Delancey (1981)

An Interpretation of Split Ergativity and Related Patterns
Author(s): Scott DeLancey

Source: Language , Sep., 1981, Vol. 57, No. 3 (Sep., 1981), pp. 626-657
Published by: Linguistic Society of America
Stable URL: https://www.jstor.org/stable/414343
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
is collaborating with JSTOR to digitize, preserve and extend access to

Language
This content downloaded from

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
All use subject to https://about.jstor.org/terms
AN INTERPRETATION OF SPLIT ERGATIVITY AND RELATED
PATTERNS
SCOTT DELANCEY
University of Colorado
Nominative/absolutive case and verb agreement are, in many languages, indicators of
a category which is here called VIEWPOINT: the perspective from which the speaker
describes the event. The order of NP constituents in a sentence encodes ATTENTION
FLOW, which is the order in which the speaker expects the hearer to attend to them.
Split ergative case-marking patterns are shown to reflect conflicts between the most
natural viewpoint and attention-flow assignments. It is argued that the characterization
and grammatical marking of an event as first-hand or inferred knowledge for a speaker,
and as intentional or inadvertent for an actor, can be described in terms of whether the
entire event or only its terminal phase is directly accessible to the conscious mind of the
speaker and the actor, respectively; and that these categories can also be described in
terms of attention flow and viewpoint.*
The study of case-marking patterns in language has aroused considerable

interest over the last decade, in the wake of attempts to describe grammars in
terms of a set of underlying semantic case-roles, and later in terms of putatively
prime grammatical relations. Among the more recalcitrant, and hence more
interesting, case-marking patterns encountered in natural languages are those
in which the assignment of particular case-markers is partly determined by
some factor other than semantic or syntactic role, so that choice of case-
marking is partly independent of the semantic case-role or the grammatical
function of the NP which receives the marking. In this paper I will discuss
three general types of split case-marking patterns: the two so-called 'split
ergative' (SE) patterns, in which a transitive agent is marked for ergative case
or left unmarked depending on its position on the 'animacy' or 'empathy'
hierarchy, or on the tense/aspect of the clause; and the 'active/stative' split
pattern, in which the subject of an intransitive verb is marked like a transitive
agent or patient, depending on whether or not it engages in the act described
on its own volition. All three patterns are discussed and exemplified in Comrie
1978a and Dixon 1979.
It is my purpose here to provide a unified, semantically-based account of
these three patterns, and in doing so to introduce a hypothesis concerning the
organization of morpho-syntax which may eventually provide answers to a
number of problems besides those addressed here. In ? 1, I describe the patterns
which are to be explained. In ?2, I deal in a general way with the significance
of the morphological alternations observed in ?1. In ?3, I describe two notions,
attention flow and viewpoint, in terms of which the data will be explained. In
?4, I present an analysis of SE patterns governed by the Empathy Hierarchy
* The ideas presented in this paper were developed in the course of work done in collaboration
with LaRaw Maran and Lon Diehl; credit for any merit which the hypothesis advanced here may
possess is at least as much theirs as mine. The deficiencies of the paper are, of course, my own
responsibility. Part of this paper was presented at the 1979 Winter LSA Meeting under the title
'Viewpoint, attention flow, and subject-coding properties'.
626

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
AN INTERPRETATION OF SPLIT ERGATIVITY 627
(EH), and of similarly governed direct/inverse systems, in terms of th

pothesis presented in ?3. In ?5, I extend the analysis to the aspectually go
SE pattern, and in ?6 to the active/stative alternation.
SPLIT CASE-MARKING PATTERNS
1.1. THE SPLIT ERGATIVE PATTERN. An SE language is one in which some

transitive clauses, but not all, are ergative constructions. For the purposes of
this paper, we are concerned only with morphological manifestations of er-
gativity: I will define an ergative construction as a transitive clause in which
a special case-form or adposition marks the semantic agent,' or verb-agreement
is with patient in preference to agent. This definition is broad enough to class
agentive passive constructions as ergative, and hence voice alternations as SE
patterns. There are reasons for distinguishing voice from SE alternations (see
Comrie 1978a for some discussion of their similarities and differences); the
distinction most relevant to my argument here is that voice alternations typi-
cally reverse the order of the agent and patient NP's, while SE alternations do
not.2 For present purposes, however, the similarities between voice alterna-
tions and SE patterns are at least as relevant as the differences between them.
In particular, we find not only that voice and SE patterns are morphologically
congruent, but also that they are often constrained by the same semantic
factors. Thus both types of alternation will be adduced as evidence for the
hypothesis presented in ?4.
Two apparently disparate factors are attested as governing the distribution
of main-clause ergative and accusative constructions in SE languages: the
tense/aspect of the verb, and the person or semantic nature of the agent (or
of both agent and patient). Both factors, as we shall see, are also attested as
constraining voice alternations.3
1.2. THE EH-SPLIT PATTERN. The SE pattern which has received the most
recent attention is one governed by the variously-called hierarchy of 'animacy',
'agentivity', 'topicality', 'salience', or 'empathy', by which Ist and 2nd persons
(hereafter SPEECH-ACT PARTICIPANTS, or SAP's) outrank human 3rd persons,
' In much of this paper, I will use the term 'agent' quite loosely, to include roles (such as that
of a perceiver) which are not analysed as being agentive in most versions of Case Theory, but
which in many languages pattern with true transitive agents with respect to the morpho-syntactic
patterns discussed here. The theoretical status of this broad use of 'agent' will be discussed in
?3.3; agentivity in the strict sense will be discussed briefly in ?6.
2 In the light of the discussion in ??3-4, this distinction might be considered criterial for distin-
guishing voice from SE alternations; however, it doesn't correlate perfectly with the standard
criterion, by which voice alternations are those marked with a special form of the verb associated
with one voice alternant. (Cf. the Sinhalese 'passive' discussed in ?6.1.)
3The most common factor governing voice alternations, both active/passive and ergative/anti-
passive, is of course the discourse-based thematicity or 'communicative dynamism' of one or both
of the two NP's in the transitive sentence. This factor clearly is related to the factors discussed
in this paper. The interaction between the 'empathy' or 'animacy' hierarchy and topicality has
long been known; and recent studies (e.g. Hopper 1979, Hopper & Thompson 1980) have shown
an interaction between information structure and aspect. However, the questions implied here are
beyond the scope of this paper.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
628 LANGUAGE, VOLUME 57, NUMBER 3 (1981)
which outrank non-human animates, which in turn outrank inanimates. In t

EH-governed split, which is almost universal in Australian languages and wel
attested elsewhere (in North America, and in some Siberian and Tibeto-Burm
languages), there is a cut-off point somewhere along the EH; agents which
rank above that point are not marked for case, while those below it are mark
for ergative case. The most common pattern distinguishes SAP's from all oth
NP's, including 3rd person pronouns. This pattern occurs in Kham, a Tibeto
Burman language of Nepal (Watters 1973):
(1) nga: nan-lay nga-poh-ni-ke.
I you-OBJ IA-hit-2P-PERF 'I hit you.'
(2) nan nga-lay na-poh-na-ke.
you I-OBJ 2A-hit-IP-PERF 'You hit me.'
(3) nan no-lay na-poh-ke.
you he-oBJ 2A-hit-PERF 'You hit him.'
(4) no-e nan-lay poh-na-ke-o.
he-ERG you-OBJ hit-2P-PERF-3A 'He hit you.'
The morphemes glossed 1A, 2P etc. are subject and object agreement-markers;
they will be discussed in ?4.2. Case-marking on the agent NP clearly follows
an SE pattern: ergative case is marked on 3rd person agents, but not on SAP
agents. Note, however, that none of the sentences given here is morphologically
ergative according to the standard definition of an ergative construction as one
in which the patient (but not the agent) shares the same case-marking as an
intransitive subject. In Kham, intransitive subjects are not marked with -lay:4
(5) no: thala-tin zo:-ke.
he roof-ABL jump-PERF 'He jumped from the roof.'
For our purposes, and perhaps in general, a definition of 'ergative construction'
based solely on transitive agent-marking is more useful than the standard def-
inition in terms of identity of marking for patient and for intransitive subject.
Thus, of the Kham examples, 4 is an ergative construction, because the tran-
sitive A is marked for ergative case; but 1-3 are not.
1.3. THE ASPECTUAL SPLIT. The other SE pattern is the aspectual split, in
which ergative morphology is associated with perfective aspect or past tense,
and accusative morphology with imperfective aspect, or with present or future
tense. Attested in a few Australian, Austronesian, and Mayan languages, this
pattern also occurs in an area extending from North India to the Caucasus-
including a number of Indo-Iranian languages, several Caucasian languages,
a few Tibeto-Burman languages, and Burushaski, a language isolate spoken in
northern Pakistan. A typical case is Gujarati (Mistry 1976):
(6) ramesh pen kharid-t-o ha-t-o.
(masc.) (fem.) buy-IMPF-MASC AUX-IMPF-MASC
'Ramesh was buying the pen.'
4 Not all patients are marked with -lay; however, its distribution is not, as might appear, an
instance of the widely-reported pattern in which all and only patients high on the EH are mark
for accusative case. Rather, the -lay postposition is used generally with definite patients (Watte
1973:199-202).

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
(7) ramesh-e pen khdrid-y-i.

(masc.)-ERG buy-PERF-FEM
'Ramesh bought the pen.
In 6, with its imperfective aspect, the agent NP is not marked for case, and
governs verb-agreement; but in 7, which is perfective, the agent is marked for
ergative case, and agreement is with the feminine patient.
This pattern is reminiscent of the well-known morphological syncretism of
perfect and passive in Western European languages, where the same form of
the verb serves as both 'past' and 'passive' participle. A parallel can also be
drawn to the restriction on the co-occurence of passive and progressive in
Early Modern English, where sentences like 8-9, but not 10 were possible:
(8) He is building (of) a house.
(9) The house has been built.
(10) *The house is being built.
Here the distribution of voice alternants is constrained by an aspectual dis-
tinction similar to that which determines the distribution of ergative and ac-
cusative patterns in Gujarati.
1.4. THE ACTIVE/STATIVE SPLIT. A third type of split case-marking pattern,

sometimes rather misleadingly referred to as SE,5 is that found in 'active'-type
languages (Klimov 1973, 1974, Comrie 1976a). In SE languages, the A NP is
marked like the intransitive subject in some constructions, but differently in
others. In 'active/stative' languages, however, it is the case-marking of the
intransitive subject which varies, agreeing sometimes with the transitive agent,
sometimes with the patient. An example is the Northeast Caucasian language
Batsbi (Deseriev 1953; cf. Comrie 1978a, whose glosses I use here):
(11) txo naizdrax kxitra.
we (ABS) to-ground fell
'We fell to the ground (unintentionally, not our fault).'
(12) a-txo naizdrax kxitra.
ERG-we to-ground fell
'We fell to the ground (intentionally, through our own carelessness)
Here the intransitive subject is marked like a transitive agent if the ev
occurred as a result of the subject's action or inaction, but like a patient if
causes of the event were entirely external to the subject. We shall see that
pattern is better discussed in connection with the inferential interpretation
perfect aspect than with the true SE patterns.
ON THE FUNCTIONS OF MORPHOLOGY
2.1. THE SIGNIFICANCE OF SPLIT ERGATIVITY. The patterns described ab

