 2006. xiv, 426 pp.
63 Filimonova, Elena (ed.): Clusivity. Typology and case studies of the inclusive–exclusive distinction.
2005. xii, 436 pp.
62 Couper-Kuhlen, Elizabeth and Cecilia E. Ford (eds.): Sound Patterns in Interaction. Cross-
linguistic studies from conversation. 2004. viii, 406 pp.
61 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 2.
2004. xii, 319 pp.
60 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 1.
2004. xii, 325 pp.
59 Fischer, Olga, Muriel Norde and Harry Perridon (eds.): Up and down the Cline – The Nature of
Grammaticalization. 2004. viii, 406 pp.
58 Haspelmath, Martin (ed.): Coordinating Constructions. 2004. xcv, 578 pp.
57 Mattissen, Johanna: Dependent-Head Synthesis in Nivkh. A contribution to a typology of
polysynthesis. 2003. x, 350 pp.
56 Shay, Erin and Uwe Seibert (eds.): Motion, Direction and Location in Languages. In honor of Zygmunt
Frajzyngier. 2003. xvi, 305 pp.
55 Frajzyngier, Zygmunt and Erin Shay: Explaining Language Structure through Systems Interaction.
2003. xviii, 309 pp.
54 Aikhenvald, Alexandra Y. and R.M.W. Dixon (eds.): Studies in Evidentiality. 2003. xiv, 349 pp.
53 Givón, T. and Bertram F. Malle (eds.): The Evolution of Language out of Pre-language. 2002. x, 394 pp.
52 Güldemann, Tom and Manfred von Roncador (eds.): Reported Discourse. A meeting ground for
different linguistic domains. 2002. xii, 425 pp.
51 Newman, John (ed.): The Linguistics of Sitting, Standing and Lying. 2002. xii, 409 pp.
50 Feigenbaum, Susanne and Dennis Kurzon (eds.): Prepositions in their Syntactic, Semantic and
Pragmatic Context. 2002. vi, 304 pp.
49 Wischer, Ilse and Gabriele Diewald (eds.): New Reflections on Grammaticalization. 2002.
xiv, 437 pp.
48 Shibatani, Masayoshi (ed.): The Grammar of Causation and Interpersonal Manipulation. 2002.
xviii, 551 pp.
47 Baron, Irène, Michael Herslund and Finn Sørensen (eds.): Dimensions of Possession. 2001.
vi, 337 pp.
46 Aikhenvald, Alexandra Y., R.M.W. Dixon and Masayuki Onishi (eds.): Non-canonical Marking of
Subjects and Objects. 2001. xii, 364 pp.
45 Bybee, Joan and Paul J. Hopper (eds.): Frequency and the Emergence of Linguistic Structure. 2001.
vii, 492 pp.
44 Voeltz, F. K. Erhard and Christa Kilian-Hatz (eds.): Ideophones. 2001. x, 436 pp.
43 Gildea, Spike (ed.): Reconstructing Grammar. Comparative Linguistics and Grammaticalization. 2000.
xiv, 269 pp.
42 Diessel, Holger: Demonstratives. Form, function and grammaticalization. 1999. xii, 205 pp.
41 Frajzyngier, Zygmunt and Traci S. Curl (eds.): Reciprocals. Forms and functions. Volume 2. 2000.
xii, 201 pp.