A usage-based theory of grammatical status and grammaticalization

Kasper Boye
Peter Harder

Language, Volume 88, Number 1, March 2012, pp. 1-44 (Article)

Published by Linguistic Society of America

DOI: 10.1353/lan.2012.0020

For additional information about this article

http://muse.jhu.edu/journals/lan/summary/v088/88.1.boye.html

Access Provided by University of Sussex at 06/15/12 3:44PM GMT

 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND
GRAMMATICALIZATION

KASPER BOYE PETER HARDER

University of Copenhagen University of Copenhagen
This article proposes a new way of understanding grammatical status and grammaticalization as
distinctive types of linguistic phenomena. The approach is usage-based and links up structural and
functional, as well as synchronic and diachronic, aspects of the issue. The proposal brings a range
of previously disparate phenomena into a motivated relationship, while certain well-entrenched
criteria (such as ‘closed paradigms’) are shown to be incidental to grammatical status and gram-
maticalization. The central idea is that grammar is constituted by expressions that by linguistic
convention are ancillary and as such discursively secondary in relation to other linguistic expres-
sions, and that grammaticalization is the kind of change that gives rise to such expressions.*
Keywords: grammatical vs. lexical status, grammaticalization, ancillariness, discourse promi-
nence, dependency

1. INTRODUCTION. The distinction between lexical and grammatical expressions has

been a cornerstone in linguistic theory from the earliest beginnings. Via the distinction
in Priscian and medieval logic between CATEGOREMATA and SYNCATEGOREMATA, it can
be traced all the way back to Aristotle (Lyons 1968, Klima 2005). Today it plays an im-
portant role in all major approaches to the study of language (cf. Slobin 1997:265–
66), formal as well as functional, and it serves to define a number of important theoret-
ical notions. For instance, grammatical rather than lexical status is often taken to be
what distinguishes aspect from aktionsart (e.g. Comrie 1976:6–7, n. 2) and mood from
modality (e.g. Palmer 1986:7), as well as tense from time, and number from enumera-
tion (cf. Palmer 1986:7).
The distinction is thus very much part of the furniture of linguistics. Yet there is a
striking lack of clarity about what it means to be a grammatical or a lexical expression.
It seems as if there has been a feeling that we all know what a grammatical expression
is, and because of this the distinction has remained pretheoretical and intuition-based.
On this background, it is natural to ask whether the distinction between lexical and
grammatical expressions can (or should) be maintained at all. Perhaps standard exam-
ples of lexical and grammatical expressions are better conceived of as belonging to dif-
ferent poles of a continuum in which not even an approximate dividing line can be
drawn? This position has been gaining ground since the early 1980s, especially in stud-
ies dealing with grammaticalization. Drawing on ideas that are often traced back to
Meillet (1912), a number of scholars have explored diachronic changes in which nouns,
verbs, and other lexical expressions develop into grammatical expressions such as af-
fixes, clitics, and auxiliaries (e.g. Traugott & Heine 1991, Lehmann 2002a, Hopper &
Traugott 2003). In support of the idea of a continuum, they have demonstrated that such
changes are gradual, and the stages can be arrayed along ‘grammaticalization clines’
with prototypical lexical expressions at one end and prototypical grammatical expres-
sions at the other.

* Kasper Boye’s work on this article was supported by a grant from the Carlsberg Foundation and by the
LANCHART center at the University of Copenhagen. We are grateful to Henning Andersen, Jürgen Bohne-
meyer, Niels Davidsen-Nielsen, Elisabeth Engberg-Pedersen, Nina Grønnum, Bernd Heine, Johan Pedersen,
Lene Schøsler, Ilja Serzants, Elizabeth C. Traugott, John Tøndering, Greg Carlson, and Jürgen Bohnemeyer,
the audiences for whom we have presented the theory, and two anonymous referees for comments on a previ-
ous version of the article.
1
2 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

What can be concluded beyond doubt as a result of this type of grammaticalization

research is that grammar and lexicon are not worlds apart. The lexicon feeds diachroni-
cally into the grammar in a way that excludes a conception of grammar as autonomous
(in the strong sense of ‘sealed’ and ‘independent’). In this respect grammaticalization
research may be seen as a challenge in particular to mainstream generative linguistics
(as acknowledged by Roberts and Roussou (2003)).
Arguably, this challenge would be strongest if some sort of systematicity in the rela-
tion between lexicon and grammar could be demonstrated. Accordingly, works on
grammaticalization have repeatedly stressed the uniformity of the various pathways.
Standard examples of grammaticalization changes have been found to be bound up with
a recurrent cluster of diachronic processes—phonological, morphosyntactic, and se-
mantic—and grammaticalization has been claimed to be a distinct, sui generis type of
phenomenon.1
In recent years, however, this claim has itself been challenged (see especially Camp-
bell 2001, Janda 2001, Joseph 2001, and Newmeyer 2001). While some phonological,
morphosyntactic, and semantic processes perhaps typically do accompany grammati-
calization changes, they do not always accompany them, and they are frequently found
also with changes that cannot be claimed to involve grammaticalization (e.g. Campbell
2001). They can hardly serve to define grammaticalization, then. So paradoxically,
those who wish to claim that grammaticalization is a distinct type of change that moves
across the continuum from lexicon into grammar find themselves entangled in the self-
same problem that they brought into focus in the first place: the problem of defining in
a satisfactory manner the distinction between lexical and grammatical expressions. The
problem simply will not go away.
The present article proposes a solution to the problem. It argues that the distinction
between lexical and grammatical expressions should be maintained, and proposes a
specific conception of it. After a brief discussion of the problems connected with stan-
dard definitions of grammatical status and grammaticalization, it outlines a unified the-
ory of the two notions. The theory is usage- or discourse-based, and it is functional in
the European sense that it focuses on the functional role of grammatical expressions in
relation to other linguistic expressions (cf. Engberg-Pedersen et al. 1996), but not in the
sense that functional factors outside of language determine what expressions are gram-
matical. The central idea behind the theory is that grammar is constituted by expres-
sions that by linguistic convention are ancillary and as such discursively secondary in
relation to other expressions—and that grammaticalization consists in the diachronic
change that leads to such expressions. Conversely, lexical expressions are by linguistic
convention potentially primary in terms of discourse prominence. The concept of dis-
course prominence is understood in terms of a core idea that we regard as essentially
uncontroversial: in entertaining complex mental content, there is always a priority di-
mension involved, so that some parts of the content are more highly prioritized than
others (cf. §3.1).
In subsequent sections, the article offers six arguments in support of the theory. The
arguments for the theory are: it allows a generalization over standard as well as non-
standard examples of grammatical expressions; it allows a generalization over standard
as well as nonstandard examples of grammaticalization; it provides a qualification of
and a preliminary answer to the question of what qualifies a lexical expression for un-

1 For instance, many authors regard grammaticalization as a ‘process’ itself (e.g. Heine & Kuteva 2002:4,

Lehmann 2002a:vii, Traugott 2003:645; see Newmeyer 2001:189–90 for discussion).

 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 3

dergoing grammaticalization; it provides a unified account of the features normally as-

sociated with grammaticalization; it makes possible a precise characterization of gram-
maticalization clines and of the relation between grammaticalization on the one hand
and regrammaticalization and degrammaticalization on the other; it allows a precise
characterization of the relationship and differences between grammaticalization on the
one hand and lexicalization and ‘constructionalization’ on the other. The article ends
with an attempt to situate the proposed theory in the theoretical landscape of grammat-
icalization and a brief summary.
2. STANDARD DEFINITIONS OF GRAMMATICAL EXPRESSIONS AND GRAMMATICALIZATION.
When grammatical expressions and grammaticalization are defined at all, they are usu-
ally defined in one of three ways.
ii(i) ostensively, with reference to uncontroversial examples
i(ii) in terms of sets of features found with uncontroversial examples
(iii) in terms of one feature found with uncontroversial examples
All three types of definitions, however, give rise to serious problems.
An example of a definition that is ultimately ostensive is found in Matthews 2007.
Matthews’s definitions of ‘grammatical morpheme’ and ‘grammatical word’ redirect
one to his definition of ‘grammatical meaning’.
(1) a. ‘grammatical morpheme’: ‘A morpheme which has grammatical mean-
ing.’ (Matthews 2007:165)
b. ‘grammatical word’: ‘One which has grammatical meaning.’ (Matthews
2007:165)
In turn, he defines grammatical meaning as follows.
(2) ‘grammatical meaning’: ‘Any aspect of meaning described as part of the syn-
tax and morphology of a language as distinct from its lexicon. Thus espe-
cially the meanings of CONSTRUCTIONS and INFLECTIONS, or of WORDS WHEN
DESCRIBED SIMILARLY.’ (Matthews 2007:164, emphasis added)
Matthews’s definitions of grammatical morpheme and grammatical meaning avoid com-
plete circularity by ultimately defining grammatical expressions in terms of typical
examples: ‘constructions and inflections’ and ‘words when described similarly’. The
problem with definitions based on typical examples is that they leave one in the dark
when it comes to linguistic expressions that are not uncontroversially grammatical or lex-
ical. In 3, for instance, are nothing and him lexical or grammatical expressions, is seem a
lexical verb or an auxiliary, and is down a lexical adverb or a grammatical particle?
(3) Almost nothing seems to get him down.
Examples of definitions in terms of sets of typical features are found in Heine & Reh
1984 and Lehmann 2002a. Heine and Reh see grammaticalization as ‘an evolution
whereby linguistic units lose in semantic complexity, pragmatic significance, syntactic
freedom, and phonetic substance’ (1984:15). Lehmann takes grammaticalization to be
‘a complex phenomenon which is constituted by … and has no existence independently
of’ changes along six parameters: integrity, paradigmaticity, paradigmatic variability,
structural scope, boundness, and syntagmatic variability (2002a:110–11). Definitions
like these often refer to one or more of the following features.
• boundness (also known as ‘boundedness’ or ‘bondedness’)
• phonological reduction
• semantic reduction (bleached, generalized, or abstract meaning)
• closed-class membership
• obligatoriness
4 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

There are two problems in relying on such features in definitions of grammatical ex-
pressions and grammaticalization (cf. Campbell 2001, and §7). First, even with uncon-
troversial cases of grammatical expressions and grammaticalization, often only a subset
of these features is found (e.g. Fischer 2007:119–20). Second, most—if not all—of the
features are frequently found also with uncontroversial examples of lexical expressions,
and with changes that may be considered entirely lexical. To begin with phonological re-
duction, one form is the type known as haplology, which is found in the development of
Engla-lond ‘land of Angles’ into England. Semantic reduction is found in the extension
of the verb fly from having the meaning of ‘move quickly through the air’, as in birds
fly, to having the meaning of ‘move quickly’, as in Flying Scotsman. In the complex-
predicate construction put away, boundness and obligatoriness are arguably properties
not only of away but also of the unquestionably lexical verb form put. Finally, closed-
class membership is arguably a property of all lexical verbs that can be construed with an
accusative-with-bare-infinitive complement: I saw/heard/*persuaded him run.
Alternatively, one might fall back on prototype theory and conceive of the features as
defining prototype examples, rather than being necessary or sufficient criteria. Like
Croft (2003:224–25), for instance, one might say that prototype examples of grammat-
ical expressions and grammaticalization have all or a certain subset of features. Again,
however, this would not solve all problems. First, it would still be unhelpful when it
comes to grammatical expressions and cases of grammaticalization that are not proto-
typical. Second, some cases that intuitively ought to belong among the standard cases in
fact seem to possess very few of the familiar features, if any at all. Schematic construc-
tions like the English interrogative clause are a case in point. They figure as uncontro-
versial examples of grammatical expressions in Matthews’s definition of grammatical
meaning cited above. But they can hardly be described as phonologically or semanti-
cally reduced (what would they be reduced in relation to?), they are bound only in the
sense that they require some lexical and morphological material to fill them, and it is
not at all clear in what sense one could talk about them in general as obligatory and as
belonging to a closed class.
As for definitions in terms of a single feature only, two types may be distinguished
(cf. Kiparsky 2005): those that refer to a morphosyntactic feature, and those that refer to
meaning (or function). Haspelmath (2004:26) provides a definition of the first type: ‘A
grammaticalization is a diachronic change by which the parts of a constructional
schema come to have stronger internal dependencies’. This definition suffers from at
least the first of the two problems discussed in connection with definitions in terms of
sets of typical features: it seems that even uncontroversial cases of grammaticalization
are not necessarily accompanied by a strengthening of the internal dependencies be-
tween the parts of a constructional schema (cf. Kiparsky 2005). For instance, it is not
obvious that the dependency that holds between the slot filled by the auxiliary have and
the slot filled by the main verb in an English perfect construction is in any sense
stronger than the dependencies between the slot filled by the lexical verb have and the
slots filled by its complements (subject, object) in a transitive construction. Hence, it is
not obvious that the grammaticalization of lexical have into an auxiliary involves a
strengthening of dependencies.
Examples of definitions in terms of ‘grammatical meaning’ (or ‘grammatical func-
tion’) are found in Heine et al. 1991, Traugott 2003, and Diewald 2006. According to
Heine and colleagues, grammaticalization is a change ‘where a lexical unit or structure
assumes a grammatical function, or where a grammatical unit assumes a more gram-
matical function’ (1991:2). According to Traugott and Diewald, likewise, grammatical-
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 5