are of particular interest because they are 'universal' in the weak sense th
while they do not occur in all languages, only these patterns occur in
5 By the standard definition of ergative and accusative patterns, which depends on whe
transitive agent or patient is marked like the intransitive subject, these languages are SE in a
of backward way; but they clearly represent a distinct type. A problem with the standard defini
is that it assumes a fixed marking for intransitive subject, with which either agent or patient mar
may then be identical.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
630 LANGUAGE. VOLUME 57, NUMBER 3 (1981)
language. Inverse patterns of EH or aspectual split, with ergative morphology

associated with agents high on the EH, or with imperfective as opposed to
perfective aspect, apparently do not exist. Thus general linguistic theory must
account for the association of ergative morphology with, on the one hand,
perfective aspect, and, on the other, lower position of the agent on the EH;
and for the association of accusative morphology with imperfective aspect and
high EH (especially SAP) agents.
2.2. CASE-MARKING. From the way that case-marking patterns vary with
grammatical distinctions not obviously relevant to semantic role, it is apparent
that the presence or absence of case-marking, as opposed to its content, is
significant in these languages. The presence or absence of ergative case with
a transitive agent (in languages where it occurs) has nothing to do with semantic
agentivity.6 It is, however, related somehow to aspect and person. This situ-
ation contrasts with that in a language where transitive agent is marked con-
sistently in all constructions. Thus, in a consistently ergative language, ergative
case-marking contrasts only with case-markers which indicate some semantic
role other than agent; and the only content of the ergative-marker is the notion
of agency. In an SE language, or a language with a voice alternation, transitive
agent is sometimes marked with ergative case (or some marker of agentivity
such as Eng. by), and sometimes not. In such a language, ergative case-marking
contrasts not only with case-markers indicating other semantic roles, but also
with the zero-marking which occurs with some agents. My claim here is the
traditional one that these represent two distinct contrasts, and thus that ergative
case in an SE language (or agent-marking in a passive construction) carries
two levels of information: the fact that it is present (i.e. its contrast with zero)
carries one message, and its identity (i.e. ergative as opposed to accusative or
some other marking) carries another. This implies the traditional idea that the
maximally unmarked nominative or absolutive case has a special status as
opposed to other cases-and that, as Trubetzkoy 1939 suggests, 'nominative'
and 'absolutive' are in some sense the same category. (As regards the discus-
sion immediately below, note that it is most often nominative or absolutive
NP's, rather than agent or patient NP's per se, which govern verb-agreement.)
2.3. VERB-AGREEMENT. A similar argument applies to verb-agreement.

There are languages in which agreement is tied to one case-role-e.g. Nepali,
where case-marking follows a complicated SE pattern, sensitive to both aspect
and the EH (see Verma 1976b for further details); but agreement is always with
agent in a transitive clause.7 Examples are from Bandhu 1973:
6 Even in the loose sense of fn. 1.

7 Nepali cannot easily be described as requiring agreement to be with subject-since, in dative
subject constructions, agreement is NOT with subject (example from Abadie 1974):
ma-lai hachuw a-io.
I-ACC sneeze come-PAsT/3rd 'I sneezed.'
This shows that, in terms of the analysis to be presented below, Nepali agreement is with semantic
Source rather than with starting-point. This seems to be a fairly unusual pattern.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
(13) md hdri-lai kitap din-chu.

I Hari-oBJ book give-NONPAST/ist
'I give Hari a book.'
(14) mai-le hdri-lai kitap di-e.
I-ERG Hari-oBJ book give-PAST/Ist
'I gave Hari a book.'
More commonly, in a language which allows agreement with only o
a clause, it may, in different constructions, be with agent or patien
with dative (as in English) or benefactive Goal.
In some languages, case-role is completely overruled by the EH
mining agreement. In English and similar languages, agreement is w
all other things being equal (i.e., active voice is the unmarked voice a
We find a very different pattern in a number of Tibeto-Burman l
where agreement must always be with an SAP in preference to a 3r
regardless of their respective semantic roles (DeLancey 1980).8
example is Tangut, an extinct Tibeto-Burman language, which h
person agreement at all (Kepping 1980; cf. Comrie 1978b):9
(15) a. ni tin nga In Idia thi-nga,
you if I ACC indeed chase-ist
'If indeed you are chasing me,'
b. ku tha tsi vid-thi-na.
then her also chase-2nd
'then chase her too.'
(16) ni pha ngi-mbin ndi-siei-na.
you other wife choose-2nd
'You choose another wife!'
(17) mei-swen mana na khe-na.
formerly you hate-2nd
'Mei-swen formerly hated you.'
Here 15a shows that if both agent and patient are SAP's, agreement is
patient; while 15b and 16-17 show that, otherwise, person rather t
determines agreement. (There is no agreement when both agent and p
are 3rd person.) The interaction of person and role in determining ag
is clearly seen in the contrast between patient agreement in 15a, and
agreement in 15b.
Similarly in Gujarati (exx. 6-7), though agreement is preferentially wi
(Mistry 1976; cf. DeLancey 1979), it correlates more with case-markin
with case-role; agreement is always with an unmarked NP if one is pr
X Dixon (1979:90) suggests that this is an unlikely situation, and it does seem to be rat
Comparing this pattern with the data presented in Giv6n 1976 we find another point o
between viewpoint and topicalization phenomena.
9 It is, of course, quite common for 3rd person 'agreement' to be 'realized' by zer
the verb to carry an agreement-marker when subject or agent is an SAP, but not when
person. It is debatable whether we should speak of 3rd person agreement in such a lang
Tangut, at any rate, where person rather than role or grammatical function is the prim
minant of agreement-marking, it is clear that the category of agreement applies only t

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
Both case-marking and verb-agreement are, as we have seen, primarily de-

pendent on aspect."' Once again, semantic role alone doesn't determine agree
ment-which is to say that agreement, like case-marking, reflects more tha
case-role. Gujarati (like English) is typical of a large number of languages in
which agreement and lack of case-marking coincide;" the pattern typified
Nepali, in which they vary independently, seems less common. This sugges
that, in many languages, agreement and nominative/absolutive case serve t
same or similar functions, and that their common function is specifically co
nected with aspect and the EH.
VIEWPOINT AND ATTENTION FLOW
3.1. The explanation which I propose for the patterns described above is
based on two ultimately psychological notions, ATTENTION FLOW (hereafter AF)
and VIEWPOINT. These notions are invoked within a view of semantics which
takes a significant part of the semantic structure of a language to be a list o
prototype scenes, each specified for a canonical set of participants (cf. Fillmor
1977a,b). A sentence describes a real or imagined event by invoking th
prototype scene of which it counts as an instance, and by identifying th
participant roles in the prototype with entities which exist in the universe of
discourse. In actual communication, not all aspects of the prototype event ar
of equal interest, and all languages have mechanisms for marking the relativ
communicative importance of the various entities and events in a sentence o
discourse. Viewpoint and AF are fundamentally parameters which contribute
to determining the relative interest of various entities involved in an actual
witnessed event; but the terms are also applicable to linguistic mechanisms
which indicate values for these parameters in a sentence, thus allowing it to
be interpreted in a manner analogous to that of an actual event. I will distinguish
these two uses of the terms, where necessary, as NATURAL VS. LINGUISTIC A
and viewpoint-the former referring to perceptual strategies, the latter to lin
guistic mechanisms. As we will see, case-marking, verb-agreement and voice
marking, and constituent order (the mechanisms involved in SE, voice, and
other alternation patterns) are the chief markers of linguistic viewpoint and
AF.
3.2. ATTENTION FLOW determines the linear order of NP's. The NP's in a
sentence are presented in the order in which the speaker wishes the hearer to
attend to them. Alternate NP orderings, as found in voice alternations and
topicalizing shifts, are mechanisms for managing AF.
Events have an inherent natural AF, which recreates the flow of attention
involved in actually witnessing the event. The basis of this natural AF is the
'o Mistry argues that agreement is dependent on case-marking; but without an explanatory
account of the case-marking pattern, this cannot be considered an explanation for either. Mistry's
evidence suggests that, in fact, case-marking and agreement are independently governed by aspect,
rather than one being contingent on the other (cf. fn. 11, below).
" Gujarati has a 'split accusative' case-marking pattern (cf. fn. 4), with the result that, in perfect
clauses with animate patients, both agent and patient are marked for case. In such clauses, agree-
ment is still with patient, which suggests that agreement and case-marking are independent of one
another, though subject to some of the same governing factors (in particular, aspect).

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
temporal ordering of phases of the event; other things being equal, th

of NP's in a sentence will reflect this temporal ordering. Compare the
(18) I drove from Bloomington to Philadelphia.
(19) 1 drove to Philadelphia from Bloomington.
Here 19 is a perfectly good English sentence, but it is clearly less nat
18, i.e., it requires more context to make it a likely utterance. To sa
'requires more context' is to say that the speaker must have a specia
to make Philadelphia the starting-point of linguistic AF, in order to o
the natural tendency to place it in a position reflecting its place in th
structure of the event. In other words, special motivation is require
chooses a sentence pattern in which linguistic AF does not recapitulat
AF.
In fact, Philadelphia cannot be the actual linguistic starting-point in a sen-
tence describing this event. In both 18 and 19, the starting-point of linguistic
AF is neither the Source nor the Goal, but rather the subject, which is se-
mantically (among other things) the Theme.'2 In English, as in many other
subject-forming languages,'3 subject is ordinarily the linguistic starting-point
(and this fact constitutes part of the definition of subject-cf. MacWhinney
1977); but it needn't be. Consider the following:'4
(20) From Bloomington I drove to Philadelphia.
(21) *To Philadelphia I drove from Bloomington.
Of the two locational prepositional phrases used in describing the event, the
one which represents the semantic Source can be fronted and made the starting
point of linguistic AF; but the one which represents the Goal cannot. This is
because the Source represents the actual starting-point of the event, and is
thus a natural candidate for linguistic starting-point.
3.3. ATTENTION FLOW IN TRANSITIVE SENTENCES. Like motion events, dative

and transitive events define space/time vectors; they prototypically are events
which begin at one point in space, and subsequently terminate at another. Just
as unmarked linguistic AF in a sentence describing a motion event is iconic,
following natural AF from Source to Goal, so unmarked linguistic AF in a
dative sentence is from giver to receiver, and in a transitive sentence is from
agent to patient."5 Thus it is not an arbitrary fact that SO is the most common
12 The terms Source, Goal, and Theme are taken from the work of Gruber (e.g. 1976): in a
motion event, they refer to the onset point, the terminal point, and the moving entity, respectively.
In other words, the prototype motion event involves a Theme which moves from Source to Goal.
'3 The term 'subject-forming language' is from Anderson 1979; like Li & Thompson's (1976)
'subject-prominent language' and Hale & Watters' (1973) 'subject-object language', this term im-
plies the claim that 'subject' is not a relevant category in all languages.
14 I owe these examples to Lon Diehl.
'5 There is considerable evidence for considering that Agent and giver are subcategories of a
fundamental case-category which also includes Source, and that Patient and receiver constitute
a single category with Goal (see Anderson 1971, 1977, Diehl 1975, Fillmore 1977a). I will use
Source and Goal to refer to these broader categories (as in fn. 7), as well as in their narrower
spatial sense.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
word-order type in human languages.16 There is some evidence (cited and

discussed by MacWhinney, MS, who points out that the evidence is far from
conclusive) that young children tend to interpret the first NP in a transitive
sentence as agent, disregarding morpho-syntactic cues to the contrary; this
suggests that children expect linguistic structure to recapitulate perceptual
structure.
Further evidence that unmarked linguistic AF is that which recapitulat

natural AF can be found in the study of voice alternations. Voice alternati
like the order alternations in 18-20, are mechanisms for managing AF. C
linguistic studies of voice alternations show that unnatural AF is highly ma
This is reflected in the fact that, although passives in which the agent is m
tioned are impossible in some languages, we find no languages which perm
agentive passives but exclude agentless ones (Eckman 1974). Even in langu
like English, which allow agentive passives, they seem much less frequent
the agentless type (Giv6n 1979a, esp. 59-60). Moreover, there is evidence
in English, at least, agentless passives are acquired well before agentive p
sives (cf. Maratsos 1978). All this suggests that passives in which the agen
mentioned are considerably less natural than agentless passives. This is co
sistent with the AF Hypothesis, since an agentive passive construction rev
natural AF-but if the agent is not mentioned, then the passive sentence
sents only one end of the event, and there is no unnatural patient-to-agent
In most languages which have an ergative or agent case-marker, it does
indicate agentivity in the strict sense of the term (referring to the self-contro
deliberate initiation of an action), but rather the starting-point of natural
In a transitive event, this is by definition the first entity in the scene to m
Consider this sentence:
(22) The driver's chest was crushed by the steering column.
Here (or in the active counterpart) the steering column counts as agent for all
morpho-syntactic purposes. This explains the extremely common syncretism
of ergative or passive agent case with instrumental; the case-form does not
refer to agentivity (on which supposition its use for non-agentive instruments
is anomalous), but rather to activity in the initial phase of the event, which
notion is equally applicable to agents and instruments. Similarly explained is
the common syncretism of ergative/agent case with ablative (see, e.g., An-
derson 1971); both cases mark natural starting-points.
If no moving entity is involved, e.g. in events of perception, natural AF will
take as starting-point the most salient object. Salience correlates with position
on the EH; hence the perceiver, which must be animate and is prototypically
human, is selected over the perceived as natural starting-point. Thus in Che-
pang, a Tibeto-Burman language of Nepal, perceiver is always in ergative case,
and agentivity on the part of the perceiver is marked by dative/accusative case
on the perceived NP (Caughley 1978):
(23) ngaa- i waa saay -naa-ng.
I-ERG bird hear-PRES- st
'I hear a bird.' (non-agentive)
16 I have no explanation for the fact that SO is not the universal pattern.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
(24) ngaa- Pi waa 2-kaay ? saay 2-naa-ng.

I-ERG bird-DAT hear-PRES- 1st
'I listen to a bird.' (agentive)
Clearly, ergative case in Chepang does not reflect agentivity, but rather natural
starting-point. Hence Case Theory has no particular problem in languages like
Tibetan, where some dialects (e.g. Lhasa) mark perceivers with ergative case,
while others (e.g. Balti) mark them as dative. The two dialects are not assuming
radically different semantic analyses of the prototypical perceptual event;
rather, Balti marks perceivers for their case-role, i.e. as Goal (as would be
predicted by the analysis of perception in, e.g., Fillmore 1968, Anderson 1971,
or Gruber 1967), while Lhasa assigns case-marking on the basis not of semantic
role but of natural AF.
3.4. VIEWPOINT. Since linguistic AF is strongly tied to the semantic (and

cognitive) structure of the event being described, while non-natural AF is (at
least in many languages) highly marked, we must explain what motivates the
existence of mechanisms for reversing it. Many if not all of the relevant
motivating factors are included in the category of viewpoint.'7 We are accus-
tomed to think of a sentence as describing an event from an external, objective
point of view;'8 but this is not always true. Novelists, film-makers, and lan-
guage-users are all aware that a scene may be described from a number of
points of view. Prototype scenes have, at most, three participant roles which
can be filled by human actors-and ordinarily no more than two.'9 (Note how
difficult it is to construct a simple sentence with even three human participants.
The only real productive possibilities are causative and benefactive construc-
tions; and it is significant that, in many languages, one or both of these can be
expressed only in syntactically complex sentences.) Thus there are, a priori,
several possible viewpoints from which such a scene can be described: the
external viewpoint of a disinterested observer, and a viewpoint associated with
each participant. Recent work (e.g. Fillmore 1977a,b, Kuno & Kaburaki 1977,
DeLancey 1978, Maran 1978, Zubin 1979) has shown that languages allow-
or require-a speaker to specify which of these viewpoints he is taking in
reporting an event, and that grammatical and lexical mechanisms exist, pre-
sumably in all languages, for specifying the viewpoint of a sentence.
The notion of viewpoint is most easily exemplified with simple motion events.
When we speak of movement from one point to another, we generally specify
whether or not it is toward the location of the speaker and hearer-i.e. whether
the viewpoint from which the motion is described is the terminal point of the
motion. If it is, an English speaker uses the verb come rather than the unmarked
17 As suggested in fn. 3, there are grounds for supposing that the complex of factors called
topicality or thematicity may be analysable in terms of the viewpoint category.
l' This is perhaps in part because linguists have become accustomed to analysing artificial
sentences having no pragmatic connection to any actual event or discourse context. It is interesting
that linguists' examples very seldom involve 1st or 2nd person participants.
'9 Cf. the discussion by Tesniere 1959 of valence, and of causativization as an increase in
valence-in which it is pointed out that all three-place verbs can be considered, at least seman-
tically, as causative versions of two-place verbs; thus give is equivalent to cause to have.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
alternative go. This is the most concrete possible instance of viewpoint spec-
ification: the speaker describes an event from a particular location, which is
his actual location in space. Of course, Eng. come and its equivalents in many
other languages are not limited to this concrete use as markers of natural
viewpoint. For various social and narrative purposes (cf. Fillmore 1966, 1975),
a speaker will often take some contextually recoverable location other than
his own as his linguistic viewpoint, and mark a motion event as oriented toward
that point by the use of come.
In English and many other languages, the marking of viewpoint for simple
motion events is entirely lexical. Motion verbs which are semantically more
complex than come/go and bring/take are not specified for viewpoint orien-
tation, though many of them have lexical alternatives, built on go and come,
which are used if viewpoint specification is necessary. It is worth noting that
these specified alternates, e.g. go/come back or go/come out, are much more
colloquial than their neutral counterparts, return and exit; this reflects their
much greater frequency in actual pragmatically constrained speech. Some lan-
guages use morphological or syntactic, rather than lexical, mechanisms for this
function.20 For example, in Jinghpaw, a Tibeto-Burman language of Yunnan
and northern Burma, the deictic orientation of sentences like 25-26 is provided
not by the motion-verb sequence sa wa, which is identical in the two examples
but by the r- and n- morphemes which precede the agreement-markers:
(25) MaGam gat de sa wa n-u 2 ai.
market to go n-3rd IND
'MaGam {is going/has gone} to market.'
(26) MaGam gat de sa wa r-u ? ai.
market to go r-3rd IND
'MaGam {is coming/will come} to market.'
Here the r- morpheme specifies terminal viewpoint; i.e., the speaker's view-
point is the terminal point of the motion. The n- morpheme, like Eng. go,
permits either onset or external viewpoint interpretation. (This and my other
Jinghpaw examples are discussed at greater length in DeLancey 1978, 1980;
see also Maran 1978, 1979.)
Viewpoint considerations can provide the motivation for alternative order-
ings of spatial Source and Goal, as in 18-19. One possible motivation for the
Goal-first order in to Philadelphiafrom Bloomington would be that Philadelphia
was the location of the speech act. This distribution of sentence alternants with
respect to speech situations is parallel to that of go and come. The more marked
lexical alternant come is used when the actual location of the speech act, or
the contextually established viewpoint, is the Goal of the motion event; the
same conditions can motivate the more marked Goal-Source order of con-
stituents. Such a use in 19 represents the sacrifice of natural AF to natural
viewpoint.
20 A few languages have no specific mechanism for indicating viewpoint with verbs of motion;
Russian is a well-known example.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
3.5. VIEWPOINT IN TRANSITIVE SENTENCES. The notion of viewpoint

can be profitably extended from concrete motion to dative and t
events. Some languages provide strong evidence for such an exte
some dialects of Jinghpaw, for example, the n- and r- morphemes of
25-26 alternate with (among others) two morphemes, d- and m-, whic
a less concrete subcategory of viewpoint:
(27) nang shi-hpe ndaijaw d-it dai.
you he-DAT this give d-2nd IND
'You gave him this.'
(28) nang shi-hpe ndaijaw m-uV ai.
you he-DAT this give m-3rd IND
'He was given this by you.'
This suggests that dative and motion viewpoint may be varieties of a single
category. Similarly in Sizang Chin (Stern 1963) and Tiddim Chin (Henderson
1965), two closely related Tibeto-Burman languages of western Burma, a mor-
pheme -(h)ong-,2' between the prefixed subject agreement-marker and the verb,
indicates that a motion event is hither-directed-i.e. that it is described from
the terminal point of the motion vector (examples from Stern):
(29) kei ka-pai hi.
I Ist-go IND 'I go/went.'
(30) kei k-ong-pai hi.
I lst-ong-go IND 'I come/came.'
This morpheme can also specify the orientation of a dative or benefactive
sentence:
(31) hong-pe-tu hi.

give-FUT IND
'(She/He) will give it to me.'
(Third person subject is 'marked' by zero.) Here the recipient of the gift is not
explicitly mentioned, but is recoverable from the (h)ong-prefix, which specifies
that the event is being described from the viewpoint of the Goal, i.e. the
recipient, who must therefore be the speaker who is describing the event. (Cf.
Eng. Give it here!, which is likewise unambiguous as to the identity of the
intended recipient.)
The parallelism between motion and dative sentences is hardly surprising,
in view of the obvious and long-noted grammatical and semantic parallelism
of the two types. The dative scene is, after all, prototypically a motion event,
in which an object physically moves from one location to another. Even when
no concrete motion is involved, the identity of marking of giver and Source,
and of receiver and Goal, is widespread:
(32) I got my rotten temperament from my grandfather.
(33) He sold his house to the University.
21 This is etymologically a grammaticalized verb 'come'. F. K. Lehman (p.c.) points out that
ex. 31 can also mean 'He will come and give it.'