ization is ‘[t]he process whereby lexical material in highly constrained pragmatic and
morphosyntactic contexts is assigned grammatical function, and, once grammatical, is
assigned increasingly grammatical, operator-like function’ (Traugott 2003:645; cf. Die-
wald 2006:2–3). Definitions like these are no less problematic than those already dis-
cussed. For one thing, we have already seen that unless they are supplemented by a
definition of grammatical meaning (or grammatical function) that is independent of the
notions of grammatical expressions and grammaticalization, there is a risk of circular-
ity: grammatical expressions are defined in terms of grammatical meaning, while gram-
matical meaning is defined in terms of grammatical expressions.
This leaves definitions in terms of the kind of meaning that is characteristic of gram-
matical expressions. There are three main ways of defining grammatical meaning in the
literature. One way, as already noted, is to define it as reduced (bleached, generalized,
or abstract) or even nonexistent. For instance, Trask defines a ‘grammatical morpheme’
as a ‘morpheme which has LITTLE OR NO SEMANTIC CONTENT and which serves chiefly as
a grammatical element’ (1993:123, emphasis added). This way of thinking about gram-
matical expressions is pervasive in terms like ‘form word’ and ‘function word’ (e.g.
Crystal 2008). Another way of defining grammatical meaning invokes an assumption
that grammatical meaning is of an ontologically distinct type. For instance, Nicolle
(1998), within the framework of RELEVANCE THEORY, claims that grammatical expres-
sions have procedural, as opposed to conceptual, meaning. The third way of defining
grammatical meaning exploits the finding that, crosslinguistically, the meanings that
grammatical or grammaticalizable expressions can have seem to fall within a notionally
limited range (e.g. Slobin 1997, Talmy 2000, Croft 2003:225). Accordingly, one might
envisage that grammatical expressions and grammaticalization can be defined simply in
terms of this notional range. Either a list of the notions covered by the range can be pro-
vided, or some common notional denominator. For instance, Diewald (2010) suggests
that it is common to meanings of grammatical expressions that they are weakly deictic
(see also Traugott 1989 on subjectification).
However, all three ways of defining grammatical meaning are problematic. Some
definitions, like Diewald’s or Nicolle’s, do not obviously cover all uncontroversial ex-
amples of grammatical expressions. It is not clear, for instance, in what sense noun clas-
sifiers, gender distinctions, and derivational affixes like English -ing are ‘weakly
deictic’ or procedural.2 Other definitions are insufficient because they also apply to cer-
tain lexical expressions. This includes definitions in terms of the most intuitively ap-
pealing feature, the property of being ‘empty’ or ‘bleached’, or ‘abstract’: in one sense,
the word thing is virtually empty of descriptive content (cf. it’s just one of those things
as opposed to thing in the sense of ‘physical object’), but the word is normally consid-
ered lexical in both usages (but see Fronek 1982). All definitions in terms of notional
domain suffer from this problem, as can be illustrated with the domain of possession.
Compare the expressions emphasized in 4–6.
(4) John HAS a mother.
(5) the mother OF John
(6) John’S mother

2 By contrast, if the notion of ‘procedural’ is interpreted broadly (i.e. as ‘includes procedural meaning’

rather than ‘has nothing but procedural meaning’), Nicolle’s (1998) proposed generalization can be contrasted
with the position that all meaning is basically procedural (cf. Harder & Togeby 1993, Harder 2007 on ‘in-
structional semantics’, and e.g. Evans & Green 2006, Evans 2009 on linguistic meanings as ‘prompting’ or
‘providing access to’ conceptual knowledge).
6 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

While there would probably be some disagreement about the status of of in 5, most
scholars would recognize has in 4 as a lexical expression, and -s in 6 as a grammatical
expression. But this distinction cannot be accounted for with reference to differences in
terms of notional domain: 4–6 all express ‘possession’, and would all have to figure in
an account of how possession is expressed in English (irrespective of the differences in
meaning between the three expressions). More generally, one can refer to the glosses
standardly provided for grammatical categories: the grammatical difference between
hat and hats is standardly glossed by the lexical contrast ‘singular’ vs. ‘plural’.
This does not contradict generalizations about what notional domains are liable to be
grammaticalized, such as Talmy’s (2000:24) observation that the domain of color never
is. The unsolved problem is the contrast between grammatical and lexical expressions
WITHIN those domains that sometimes grammaticalize.
These problems notwithstanding, we claim that there is such a thing as grammatical-
ization as a distinct type of linguistic change (pace Campbell 2001 and other contributors
to Language Sciences 23), and that there is such a thing as a distinct class of grammati-
cal expressions, which are the result of grammaticalization (pace Hopper 1991:33). This
claim is not at variance with the existence of gray zones, or clines (including grammati-
calization clines). As pointed out by Putnam (2001:38), the fact that a clear distinction
cannot be made in all cases does not entail that the distinction is invalid; his example is
that the distinction between ‘bald’ and ‘not bald’ is not rendered invalid by the existence
of people with fuzz on their heads. In a similar vein, Aarts (2007) offers a meticulous ar-
gument showing how gradient properties both within and across categories can be cap-
tured in a framework that keeps the categorical boundaries precise while recognizing that
instantiations constitute a cline. In contrast, an account that recognizes ONLY gradience
has difficulties in being precise about the role (if any) of categories.3 We propose a dis-
course-based theory of grammar that at the same time provides a clear delimitation and
a motivation for a cline between marginal and central instances.
3. OUTLINE OF A DISCOURSE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATI-
CALIZATION. The theory we wish to propose is panchronic in that it covers both gram-
matical expressions considered synchronically and grammaticalization conceived of as
the type of change that diachronically gives rise to grammatical expressions. The central
idea is that while grammatical expressions and grammaticalization cannot be defined in
terms of specific phonological, morphosyntactic, or semantic features, alone or in com-
bination, they can be defined in terms of the ancillary status that grammatical expressions
by linguistic convention have in relation to other expressions. Whereas lexical expres-
sions may or may not, in actual communication, convey the main point of a linguistic
message, grammatical expressions (morphemes, words, constructions) are convention-
ally specified as noncarriers of the main point, serving instead an ancillary communica-

3 Aarts’s topic is gradience among distribution-based syntactic categories. He argues (in continuation of

Ross 1969, Huddleston 1976, and Huddleston & Pullum 2002) that verbs constitute one overarching category,
with no clear categorial boundary between main verbs and auxiliaries. But at the same time he recognizes that
there are clear intracategorial differences between more or less prototypical verbs. This might appear to go
against our proposal for a clear distinction between grammatical and lexical expressions, including a clear
distinction between auxiliaries as grammatical expressions and main verbs as lexical expressions. But that
would only be the case if we claimed that a criterion for grammaticalization was the rise of a new, clear-cut
distributional class. This is not what we claim: like classic grammaticalization research, we assume that when
lexical expressions undergo grammaticalization, it happens one expression at a time. Some historical devel-
opments may create a new distributional class, but there is no one-to-one relation between grammaticalization
of individual lexical expressions and the rise of new syntactic categories.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 7

tive purpose as secondary or background elements. The fact that this is in ‘European’
terms a functional definition (cf. §1), rather than a definition in terms of inherent features,
can be illustrated with an analogy: a second violin is not inherently different from a first
violin—it just has a secondary function in relation to that of the first violin.
(7) LEXICAL EXPRESSIONS: Lexical expressions are by convention capable of
being discursively primary.
(8) GRAMMATICAL EXPRESSIONS: Grammatical expressions are by convention an-
cillary and as such discursively secondary.
(9) GRAMMATICALIZATION: Grammaticalization is the diachronic change that
gives rise to linguistic expressions that are by convention ancillary and as
such discursively secondary.
With the definitions provided in 7–9, the theory links up function and language use with
linguistic structure (§3.2), it pinpoints general similarities and differences between lex-
ical and grammatical meaning (§3.3), and it makes possible explicit, decidable tests for
grammatical status (§3.4). Before we go on to show this, however, we need to clarify
what we mean by the notions drawn upon in the definitions.
3.1. THEORETICAL PREREQUISITES.
The theory we propose depends on the notion of
DISCOURSE PROMINENCE, which forms the basis for the notions of DISCURSIVELY PRIMARY
vs. DISCURSIVELY SECONDARY status. Furthermore, our account of the relation between
discourse and grammar depends on the notion of LINGUISTIC CONVENTION. Finally, in
order to make clear what we understand by discourse prominence as a conventional
property of linguistic expressions, we need to be precise about how it differs from and
relates to the notion of FOCUS.
DISCOURSE PROMINENCE. We understand PROMINENCE as a universal feature of human
understanding of complex mental content. In understanding incoming information there
is always a priority dimension involved, so that some parts of the information are more
highly prioritized—more prominent—than others. This idea enters, in different modula-
tions, into the foundation for various linguistic and nonlinguistic theories—for instance,
Sperber and Wilson’s (1995) relevance theory, Langacker’s COGNITIVE GRAMMAR (cf.
Langacker 2008:12), and Bundesen’s (1990) theory of visual attention. In actual pro-
cessing, the prioritization may be seen as involving a ‘race’ (cf. Bundesen 1990).4 We
therefore assume that differential assignment of prominence is going on in all cases of
information processing, including for instance the processing of visual information.
When incoming information is conveyed by means of language, we speak of DISCOURSE
prominence: some aspects of linguistically conveyed mental content are more prominent
than others.
Discourse prominence is a scalar phenomenon. This means that a given element can
be ranked as higher or lower on a scale of prominence than another expression. In prin-
ciple, at least, the discourse prominence of any given element can also be measured and
assigned a value on a scale. The theory that is advocated here, however, depends only
upon the ability to COMPARE different elements in terms of discourse prominence. More
precisely, it depends on the relative discourse prominence of linguistic expressions that
are syntagmatically related and belong to the same utterance. This is where the distinc-
tion between primary and secondary status comes in: to be discursively primary is to

4 The term ‘race’ is used of different processing phenomena. For instance, Frauenfelder and Schreuder (e.g.

1992) refer to the choice between two processing routes for words (as single units vs. as morphologically
complex units) as a ‘race’. The race for prominence is in principle independent of such other types of race.
8 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

have higher discourse prominence than all other syntagmatically related elements in the
utterance; to be discursively secondary is to have lower prominence than one or more
syntagmatically related expressions in the utterance. It follows that an element can be
secondary in relation to another element that is itself secondary in relation to a third
element—but a primary element cannot be primary (or secondary) in relation to another
primary element.
Discourse prominence as described above is what in European linguistics is called a
‘substance domain’, or in Pike’s terminology an ‘etic’ as opposed to an ‘emic’ phenom-
enon. That is, it is nonconventional and not inherently associated with specific struc-
tural features at all. In addition to being a prelinguistic feature of human life, however,
it also serves as a basis for linguistic conventions that associate prominence properties
with specific linguistic expressions. In this, discourse prominence is analogous to
‘time’ or ‘possibility’: these notions also exist independently of any specific linguistic
expressions, but are conventionally associated with linguistic elements such as tense
and temporal adverbs (in the case of time), and modal verbs and modal adverbs (in the
case of possibility).

LINGUISTIC CONVENTION. We understand the notion of LINGUISTIC CONVENTION in ac-

cordance with the tradition of Lewis (1969), Clark (1996:70–72), and Croft (2000:
95–99) as reflecting a state of coordination between members of a community that goes
beyond individual instances of linguistic communicative interaction. Like other kinds
of conventions, linguistic conventions can be understood in terms of ‘status assign-
ments’ that are shared between the members of a community (cf. Searle 1995:37–43 on
status functions and collective intentionality).
Because linguistic conventions depend on a state of coordination in the speech com-
munity, they are distinct from the properties of any single usage event and the mental
representations that go with it in the minds of the specific language users involved. To
see this, consider the following example, which is relevant for grammaticalization: A
person with limited command of Spanish might say *ho dos casas instead of tengo dos
casas ‘I have two houses’, taking Spanish haber to convey the same possessive mean-
ing as English have. In such a case there would be a recoverable usage meaning of ‘pos-
session’, and that meaning would then be part both of an actual communicative event
and of a mental representation in the minds of the interlocutors, but it would be at odds
with the conventional meaning potential of the verb haber, which is limited to mean-
ings associated with the auxiliary function of the verb.
Although they are distinct, usage and conventions interact. On the one hand, conven-
tions are dependent on usage: conventions can remain stable only if they continue to be
reflected in actual usage and may therefore change as a result of changes in usage. On
the other hand, usage is dependent on conventions: when conventions are in operation,
they constrain actual usage events. Conventions do not operate REGARDLESS of circum-
stances of use, as sometimes claimed by overeager structuralists, but they INTERACT
with circumstances of use. This is why conventions may be overridden in specific situ-
ational circumstances, as illustrated in the *ho dos casas case discussed above. Over-
riding may in the long run result in new conventions, and in this way usage properties
may over time come to be conventional properties of the language. But individual cases
of overriding do not in themselves constitute language change (i.e. change in linguistic
conventions in the community).
Linguistic convention is what underlies all language-specific properties, from pho-
netics via structural relations to meaning. It also applies to social meaning in cases like
honorifics, and to attitudinal stance in cases like positively vs. negatively charged
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 9

words such as thrifty vs. stingy. On this background it is not surprising that it should
also apply to discourse prominence. Within the overall situational process of under-
standing a linguistic message, conventions about what is prominent are important for
the same reason that other linguistic conventions are important: speakers and hearers
need to align with each other, also in terms of prominence assignment. If interlocutors’
assumptions about what is the main thing do not coincide, they are literally talking at
cross-purposes. Prominence assignment as an etic or usage phenomenon takes place all
the time in all speakers, irrespective of what the precise conventions about prominence
may be—but if the emic or conventional prominence properties function efficiently,
they reduce the percentage of cases in which speaker and hearer assignments do not co-
incide. As will be apparent, our theory depends on understanding discourse prominence
both as a substance or etic phenomenon and as a conventional or emic phenomenon. If
this dual mode of existence is missed, the point of our article gets lost.
This view of discourse prominence as a conventional phenomenon forms the back-
ground for our claim that the contrast between lexical and grammatical expressions
is one way in which prominence properties are conventionalized. The most familiar
manifestation of prominence in language, however, is the notion of FOCUS, to which we
now turn.
FOCUS. The natural point of departure for understanding the conventionalization of
discourse prominence is the notion of focus. In broad terms, being prominent means
more or less the same as being in focus (cf. Langacker 2008:66–73, 418). Since we
claim that there is also a conventional link between prominence and lexical status, we
need to make clear why focus in itself does not exhaust the prominence issue, which re-
quires a more precise account of the nature of focus. Based on this account, we first try
to describe precisely under what narrowed-down conditions focus and primary status
coincide; second, we demonstrate where they may diverge; and finally, we use the foun-
dations established to address the distinction between lexical and grammatical status.
The account we present is based on assumptions shared between Lambrecht (1994)
and ‘alternative semantics’ (cf. Rooth 1992). Citing Akmajian, Lambrecht (1994:212)
describes the focused element as serving to fill out the missing argument in a proposi-
tion; thus in 10, Mitchell is the focused element because the information conveyed by
10 can be analyzed as in 11.
(10) MITCHELL urged Nixon to appoint Carswell.
(11) [x urged Nixon to appoint Carswell], [x = Mitchell]
The general idea is that the focus is a semantic ‘slot’ or ‘role’ that might also have been
filled out by a range of other options (in the case of Mitchell, e.g. Haldeman, Ehrlich-
man, or Haig). Thus conceived, focus constitutes an extra semantic element over and
above sentence meaning as standardly understood: it adds the property of presupposing
a contextual set of alternative propositions, all identical except for the value of some
variable. The focus construction then (in an assertive utterance) specifies that the refer-
ent of the focalized constituent is the actual value.
This account makes it possible to specify both why prominence and focus differ and
how they may coincide. Focus assignment is not basically a matter of prominence as-
signment, but of specifying information structure—but it is obvious why the focal part
of the information GETS higher prominence: the nonfocal part has presupposed status,
and thus makes no dynamic contribution to the message. Further, a central feature is
that focus assignment brings about a bipartition of utterance meaning (into the focal and
the nonfocal part). In so doing, it superimposes a digital element on the basically gradi-
ent nature of prominence as a prelinguistic phenomenon (cf. Bolinger 1968:17).
10 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

In spite of the differences, these properties also demonstrate that there is a point of
convergence, which is crucial for our purposes: at the privileged binary level imposed
by focus assignment, the focal constituent, viewed as a whole, is the most prominent
part of utterance meaning, and the rest of it, viewed as a whole, is less prominent. With
the terms introduced above, the focal constituent has DISCURSIVELY PRIMARY status
while the nonfocal part has DISCURSIVELY SECONDARY status. For that reason, focus can
be used to diagnose whether particular expressions have a potential for primary status
(cf. §3.4).
Although focus is a property of linguistic utterances, its relationship with linguistic
conventions is complex, and a full account would take us well beyond the scope of this
article (cf. Lambrecht 1994 for a painstaking discussion of the issue). In order to be pre-
cise about the relations between focus and the lexical-grammatical distinction, how-
ever, it is necessary to go a little bit further into the issue of the coexistence of different
levels of granularity and of conventionalized and nonconventionalized aspects of
prominence. A key issue is the distinction between broad and narrow focus, and the
very indirect relation between auditory prominence and discourse prominence that it
entails (cf. Ladd 1980, 2008, Lambrecht 1994:17).5 Consider first example 12 (from
Ladd 2008:214; the small capitals are ours).
(12) I gave him five FRANCS.
Following Ladd, in the BROAD-focus reading, auditory prominence (accent placement)
goes on the last word in the focused part of the sentence—which may be the whole mes-
sage ‘I gave him five francs’ or part of it, such as ‘five francs’ (Ladd 2008:215, n. 2; see
also Lambrecht 1994:247). In contrast, the NARROW-focus understanding is one in
which focus (and hence nonpresupposed status) is on the accented word only (i.e. the
currency ‘francs’).6
Based on the distinction between broad and narrow focus, it can be seen that there is
a micro level of granularity at which prominence is not reducible to focal vs. nonfocal
position. For illustration, we take another example from Ladd (1980:75, discussing
Schmerling 1976), understood in a broad-focus reading where the whole NP a wonder-
ful man is understood as being in focus.
(13) John is a wonderful MAN.
5 Auditory prominence is strictly a property of the ‘sound’ side of the linguistic sign, as opposed to the con-
tent side. Although there is an iconic element, such that prominence on the content side (i.e. discourse promi-
nence) in certain (and language-specific) cases covaries with auditory prominence, the full story is much
more complex. First of all, one type of auditory prominence has little to do with the issue, namely word (or
lexical) stress. In the case of the segmentally identical verb and noun import, relative auditory prominence of
one syllable over the other signals word class, not which syllable is more discourse-prominent. Second, there
is by contrast a prosodic level at which auditory prominence enters into the linguistic marking of focus (with
associated discourse prominence). This is what is known as sentence stress (or sentence accent). In 10
(MITCHELL urged Nixon to appoint Carswell), for example, Mitchell has sentence stress. The sentence stress
might alternatively be assigned to Carswell—in which case focus would be assigned to a different part of the
sentence. Mitchell also has lexical stress on the first syllable Mitch- (while other names may have lexical
stress on the second syllable, e.g. Liddell). The two types, word stress and sentence stress, are thus clearly
distinct.
6 The difference between the broad and narrow reading is a difference in terms of the type of alternatives

that would be relevant (cf. §3.1 on alternative semantics). In the narrow-focus reading, it is only alternatives
to the meaning of the stressed word that are relevant—in 12, for instance, ‘dollars’, ‘euros’, or ‘rubles’. In the
broad-focus reading, the range of alternatives may be broader, up to alternatives to the whole sentence mean-
ing (in the case of what Lambrecht calls sentence-focus structure; Lambrecht 1994:233). Alternatives would
then be of the form ‘Joe gave him a hard time’, ‘Jill stripped him of everything’, ‘Bill lent him five pounds’—
for instance, in answer to the question What happened?.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 11

The auditory prominence (accent) is on man; the stipulated focus marks as discursively
primary the NP a wonderful man. In other words, the range of alternatives, defined in
terms corresponding to the noun phrase a wonderful man, may include ‘my old friend’,
‘a complete idiot’, ‘your second cousin’, and so forth. But below the binary level, at the
micro level of discourse prominence, the most natural reading of 13 is one in which the
most prominent element is wonderful (man being an unsurprising category in the situa-
tion). On that assumption, at a micro level of granularity wonderful would be discur-
sively primary, and all other elements secondary in relation to it.7
This shows that a description in terms of focus alone does not cover everything there
is to say about discourse prominence as a property of linguistic utterances. This means,
first of all, that there is room for asking how other linguistic factors may reflect promi-
nence differences. Second, it means that in using focus properties to throw light on the
distinction between lexical and grammatical expressions, it is practical to think in terms
of NARROW focus, where stress, focality, and prominence are most tightly correlated.
DISCOURSE PROMINENCE AND LEXICAL STATUS: A COMPLEX RELATIONSHIP. By our defini-
tion, the prominence properties associated with the distinction between grammatical
and lexical status differ in two ways from the prominence properties associated with
focus. First of all, grammatical or lexical status is not bound to a particular utterance—
it is a property of items in the inventory of conventional expressions of the language.
Second, this entails that the crucial property is one involving potential rather than actu-
alized meaning: viewed as an item in the inventory, an expression does not have any ac-
tualized discourse prominence.
A lexical expression, on that view, is one with the potential for contributing the pri-
mary element of utterance meaning. What this means can be illustrated with an exam-
ple that parallels the wonderful man case, but with a grammatical expression involved.
For example, in the announcement Joe has died, the present perfect is naturally under-
stood as part of the focus (relevant focal alternatives might be ‘will return’, ‘is missing’,
‘cannot cope’, etc.), but from that it does not follow that the perfect is as prominent as
the meaning of the verb die (which would be the obvious candidate for discursively pri-
mary status).
Further, consider 14.
(14) Michael has always loved to swim.
If our theory is correct, the distinction between grammatical and lexical expressions
means that there is something we know about prominence assignment in 14, irrespec-
tive of sentence stress and focus assignment, and even if it is understood with a broad-
focus reading that includes the entire utterance (sentence focus; see Lambrecht
1994:336): by convention, the discursively primary element in 14 is not ‘an anterior
state of affairs with current relevance has occurred’ (or whatever the best paraphrase of

7 It may be useful to spell out in more painstaking detail (i) why neither the basic notion of discourse

prominence nor the more specific notion of discursively primary status reduces to being in focus or being au-
ditorily prominent, and (ii) under what circumstrances they converge. (i) Sentence stress (auditory promi-
nence) would be on man also if for some reason man was the most discourse-prominent term; hence,
discourse prominence could vary without change of stress. Also, a broad-focus interpretation entails that the
meanings of BOTH words are in focus—which means they are included in the most prominent part of the ut-
terance, without for that reason being equally prominent. (ii) However, in the case of narrow focus, as in 10,
all prominence-related factors converge: Mitchell has (prosodic) auditory prominence, is the sole focus ele-
ment, and is discursively primary—and this is why cases of narrow focus can have a diagnostic function also
when it comes to the prominence that we claim is associated with grammatical status (cf. §3.4).
12 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

the present perfect construction might be). As a grammatical expression, the present
perfect construction is by convention secondary in relation to the lexical verb love.
In order to find out what the primary elements might be, we can restrict our attention
to lexical expressions. Lexical expressions differ from grammatical expressions in that
by convention, they have the potential to be discursively primary, although in actual
communication they need not be primary. All lexical expressions (love, swim, Michael,
and always) are thus potential candidates for primary status. What is actually primary at
the micro level depends on the individual utterance; it may be any of them, or a combi-
nation such as ‘loves to swim’: there may be no differentiation below the binary level of
focus vs. nonfocus. The point here is that just as prominence per se cannot be reduced
to focus, so are the prominence properties associated with the distinction between lexi-
cal and grammatical expressions irreducible to focus properties alone.
It may appear to be a problem that we base the categorical distinction between lexi-
cal and grammatical expressions on discourse prominence, which is a scalar phenome-
non. Lexical expressions are not defined in terms of ‘high prominence’, however, nor
are grammatical expressions in terms of ‘low prominence’. Rather, they are defined in
terms of, respectively, being or not being potentially discursively primary (i.e. the main
point of an utterance).
3.2. DISCOURSE PROMINENCE AND LINGUISTIC STRUCTURE. The theory finds a natural
place within a general theory of usage-based structure. The defining property of gram-
matical expressions—their conventional discursively secondary status—links up prop-
erties pertaining to the usage and function of grammatical expressions with structural
properties.
On the one hand, the conventional property must be understood in relation to usage:
it is distilled out of usage, but as soon as it has been established it also restricts subse-
quent usage patterns, because convention now also plays a role in where prominence
goes in actual communication.
On the other hand, the conventional property unique to grammatical expressions
links up with structural properties. In particular, it follows from the conventional prop-
erty that grammatical expressions are DEPENDENT upon one or more cooccurring ex-
pressions (lexical or grammatical). Since they are by convention secondary, they
require something with respect to which they can be secondary in actual communica-
tion. This goes for grammatical expressions of all levels of morphosyntactic complex-
ity. In 14 above (Michael has always loved to swim), for instance, the morphologically
simple present-tense suffix -s is dependent upon the auxiliary have with respect to
which it is secondary, but the auxiliary is itself dependent upon the verb love with re-
spect to which it is secondary. Ultimately, schematic constructions are dependent upon
the morphological material they require to be filled in with, and with respect to which
they are secondary. For instance, the declarative construction in English conventionally
expresses, as discursively secondary, the illocutionary meaning of ‘assertion’ (in the
weak sense that contrasts with ‘question’, rather than a strong sense that would contrast
with e.g. ‘suggestion’). But in order to convey this meaning, it requires its slots to be
filled in with morphological material with respect to which it can be secondary.
Since lexical expressions are by convention potentially primary, they are not (neces-
sarily, at least) dependent upon cooccurring expressions. As a special case, they may be
primary by virtue of the fact that they occur as the only expression in an utterance. The
complex expression the house is on fire! can thus be reduced to the lexical expression
fire!—by paring away all other meanings while keeping the primary element of the
message intact.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 13

3.3. GRAMMATICAL MEANING AND LEXICAL MEANING. The definitions of grammatical

and lexical expressions central to the proposed theory are bound up with definitions of
the functions or content sides of the two kinds of expressions. We refer to these func-
tions or content sides as GRAMMATICAL MEANING and LEXICAL MEANING respectively,
without a priori assuming that they differ in any particular respect (for instance, such
that lexical meaning is ‘conceptually rich’ while grammatical meaning is ‘abstract’).
The definitions are as follows.
(15) LEXICAL MEANING: Lexical meaning is by convention capable of being dis-
cursively primary.
(16) GRAMMATICAL MEANING: Grammatical meaning is by convention discur-
sively secondary.
With these definitions, the theory highlights one similarity and one difference between
grammatical and lexical meanings.
The similarity is that both grammatical and lexical meanings are conventional.
Grammatical and lexical meanings thus contrast with conversational implicatures that
are entirely situationally determined. Just like conventional meanings, however, con-
versational implicatures can be either discursively primary or discursively secondary.
Among the classic examples, it’s cold in here can be mentioned as a case where the rel-
evance-triggered implicature ‘perhaps we should close the window’ is the speaker’s
primary point. In contrast, there’s a garage round the corner triggers an implicated
meaning, ‘the garage is open now’, that is not the primary point of the message.
The difference between grammatical and lexical meanings has to do with the expres-
sions they are found with. It follows from the definition above that grammatical mean-
ing can be conventionally associated not only with grammatical expressions, but also
with lexical expressions. Lexical expressions need to have lexical meaning—otherwise
they would not be capable of being discursively primary, and this would by definition
disqualify them as lexical expressions. But lexical expressions can additionally have
grammatical meaning. Consider, for instance, the English verb form ran. This form can
be analyzed as having (at least) two conventional meanings: ‘run’ and ‘past’. Arguably,
the former of these meanings is lexical in that it is potentially primary, but the latter is
grammatical in that it is secondary by convention:8 it cannot, as long as conventions are
adhered to, be used to communicate the meaning ‘past’ as the main point of a linguistic
message. Similarly, tree, in addition to its lexical meaning, conventionally expresses its
membership in the grammatical class ‘countable noun’ (i.e. ‘individuated member of a
category’), which cannot constitute discursively primary meaning. In contrast to gram-
matical meaning, lexical meaning can be found only with lexical expressions. The pos-
sibility of a grammatical expression with lexical meaning is excluded because lexical
meaning makes the expression by convention potentially primary.
3.4. SYMPTOMS OF AND TESTS FOR GRAMMATICAL STATUS. In many cases, speakers can
make fairly clear intuitive judgments about whether a given linguistic expression is
conventionally suited to convey the primary point of an utterance. For instance, it is
hard to imagine the suffix -ly in happily as conveying the main point (cf. §4). Appeals
to such intuitive judgments enter into the discussion below. The empirical base of the
theory, however, will be strengthened to the extent that such judgments can be sup-
ported by clear-cut diagnostic procedures.