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
638 LANGUAGE. VOLUME 57, NUMBER 3 (1981)
The extension of such a localist analysis to transitivity is somewhat m

controversial, but much the same sort of evidence can be adduced for it.
particular, the syncretism of ablative with ergative, and of accusative wi
dative and/or allative, is extremely frequent (see Anderson 1971, Diehl 19
For our purposes it is sufficient to note similarity of behavior with respec
viewpoint-switching. For example, compare the following:
(34) a. She gave me a check.
b. I got a check from her.
(35) a. She kicked me. (cf. She gave me a kick.)
b. I was/got kicked by her.
Here the (b) sentences differ from the (a) alternants in two significant ways:
Goal rather than Source has been selected as subject,22 and the Source (giver
in the dative sentences, and agent in the transitive) is prepositionally marked
to indicate its role.23 In English dative and transitive sentences, one NP is
ordinarily selected to be both linguistic viewpoint and linguistic starting-point;
this double selection is the basis of the category of subject. The (b) sentences,
by selecting Goal as viewpoint-and thus starting-point-reverse natural AF;
therefore Source must be marked for its role.
There is an important difference between the interpretations of viewpoint
in these examples vs. the simple motion cases of ?3.4. In the motion examples,
viewpoint has an extremely literal interpretation: the viewpoint is an actual
point, a spatial location. The prototypical viewpoint is the location of the
speech act, and motion events are specified as being directed (or not) toward
that locus-or a contextually established substitute. In the dative and transitive
examples, however, viewpoint is associated not with a location as such, but
with a participant (although typically the location of each participant does in
fact mark an endpoint of an actual spatial vector defined by the event). View-
point is still a fundamentally deictic notion, however; and if an SAP is also a
participant in the event being reported, then the most natural viewpoint for
the sentence is with the SAP. This is the explanation for the difference in
naturalness between passive sentences such as these:24
(36) I was flunked by Prof. Summers.
(37) Mary Summers was flunked by me.
The reason for forsaking natural AF in a passive is to place the viewpoint NP
first, i.e. to make it the starting-point of linguistic AF.25 But I, as an actor in
22 Goal and Source are here used in the extended sense of fn. 15.
23 The vagueness of sentences like 34b, with regard to whether the Source was actually a willing
giver, is irrelevant here-though it is relevant to a complete discussion of the choice of give or
get as natural or inverted AF. I suspect that get sentences in which Source is not mentioned are,
like the analogous agentless passives, far more common than sentences like 34b in which AF is
aztually reversed.
24 This has sometimes been claimed to be a difference in grammaticality, but it certainly is not
(see Kato 1979).
25 The interaction between these and other factors in determining linguistic starting point in
English is discussed in illuminating detail in MacWhinney 1977.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
an event, am a much more natural locus of viewpoint in a sentence wh

utter than any 3rd person. The difference in inherent eligibility for v
status is in itself sufficient to motivate the choice of 36 in preferenc
active alternant, but it makes 37 a highly unlikely choice (though not im
at least in English).
We thus have the high end of the EH, i.e. the split between SAP's an
other NP's, reflected in what we might, in current terminology, call
availability for demotion'. This distinction between SAP's and other N
simply an instance of the fundamentally speech-act-centered nature o
point-the distinction between here, where you and I are, and ever
else, where everybody else is. (The deictic nature of the distinction be
SAP's and other NP's has long been noted; cf. Kurylowicz 1964, and esp
Benveniste 1946, 1956). This is exactly the split reflected by Sizang/Ti
(h)ong-, discussed above (exx. 29-31). This prefix can indicate orient
an event toward either SAP, as in the Tiddim sentence (Henderson):
(38) k-ong-mat-sak hi.
st-ong-catch-BENEF IND
'I've caught it for you.'
Here again the unspecified Goal is unambiguous, since ong- requires either 1st
or 2nd person Goal, while k- identifies Ist person as subject and hence Source.
THE EH-SPLIT PATTERN
4.1. SUBJECT-CODING PROPERTIES. As noted in the previous section, most

English sentences select the same NP as both the viewpoint and the starting-
point of AF. This NP is then the leftmost NP, the NP unmarked for case, and
the NP which governs verb-agreement. Leftmost position is, by definition, a
property of the starting-point of linguistic AF. Nominative/absolutive case and
(in many languages) verb-agreement are properties of the viewpoint NP (cf.
Zubin). Thus, if there are languages in which viewpoint and starting-point need
not coincide, we would expect that these three 'subject' properties might be
distributed between two different NP's.26 This is precisely what we find in the
SE languages, where agreement and nominative/absolutive case are sometimes
associated with the leftmost NP, depending on whether natural viewpoint and
natural starting-point are the same NP.
4.2. THE SAP/3RD PERSON SPLIT. Consider again Kham sentences 1-4:
(1)' nga: nan-lay nga-poh-ni-ke.
I yOU-OBJ IA-hit-2P-PERF 'I hit you.'
(2) nan nga-lay na-poh-na-ke.
you I-OBJ 2A-hit-IP-PERF 'You hit me.
(3) nan no-lay na-poh-ke.
you he-OBi 2A-hit-PERF 'You hit him.
(4) no-e nan-lay poh-na-ke-o.
he-ERG you-OBJ hit-2P-PERF-3A 'He hit you.
In all these examples, the leftmost NP-the starting-point of linguistic AF-
26 A split may also occur when, for example, agreement codes neither viewpoint nor starting-
point, but a deep case-role, as in Nepali (?2.3).

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
is also the agent, i.e. the natural starting-point. When it is also an SAP, i.e.
a natural viewpoint locus, it is so marked by being in the nominative ca
Otherwise, it must be marked for ergative case, which identifies it as the natural
starting-point.
Since the Kham verb agrees with two NP's, agreement cannot be associate
with viewpoint in the same straightforward manner as in English or Gujar
(cf. ?6.1). Nevertheless the Kham agreement pattern reflects natural viewpoi
and in a particularly interesting way. SAP patient-markers (glossed 1P, 2P) a
suffixed directly to the verb root; there is no 3rd person patient-marker. S
agent-markers (IA, 2A) are prefixed (as in exx. 1-3). The 3rd person agent-
marker, however, occurs to the right of all other suffixes (including the pati
suffixes), as in 4. Thus unmarked constituent order places natural starting
point first,27 regardless of natural viewpoint; but the order of agreement affixes
places natural viewpoint first, regardless of natural AF. Note that this analy
requires that 1st and 2nd persons count as equally natural viewpoint loci, a
no ranking is reflected in affix order. This is what is predicted by the Viewpoint
Hypothesis, since both SAP's are prototypically located at the deictic cente
of the speech act.28 Note that no ranking of the SAP's is indicated by the ca
marking pattern either; exx. 1-2 show that either SAP, when it is the natu
starting-point, is marked as natural viewpoint (by lack of case-marking), eve
when the other SAP is present. The same pattern is reflected in the distribut
of Sizang/Tiddim (h)ong-, which can occur whenever there is an SAP Go
even when Source is the other SAP. These facts are among the data indicati
that the most natural situation is one in which viewpoint and starting-poin
coincide.
The explanation given above for the Kham affixation pattern receives con-
firmation from an alteration of the pattern which occurs in the passive con-
struction. Here constituent order is reversed,29 and the verb is marked with
an -o passive suffix (examples from Watters 1973):
(39) nga: ao zihmld nga-li-ke.
I this house lA-stay-PERF
'I stayed in this house.'
(40) ao zihmld nga: nga-li-o.
this house I lA-stay-PAss
'This house was lived in by me.'
27 The critical reader will have noted already that my remarks on English in ?4.1 do not consider
topicalization phenomena such as those termed 'secondary' by Fillmore 1968 or 'sentence-level'
by Foley & Van Valin 1977. In sentences like Him I won't listen to, the leftmost NP is marked
for case and does not control agreement. Similar exceptional instances occur in Kham and in most,
if not all, of the other languages discussed in this paper. The relationship of such phenomena to
the primary topicalization phenomena (e.g. passivization) in terms of the framework proposed here
remains to be elucidated (see fn. 3).
28 There are languages which rank the SAP's in one or the other order (Silverstein 1976, Dixon
1979, and see ?4.3 below). I assume such rankings to be arbitrary.
29 The passive verb form also has several uses in which natural constituent order is retained
(see Watters 1973, 1975).

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
In this construction, the 3rd person agent-marker occupies the same pr

slot as the SAP agent-markers:
(41) ao be:h ram-e o-jay-o.
this basket Ram-ERG 3A-make-PAss
'This basket was made by Ram.'
Thus in all passive sentences, affix order is agent-patient, regardless of person.
This can be interpreted as compensating for the unnatural AF in constituent
order: when linguistic AF is patient-agent, affix order then encodes the natural
AF which is not recoverable from constituent order, rather than encoding
natural viewpoint as it does in active sentences.
4.3. DIRECTION-MARKING AND SPLIT ERGATIVITY. A number of languages use

a different mechanism for marking the identity or non-identity of natural view-
point and natural starting-point-coding it directly on the verb, rather than
marking on NP's. This mechanism is best known from the Algonquian lan-
guages, but is also attested in Australia (Heath 1976), in Nootkan (Whistler
1980), and in Tibeto-Burman (DeLancey 1980). The structure of these direction-
marking systems is perfectly congruent to that of an EH-split system like
Kham, as Dixon 1979 and Whistler 1980 have pointed out. The essential feature
of a direction-marking system is that the verb in a transitive sentence is mor-
phologically marked when P is an SAP and A is not. This is called the inverse
configuration. Some direction-marking languages also mark the direct config-
uration, in which A is an SAP and P is not. Some also have distinct marking
for 1st person A -> 2nd person P, or for 2nd A - 1st P, or both; others class
one of these as direct, the other as inverse.
An example of a fairly simple direction-marking system is Nocte (or Nam-
sangia), a Tibeto-Burman language of easternmost Assam. Nocte marks agree-
ment on the verb in a pattern very similar to that of Tangut (?2.3). Agreement
is always with an SAP in preference to a 3rd person; unlike Tangut, Nocte
does have a 3rd person agreement-marker which occurs when no SAP is in-
volved. Like Tangut, Nocte marks agreement with 1st person patient when
agent is 2nd person; however, with 1st person agent and 2nd person patient,
agreement is for 1st person plural, thus avoiding commitment to either of the
more or less equal-ranking SAP's. This distribution suggests that Nocte ranks
1st person slightly above 2nd, and this supposition is borne out by the distri-
bution of the inverse marker. This is an h suffix placed between the verb and
the agreement suffix in the 3A-2P, 3A--IP, and 2A--P configurations (data
from Das Gupta 1971):
(42) nga-ma nang hetho-e.
I-ERG you teach-lpl. 'I will teach you.'
(43) nang-ma nga hetho-h-ang.
you-ERG I teach-INv-lst 'You will teach me.'
(44) nga-ma ate hetho-ang.
I-ERG he teach-lst 'I will teach him.'
(45) ate-ma nga-nang hetho-h-ang.
he-ERG I-ACC teach-lNv-lst 'He will teach me.'

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
Given a language-specific ranking of Ist>2nd>3rd person, the distribution o

the h inverse suffix can be described in exactly the same terms as that
ergative case in Kham: when natural viewpoint and natural starting-point agr
there is no mark; when they conflict, the verb is marked with h.30 In a 3A->
configuration, there is no conflict, as neither participant is a more natural
viewpoint locus than the other; hence the inverse marker does not occur.3'
The facts concerning the distribution of case-marking are far from clear. Se
tences 42-45, which were obviously elicited as a paradigm and hence mu
count as artificial examples, indicate consistent ergative marking, with an
cusative split in which SAP's are marked as accusative when A is 3rd person
In fact, neither pattern holds up consistently throughout Das Gupta's data;
particular, ergative marking is quite inconsistent, though its distribution su
gests a typical SAP>3rd split complicated by some other unidentifiable fact
The existence of both regular EH-split marking and direction-marking in a
single language would seem to be redundant, and I know of only one cl
example. Jyarong, a Tibeto-Burman language of Szechwan, as described
Jin et al. 1958, manifests both patterns;32 and they reflect exactly the sam
hierarchy in which 1st person slightly outranks 2nd, while both strongly outrank
3rd:
(46) no-ks nga kd-u-nasno-ng.

you-ERG I T-INv-scold-lst 'You will scold me.'
(47) md-kd nga u-nasno-ng.
he-ERG I INv-scold-lst 'He will scold me.'
(48) nga ma nasno-ng.
I he scold-lst 'I will scold him.'
(49) nga no td-a-nasno-n.
I you T-A-scold-2nd 'I will scold you.'
(50) ma-kd no td-u-nasno-n.
he-ERG you T-INv-scold-2nd 'He will scold you.'
Note that the distribution of the inverse prefix u- and the ergative postposition
ka is exactly the same;33 both occur when and only when the more natural
30 In the perfective and negative paradigms, Nocte shows traces of an earlier direct-making
suffix which came after the agreement-marker; unfortunately the traces are not sufficient to indicate
its distribution (in particular, there is no evidence as to whether it marked the lst->2nd or the
3rd-*3rd configurations). Available details are given in DeLancey 1980.
31 In some North American languages, 3rd->3rd configurations are marked. In the Algonquian
languages, proximate A --- obviative P counts as a direct configuration, and obviative A -> prox-
imate P as inverse. Whistler describes a direction-marking system in Nootka which seems to be
sensitive to discourse-based thematicity when two 3rd person participants are involved; this pro-
vides further evidence for the relevance of the Viewpoint Hypothesis to the study of topicality.
32 The data as presented in Jin et al. are under-analysed; the analysis presented here is justified
in DeLancey 1980.
33 It is not exactly the same when dual and plural NP's are involved; according to the paradigms
presented by Jin et al., the distribution of the ergative marker follows a very complicated pattern-
so that, for example, Ist person dual (but not singular or plural) agents take ergative case with 2nd
person dual or plural (but not singular) patients. The distribution of the inverse marker shows no
such sensitivity, but 3rd--3rd configurations are marked as inverse when agent is dual or plural,
but direct when agent is singular (the ergative marker remains throughout).