8 The meaning element ‘past’ may be understood as indexically rather than symbolically coded (cf. Ander-

sen 2010).
14 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

It follows from the definitions given in the beginning of §3 that identification of an

expression as lexical requires evidence that it can be discursively primary in actual
usage, while identification of a given expression as grammatical requires evidence that
it cannot be (because it is secondary by convention). In this section, we first point out
two diagnostic symptoms of grammatical status, and then suggest two concomitant sets
of tests that can be used to provide such evidence. Both sets of tests exploit the fact that
languages possess conventional means for singling out which expression is intended to
have discursively primary status in a given utterance (cf. §3.1). Subsequently, we dis-
cuss limitations and exceptions to the tests.
The first diagnostic symptom involves ‘focalizability’. As described in §3.1, the focal
expression is the primary element of the utterance (at the appropriate level of granularity).
Focalizability is thus a symptom of lexical status: only lexical—potentially primary—ex-
pressions can be assigned discursively primary status by focalizing expressions. It follows
that nonfocalizability is a symptom of grammatical—conventionally secondary—status.
(17) NONFOCALIZABILITY AS SYMPTOM OF GRAMMATICAL STATUS: Grammatical
expressions cannot be assigned discursively primary status by focalizing
expressions.
This symptom can be brought out in a range of different tests that depend on language-
specific means of marking focus. In English, focalizing expressions include cleft con-
structions, pseudo-cleft constructions, narrowly focal stress, and focus particles like
only, just, and even. For English, a list of focus tests thus includes those in 18.
(18) FOCUS TESTS (English)
a. Grammatical expressions cannot independently occur in the focal posi-
tion of cleft constructions.
b. Grammatical expressions cannot independently occur in the focal posi-
tion of pseudo-cleft constructions.
c. Grammatical expressions cannot independently receive narrowly focal
stress (this is a familiar feature of grammatical expressions; cf. Bybee at
al. 1994:7).
d. Grammatical expressions cannot independently occur in the semantic
scope of focus particles like only, just, and even.
It is an empirical question to what extent one can find counterparts to these tests in other
languages. For instance, languages may lack cleft and pseudo-cleft constructions. How-
ever, as long as a given language has sentence-internal means for indicating focus (for in-
stance, focus particles), it will be possible to establish such tests for grammatical status.
The second symptom of grammatical status involves what we call ‘addressability’: to
address (the meaning of) a linguistic expression is to take it up in a subsequent utter-
ance. Like focalizability, addressability is a symptom of lexical status: only lexical—
potentially primary—expressions can be assigned primary status by being addressed
in subsequent discourse. Unlike focalizability, however, addressability is a sentence-
external and discourse-based symptom. It trades on linguistic means for questioning,
confirming, denying, responding to, or otherwise commenting upon or ‘addressing’ an
expression belonging to a previous utterance in a subsequent utterance, and rests on the
assumption that addressing an expression belonging to a previous utterance entails as-
signing it discursively primary status. This assumption reflects the same intuition as the
loi d’enchainement (‘law of linking’) suggested by Ducrot (1972:81) in his speech acts-
inspired semantics, which says that presupposed elements cannot form the link to the
next utterance, and also as the idea, central to Posner’s (1972) theory of Kommentieren
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 15

(‘commenting’) and Kommentierbarkeit (‘commentability’), according to which the

fact that something in an utterance is commented upon shows that this element has the
‘highest communicative relevance’ (Posner 1972:11). The same idea is found in Dahl’s
(2004:86) remark on the ‘inaccessibility’ of inflectional meaning. In accordance with
this assumption, nonaddressability is a symptom of grammatical—conventionally sec-
ondary—status.
(19) NONADDRESSABILITY AS SYMPTOM OF GRAMMATICAL STATUS: Grammatical ex-
pressions cannot be assigned discursively primary status by being addressed
in subsequent discourse.
This symptom can be exploited in a number of addressability tests that depend on
language-specific means of addressing and thus assigning discursively primary status to
an expression in a previous utterance. For English, a list of addressability-based tests
includes those in 20.
(20) ADDRESSABILITY TESTS (English)
a. Grammatical expressions cannot independently be questioned by WH-
questions (the close relation between addressability and focalizability can
be seen from the fact that WH-constituents are structurally marked as
focal: by being FOCALIZED in the subsequent clause, the element in the
previous clause is thus ADDRESSED as primary).
b. Grammatical expressions cannot independently be referred to anaphori-
cally or cataphorically (i.e. they do not introduce discourse referents that
can be picked out by indexical expressions).
c. Grammatical expressions cannot independently be questioned by yes-no
questions.
As in the case of focus tests, addressability tests do not work in the same way in all
languages.
Let us illustrate how the tests work by applying them to an uncontroversial example
of a grammatical expression: the English auxiliary gonna, illustrated in 21.
(21) Jones is gonna call her tomorrow.
Gonna comes out as a grammatical expression by the addressability tests. In hearing 21,
one cannot reply how? and intend to be understood as meaning ‘how, gonna?’. The
reply how? addresses the manner in which a state of affairs takes place, and can only be
understood as addressing call: ‘how, call? (—by cellphone or by Skype?)’. By the ad-
dressability tests, accordingly, call differs from gonna in being lexical. The addressabil-
ity tests go naturally with the fact that in every language there appears to be a limited set
of interrogative devices, and this set never includes devices for questioning the mean-
ings of grammatical expressions (cf. Boye & Harder 2007:575). Gonna is grammatical
also by the focus tests. For instance, it cannot be brought into narrow focus by means of
sentence stress (but see the discussion of putative exceptions below). This is a familiar
feature of grammatical expressions (cf. Bybee et al. 1994:7). Thus, whereas it would be
straightforward to give Jones narrow focus (analogously to Mitchell in 10) and also
straightforward to read sentence stress on tomorrow as indicating narrow focus, gonna
is a bad candidate for narrow focus.
In contrast to gonna, a full verb like kill comes out as lexical by both the addressabil-
ity tests and the focus tests. It can be put in narrow focus by means of stress (22), be ad-
dressed by how (23), and be inserted in the focal slot of a pseudo-cleft construction (24)
or a focus particle (25).
16 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

(22) I’ll KILL anyone who insults my mother.

(23) —I am fully prepared to kill.
—How? (‘how, kill?’; alternatively ‘how, prepared to kill?’)
(24) What I want is to kill (him).
(25) I am ready to even kill.
Compare also ‘parenthetical’ I think in 26 and Jill regrets in 27.
(26) Jill left her husband, I think.
(27) Jill regrets that she left her husband.
By the addressability test, parenthetical I think comes out as grammatical (Boye &
Harder 2007): in hearing 26, one cannot ask a yes-no question such as really? and in-
tend to be understood as meaning ‘really, do you think so?’; yes-no questions address
the clause Jill left her husband. In contrast, Jill regrets comes out as lexical: in hearing
27, it is perfectly possible to ask really? and intend to be understood as meaning ‘really,
does she regret?’.9
As mentioned, there are limitations to some of the tests. They result from the fact that
focus markers and linguistic means for addressing expressions in previous utterances
are often bound up with structural constraints. For instance, WH-questions and
anaphoric pronouns address (sub)constituents only, and yes-no questions apply only to
clauses. Similarly, cleft and pseudo-cleft constructions are structurally constrained to
operate at or above the phrasal level (cf. Huddleston & Pullum 2002:1417–19). This
means that the tests cannot in all cases serve their diagnostic purpose. For instance,
focus tests 18a and 18b, which draw on cleft and pseudo-cleft constructions respec-
tively, fail to distinguish grammatical expressions from lexical expressions that are not
(sub)constituents. Their direct diagnostic value is thus limited to cases where the ex-
pression under investigation is not for obvious distributional reasons unqualified for
filling the focal slot of cleft and pseudo-cleft constructions.
The diagnostic value is obviously higher the more widely applicable the test is, and
this makes focus test 18c, which draws on focal stress, the most useful test. Thanks to
stress-signaled narrow focus, there is a possibility of diagnosing all lexical expressions
as such, since subphrasal lexical expressions can be brought into narrow focus by this
means. By focus test 18c, for instance, attributive adjectives like wonderful are lexical,
whereas the definite article the is grammatical.
(28) Joe is a WONDERFUL man.
(29) ?What is THE weather like?
But we cannot simply identify all grammatical expressions as those that fail the test—
because for distributional reasons not all grammatical expressions can be naturally put
to the test. Take, for instance, the English plural suffix -s in 30.
(30) The streets are wet.
Obviously, because it is not a constituent, it cannot be focalized by cleft or pseudo-cleft
constructions or by focus particles, and since it is not an independent syllable it is im-
9 It is important that addressability and nonaddressability are understood as applying to linguistic elements.
In hearing 26 (Jill left her husband, I think), it is of course perfectly possible to question the speaker’s epis-
temic stance by asking do you really think so?. Such a question does not, however, address the linguistic ele-
ment I think in 26: it does not tie up with the meaning ‘I think’ AS EXPRESSED BY THE PARENTHETICAL. The same
question may follow an utterance like Jill left her husband, in which there is no I think. Thus, locutors are free
to pick up and dwell on all kinds of meaning expressed or implied in the preceding discourse (cf. Bohnemeyer
2009 on temporal anaphora in tenseless languages), and independently of preceding discourse they may in-
troduce new meanings. Outside of metalinguistic and contrastive contexts (cf. §3.4), however, they are not
free to address all linguistic elements in the preceding discourse.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 17

mune even to narrow-focal stress. Thus, none of the focus tests discussed above can re-
ally be applied. Therefore, we are left with intuitions to tell us whether -s is potentially
discursively primary. We believe intuitions are firm enough to warrant the answer ‘no’.
One might choose to claim that the mere fact that such cases cannot be brought to pass
the test for lexical status shows that they are grammatical—but we prefer to say that in
certain cases we have to rest content with intuitive judgments unsupported by tests.
There is also a set of exceptions to the tests. These exceptions are well defined, how-
ever, as the set of contexts in which focalizing and addressing serve metalinguistic and
contrastive purposes. In both metalinguistic and contrastive contexts, grammatical ex-
pressions can be brought into narrow focus, and at least in metalinguistic contexts, they
can also be addressed (cf. Boye & Harder 2007:575). What makes these contexts ex-
ceptional is that here focalizing and addressing do not assign discourse prominence in
the sense in terms of which grammatical and lexical expressions are defined: they do
not serve the purpose of singling out an element as discursively primary in relation to
SYNTAGMATICALLY related elements (cf. §3.1).
Consider first two cases of metalinguistic focus.
(31) I said Smith hatED her—not hates.
(32) —Has he called her?
—No, he is GONNA call her.
Examples 31 and 32 differ in a clear-cut way from standard cases of narrow focus of the
kind discussed in §3.1 in that the focal expressions, -ed and gonna, are not just used, but
also mentioned. Something similar can be said of 33 in which the question what? can be
read as metalinguistically addressing gonna: the question concerns not just the meaning
of gonna, but also the appropriateness of using the word gonna.
(33) —He is gonna call her.
—What? He has called her already, hasn’t he?
In any case, focalizing gonna in 32 and addressing it in 33 is clearly not a matter of sin-
gling gonna out as discursively primary in relation to its syntagmatic context. Rather, it
is a matter of drawing attention to the appropriateness of the expression used—possibly
in relation to PARADIGMATICALLY related expressions. An analogous example is when
metalinguistic stress overrides conventional lexical accent, as in 34. This is clearly not
a matter of assigning primary status to the stressed element (im-) in relation to its syn-
tagmatic context (-port), but rather a matter of drawing attention to the appropriateness
of the stressed element im- as contrasting with the paradigmatic alternative ex-.
(34) I said they IMport (not EXport) toxic waste.
Consider now a case of contrastive-focus readings.
(35) This is what we HAVE done; now I’ll talk about what we’re GONNA do.
Contrastive focus can be described as superimposed on ordinary focus. The superim-
posed property of contrastive focus involves two characteristic properties (Erteschik-
Shir 2007:48–49). First, unlike ordinary focus, it can be assigned to something that is
not a syntactic category. Second, it semantically invokes the contrast set as part of the
interpretation (in 35, gonna is understood as ‘rather than have’).10 Like metalinguistic

10 This is different from the sense in which alternative semantics (cf. §3.1) describes focus generally as in-

voking the presence of potential alternatives, also in the case of noncontrastive focus. The crucial property of
focus is that it brings the information state forward. The focused part of the message is privileged because
something other than the actual expression might have been inserted here, while the rest of the message is
presupposed. In the case of contrastive focus, the point includes denying the alternatives, not simply specify-
ing the ‘true’ value.
18 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

focus, then, contrastive focus does not serve the purpose of highlighting an expression
in relation to its syntagmatic context. Rather, it highlights the focal expression in rela-
tion to its PARADIGMATIC alternatives. This is why contrastive-focus contexts represent
an exception to the application of the focus tests.11 The theory proposed in the present
article thus offers empirical procedures for determining whether a given expression is
lexical or grammatical—without claiming that the identification of grammatical expres-
sions is in all cases a straightforward matter.
There is an extra dimension of the problem of identification, which would be present
no matter what criterion was involved: identifications depends on the individuation of
conventionalized expressions. How many different conventional expressions are there
with the phonological form have, for instance? The obvious difficulties lead some au-
thors to abandon all hope of making any distinctions and to recognize only gradience,
but as discussed in §2, this does not really solve the problem. The problem also has a di-
achronic counterpart: at what precise point in the history of English was it possible to
distinguish auxiliary from lexical have? Some scholars might wish to introduce a split
between expressions as soon as even the most subtle distributional differences arise.
Others might wish to lump them together all the way through. As will be apparent, the
more disposed one is to recognize as a special, conventional expression a usage type in
which an expression has secondary status, the more grammatical expressions one will
recognize in a language. Conversely, the more one is disposed to maintain a monolex-
emic interpretation that includes the lexical, potentially primary, source, the fewer
grammatical expressions there will be. We take no stand on this issue, because the the-
ory we propose does not depend on it.
4. THE THEORY PROVIDES A GENERALIZATION OVER EXAMPLES OF GRAMMATICAL EXPRES-
SIONS. The first of the arguments that we present for the theory is that it covers standard
and uncontroversial examples of grammatical expressions. This is a fundamental argu-
ment for the theory. If there were not a considerable degree of overlap between what
everybody agrees is grammatical and what is classified as grammatical on the basis of
the theory, it might be claimed that the theory is not a theory of grammatical status, but
of something else.
It has already been demonstrated, in §3.4, that the English auxiliary gonna and the ar-
ticle the pass the tests for grammatical status that are part of the theory. Similarly, af-
fixes such as past-tense -ed, clitics such as the ‘group genitive’, and auxiliaries such as
progressive be come out as grammatical by the proposed theory: intuitively, at least (cf.
§3.4 on the limitations to the proposed tests for grammatical status), they cannot consti-
tute the main point of an utterance. For instance, the main point of uttering 36 could not
plausibly be to convey the meaning of ‘past’.
(36) She hated him.
The same goes for schematic constructions. By the proposed theory, they are unequivo-
cally grammatical. As discussed in §3.2, for instance, the English declarative as a
schematic grammatical expression has the illocutionary meaning ‘assertion’, but the
main point of uttering a declarative sentence like that in 36 is not to convey that an as-

11 There is also another type of case that may appear to constitute a counterexample: replies of the form he
HAS or they DID to observations like has he done his homework? or they didn’t make it. In such cases, however,
the auxiliaries has and did do not convey merely their secondary grammatical content; they also stand as pro-
forms for the whole asserted proposition (cf. also the discussion in n. 12 of the grammatical status of words
like probably when they stand alone as answers to a question).
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 19

sertion is being made. Since the theory thus captures cases that everybody considers
grammatical, it provides an explicit rationale for what used to be an implicit consensus.
As an example of where this rationale can contribute to the discussion, consider de-
rivational morphemes. Hopper and Traugott (2003) make a distinction between lexical
and grammatical derivational morphemes. The former, for instance English -ling in
duckling, are said to ‘add a meaning component without affecting the category in ques-
tion’. The latter, for instance -ly in happily, ‘change the category of the word’ (Hopper
& Traugott 2003:5). From the point of view of the theory proposed here, both types of
derivational morphemes are clearly grammatical expressions, secondary and ancillary
to the stems to which they are attached. Neither can be independently addressed or fo-
calized outside of metalinguistic and contrastive contexts. They differ in terms of their
meanings, not in terms of their conventional discourse-prominence potential.
However, the theory also includes among grammatical expressions cases that are not
standardly associated with grammatical status (cf. also Egan 2010:136–39 on the possi-
ble grammaticalization of fail to into an auxiliary). Adverbs as a category are often
taken to be lexical expressions. In FUNCTIONAL GRAMMAR and FUNCTIONAL DISCOURSE
GRAMMAR (Dik 1997, Hengeveld & MacKenzie 2008), for instance, they belong to the
group of lexical ‘satellites’ as opposed to the group of grammatical ‘operators’. With
some adverbs, this view is perfectly compatible with the theory proposed here. Manner
adverbs like quickly, temporal adverbs like yesterday, and locational adverbs like here
are all potentially primary. They can be addressed by how, when, and where respec-
tively. But by our definitions, wide-scope adverbs like however, probably, and evidently
are clearly grammatical expressions. They cannot be singled out as discursively pri-
mary by being addressed or focalized (cf. §3.4). They are by convention secondary and
thus dependent upon an accompanying expression in relation to which they can be sec-
ondary in actual communication.12
In other cases, the theory excludes from the class of grammatical expressions those
that have been associated with grammatical status. A case in point is the Danish modal
verb kunne ‘can’. Danish grammarians have frequently treated kunne as an auxiliary—
a grammatical verb (Wiwel 1901:151, Diderichsen 1946:169, Davidsen-Nielsen 1990:
22). As argued in Boye 2010, however, kunne is frequently discursively primary. An-
other example is demonstratives. Mainly because they constitute a closed class, demon-
stratives are considered grammatical and mentioned as illustration cases of the relation
between grammatical and lexical types of meaning by Talmy (2000:25). However, as
pointed out by Diessel (1999:150–52, 2006), the conception of demonstratives as gram-
matical does not capture their unique properties. Among these are their very early ap-
pearance in acquisition, their nonderived status, and their universal occurrence as
means of coordinating joint attention. Demonstratives, moreover, are the SOURCES of a

12 The fact that, for instance, probably can be used on its own, as a reply to a yes-no question and a signal
of assent, might be viewed as a problem for this analysis.
(i) —Is he right?
—Probably!
Even in (i), however, the meaning of the adverb probably does not express the main point. Rather, probably!
must be understood as a short form of Yes, probably he is right, which expresses as its main point a confirma-
tion of a proposition, and in which probably adds only a secondary and backgrounded qualification. This ac-
count tallies with the fact that *improbably would not be a possible answer, while probably not is okay, with
not as an intervening operator between probably and the elided proposition (adjectives, in contrast, remain
primary and foregrounded, since a negative adjectival reply is okay: not bloody likely!).
20 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

variety of grammaticalization chains, including cases ending up in definite articles and

complementizers. Rather than ‘markers’, Diessel suggests (2006:473, following Brug-
mann & Delbrück 1911:311, Bühler 1934:144) that demonstratives should be under-
stood essentially as ‘particles’ and defined in terms of their unique discourse function
rather than any particular grammatical alignment.13 This discourse definition is suffi-
cient to delimit demonstratives as a closed class.
In support of Diessel’s analysis, demonstratives are clearly nongrammatical by the
addressability and the focus tests. For instance, the demonstrative that can be both ad-
dressed and focalized.
(37) —Look at that.
—What? (Oh, that!)
(38) What I like about him is exactly that.
This is the case also with adnominal demonstratives.
(39) —Look at that picture over there.
—Which picture?
(40) Was it that one you meant?
Thus, demonstratives are by convention potentially primary. Accordingly, they qual-
ify as lexical expressions (cf. Woodworth 1991:285 on demonstratives as lexical
expressions).
Like demonstratives, personal pronouns are often considered grammatical expres-
sions. They make up a closed class, and they can be derived diachronically from nouns
or noun phrases. This is the case, for instance, with the Spanish pronoun usted ‘you’
(singular, polite), which developed from vuestra merced ‘your grace’. However, while
usted and, for instance, the English pronoun he are certainly not prototypical nouns,
neither this nor their closed-class membership or diachronic sources makes them gram-
matical by our theory (cf. §2 on gradience). They clearly retain the potential of being
discursively primary. They can fill out a whole argument slot on their own, which
makes them part of the semantic core of a simple clause. The pronoun he may be ad-
dressed as well as brought into focus (cf. it was he who did it), which in accordance
with the tests introduced in §3.4 classifies it as lexical, even if it may typically be dis-
cursively secondary.
Our noncommittal stance on how to individuate expression types (cf. §3.4), however,
entails that even though the distribution of pronouns includes lexical subtypes, other sub-
types may be considered grammatical. The two sets of pronominal forms in French may
throw light on this: moi, toi, lui (corresponding to emphatic me, you, he), and so forth are
potentially primary and thus lexical, while the clitics (or affixes) je, te, il are by conven-
tion secondary and thus grammatical. In English, rather than lumping all instances of he
into one (nongrammatical) expression, one might choose to distinguish grammatical and
lexical subtypes of personal pronouns on distributional grounds. The grammatical vari-
ant of he (which would show familiar signs of grammatical status such as phonological
reduction) would be the translation equivalent of French il, and so forth.

13 Heine and Kuteva (2007) challenge Diessel’s claim that demonstratives are typically nonderived, point-

ing out that they are frequently derived from locational expressions; but if words like there are included in the
demonstrative class, presumably this observation could be rephrased as saying that nominal demonstratives
are frequently derived from a combination of a nominal element and a ‘demonstrative’ adverbial, as in Dan-
ish den der ‘it there’ and den her ‘it here’.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 21

Just as a distinction can be made between lexical and grammatical pronouns, a dis-
tinction can be made between lexical and grammatical prepositions (cf. also Lehmann
2002b:8). The English prepositions of and off have the same source, but while the for-
mer, at least as a possessive marker, is arguably grammaticalized, the latter seems not to
be. Thus, possessive of cannot be addressed, but off can.
(41) —This will be the death of me.
—*Why (of )?
(42) A: —Keep your hands off me.
B: —Why (off )?
A: —Touching implies intimacy!
Just as for pronouns, however, it would be possible in principle to set up distribu-
tional subtypes for other prepositions, and thus to distinguish between grammatical and
lexical cases.
It is natural to ask, finally, what the proposed theory implies for the status of so-called
‘function words’. Function words are generally regarded as a subset of grammatical ex-
pressions, but the class is beset by the same traditional problem of fuzzy boundaries that
we saw in the case of the lexical-grammatical continuum (cf. van Gelderen 2005 on the
cline between lexical and functional words). In our theory, however, it is an empirical
question whether all expressions referred to as ‘function words’ are in fact grammati-
cal—that is, whether they are all by convention discursively secondary. While most func-
tion words may turn out to be grammatical, there are exceptions such as, for instance,
indexicals that can head NPs/DPs. We believe it would be possible to define the category
of function words so as to include words that are lexical in addition to words that are
grammatical by our criterion.
To sum up, we believe the proposed theory provides a generalization over all uncon-
troversial cases of grammatical expressions as well as a means of clarifying the status of
expressions that are found in between uncontroversial cases of grammatical expressions
and uncontroversial cases of lexical expressions. Sometimes, the status assigned to du-
bious cases by the theory may be in conflict with traditional ideas of what is lexical and
what is grammatical, as in the case of demonstratives and the Danish modal verb kunne.
Since, however, these traditional ideas are not based on a coherent theory but remain
pretheoretical and intuition-based, such conflicts may be regarded as a problem for the
tradition rather than necessarily for the theory proposed here (cf. Boye 2010:98–99).
5. THE THEORY PROVIDES A GENERALIZATION OVER EXAMPLES OF GRAMMATICALIZATION.
Just as the theory generalizes over uncontroversial examples of grammatical expres-
sions, it generalizes over uncontroversial examples of grammaticalization. This is an-
other fundamental argument for the theory. As in the case of grammatical expressions,
if there were not a considerable degree of overlap between what everybody considers
grammaticalization and what is classified as grammaticalization on the basis of the the-
ory, it might be claimed that the theory is not a theory of grammaticalization, but of
something else.
Consider again the proposed definition of grammaticalization.
(43) GRAMMATICALIZATION: Grammaticalization is the diachronic change that
gives rise to linguistic expressions that are by convention ancillary and as
such discursively secondary.
In accordance with this definition, a distinction can be made between three basic types
of grammaticalization (cf. Andersen 2008 for an alternative detailed typology of gram-
matical changes).
22 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

The first type of grammaticalization has its source in a lexical expression. Here,
grammaticalization consists in ANCILLARIZATION, a CHANGE IN EXISTING DISCOURSE-
PROMINENCE CONVENTIONS: the potentially primary status of a lexical expression is re-
placed with the secondary status of a grammatical expression.
The second type has its source in a ‘pragmatically conveyed’ meaning with second-
ary status. Here, grammaticalization consists in a CONVENTIONALIZATION OF A DISCUR-
SIVELY SECONDARY MEANING as a property of a new linguistic expression: a linguistic
expression—for instance fixed word order—becomes conventionally associated with a
secondary meaning that was originally part of a pragmatic total message, but not con-
ventionally associated with any linguistic expression.
In the third type of grammaticalization the source is an already existing grammatical
expression. Here, grammaticalization consists in the development of such an expression
into a new grammatical expression distinct from its source.
The third type is dealt with in §8. At present we are concerned with the two other
types. The first type of grammaticalization is the standard or prototype case, captured
by the classical definitions of Meillet (1958 [1912]:131) and Kuryłowicz (1975 [1965]:
52), and in the dictionary definition of Matthews (2007:164). The second, ‘nonstan-
dard’ type was recognized already by Meillet (1958 [1912]:147–48), who mentions fix-
ation of word order as a case of grammaticalization, but some scholars are reluctant to
include it under grammaticalization proper (e.g. Hopper & Traugott 2003:60). As
should be clear, both types of grammaticalization find a natural place in the theory ad-
vocated here. The theory provides a generalization over both types, and at the same
time pinpoints the difference between them. In what follows, we illustrate this in more
detail, dealing first with standard cases of grammaticalization (§5.1) and then with non-
standard cases (§5.2).
5.1. STANDARD CASES OF GRAMMATICALIZATION. As an example of the standard type,
we return to the development of the English prospective auxiliary going to/gonna.
(44) I am goingLEX (in order) to eat. >
I am goingGRAM to (= gonna) eat.
From the point of view of the proposed theory, as mentioned, developments like this in-
volve a change in existing discourse-prominence conventions. The crucial difference
between the input to the change, a progressive form of the verb go, and the output of it,
the auxiliary going to/gonna, is that while the former as a lexical expression can by con-
vention convey its meaning either as primary or as secondary, the latter can only convey
its meaning as secondary. As argued in §3.4, the main point of saying I am going
to/gonna eat is not to indicate prospectivity. Rather, as opposed to lexical go, which
need not be in construction with a following verb form, going to/gonna is dependent on
a verb stem such as eat because it requires an expression in relation to which it can be
secondary in actual communication.
According to the proposed theory, grammaticalization of lexical expressions consists
in the rise of this crucial difference. The bridge between lexical going and grammatical
going to/gonna is the secondary uses of the former. In examples like I am going to eat in
44 (cf. Hopper & Traugott 1993:2), go is accompanied by another verb that can compete
with it for discourse prominence. In actual discourse, the construction in 44 can be used
either to convey ‘THAT I am going’, in which case eat is secondary in relation to go, or to
convey ‘WHY I am going’, in which case go is secondary in relation to eat. Grammatical-
ization is the conventionalization of going as having secondary status. The competition
scenario is illustrated in 45, where boldface marks the expression that comes out as pri-
mary, while LEX and GRAM mark an expression as respectively lexical or grammatical.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 23