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
viewpoint is not the starting-point. Note further that Jyarong has a

rection-marker,34 the a- prefix in 49. This occurs only with lst->2n
urations. Now, 2nd->lst configurations count as inverse, as show
presence of the inverse and ergative morphemes in 46. Thus we sho
49 to count as a direct configuration. But the present account of vi
suggests that either or both of the configurations involving both SA
count as slightly ambiguous, even in a language which imposes a ran
them. Thus the existence of a special marking for one or both categor
as confirming my hypothesis.35
Also confirmatory is the existence of languages with 2nd> st ranki
side languages with Ist>2nd and Ist=2nd; this demonstrates that
versal principle determines the SAP>3rd ranking, but no such princ
termines the ranking of SAP's. The 2nd> 1 st>3rd hierarchy is found t
the Algonquian family-where, typically, agreement prefixes index 2n
if there is one, 1st person otherwise, and 3rd person only if there
participant. Potawatomi (Hockett 1966) can be taken as typical; the
parts of the transitive animate paradigm are (V represents the verb)
(51) 2-1 k-V 3->2 k-V-uk
1--2 k-V-un 1->3 n-V-a
2--3 k- V-a 3-> 1 n- V-uk
The prefixes are personal agreement-markers

direction-markers. Note that the distributio
reflects a 2>1>3 hierarchy; but the direction
the special status of the 'local' SAP->SAP categ
with the SAP---3rd configurations, but not w
occurs with 3rd->SAP, but not with lst->2nd,
-un. We may compare the Potawatomi directi
Algonquian languages, Parry Island Ojibwa (R
lor 1969),36 as shown in Table 1.
POTAWATOMI OJIBWA BLACKFOOT
direct a a a
1->2 un ed o
2--- 1 0 i oki
inverse uk ego oki
TABLE 1.
Both Blackfoot and Ojibwa assign agreement prefixes according to the same
system as Potawatomi; but again, the direction-marking system reflects a less
clear-cut hierarchy. Ojibwa distinctly marks each of the four categories, show-
34 On my analysis, Jyarong also has a direct morpheme, but it does not occur in the examples
cited here, and its explication would be irrelevant. It is described in DeLancey 1980.
35 The 2nd-,1st configuration also has a separate mark. the substitution of kd- for ta-. One of
these occurs in every configuration, transitive or intransitive, which involves a 2nd person. This
series is discussed at greater length in Bauman 1975 and DeLancey 1980.
36 An earlier version of this paper, circulated by the Indiana University Linguistics Club, cited
Blackfoot forms from Uhlenbeck 1938 which were not correct. I am grateful to Allan Taylor for
pointing this out to me and furnishing the correct paradigm.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
ing clearly that any ranking of the two SAP's is of a different order than t
ranking of SAP's over other NP's. Of particular interest to the present argum
is the Blackfoot system, which marks 2nd-> st as inverse, so that the direct
system reflects a Ist>2nd ranking, rather than the 2nd> st ranking sugges
by the order of agreement prefixes. Blackfoot thus presents evidence that t
ranking of 2nd>lst not only lacks universal validity (as suggested by Silver
stein), but also lacks strong motivation even in those languages which manif
it. We may conclude that both 2nd>lst and lst>2nd are possible variatio
on the universal SAP>3rd theme.
4.4. VIEWPOINT AND THE EH. I have so far accounted for only the high en
of the EH-the distinction between SAP's and all other NP's, which is equiv-
alent to the distinction between the spatial deictic center and everywhere el
The EH also encompasses a ranking of various types of full NP's. A ful
statement of the hierarchy, incorporating all the widely-attested distinctio
with which I am familiar, is:
SAP's > 3rd pronouns > human > animate > natural forces > inanimate
Only the two highest splits-SAP's over everything else and pronouns o
full NP's-are widely-attested as governing SE marking; and so far as I khow
only the SAP>3rd split ever governs direction-marking. Nevertheless, a nu
ber of other apparently viewpoint-related phenomena are governed by the r
of the EH. Consider, for example, the account given by Hawkinson & Hym
1974 of agreement in Shona, a Bantu language of Zimbabwe. In Shona, as in
many African languages, the verb is marked with a prefix reflecting the no
class of the subject. If the subject is a conjoined NP including nouns of two
different classes, agreement will be plural, but noun-class agreement is wi
the noun which is higher on the EH:
(52) murlume ne imbwa va-ka-famba.
man and dog NC-PAST-walk
'The man and the dog walked.'
(53) *murume ne imbwa dza'-ka-fJmba.
man and dog NC-PAST-walk
The conjoined NP 'man and dog' can be cross-referenced only by the plura
human prefix va, not by the animate plural prefix dza. Moreover, in such
conjoined NP, the noun higher on the EH must be the starting-point of linguisti
AF:
(54) *imbwa ne murutme va-ka-fdmbd.

Kuno & Kaburaki provide a number of examples of patterns in English which
reflect the EH; particularly relevant to my present concerns is its reflection
in the English voice alternation. Consider the following pairs of sentences:
(55) a. Many terrorists are motivated by patriotism.
b. Patriotism motivates many terrorists.
(56) a. A woman was struck by lightning.
b. Lightning struck a woman.
(57) a. My father was crippled by arthritis.
b. Arthritis crippled my father.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
In each case, the active alternant is considerably less natural; i.e., it

considerably more context, to make it a plausible utterance, than th
logically more marked passive alternant. Thus both English and Sho
a tendency to make an NP higher on the EH the starting point, in
to one lower.
Clearly, distinctions between human and non-human, or animate and inan-
imate, are not susceptible to the same kind of deictic explanation as I have
suggested for the SAP>3rd split. Nevertheless there is abundant evidence that
higher position on the EH counts as higher eligibility for viewpoint placement.
Kuno & Kaburaki suggest the notion 'empathy' (whence the term 'Empathy
Hierarchy') as a solution to this problem. When the event being reported does
not involve either SAP as a participant, viewpoint may be placed with either
participant. Kuno & Kaburaki suggest that speakers, being animate and human,
are more likely to 'empathize with' (i.e. take the viewpoint of) human beings
than animals, and of animals than inanimates. (The same solution is suggested
by Zubin, who describes the EH as a scale of 'egocentrism'.) On this account,
then, the entire EH can be interpreted in terms of relative eligibility for view-
point placement.
4.5. THE EH AND NATURAL AGENTIVITY. An explanation of the EH-split pat-

tern which has gained some currency in the past few years is based on an
interpretation of the EH in terms of a notion of natural agentivity. This account,
proposed by Silverstein 1976 and Comrie 1978a, and seconded by Dixon 1979,
maintains that each NP type is a more natural agent than any NP lower on the
hierarchy, and a less natural agent than any higher. This distribution of case-
marking is thus susceptible to a functional explanation, according to which
NP's that are inherently likely agents needn't be marked to show their role;
but those that are inherently less likely to function as agents must, when they
do occur as agents, be marked for their role.
As we shall see in ?6.2, the interpretation of at least part of the EH in terms
of natural agentivity is attested in languages. Nevertheless, a number of ob-
jections must be raised to this hypothesis as an explanation of the SE pattern.
To begin with, it does not provide a convincing interpretation of the entire
EH. While it is intuitively clear that, in general, animate beings are more likely
first movers than inanimates-and, at least in the world of human discourse,
that humans are more likely agents than non-humans-it is far from clear that
any such ranking will operate between SAP's and human 3rd persons. Thus,
while the Natural Agentivity Hypothesis allows for a split at any point along
the hierarchy, we might expect the split in ergative-marking to occur more
commonly between human and non-human, or between animate and inanimate
NP's, than between SAP's and other NP's. In fact we find just the opposite:
the majority of attested SE patterns follow the SAP>3rd split. The only other
widespread pattern is the split between pronouns and full NP's; other patterns
are attested only rarely. This is, of course, exactly what is predicted by the
Viewpoint Hypothesis, according to which SAP>3rd is the most natural pos-
sible split.
A similar problem concerns the ranking of the SAP's. Presumably, if human

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
646 LANGUAGE. VOLUME 57. NUMBER 3 (1981)
linguistic ability or cognition, or the realities of existence, involve some

versal ranking of likely agents, 1st and 2nd persons should be univers
ranked in one or the other order. Silverstein's markedness analysis predic
2nd>lst ranking; Dixon 1979, apparently on intuitive grounds, suggests th
lst>2nd is correct. Dixon notes that the question is controversial, and sugge
the possibility that the SAP's constitute a single type on the hierarchy. T
as we have seen, is correct; 1st and 2nd person do prototypically constitu
a single deictic category, and hence a single viewpoint category.
A further objection to the Natural Agentivity Hypothesis as an explana
of the SE pattern is that it fails on two counts of insufficient generality: it
explain neither all the manifestations of the EH nor all the manifestation
the SE pattern. For example, no explanation in terms of inherent agentiv
will adequately account for the differences in naturalness among exx. 55-
It is true that the low-EH agents occur with the agent-marker by in the m
natural passive versions; but it is hardly plausible that the function of pas
in such cases is to apply agent-marking to inherently unlikely agents. Inde
lightning and arthritis are very likely sources of transitive events. Clearly,
significant effect of the choice of the passive construction is to place the m
animate NP in leftmost position; agentivity is not relevant.
The Natural Agentivity Hypothesis not only fails to account for the full rang
of manifestations of the semantic aspect of the EH-split pattern (i.e. the
it also fails to account for the range of manifestations of its morphologi
aspects, i.e. the variation in case-marking. Recall the suggestion in ?2.2 th
the absence of case-marking seems to have a function of its own. The hypot
presented by Silverstein and by Comrie doesn't allow this interpretation;
supposes simply that case-marking is present when it is likely to be need
and absent when superfluous. On this account, no connection exists betwe
the two attested SE patterns; it must be purely coincidental that the aspec
and EH-split patterns manifest precisely the same morphological alternat
But an account of the EH-split pattern which permits an explanation of t
parallelism is more satisfactory than one which does not. As we shall see
the next section, my hypothesis can capture the parallelism between the
patterns.
VIEWPOINT AND ASPECT
5.1. PERFECTIVE AND IMPERFECTIVE.37 The aspectual SE pattern associates

what I have identified as viewpoint-marking morphology (i.e. nominative/ab-
solutive case and verb-agreement) with agent in imperfective aspect, and with
patient in perfective. Thus a unitary account of split ergativity requires a prin-
cipled association between terminal (patient) viewpoint and perfective aspect.
(If we assume that onset or agent is the unmarked viewpoint choice-i.e. that,
in the unmarked configuration, viewpoint and starting-point coincide-then
accusative morphology in imperfective requires no special explanation. Note
that this assumption is borne out in all the patterns discussed so far.)
37 The argument in this section is given in more detail and with further exemplification in
DeLancey 1979.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
I have described natural AF as reflecting the temporal structure of

described. This can, again, be most clearly illustrated with an ex
concrete motion. Compare ex. 18, 1 drove from Bloomington to Phil
with 58 as describing the same event:
(58) I drove from 8 A.M. till 7 P.M.
At 8 A.M., I am in Bloomington; at 7 P.M., I am in Philadelphia. The
time vectors are co-extensive, and their endpoints coincident; physic
perhaps young children-see Piaget 1946) would consider them two
tions of a single four-dimensional space/time vector. Clearly the sam
to (at least prototypical) transitive events; the time elapsing from
termination may be so brief as not to be worth mentioning, but a t
event does indeed originate with the agent at one point in time, and t
at the patient at a later point.
It therefore seems natural to interpret aspect as the temporal analog
spatial and transitive viewpoint specification discussed above,38 and
perfective aspect as a specification of terminal viewpoint, analogo
passive (terminal viewpoint with respect to the transitive vector) and
verbs (terminal viewpoint with respect to a purely spatial vector). The
SE languages, then (as I have argued in detail elsewhere, DeLanc
don't permit a conflict between aspect and viewpoint assignment: pe
aspect requires that viewpoint be with the NP associated with the te
terminal point, i.e. the patient. Then the morphological alternation in
(exx. 6-7) is precisely what we would expect; in the perfective, ag
marking viewpoint, is with patient, and the agent is marked for case-
that it is NOT the viewpoint, even though it is the starting-point. The
and EH-split patterns are thus parallel: in both, viewpoint placement
strained; and when the constraints prevent the natural starting-point from
selected as viewpoint, it must be marked as starting-point by ergati
marking.
5.2. PERFECT AND INFERENTIAL. The account of perfective aspect g

provides a natural interpretation of the distinction between the catego
FECTIVE and PERFECT (Comrie 1976b) as parallel to the difference be
'loose' and a strict interpretation of a 'come' verb. Perfective views a
from its terminal point, but there is no necessary relation between that te
point and the speech act. Perfect, by contrast, takes NOW, the tempor
of the speech act, as viewpoint. Thus it comes as no surprise to find
verbs used as markers of the perfect, as in Thai:39
(59) khaw lap maa haa chuamooy le'w.
he sleep come five hour PERF
'He's been sleeping for five hours [now].'
This in turn allows an illuminating account of the inferential interpretation
of perfect aspect discussed by Comrie 1976b. In a number of languages, a
38 For present purposes, the term 'aspect' refers only to the perfective/imperfective distinction.
39 My understanding of various aspectual phenomena in Thai has benefited greatly from dis-
cussions with Sarojini Huvanandana.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
perfect construction is ordinarily interpreted as reporting inference rather than

direct knowledge. Georgian has such a construction as part of a three-way SE
pattern (examples from Boeder 1967; the over-all pattern of case-marking and
agreement is discussed in DeLancey 1979):
(60) kaceb-i cer-en ceril-s.
men-NOM write letter-DAT
'The men are writing a letter.' (imperfect)
(61) kaceb-ma da-cer-es ceril-i.
men-ERG AoR-write-3pl. letter-NoM
'The men wrote a letter.' (aorist)
(62) kaceb-s u-cer-ia-t ceril-i.
men-DAT 3rd-write-PERF-PL letter-NoM
'The men have [apparently] written a letter.' (perfect)
Here 62 would be used if the speaker has inferential reason to believe the truth
of the proposition-perhaps he finds a letter in the room which the men have
just vacated-but it does not report an event which the speaker has witnessed.40
Such a construction is the temporal equivalent of an agentless passive, in that
both reflect a constraint on reversed AF. In a language which allows only
agentless passives, selection of the transitive terminal point (the patient) as
viewpoint makes it impossible to include the transitive starting-point (the agent)
in the picture presented; speaker cannot ask hearer to 'look backward' down
the transitivity vector. Likewise, in languages which place an inferential in-
terpretation on the perfect, the placement of viewpoint at or beyond the tem-
poral terminal point excludes an interpretation in which the onset point is
actually included in the picture presented. Again, we can interpret this as a
constraint on directing the hearer's attention 'backward' against the flow of
the time vector.
This view of the structure of inferential constructions fits well with Slobin
& Aksu's analysis of the Turkish perfect/inferential particle mi~; they accom-
modate a wide range of uses within a general definition according to which
mis indicates that a sentence reports an event known to the speaker only by
inference from a resultant state, with no first-hand knowledge of the process
which has led up to that state. It is significant, for purposes of the argument
to be presented below, that the speaker may indeed have first-hand knowledge
of the terminal point of the event, so long as it takes him by surprise, i.e. if
he is unaware of any phases of the event preceding its culmination. Thus the
sentence
(63) Kemal gel-mis.

come-mis 'Kemal came.'
This could be spoken if the speaker sees Kemal's coat, or has been informed
40 Some readers of an earlier draft of this paper objected, correctly, that since the inferred age
in this construction is in dative rather than ergative case (note that Georgian has a distinct ergati
case), the perfect should not be considered an ergative construction. However, the important p
for my purpose is that the agent is marked for a case other than nominative, and thus is mar
as not being the viewpoint NP.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
of Kemal's arrival, but has not yet seen Kemal himself; but it could
used if the speaker opens the front door to Kemal's knock and
standing there, providing that Kemal's visit comes as a surprise-i.e.
he opened the door, the speaker had no idea that Kemal might be c
Syuwa, a Tibetan language of Nepal, has a similar semantic ca
which an inferential construction 'is also used when the speaker re
event whose temporal origin is inaccessible to him' (Hoehlig 1978:21
(64) dang kongmu nuphela singha durbar nangla mei chii-d
yesterday night midnight (place name) inside fire burn-P
'Yesterday evening in the middle of the night a fire broke
Singha Durbar [it seems].'
This is spoken by someone who had gone to watch the fire, and thus
hand knowledge that a fire had broken out, but hadn't been there to
VIEWPOINT, ATTENTION FLOW, AND AGENTIVITY
6.1. INTENTION AND INADVERTENCE. The Sinhala language (Gair

a 'passive' construction,41 the most common interpretation of which
precipitating of the event by the agent was inadvertent:42
(65) mam pingaan binda.
I(NOM) plate broke
'I broke the plate [on purpose].'
(66) man-atin pingaan binduna.
I(OBL)-by plate broke(P)
'I broke the plate [accidentally].'
There is an obvious analogy here with an inferential construction used when
only the terminal point of an event is directly perceived by the speaker: the
difference between an accidental and a purposeful act is precisely in whether
the actor is aware of all phases or only of the act's termination. In a deliberate
act, all phases, from inception to completion, are present to the consciousness
of the agent; but in an inadvertent occurrence, only the termination is present
(even though an objective phase-by-phase description of the two events might
be identical: the fingers relax, the plate falls ...) The pattern of case-marking
in the Sinhala alternation can thus be considered analogous to the SE patterns
discussed above: unmarked viewpoint is the onset point of the event; the
assignment of linguistic viewpoint to the terminal point is indicated morpho-
logically by marking the natural starting-point for case. The interpretation of
the sentence as reporting an inadvertent event, like the inferential interpretation
of a perfect, results from a constraint on reversed AF; the terminal phase of
41 The characterization of this verb form as 'passive' is quite misleading, as pointed out by Gair
1970. What Gair calls the 'P' form of the verb can occur with intransitive as well as transitive
verbs, with the same reading of inadvertence. This alternation of verb form in Sinhala is thus
functionally equivalent to the case-marking alternation in the 'active' languages discussed in ?6.3.
42 There is also a capabilitive interpretation, reminiscent of the use of the Ilocano passive
described by Schwartz 1976, the reading of which is that the agent is capable of or good at the act
described.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
the event can be taken as the viewpoint only if earlier phases took place outside
the actor's awareness.
6.2. ATTENTION FLOW AND CONTROL. We noted in ?3.3 that considerable

cross-linguistic evidence exists for an association between agentivity and left
most position. This association is predicted by the interpretation of agent a
the first mover in a transitive event, i.e. the starting point of natural AF.
?6.1, I have suggested an interpretation of the stricter sense of the notion
agentivity, based on the AF concept. We find a most interesting confirmatio
of this latter interpretation in an EH-based constraint on word order in Nav
transitive sentences.
Navajo has a now-famous alternation, in transitive sentences with full NP
actors, between a construction in which transitive agent precedes patient and
a 'passive' in which patient precedes agent. The roles of the two NP's are not
marked on the NP's themselves, but are reflected in the choice of agent agree-
ment prefix on the verb; the 3rd person agent prefix yi- indicates that agent
precedes patient, bi- that patient precedes agent. The choice of order is fac-
ultative only when agent and patient are of the same rank on the Navajo
version of the EH (see, among others, K. Hale 1973, Creamer 1974,
Witherspoon 1977, Saville-Troike & McCreedy 1979). Otherwise it must op-
erate so as to place whichever NP is higher on the EH in leftmost position,
regardless of semantic role:
(67) a. ti? dzaanez yi-ztat.
horse mule 3rd-kicked it
'The horse kicked the mule.'
b. dzaaneez t1i bi-ztal
mule horse 3rd-kicked it
'The mule got kicked by the horse.'43
(68) a. hastiin 1tf? yi-ztat
man horse 3rd-kicked it
'The man kicked the horse.'
b. *tii2 hastiin bi-ztal.
(69) a. */((? hastiin yi-ztat.
horse man 3rd-kicked him
b. hastiin lti bi-ztal.
man horse 3rd-kicked him
'The man got kicked by the horse.'
Now, it is clear from 67 that Navajo does not require linguistic AF to follow
natural AF; a patient can in some circumstances precede an agent in a sentence
Thus the unacceptability of 68b must be connected with the higher ranking o
men than of horses on the EH. This conclusion is confirmed by the unac
ceptability of agent-patient order in 69a; in both 68 and 69, the higher-rankin
NP must precede.
43 The 'passive' examples might well be glossed 'got himself V-ed', in order to approximate th
actual force of the Navajo sentences; cf. below and Witherspoon.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
According to Witherspoon, this is not simply an instance of natural