(45) The grammaticalization of going

LEXICAL STATE
goingLEX to eat Competition for discourse prominence, and going
wins the competition (so that eat is secondary)
<> Synchronic usage alternation
goingLEX to eat Competition for discourse prominence, and eat wins
the competition (so that going is secondary)
> Grammaticalization: conventionalization of going as
secondary
GRAMMATICAL STATE
goingGRAM to eat Result of grammaticalization: a grammatical descen-
dant of lexical going that is by convention second-
ary
Of course, this scenario does not bring us all the way to gonna. In order to get there, go-
ingGRAM needs, among other things, to fuse with to and subsequently undergo phono-
logical reduction. Moreover, the scenario does not say anything about the change to
prospective meaning. The scenario in 45 concerns only the criterial change whereby a lex-
ical expression acquires grammatical status (cf. §8.1 on continued grammaticalization).
We suggest that all standard cases of grammaticalization of lexical expressions con-
form to a generalized version of the scenario in 46.
(46) Generalized scenario for standard cases of grammaticalization
LEXICAL STATE
XLEX YLEX Competition for discourse prominence, and X wins
the competition (so that Y is secondary)
<> Synchronic usage alternation
XLEX YLEX Competition for discourse prominence, and Y wins
the competition (so that X is secondary)
> Grammaticalization: conventionalization of X as
secondary
GRAMMATICAL STATE
XGRAM YLEX Result of grammaticalization: grammatical descen-
dant X of lexical X that is by convention second-
ary
Compare, for instance, the grammaticalization of going to into gonna with the develop-
ment of adnominal demonstratives into articles (47), the development of cataphoric
demonstratives into complementizers, as in Faroese tað > at (48), and the development
of Afrikaans ek glo ‘I think’ into the evidential particle or adverb glo (49).
(47) that man > the man
(cf. Traugott 1980:49, Diessel 1999:128–29, Heine & Kuteva 2002:109–11)
(48) Faroese (Lockwood 1968:223, from Diessel 1999:124)
eg sigi tað, hann kemur. > eg sigi, at hann kemur.
I say that he comes I say that he comes
‘I say that: he comes.’ ‘I say that he comes.’
(49) Afrikaans (Boye & Harder 2007:591–92; cf. Thompson & Mulac 1991:318)
ek glo hy ryk is. > hy is glo ryk.
I think he rich is he is EVIDENTIAL rich
‘I think that he is rich.’ ‘He is said (supposed, believed) to be rich.’
The four changes—44 and 47–49—differ with respect to both source and result. But in
all cases, the source is an expression that is lexical in the sense of being potentially pri-
24 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

mary, and the result is an expression that is grammatical in the sense of being secondary
by convention (cf. §4 on the lexical status of demonstratives). And in all cases, a pre-
requisite for the change is a construction in which the lexical expression competes with
another lexical expression for discourse prominence—and eventually loses the compe-
tition: the adnominal demonstrative in 47 competes with its head noun (man). The cat-
aphoric demonstrative in 48 competes with the clause to which it refers (hann kemur),
and the matrix clause ek glo in 49 competes with its complement clause (hy ryk is).
We hope to have demonstrated by now that the proposed theory generalizes over ex-
amples of grammaticalization of lexical expressions. The scenario illustrated in 46 is
not just an adequate generalization, however. It also inherently emphasizes the fact that
grammaticalization of lexical expressions occurs in specific constructions (e.g. Leh-
mann 1992:406, Croft 2003:253, Hopper & Traugott 2003:2–3; cf. also §9), and in ac-
cordance with the functional-cognitive view of linguistic structure as arising from
usage, it pinpoints the discourse basis of grammaticalization (cf. §6).
The scenario illustrated in 46 does not, however, say anything about the specific mech-
anisms by which a grammatical expression arises. Two possibilities may be mentioned.
First, there may be a change whereby a particular meaning goes from being potentially
primary to being by convention secondary—the simplest scenario in the standard type.
Second, a secondary meaning of a lexical expression may oust a distinct, potentially pri-
mary meaning of the same expression. A subcase of such a change involves a secondary
meaning that is pragmatically, rather than conventionally, associated with the lexical
source expression and consists in what has been called ‘conventionalization of implica-
ture’ or ‘pragmatic strengthening’ (cf. Traugott 1988).
5.2. NONSTANDARD CASES OF GRAMMATICALIZATION. Nonstandard cases have been
widely discussed (e.g. Hopper 1991:17, Lehmann 2008:207) in relation to a scenario
suggested by Givón (1979:207–22) under the term ‘syntacticization’. Syntacticization
consists in the change from ‘loose parataxis’ to ‘tight syntax’ (Givón 1979:208), or in
the words of Traugott and Heine (1991:2–3), ‘the evolution of syntactic … structure
through fixing of discourse strategies’ (cf. Heine & Reh 1984:28–31 on ‘permutation’,
Lehmann 2002a:153–59 on ‘fixation of word order’, Lehmann 2008 on information
structure and grammaticalization, and Croft 2003:257 on ‘rigidification’ and ‘permuta-
tion’). One of Givón’s examples of such change is the development in ‘noneducated
American English’ from topic to subject, accompanied by the development of a re-
sumptive pronoun into an agreement marker (Givón 1979:209).
(50) My ol’ man, he rides with the Angels. > My ol’ man he-rides with the Angels.
TOPIC PRO SUBJECT AGR-
Another example is the development in Kimbundu of a left-dislocated object topic into
a subject of passive.
(51) Kimbundu (Givón 1979:211)
Nzua, a-mu-mono > Nzua a-mu-mono (kwa meme)
John they-him-saw John they-him-saw (by me)
‘John, they saw him.’ ‘John was seen by me’
From the point of view of the proposed theory, changes like these differ from the standard
cases of grammaticalization discussed in §5.1 in that they cannot be said to involve a
change in existing discourse-prominence conventions. As described by Givón, the
changes in 50 and 51 are changes from a discourse structure into a grammatical syntactic
structure. We may conceive of the input to the change as a pragmatic phenomenon: the as-
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 25

sociation of initial position with the ‘topic-of’ relation follows from the principle that
‘what comes first is what it is all about’. The output, by contrast, is a structural phenome-
non: the association of initial position with the ‘subject-of’ relation is conventionalized.
Clearly, both the meaning of ‘topic-of’ and the meaning of ‘subject-of’ are secondary
by convention. For instance, neither of the two sentences in 50 can be used with the
main purpose of communicating that ‘my old man is the topic/subject of this clause’.
Arguably, then, the developments in 50 and 51 involve the conventionalization of a
hitherto pragmatically conveyed secondary meaning. Clause-initial position arises as a
linguistic expression in that it comes to be a conventional marker of a ‘subject-of’ rela-
tion (in 50 a new agreement marker, he-, simultaneously arises to index the same rela-
tion). In the proposed theory the two cases unequivocally qualify as (nonstandard)
examples of grammaticalization. Example 52 illustrates the scenario for 50.
(52) Syntacticization of the ‘subject-of’ relation in ‘noneducated American
English’
My ol’ man, he rides with the Angels Secondary ‘topic-of’ relation conveyed
TOPIC PRO as pragmatic meaning according to
the pragmatic principle that what
comes first is what it is all about
> Grammaticalization: conventionaliza-
tion of secondary meaning as associ-
ated with a new linguistic expression
(clause-initial position)
My ol’ man he-rides with the Angels Result of grammaticalization: linguis-
SUBJECT AGR- tic expression (clause-initial posi-
tion) that conventionally expresses
its meaning (a ‘subject-of’ relation)
as secondary
As in the case of the scenario illustrated in 45 in §5.1, the scenario illustrated in 52 does
not say everything there is to say about the development it covers. It does not capture,
for instance, the fact that the grammaticalization of clause-initial position in 50 is ac-
companied by a change of meaning from ‘topic’ to ‘subject’. Nor does it pinpoint the
exact mechanisms by which pragmatic discourse structure gives rise to syntactic struc-
ture. However, it is not intended to capture this. The scenario in 52 is only intended to
capture what makes the change in 50 a case of grammaticalization.
We hope to have demonstrated that the proposed theory generalizes also over non-
standard cases of grammaticalization. Furthermore, it provides a precise characteriza-
tion of what is common to and what is different in standard and nonstandard cases. Both
types of grammaticalization have as their output an expression that is grammatical in
that it is by convention discursively secondary. But they have different types of input
and thus different ways of arriving at the output.

6. WHAT QUALIFIES A LEXICAL EXPRESSION FOR GRAMMATICALIZATION? The proposed

theory allows us to qualify and provide a preliminary answer to the question of what
qualifies a lexical expression for grammaticalization—that is, an answer to the question
of what conditions standard cases of grammaticalization like the ones discussed in §5.1.
As discussed above, it has often been noted that standard examples of grammatical
expressions have meanings that belong to a limited range of notional domains (e.g.
Heine et al. 1991:32–39, Bybee et al. 1994:10, Slobin 1997, Croft 2003:225; cf. §2).
26 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

Underlying such definitions is a view according to which the motivation for grammati-
calization is notional (cf. phrases like Slobin’s 1997 ‘grammaticizable notions’ or
Croft’s 2003:225 ‘grammatical concepts’). A few attempts to narrow down what quali-
fies a notion for grammaticalization have been made. Lyons (1968:438) takes general-
ity to be a crucial feature. By contrast, Heine and colleagues (1991) characterize the
notional ‘source structures’ (‘concepts’ and ‘propositions‘) of grammaticalization as
having to do with ‘some of the most elementary human experiences’ (p. 33) and as
being ‘basic to human experience’ (p. 36). Arguing convincingly against granting fre-
quency the ultimate responsibility for grammaticalization (pp. 38–39), they hypothesize
that what makes ‘source concepts’ ‘eligible’ for grammaticalization ‘is the fact that they
provide “concrete” reference points for human orientation that evoke associations and
are therefore exploited to understand “less concrete” concepts’ (p. 34). In a similar vein,
Bybee and colleagues speculate ‘that rather than generality, it is the reference plane of
basic, irreducible notions—whether they concern existence or movement in space or
psychological or social states, perspectives, and events—which serves as the basis for
grammatical meaning in human languages’ (1994:10; cf. Traugott 1982:246).
These characterizations are all quite vague, and, as pointed out by Slobin (1997:295–
96), they do not account for the fact that notions found with grammatical expressions
are also found with lexical expressions (cf. Lyons 1968:438, and §2). As for the ques-
tion of what makes a notion grammaticalizable, it seems appropriate to conclude with
Slobin, then: ‘At present there is no useful answer to this question beyond an empiri-
cally based list of the notions that receive grammatical expression in the languages of
the world’ (1997:295).
The theory we propose allows us to come up with a preliminary answer, which may
at least set the stage for further research. As discussed in §3.2, the theory links up func-
tion and language use with linguistic structure. In accordance with the theory, we argue
below that not only function and usage but also structural considerations play a role in
grammaticalization. In other words, a lexical expression must qualify not only func-
tionally or notionally but also structurally for grammaticalization.
The structural qualification ultimately derives from the definitions of lexical expres-
sions as potentially primary and grammatical expressions as secondary by convention.
As discussed in §5.1, it is a prerequisite for standard cases of grammaticalization that
there is a structural possibility of competition for discourse prominence; and when
grammaticalization occurs, this competition results in a new structural relation. Con-
sider the generalized scenario for standard cases of grammaticalization in 46, repeated
here for convenience as 53.
(53) Generalized scenario for standard cases of grammaticalization
LEXICAL STATE
XLEX YLEX Competition for discourse prominence, and X wins
the competition (so that Y is secondary)
<> Synchronic usage alternation
XLEX YLEX Competition for discourse prominence, and Y wins
the competition (so that X is secondary)
> Grammaticalization: conventionalization of X as
secondary
GRAMMATICAL STATE
XGRAM YLEX Result of grammaticalization: grammatical descen-
dant X of lexical X which is by convention sec-
ondary
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 27

In accordance with this scenario, in order for a lexical expression, XLEX, to develop into
a grammatical expression, XGRAM, it needs to be in construction with another lexical ex-
pression, YLEX, with which it can compete for discourse prominence. This is the struc-
tural qualification.
(54) STRUCTURAL QUALIFICATION FOR THE GRAMMATICALIZATION OF A LEXICAL EX-
PRESSION: In order for a lexical expression to qualify for grammaticalization,
there has to be a structural possibility for competition for discourse promi-
nence in the form of an adequate accompanying lexical expression.
It is an empirical question which classes of expressions may compete with each other
for discourse prominence. Grammaticalization of verbs into auxiliaries obviously pre-
supposes a competition between lexical verbs. This was in effect noted almost thirty
years ago by Bolinger (1980:297): ‘The moment a verb is given an infinitive comple-
ment, that verb starts down the road to auxiliariness’. Accordingly, English going could
not grammaticalize into an auxiliary in the construction in 55.
(55) I am going (to Rome).
The reason is that in 55 there is no other verb with which going can compete for dis-
course prominence.
Bolinger’s observation is not entirely accurate, however. A verb does not start down
the road to auxiliariness as soon as it is given a verbal complement. The possibility of
competition for discourse prominence is necessary but not sufficient for a lexical ex-
pression to grammaticalize. In order to grammaticalize, the lexical epression also has to
lose the competition. And it has to lose so badly that it is eventually conventionalized
as a loser.
Only losers qualify for grammaticalization, then, and this brings us to the next ques-
tion: What qualifies a lexical expression functionally or notionally for grammaticaliza-
tion? We are now in a position to reformulate this in a more precise way: What makes a
lexical expression come to be conventionalized as secondary?
Clearly, frequency plays a role here. One role of frequency has been, and would have
to be, acknowledged in any theory of grammaticalization: in order for an expression to
come to be conventionalized as grammatical, it must be used with a certain frequency.
In light of the proposed theory, a second role may be hypothesized that is entirely over-
looked in the literature. It may be hypothesized that in order for an expression to come
to be conventionalized as secondary, it must be used with a high frequency with sec-
ondary status RELATIVE TO ITS USES WITH PRIMARY STATUS. Consider again English going
in its lexical ‘movement’ sense and in combination with an adverbial of purpose.
(56) I am going to eat.
Even if going in this construction were frequently used with secondary status, this pos-
sibly would not make the verb grammaticalize into a prospective marker if it were also
frequently used with primary status. In order to be conventionally associated with sec-
ondary status, an expression may need to occur predominantly with this function.
The question now is: Which functional or notional property qualifies a lexical expres-
sion for being used frequently, in both of the two aforementioned ways, as secondary?
We agree with Lyons that generality is part of the answer. In order to be used with some
frequency, in the first of the two aforementioned ways, a lexical expression cannot be too
semantically specific and idiosyncratic (it is not an accident that words like gargoyle or
defenestrate are not among attested instances of grammaticalization). Generality cannot
be the whole answer, though. Food, fish, and furniture are all semantically at a general
(superordinate) level, but nevertheless disinclined to undergo grammaticalization.
28 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

What our theory suggests is that grammaticalization candidates must be generally ap-
plicable with SECONDARY status. In order to undergo grammaticalization, they must be
suitable for enhancing, in a generalizable way, the functional potential of an accompa-
nying lexical expression. This, we suggest, is the other part of the answer. With the ‘sec-
ond violin’ analogy: only instruments that serve well as secondary accompaniments to
the lead instrument get recruited for such functions.
(57) FUNCTIONAL QUALIFICATION FOR THE GRAMMATICALIZATION OF A LEXICAL EX-
PRESSION: In order for a lexical expression to qualify for grammaticalization,
it must have a functional or notional property that makes it enhance the func-
tional potential of an accompanying expression in relation to which it is
secondary.14
If we look at the staple examples of grammatical categories such as tense, aspect, mood,
number, case, their semantic contribution has an obvious ancillary ‘enhancement’ poten-
tial: a proposition such as ‘the house BE on fire’ is more useful if tensed, for example,
and the information ‘Joe DRINK’ is enhanced if we know whether it is habitual or in
progress. Also, it is plausible that temporal locations (tense) and contours (aspect) will
frequently be more functionally useful as accompaniments to an event whose location
and contour they specify—rather than standing on their own, or being the primary point.
Note that there is no claim that these notional types of meaning cannot ever be the
primary point. For instance, the lexical expression more than one and the grammatical
plural add the same descriptive information. In fact, it may in some cases be practical to
have both grammatical and lexical expressions of the same notion at one’s disposal. For
a detective who has just established that a crime cannot have been committed by a sin-
gle person, it would be more appropriate to highlight this fact by using the lexical ex-
pression, as in there is more than one (criminal) involved, than to express the plurality
as secondary information by means of -s, as in there are criminals involved.
This does not necessarily contradict other theories proposed. It may well be that lex-
ical expressions that undergo grammaticalization ‘provide “concrete” reference points
for human orientation that evoke associations and are therefore exploited to understand
“less concrete” concepts’ (Heine et al. 1991:34). It may be that they have to do with ‘the
reference plane of basic, irreducible notions’ and ‘concern existence or movement in
space or psychological or social states, perspectives, and events’ (Bybee et al. 1994:10).
In light of the proposed theory, however, it is crucial that they also meet the structural
and functional qualifications described above. What in our view disqualifies color
terms from ever becoming grammaticalized, as pointed out by Talmy (cf. §2), is not that
they are not sufficiently general or basic in meaning, but that it is not sufficiently fre-
quently relevant to add a secondary accompanying specification of color to a message.
7. FEATURES STANDARDLY ASSOCIATED WITH GRAMMATICALIZATION. As argued in §2,
grammatical expressions and grammaticalization cannot be defined in terms of any
phonological, semantic, or morphosyntactic features. No such features are both invari-
antly and exclusively found with grammatical expressions and grammaticalization.
However, the following are more or less frequently found with grammatical expressions.

14 When we say ‘enhance the functional potential’, to pursue the violin analogy, we intend to describe a

‘support relationship’ analogous to the functional relation between a second violin and the first violin. Instead
of looking for what the grammatical expression does in itself, the point is to ask what a grammatical expres-
sion does for the usefulness of the element in relation to which it is secondary. A second violin does not con-
tribute a melody of its own, but provides a background that contributes to the musical effect of the whole.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 29

• boundness
• phonological reduction
• semantic reduction (bleached, generalized, or abstract meaning)
• closed-class membership
• obligatoriness
Below we argue that the proposed theory provides a coherent motivation for why this is
so (cf. Boye & Harder 2009:32–38).
7.1. BOUNDNESS. Boundness cannot in itself serve to define grammatical expressions,
since lexical material can be bound. In an English indicative sentence such as Joe left,
for instance, the verb left is bound to the subject (in this case, Joe) in the sense that it
cannot occur without it and still be an indicative sentence. In fact, with the exception of
holophrases like ouch and hurrah, which constitute a whole utterance on their own, all
expressions can be said to be bound in different ways to those expressions that they
combine with. However, whereas boundness is a variable property of lexical items, it is
an invariant property of grammatical items. The proposed theory provides a motivation
for why this is so.
From the point of view of the theory, the boundness of grammatical expressions is
a direct consequence of the fact that they are by convention secondary. Grammatical
expressions require the cooccurrence of other expressions with respect to which they
can be secondary. In our standard example, the English auxiliary gonna is dependent
upon an infinitive with respect to which it can express its ‘prospectivity’ meaning as
secondary.
(58) *I am gonna go.
(59) *I am gonna.
Likewise, as discussed in §3.2, schematic constructions as grammatical expressions are
dependent upon the lexical and morphological material they require to be filled in with,
and with respect to which they express their meaning as secondary.
The theory also allows us to see a difference between the boundness of grammatical
expressions and the boundness of lexical expressions. Lexical expressions are not
bound in the sense in which grammatical expressions are. They are not dependent upon
other expressions with respect to which they can be ancillary and secondary.
7.2. PHONOLOGICAL REDUCTION. Like boundness, phonological reduction cannot
serve to define grammatical expressions. Lexical expressions tend to undergo phono-
logical reduction when they are used frequently enough. Reportedly, for instance, a bus
driver managed to reduce Århus Amtssygehus (‘Århus County Hospital’) to one syllable
when announcing it as the next stop (cf. also the emergence of short forms of names
such as Viv from Vivian). Still, it seems that grammatical expressions that develop from
lexical expressions are often phonologically reduced in relation to their diachronic
source. Compare, for instance, the English auxiliary gonna with its diachronic source
going to. The former expression is phonologically less prominent than the latter in that
it has retained only two out of originally three syllables. The standard account of this
turns on frequency: grammatical expressions that undergo phonological reduction do so
because they are highly frequent.
We do not wish to cast any doubt upon this account. As discussed in §6, the proposed
theory is fully compatible with the assumption that frequency plays a role in grammati-
calization: in order for a lexical expression to undergo grammaticalization, it must be
used with secondary status frequently enough to be conventionally associated with this
status, and this may lead to phonological reduction.
30 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

However, the proposed theory offers a supplementary account that, unlike the pure
frequency account, provides a unique link between grammatical status and phonologi-
cal reduction. Just as boundness may be a consequence of the conventional secondary
status of grammatical expressions, so may phonological reduction: the low discourse
prominence of grammatical expressions may iconically motivate their low phonologi-
cal prominence.15
7.3. SEMANTIC REDUCTION. Semantic reduction cannot be used to define grammatical
expressions any more than phonological reduction. Just as lexical expressions can be
bound and phonologically reduced, they can also have a meaning that is reduced, gen-
eral, bleached, or abstract (cf. §6 on food, fish, and furniture). Nevertheless, there are
factors motivating why grammatical expressions tend to have such meanings. The pro-
posed theory specifies two ways in which grammaticalization interacts with semantic
reduction.
On the one hand, as discussed in §6, semantically reduced status may be seen as a
PREREQUISITE for grammaticalization. In order for a string to become conventionally as-
sociated with a secondary meaning, it must be used with some frequency with this
meaning, and the meaning must be generally applicable as secondary information. For
instance, it was a prerequisite for the grammaticalization of the verb go into the auxil-
iary gonna that it became conventionally associated with the meaning of ‘prospectivity’
because this meaning, serving to localize predications in time, is generally applicable as
a secondary companion of predications.
In addition, the proposed theory allows us to see semantic reduction as a CONSE-
QUENCE of grammaticalization. The background (i.e. the area outside focus) of a photo-
graph has low salience and lacks sharp contours. Similarly, a meaning that is by
convention secondary, as background information, must have low salience and must be
prone to lose its contours, its semantic specifics. Consider again the development of the
Afrikaans particle glo ‘allegedly, presumably, seemingly’ from ek glo ‘I think’ (49, re-
peated here as 60).
(60) Afrikaans (Boye & Harder 2007:591–92; cf. Thompson & Mulac 1991:318)
ek glo hy ryk is. > hy is glo ryk.
I think he rich is he is EVIDENTIAL rich.
‘I think that he is rich.’ ‘He is said (supposed, believed) to be rich.’
The semantic development from the meaning ‘I think’ to the meaning ‘allegedly, pre-
sumably, seemingly’ may be a consequence of the former meaning getting secondary
status. On that interpretation, what is grammaticalized to begin with is epistemic atti-
tude, marked by an explicit reference to the speaker. Initially, then, an expression arises
that is much like clause-medial English I think.
(61) He is, I think, rich.
Only later does the lack of salience associated with the secondary status of this marker
lead to a bleaching of its meaning as well as loss of the first-person pronoun.
7.4. CLOSED-CLASS MEMBERSHIP. Closed-class membership, as well as its diachronic
correlate, paradigmatization, is also a problematic definiens of grammaticalization. On
the one hand, many uncontroversially lexical expressions can be said to belong to
closed classes. Examples are certain verb classes, Latin first names, and simplex nu-

15 As pointed out to us by Ilja Serzants (p.c.), phonological reduction may be seen as ‘phonetically’ con-

nected with discursively secondary status: discursively secondary expressions tend not to have their own ac-
cent, and lack of accent is very often a prerequisite for shortening.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 31

merals. On the other hand, the notion of paradigms is far from uncontroversial. Ameri-
can functionalists, in particular, are generally skeptical toward the usefulness of the no-
tion (e.g. Bybee 1985).
According to the proposed theory, paradigmatization into closed classes is not a di-
rect or necessary consequence of grammaticalization. Still, by highlighting semantic
generality as a prerequisite for grammaticalization (cf. §6), the theory provides a place
for a link between grammaticalization and paradigmatization: the more general the
meanings involved are, the fewer it takes to cover a whole functional or notional do-
main. It is thus natural to find paradigms of grammatical expressions that carve up a
functional or notional domain into a few general regions.
7.5. OBLIGATORINESS. Like boundness, phonological reduction, semantic reduction,
and closed-class membership, obligatoriness cannot be used to define grammaticaliza-
tion. On the one hand, only a subset of the standard examples of grammatical expres-
sions can be claimed to be obligatory in any useful sense of the term. For instance,
derivational suffixes like -ness in kindness can hardly be claimed to be obligatory (cf.
§4 on the status of derivational morphemes as grammatical). You need them, of course,
if you want to use a nominalized form of the adjective, but it is true of all linguistic ex-
pressions, including lexical ones, that you need them for certain purposes. On the other
hand, there are obligatory choices that would not usually be understood as grammatical.
For instance, with very few possible exceptions, Danish first names are rigidly feminine
or masculine, so if you select a first name, you get gender in the bargain. In our view,
this does not make gender in first names part of the grammar of Danish.
In the proposed theory, obligatoriness may be seen as one possible consequence of
grammatical status, especially when it is understood as applying to CHOICES rather than
to expressions. With the conception of grammatical expressions as secondary compan-
ions of other expressions, the rise of obligatory choices can be seen as a frequency effect:
as a limiting case, when a grammatical expression accompanies another expression fre-
quently enough, its absence becomes significant, and an obligatory choice arises between
presence and absence of the expression (cf. Bybee 1994:240).
8. GRAMMATICALIZATION CLINES, REGRAMMATICALIZATION, AND DEGRAMMATICALIZA-
TION. In §5, a distinction was made between three basic types of grammaticalization: in
addition to the standard and nonstandard examples discussed in §5, a third type was in-
troduced, to which we now turn. The third type of grammaticalization consists in the
development of an already existing grammatical expression into a new grammatical ex-
pression distinct from its source. Within this type a distinction can be made between
three subtypes: (i) the development of a grammatical expression into a new expression
that is in some sense more grammatical than its source—we refer to this subtype as
CONTINUED GRAMMATICALIZATION; (ii) the development of a grammatical expression
into a new expression that is less grammatical than its source—a subcase of what is
often referred to as DEGRAMMATICALIZATION; (iii) the development of a grammatical ex-
pression into a new expression that is neither more nor less grammatical than its
source—this is what we call REGRAMMATICALIZATION.16 The first two of these three sub-
16 Our notion of regrammaticalization differs a little both from Greenberg’s (1991) notion and from Ander-

sen’s (2006, 2008) notion of ‘regrammation’. Greenberg (1991:301) uses the term ‘regrammaticalization’ to
refer to cases where a heavily grammatical expression acquires a new grammatical meaning. For Andersen
‘regrammation’ is ‘[a] change by which a grammatical expression, through reanalysis, is ascribed different
grammatical content’ (2008:21), irrespectively of whether the new content is associated with a higher or
lower or equal degree of grammatical status (cf. n. 17). By contrast, we classify as regrammaticalization only
changes in which the input and the output are grammaticalized to the same degree, whether or not there is a
semantic difference between them.
32 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