considerations being given priority in determining linguistic AF, as in
examples 55-57. Rather, it reflects a Navajo interpretation of the low
of the EH in terms of natural agentivity. To the Navajo speaker, 67a
the origin of the event of the horse's volition, while 67b attributes th
to the mule's carelessness (or perhaps masochism) in allowing its
kicked. In other words, the first NP is considered to have a causal role in
bringing about the event. Thus 68a is better than 68b because, in this world,
men can kick horses if they like, without the victim's having much say in the
matter. But 69b is better than 69a because a man who gets kicked by a lower
animal can be assumed to have let himself in for it; men with sense enough to
stay out of the way will never end up being kicked by horses.
This is a case of linguistic AF encoding what is perceived as inherent natural
AF-but tracing natural AF back, not to the literal first mover, but to what
is considered to be the first cause. This is the same interpretation of AF
discussed in ?6.1, where an event vector is considered to originate with an
actor who sets about causing the event to occur, but not with an actor to whom
the event comes as a surprise. The parallelism of the Navajo and Sinhala cases
is pointed up by the report (cited by Saville-Troike & McCreedy) that many
Navajo speakers will accept a sentence like 'The man was riding a horse' in
the bi- form, with 'horse' preceding 'man', if it is understood that the man was
drunk and thus not in control of the event. (Note that such exemptions vitiate
the notion of an absolute hierarchy of inherent agentivity; what is involved
here is not the abstract potential agentivity attributable to NP types, but the
actual degree of agentivity attributable to real entities in the universe of
discourse.)
6.3. THE 'ACTIVE'-TYPE LANGUAGES. This line of argument provides us with

an explanation for the distribution of case-forms in the so-called 'active'-type
languages, such as Batsbi and Eastern Pomo (McLendon 1978). Eastern Pomo
is an EH-split language in which pronouns are marked in transitive sentences
according to the nominative-accusative pattern:
(70) mi'p' mip-al sak'a
he(NOM) he-Acc killed 'He killed him.'
A pronominal subject of an intransitive verb which is necessarily volitional is
in the unmarked agent-case:
(71) mip' kaluhuya 'He went home.'
But that of a necessarily non-volitional intransitive (e.g. 'be burned', 'sneeze',
'forget', 'go crazy') is in the same case as a transitive patient:
(72) mi'p-al xa bakut ma.
he-Acc water fell 'He fell in the water.'
Some intransitive verbs can take either construction, depending on w
they are to be interpreted as volitional or inadvertent (the 1st and 2nd
sg. pronoun forms are suppletive; here ha' is agent case, and wi patien
(73) ha ba'tec'ki 'I got bumped [on purpose].'

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
(74) wi ba tec'ki 'I got bumped [accidentally].'

This alternation is parallel to the Sinhala example in that, when the event
is presented as accessible to consciousness only at its termination, the actor
takes the marked case-form. Both Sinhala and Eastern Pomo assign viewpoin
with respect to an event vector which has its origin not at the onset of th
perceptible event-i.e. at the inception of action on the part of the transitiv
agent or intransitive actor-but rather at the onset of a causal chain leading
to the occurrence of the event. In the event of a deliberate act, the onset poin
of the event vector is with the actor at the point in time when he first inten
the act, rather than the point at which he initiates action. When an eve
originates in a decision on the part of the actor, then the actor is both startin
point and viewpoint; this is reflected in its unmarked case-form. When an eve
originates somewhere other than with a decision on the part of the actor, th
viewpoint (still associated with the actor) and starting-point do not coincide
with the result that the actor is in marked case. Thus case assignment in the
languages is parallel to what we have seen in SE languages: the unmarked ca
indicates that starting-point and viewpoint coincide, the marked case that the
do not.
The use of patient case-marking for the inadvertent actor in Eastern Pomo
is just what would be predicted by Case Theory, since the actor is the Goal
of an event generated elsewhere, rather than the Source (agent) of the event.
The content of the case-marking is thus analogous to the dative subjects of
verba sentiendi in South Asian languages and elsewhere (including many 'ac-
tive'-type languages, according to Klimov 1974). Batsbi also has agent-case
subjects as volitional,44 and patient-case subjects as inadvertent actors (cf.
Deseriev):
(75) ax so kottov.
you(ERG) I annoy 'You're bothering me.'
(76) smiaki-ov xo vuic'no.
man-ERG you fed 'The man fed you.'
(77) ax it'ax.
you(ERG) run 'You're running.'
(78) xo maisvar.
you(ABss) were hungry 'You were hungry.'
But in Batsbi (as can be seen more clearly from exx. 11-12),
is the more marked form-thus precluding an explanation in term
assignment, as suggested for Eastern Pomo. The contradiction is
however, since Batsbi, unlike Eastern Pomo, is a consistently erg
i.e.. all transitive agents are marked with ergative case.45 Since
44 This is not an ergative case in the strict sense of the word, since it also mark
subjects.
45 However, ergative case-marking distinguishes animate from inanimate agents in that distinct
case-markers are used: one form marks animates, the other instruments and inanimate agents.
Thus Batsbi, like Navajo, recognizes a hierarchy of inherent agentivity. Note, however, that this
is not the sort of split predicted by the Natural Agentivity Hypothesis, since more plausible as
well as less plausible agents are marked as agents.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
case-form doesn't alternate with a nominative form in sentences with two

possible viewpoints, its presence cannot be connected with viewpoint speci-
fication. The Batsbi ergative marker thus carries only its own semantic content,
which is the identification of an NP as starting point. We conclude that Batsbi
marks the volitional/inadvertent distinction just like Eastern Pomo, marking
a volitional actor as the starting-point of an event vector, and an inadvertent
actor as the terminal point, depending on whether the event has its origin in
the actor's intention or elsewhere.
CONCLUSION
7. All the variable case-marking patterns which we have examined can be

explained in terms of viewpoint and starting-point of AF. Ergative case-marking
labels the starting-point when it is not also the viewpoint; when viewpoint and
starting-point coincide, the NP is not marked for case. Leftmost position marks
starting-point, while verb-agreement, in many of the languages we have looked
at, indicates viewpoint.
The three major patterns of split case-marking which we have considered-
the EH-split, the aspectual split, and the active pattern-differ from one
another in what aspect of the event vector is paramount in assigning view-
point and AF values. The two endpoints of an event vector are simultaneously
points in space, points in time, and entities in the universe of discourse. The
EH-split pattern assigns viewpoint on the basis of the identities of the occu-
pants of the two endpoints of the event vector; the typical pattern is one in
which viewpoint placement is deictically constrained,46 so that it must be
placed at the endpoint occupied by an SAP if possible. The aspectual split pat-
tern assigns viewpoint with respect to the temporal aspect of the event vector,
with terminal viewpoint corresponding to the attainment of the terminal point
by the actors in the event.
There is another difference in principle between these two approaches to
viewpoint assignment. Note that, in the EH-split pattern, viewpoint is assigned
with respect to the actual point of view of speaker and hearer, both of whom
are prototypically external (at the time of the speech act) to the event-even
if one or both were or will be participants in the event. In the aspectual split,
viewpoint is assigned with respect to the actual point of view of a participant
in the event; perfectivity as terminal viewpoint is the point of view of an actor
who has reached the end of the event. This same distinction constitutes the
difference between the expression of inference and of volition. These share a
more abstract interpretation of the event vector, in which it represents not the
actual unfolding of an event in space/time, but its development from potentiality
to realization.47 Here the starting-point represents the cause of the event's
46 There is some reason to believe that the distinction between old and new information can be
interpreted as a deictic notion, and thus that the pronoun vs. full-NP split pattern may also have
a deictic interpretation. Note, for example, in ?4.3, that some North American direction-marking
systems treat the proximal 3rd -> obviative 3rd configuration as direct, and obviate -- proximal
as inverse.
47 Other manifestations of this interpretation of the event vector have been described in unpub-

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
becoming real, and the terminal point its culmination in reality. In these
guages, viewpoint placed at the starting-point indicates that the event is tra
from its inception; terminal viewpoint indicates that the event enters consc
awareness only when it is realized. The difference between the marking o
inference vs. first-hand knowledge and inadvertence vs. volition is that t
first distinction represents the point of view of an external observer
speaker-while the second represents a point of view associated with the a
in the event. (In this connection, it is interesting that the marking of volitional
in Batsbi is applicable only to SAP's; in other words, one does not mak
insupportable inferences about the awareness of 3rd persons.)
Thus the framework suggested here allows a unitary account of sev
apparently disparate phenomena. It should be noted that the basic concept
invoked here, viz. viewpoint and AF, do not have to be taken as given in
sense of the 'primes' of Relational Grammar, or of the 'core semantico-sy
tactic relations' named by Dixon (1979:61): transitive subject, transitive obje
and intransitive subject. They are, rather, hypotheses about human cogni
and perceptual structure; as such, they are subject to empirical investigat
by methods other than those of theoretical linguistics.48
REFERENCES
ABADIE, PEGGY. 1974. Nepali as an ergative language. Linguistics of the T

Area 1.156-77.
ANDERSON, JOHN M. 1971. The grammar of case: Towards a localistic theory. C
bridge: University Press.
- . 1977. On case grammar: Prolegomena to a theory of grammatical relations.
don: Croom Helm.
. 1979. On being without a subject. Bloomington: Indiana University Linguistics
Club.
BANDHU, CHURAMANI. 1973. Clause patterns in Nepali. In A. Hale, 2:1-80.
BAUMAN, JAMES. 1975. Pronouns and pronominal morphology in Tibeto-Burman. Berke-
ley: University of California dissertation.
BENVENISTE, EMILE. 1946. Les relations de personne dans le verbe. BSLP 43.1-12.
[English translation in Benveniste 1971:195-204.]
-- . 1956. La nature des pronoms. For Roman Jakobson, ed. by Morris Halle et al.,
34-7. The Hague: Mouton. [English translation in Benveniste 1971:217-22.]
-. 1971. Problems in general linguistics. Coral Gables: University of Miami Press.
BOEDER, WILFRIED. 1967. Uber die Versionen des georgischen Verbs. Folia Linguistica
2.32-152.
CAUGHLEY, Ross. 1978. Participant rank and verbal cross-reference in Chepang. I
Grimes, 163-78.
COMRIE, BERNARD. 1976a. Review of Klimov 1973. Lingua 39.511-60.
- . 1976b. Aspect: An introduction to the study of verbal aspect and related problems.
Cambridge: University Press.
lished work by Lon Diehl. Note, for example, the use of forms meaning 'via' as agent-markers,
as in English. Uhlenbeck apparently came to a similar analysis, based in part on examination of
'active' languages, but considered it characteristic only of the primitive mind. The relevant passage
is cited in Sapir 1917.
48 Work in this direction has been undertaken; see MacWhinney 1977, Zubin 1979, and references
cited by them.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
-. 1978a. Ergativity. Syntactic typology, ed. by Winfred P. Lehmann, 329-94. A

tin: University of Texas Press.
. 1978b. Agreement, animacy, and voice. Paper presented at the LSA W
Meeting.
CREAMER, MARY H. 1974. Ranking in Navajo nouns. Navajo Language Review 1.29-38.
DAS GUPTA, K. 1971. An introduction to the Nocte language. Shillong: North East
Frontier Agency.
DELANCEY, SCOTT. 1978. Empathy and Jinghpaw agreement. Paper presented at the
LSA Winter Meeting.
-- . 1979. Aspect, transitivity, and viewpoint. Paper presented at the Symposium on
Tense/Aspect: Between Semantics and Pragmatics, UCLA.
-- . 1980. Deictic categories in the Tibeto-Burman verb. Indiana University dissertation.
DESERIEV, JUNUS D. 1953. Bacbijskij jazyk. Moscow: Nauka.
DIEHL, LON. 1975. Space case: Some principles and their implications concerning linear
order in natural languages. Working Papers of the Summer Institute of Linguistics,
University of North Dakota Session 19.93-150.
DIXON, ROBERT M. W. 1976 (ed.) Grammatical categories in Australian languages. Can-
berra: Australian Institute of Aboriginal Studies.
. 1979. Ergativity, Lg. 55.59-138.
ECKMAN, FRED. 1974. Agentive and agentless passives. Stanford Working Papers on
Language Universals 14.59-74.
FILLMORE, CHARLES. 1966. Deictic categories in the semantics of 'come'. Foundations
of Language 2.219-26.
. 1968. The case for case. Universals in linguistic theory, ed. by Emmon Bach
& Robert Harms, 1-88. New York: Holt, Rinehart & Winston.
-- . 1975. The Santa Cruz lectures on deixis. Bloomington: Indiana University Lin-
guistics Club.
-- . 1977a. The case for case re-opened. Syntax and semantics, 8: Grammatical re-
lations, ed. by Peter Cole & Jerrold Sadock, 59-81. New York: Academic Press.
. 1977b. Topics in lexical semantics. Current issues in linguistic theory, ed. by
Roger Cole, 76-138. Bloomington: Indiana University Press.
FOLEY, WILLIAM, and ROBERT VAN VALIN. 1977. On the organization of 'subject' prop-
erties in universal grammar. Proceedings of the 3rd Annual Meeting, Berkeley
Linguistics Society, 293-320.
GAIR, JAMES. 1970. Colloquial Sinhalese clause structures. The Hague: Mouton.
-- . 1976. Is Sinhala a subject language? (or, How restricted is your PNP?) In Verma
1976a, 39-64.
GIV6N, TALMY. 1976. Topic, pronoun, and grammatical agreement. In Li, 149-88.
. 1979a. On understanding grammar. New York: Academic Press.
. 1979b (ed.) Syntax and semantics, 12: Discourse and syntax. New York: Aca-
demic Press.
GRIMES, JOSEPH. 1978 (ed.) Papers on discourse. Arlington, TX: Summer Institute of
Linguistics.
GRUBER, JEFFREY. 1967. Look and see. Lg. 43.937-47.
- . 1976. Lexical structures in syntax and semantics. Amsterdam: North-Holland.
HALE, AUSTIN. 1973 (ed.) Clause, sentence and discourse patterns in selected languages
of Nepal. Norman, OK: Summer Institute of Linguistics.
--, and DAVID WATTERS. 1973. A survey of clause patterns. In A. Hale, 2:175-249.
HALE, KENNETH. 1973. A note on subject-object inversion in Navajo. Issues in lin-
guistics, ed. by Braj B. Kachru et al., 300-309. Urbana: University of Illinois
Press.
HAWKINSON, ANNIE K., and LARRY M. HYMAN. 1974. Hierarchies of natural topic in
Shona. Studies in African Linguistics 5.147-71.
HEATH, JEFFREY. 1976. Substantival hierarchies: Addendum to Silverstein. In Dixon,
172-90.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
HENDERSON, EUGENIE J. A. 1965. Tiddim Chin: A descriptive analysis of two texts.

London: Oxford University Press.
HOCKETT, CHARLES F. 1966. What Algonquian is really like. IJAL 32.59-73.
HOEHLIG, MONIKA. 1978. Speaker orientation in Syuwa (Kagate). In Grimes, 19-24.
HOPPER, PAUL. 1979. Aspect and foregrounding in discourse. In Giv6n 1979b:213-41.
, and SANDRA THOMPSON. 1980. Transitivity in grammar and discourse. Lg.
56.251-99.
JIN, PENG, et al. 1958. Jiarong yu Suomo hua-de yuyin he xingtai. [The phonology
morphology of the Jyarung language (Suo-mo dialect): Part 2, morphology.] Yuy
Yanjiu 3.71-108.
KATO, KAZUO. 1979. Empathy and passive resistance. LI 10.149-52.
KEPPING, KSENIA BORISOVNA. 1980. Elements of ergativity and nominativity in Tan
Ergativity: Towards a theory of grammatical relations, ed. by Frans Plank, 263-
London: Academic Press.
KLIMOV, GEORGIJ A. 1973. Ocerk obscej teorii ergativnosti. Moscow: Nauka.
-- . 1974. On the character of languages of active typology. Linguistics 131.11-2
KUNO, SUSUMU, and ETSUKO KABURAKI. 1977. Empathy and syntax. LI 8.627-72
KURY+OWICZ, JERZY. 1964. The inflectional categories of Indo-European. Heide
Winter.
Li, CHARLES N. 1976 (ed.) Subject and topic. New York: Academic Press.
-, and SANDRA A. THOMPSON. 1976. Subject and topic: A new typology of lang
In Li, 457-89.
MCLENDON, SALLY. 1978. Ergativity, case, and transitivity in Eastern Pomo. IJAL
44.1-9.
MACWHINNEY, BRIAN. 1977. Starting points. Lg. 53.152-68.
--. MS. Point-sharing. To appear in Communicative competence: Acquisition and
intervention, ed. by Richard L. Schiefelbusch & J. Pickar. Baltimore: University
Park Press.
MARAN, LARAW. 1978. Aspect and agreement particles in Jinghpaw. Paper presented
at the 11th International Conference on Sino-Tibetan Languages and Linguistics,
University of Arizona.
-. 1979. Localization and representational structures in Jinghpaw: A view of the
epistemological and semantic foundations. Paper presented at the Symposium on
Tense/Aspect: Between Semantics and Pragmatics, UCLA.
MARATSOS, MICHAEL. 1978. New models in linguistics and language acquisition. Lin-
guistic theory and psychological reality, ed. by Morris Halle et al., 246-63. Cam-
bridge, MA: MIT Press.
MISTRY, P. J. 1976. Subject in Gujarati: An examination of verb-agreement phenomena.
In Verma 1976a:240-69.
PIAGET, JEAN. 1946. Le developpement de la notion de temps chez l'enfant. Pari
Presses Universitaires de France.
ROGERS, JEAN H. 1975. Prediction of transitive-animate verbs in an Ojibwa diale
IJAL 41.114-39.
SAPIR, EDWARD. 1917. Review of Het passieve karakter van het Verbum Transi
van het Verbum Actionis in Talen van Noord-Amerika, by C. C. Uhlenbeck.
1.82-6.
SAVILLE-TROIKE, MURIEL, and LYNN A. MCCREEDY. 1979. Topic-prominence in Navajo.
Paper presented at the LSA Winter Meeting.
SCHWARTZ, ARTHUR. 1976. On the universality of subjects: The Ilocano case. In Li,
519-44.
SILVERSTEIN, MICHAEL. 1976. Hierarchy of features and ergativity. In Dixon, 112-71.
SLOBIN, DAN, and AYHAN AKSU. 1979. Tense, aspect, modality, and more in the use of
a Turkish particle. Paper presented at the Symposium on Tense/Aspect: Between
Semantics and Pragmatics, UCLA.
STERN, THEODORE. 1963. A provisional sketch of Sizang Chin. Asia Major n.s. 10.222-78.

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC
TAYLOR, ALLAN R. 1969. A grammar of Blackfoot. Berkeley: University of

dissertation.
TESNIERE, LUCIEN. 1959. Elements de syntaxe structurale. Paris: Klincksieck.
TRUBETZKOY, NIKOLAI S. 1939. Les rapports entre le ddtermine, le ddterminant et le
defini. Melanges de linguistique offerts a Charles Bally, 75-82. Geneva: Georg.
UHLENBECK, C. C. 1938. A concise Blackfoot grammar. (Verhandelingen der Konink-
lijke Akademie van Wetenschappen, Afdeeling Letterkunde, n.s. 41.) Amsterdam.
VERMA, MANINDRA K. 1976a (ed.) The notion of subject in South Asian languages.
(South Asian studies, 2.) Madison: University of Wisconsin.
- . 1976b. The notion of subject and the data from Nepali. In Verma 1976a:270-86.
WATTERS, DAVID. 1973. Clause patterns in Kham. In A. Hale, 1:39-202.
-- . 1975. The evolution of a Tibeto-Burman pronominal verb morphology. Linguistics
of the Tibeto-Burman Area 2.45-80.
WHISTLER, KENNETH W. 1980. Inverse person-marking in Nootkan. MS, University
California, Berkeley.
WITHERSPOON, GARY. 1977. Language and art in the Navajo universe. Ann Arbor:
University of Michigan Press.
ZUBIN, DAVID A. 1979. Discourse function of morphology: The focus system of Germ
In Giv6n 1979b:469-504.
[Received 18 April 1980;

Revision received 22 September 1980;
Accepted 15 December 1980.]

39.8.69.75 on Sun, 19 Sep 2021 02:56:17 UTC

Delancey (1981)

Uploaded by

Copyright:

Available Formats

Delancey (1981)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Delancey (1981)

Uploaded by

Copyright:

Available Formats

An Interpretation of Split Ergativity and Related Patterns

Author(s): Scott DeLancey

Stable URL: https://www.jstor.org/stable/414343

is collaborating with JSTOR to digitize, preserve and extend access to

This content downloaded from

The study of case-marking patterns in language has aroused considerable

This content downloaded from

(EH), and of similarly governed direct/inverse systems, in terms of th

1.1. THE SPLIT ERGATIVE PATTERN. An SE language is one in which some

This content downloaded from

which outrank non-human animates, which in turn outrank inanimates. In t

This content downloaded from

(7) ramesh-e pen khdrid-y-i.

1.4. THE ACTIVE/STATIVE SPLIT. A third type of split case-marking pattern,

2.1. THE SIGNIFICANCE OF SPLIT ERGATIVITY. The patterns described ab

This content downloaded from

language. Inverse patterns of EH or aspectual split, with ergative morphology

2.3. VERB-AGREEMENT. A similar argument applies to verb-agreement.

6 Even in the loose sense of fn. 1.

This content downloaded from

(13) md hdri-lai kitap din-chu.

This content downloaded from

Both case-marking and verb-agreement are, as we have seen, primarily de-

This content downloaded from

temporal ordering of phases of the event; other things being equal, th

3.3. ATTENTION FLOW IN TRANSITIVE SENTENCES. Like motion events, dative

This content downloaded from

word-order type in human languages.16 There is some evidence (cited and

Further evidence that unmarked linguistic AF is that which recapitulat

This content downloaded from

(24) ngaa- Pi waa 2-kaay ? saay 2-naa-ng.

3.4. VIEWPOINT. Since linguistic AF is strongly tied to the semantic (and

This content downloaded from

This content downloaded from

3.5. VIEWPOINT IN TRANSITIVE SENTENCES. The notion of viewpoint

(31) hong-pe-tu hi.

This content downloaded from

The extension of such a localist analysis to transitivity is somewhat m

This content downloaded from

an event, am a much more natural locus of viewpoint in a sentence wh

4.1. SUBJECT-CODING PROPERTIES. As noted in the previous section, most

In all these examples, the leftmost NP-the starting-point of linguistic AF-

This content downloaded from

This content downloaded from

In this construction, the 3rd person agent-marker occupies the same pr

4.3. DIRECTION-MARKING AND SPLIT ERGATIVITY. A number of languages use

This content downloaded from

Given a language-specific ranking of Ist>2nd>3rd person, the distribution o

(46) no-ks nga kd-u-nasno-ng.

This content downloaded from

viewpoint is not the starting-point. Note further that Jyarong has a

The prefixes are personal agreement-markers

This content downloaded from

(54) *imbwa ne murutme va-ka-fdmbd.

This content downloaded from

In each case, the active alternant is considerably less natural; i.e., it

4.5. THE EH AND NATURAL AGENTIVITY. An explanation of the EH-split pat-

This content downloaded from

linguistic ability or cognition, or the realities of existence, involve some

5.1. PERFECTIVE AND IMPERFECTIVE.37 The aspectual SE pattern associates