types are frequently discussed in the literature. Below, in §8.1, we argue that the pro-
posed theory not only is compatible with the distinction between subtypes, but also pro-
vides a theoretical basis for it, which is currently missing from the literature.
Subsequently, in §8.2, we deal with the special case of degrammaticalization that con-
sists in the development of a grammatical expression into a lexical one. Again, we argue
that the proposed theory not only is compatible with this kind of development, but also
makes possible a precise characterization of it.
8.1. DEGREE OF GRAMMATICAL STATUS. Continued grammaticalization, regrammati-
calization, and those types of degrammaticalization that do not proceed all the way to
lexical status have in common that they result in a grammatical expression. All three
types of development are thus covered by the definition of grammaticalization given in
§3. The distinction between the three types presupposes a distinction between different
degrees of grammatical status. Consider, for instance, the grammatical part of the so-
called ‘verb-to-affix cline’ in 62, discussed by Hopper and Traugott (2003:111–14).
(62) (vector verb >) auxiliary > clitic > affix
Any change that proceeds from left to right (e.g. from auxiliary to clitic) is a case of
continued grammaticalization, and any change that proceeds from right to left (e.g.
from affix to clitic) is a case of degrammaticalization. Any change that does not make
any progression on the cline (e.g. a change from one auxiliary to another, distinct auxil-
iary) would be a case of regrammaticalization. In a synchronic interpretation, 62 repre-
sents a ranking of grammatical expressions in terms of degree of grammatical status:
auxiliaries are taken to be less grammatical than clitics, which in turn are taken to be
less grammatical than affixes.
As conceived of in the literature, the distinction between different degrees of gram-
matical status often seems to be entirely based on preconceptions about grammatical
expressions.17 Theoretical arguments for conceiving of affixes as more grammatical
than auxiliaries have to our knowledge never been given. Auxiliaries may give rise to
affixes, and not vice versa, but the conception of the developments in which they do so
as cases of continued grammaticalization—a case of change from grammatical to more
grammatical—appears to be entirely a result of the fact that affixes are better (in the
sense of ‘more uncontroversial’) examples of grammatical expressions than auxiliaries
(cf. the proliferation of terms like ‘semi-auxiliaries’ for cases of grammatical verbs that
are not uncontroversial). Similarly, Kuryłowicz’s (1965:69) view of inflectional affixes
as being more grammatical than derivational ones is theoretically hollow.
In practice, degree of grammatical status is typically linked to (some of ) the features
discussed in §7: a grammatical expression is seen as more grammatical than another
one if it is more bound, more phonologically reduced, more semantically reduced, or if
it belongs to a more closed class. For instance, the main argument for identifying the
change of the Estonian question-marking suffix -s into the question-marking word es as
a case of degrammaticalization (Campbell 1991, 2001:128) is that the latter is con-
ceived of as less bound than the former. As discussed, however, phonological reduction,
semantic reduction, and so forth are not found exclusively with grammatical expres-
sions, and as long as they are not linked theoretically to grammatical status and gram-
maticalization, any use of the features as diagnostics is unwarranted.

17 Accordingly, some scholars avoid the distinction altogether. Notably, when Andersen (2006, 2008) talks

about ‘regrammation’ and ‘degrammation’, he is not concerned with degree of grammatical status, but with
development of new grammatical meaning and loss of grammatical meaning, respectively.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 33

The basic problem is that in the absence of a theory of what it means to be grammat-
ical, any claim that one grammatical expression synchronically is more or less gram-
matical than another one or gives rise to an expression that is more or less grammatical
than itself is empty.
The proposed theory offers a solution to the problem. As discussed in §7, it provides
a link between grammatical status and grammaticalization on the one hand and features
like boundness, phonological reduction, and semantic reduction on the other. With the
theory, there is thus a theoretical basis for talking about grammatical expressions that
are more bound, more phonologically reduced, or more semantically reduced than oth-
ers as ‘being more grammatical’, at least in the sense of ‘showing more features poten-
tially motivated by grammatical status’. For instance, affixes can be claimed to be more
grammatical than auxiliaries on the grounds (i) that they are more bound, and (ii) that
boundness is motivated by conventional secondary status. That is, affixes are more
grammatical than auxiliaries at least in the sense that they show MORE SYMPTOMS of
being grammatical. Caution is warranted, however. Since, as discussed, the relevant
features are found also with lexical expressions (and thus only POTENTIALLY motivated
by grammatical status), they are not diagnostic symptoms (as opposed to nonfocaliz-
ability and nonaddressability; cf. §3.4). It cannot be taken for granted that showing
more symptoms is always equivalent to being more grammatical.
With a theoretical anchoring of degree of grammatical status, the distinction between
continued grammaticalization, regrammaticalization, and degrammaticalization into a
less grammatical expression is no longer hollow. The three types of changes may be dis-
tinguished in terms of the changes in amount of symptomatic features they lead to
(whether the output of the change has more, less, or the same amount as the input). It is
not obvious to us that all changes that are accepted as cases of degrammaticalization in
the literature are actually valid cases in this theoretically qualified sense. Consider, for
instance, the change in English from the genitive marker -s attaching to noun stems to
attaching to the outermost element of an NP. This change is seen by, for instance,
Campbell (2001:127–29) and Norde (2001:247–56, 2009:4) as a case of degrammati-
calization, a change from affix to clitic. But it is not obvious that the change involves
any change in quantity of symptoms of grammatical status (including ‘degree of bound-
ness’, whatever that might be).
8.2. DEGRAMMATICALIZATION. Degrammaticalization basically consists in a change in
which a grammatical expression gives rise either to a lexical expression or to a ‘less
grammatical’ expression. However, the definitions of degrammaticalization found in
the literature differ with respect to details. Norde (2009) requires that the output of de-
grammaticalization preserves both the identity of the construction that includes the ex-
pression undergoing degrammaticalization and the identity of the place of the
expression in that construction. Andersen (2008) in his definition of what he calls ‘de-
grammation’ does not require anything like that.
The subcases of degrammaticalization that result in a (less) grammatical expression
were discussed in §8.1. Here, we focus on those subcases that result in a lexical expres-
sion. A case in point is the Welsh development in 63 in which the originally adpositional
phrase yn ol ‘after, fetch’ is reinterpreted as a verb and subsequently reduced to nôl
‘fetch’ (Norde 2009:4, referring to Willis 2007:294, 297).
(63) Yna yd aeth y gweisson yn ol y varch a ’e arueu y Arthur.
then PTCP went the lads after his horse and his weapons for Arthur
‘Then the lads went after/went to fetch his horse and his weapons for
Arthur.’
34 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

>
Nolwch y Brenin i ’w examnio.18
fetch.2PL.IMP the king to 3SG.M examine.INF
‘Fetch the king to be cross-examined.’
Existing discussions of cases like 63 suffer from the by now familiar problem that they
lack a clear theory about what it means to be a grammatical expression. As in the case
of the change from affix to auxiliary, there is no theoretically based way to classify the
change from adposition to verb as a case of degrammaticalization. In her discussion of
degrammaticalization, Norde (2009) relies on features traditionally associated with
grammatical expressions like those discussed in §7, but there is no attempt to relate
these features theoretically to grammatical status, and as argued in §7, the features are
not diagnostic.
Degrammaticalization into a lexical expression falls outside grammaticalization by
the definition given in §3. Still, the proposed theory may help to throw light on the phe-
nomenon. To begin with, the theory provides a theoretical basis for talking about de-
grammaticalization in the first place. Second, the tests discussed in §3.4 provide a
means of diagnosing cases of degrammaticalization: a diachronic change can be classi-
fied as a case of degrammaticalization into a lexical expression if the output of the
change can in actual communication be used with primary status, while the input to the
change cannot. Third, the theory entails a crucial difference between changes from lex-
ical to grammatical and changes from grammatical to lexical that may serve to motivate
why, as acknowledged also by Norde (2009), the latter are relatively infrequent. By the
theory, change from lexical to grammatical is natural and expected in the sense that it
exploits the potential of lexical expressions to have discursively secondary status, and
simply consists in the conventionalization of this status. In contrast, change from gram-
matical to lexical may be considered unnatural: the change from being by convention
secondary to being by convention potentially primary requires that conventions are
overridden in actual language use.
9. LEXICALIZATION AND CONSTRUCTIONALIZATION. Construction grammar raises a
challenge for the theory proposed above. The constructional format of description cuts
across the distinction between lexical and grammatical units, and the changes whereby
constructions arise have elements both of lexicalization and grammaticalization. We be-
lieve, however, that our theory can throw light on the differences between these two
types of change, while maintaining awareness of the shared features. In the following,
we take up two points: first, the issue of how to distinguish between lexicalization and
grammaticalization in the rise of constructions, and second, the relation between what
we referred to in §5 as standard cases of grammaticalization (i.e. the development of
grammatical expressions from lexical expressions) and what has been referred to as
‘constructionalization’.
As pointed out in the literature (e.g. Lehmann 2002b:1, Brinton & Traugott 2005:89–
90, Fischer 2007:226–29), grammaticalization is similar to lexicalization in that both
are changes whereby a new conventional expression arises—and both may involve re-
duction, phonological and morphosyntactic as well as semantic. Fischer mentions the
development of cupboard as an example of lexicalization that involves reduction. Se-
mantically, the meaning ‘cup’ has disappeared; phonologically, /p/ has disappeared, and

18 Abbreviations: INF: infinitive, M: masculine, PL: plural, PTCP: participle, SG: singular.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 35

so forth. This is fully in harmony with our account, where reduction is not criterial for
grammaticalization.
However, it leaves the question of how to tell lexical expressions and lexicalization
from grammatical expressions and grammaticalization. Fischer’s proposal focuses on
‘schematicity’ and more precisely on the difference between (in her terms) token and
type: lexical expressions are understood as tokens alone, while grammar involves types
(i.e. classes like ‘adjective’). But Fischer herself points out that her approach does not al-
ways provide a clear-cut criterion (2007:229). Thus the development of going to and later
gonna as auxiliaries, which is among the standard examples of grammaticalization, in-
volves the development of what in Fischer’s terms is a token (going to and later gonna)—
in her terms, a lexical element—as well as abstract types (be as well as an infinitive, both
of which are necessary accompaniments of going to). For such cases, where both tokens
and abstract types are involved, Fischer invokes the standard criteria (boundness, etc.) in
a version adapted from Lehmann 1985 (Fischer 2007:118), and these criteria do not make
clear what precisely qualifies as an example of grammaticalization rather than lexical
change (cf. §7 above). The problem with Fischer’s approach is evident in her discussion
of the change of Old English hwilum into the connective while. Fischer is reluctant to
class this change as grammaticalization, because she sees Brinton and Traugott’s (2005)
‘decategorialization’ account as depending on ‘a too facile distinction between lexical
and functional material’ (Fischer 2007:229). We see our theory as providing a new ra-
tionale for Brinton and Traugott’s account: when it develops into a connective, while is
conventionalized with a discursively secondary role (in relation to the clauses linked by
the connective). In our view, lexicalization and grammaticalization are both instances of
conventionalization; both create a new linguistic expression. The only difference is that
grammaticalization results in an expression that is by convention discursively secondary
(see also below on Lehmann 2002b, 2004).
We now turn to the relation between standard cases of grammaticalization and the
rise of constructions. In grammaticalization studies, the need to consider the whole con-
struction was stressed already by Lehmann (1982), but construction grammar has
brought about an increasing focus on the issue (cf. the programmatic formulation in
Bergs & Diewald 2008:3). Although the modern concept of construction has provided
an illuminating new generalization that unites previously disparate aspects, there is also
a potential source of confusion in the fact that the term ‘construction’ has been general-
ized from the classic sense of ‘pattern’ to all expressions, all ‘form-meaning pairs’ (cf.
Goldberg 2006:3). Even if we maintain focus on the core sense in which the construc-
tion means the larger syntagmatic whole, there is still a risk of a too undifferentiated ap-
proach. As pointed out by Traugott (2008:34), the ‘holistic’ focus may detract attention
from the complexity of which the overarching construction in itself is only one part.
This risk manifests itself, for instance, in Trousdale 2008:54: ‘It is important to stress
the constructional nature of the process of grammaticalization … grammaticalization
applies … to the construction as a whole’. As we saw above, however, the emergence of
a construction may also involve lexicalization: ‘As constructions grammaticalize, they
become more schematic; as they lexicalize, they become more idiom-like’ (Trousdale
2008:59). The holistic approach offers an account of ‘where grammar and lexis meet’
(cf. Brinton 2008, Trousdale 2008:58, citing Algeo 1995)—but no clear way of telling
them apart.
The theory proposed here, in contrast, provides an account of both the changes that
affect the whole construction and the changes that affect individual units within it. We
use the term CONSTRUCTIONALIZATION for the overarching change into a new whole con-
36 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

struction. Within it, several more specific developments may occur. With respect to the
rise of a fully schematic construction, see §5.2. With respect to constructions that are
less schematic—that is, constructions that are bound up with specific fillers—two
things may happen. First, constructionalization can occur without grammaticalization
of a lexical expression (cf. §5.2). Second, constructionalization may go hand in hand
with the grammaticalization of a lexical expression. In our standard example, the gram-
maticalization of going to into an auxiliary is bound up with the rise of a whole new
construction of which it forms part (cf. Hopper & Traugott 1993:82). In addition to the
change in going from verb of movement to prospective marker, the full construction in-
cludes two other syntagmatically associated elements: a form of be that goes before it,
and a to-infinitive form that comes after (as an obligatory feature of the verb that con-
veys the main lexical content). Two grammaticalization changes may thus be seen to go
hand in hand: the rise of the new auxiliary going to, and the rise of the new schematic
construction BE going to + V, both by convention discursively secondary.
The constant element in all cases is constructionalization—that is, the rise of a whole
construction. It may be combined with standard cases of grammaticalization, or with
lexicalization, or it may occur on its own. Standard cases of grammaticalization (i.e.
grammaticalization of a lexical expression), conversely, cannot occur except as part of
constructionalization: an expression never becomes grammaticalized in one fell swoop
across all construction types (cf. the lexical status of going to in front of place names
like Rome).
This, we claim, is the precise sense in which the construction is central to grammati-
calization—and that sense can only be made clear if you have a precise definition of
what constitutes grammaticalization. Otherwise, the phrase ‘where grammar and lexis
meet’ has no precise meaning.
Our position is close to that of Lehman (2004:168), who sums up his view by saying
that ‘grammaticalization pushes a sign into the grammar, while lexicalization pushes it
into the lexicon’, and explicitly links up the synchronic and diachronic dimensions in an
integrated theory of grammar and grammaticalization.19 In a two-dimensional diagram
(Lehmann 2002b:3, 2004:168–69), the cline from grammar to lexicon is one dimen-
sion, while hierarchical level is the other dimension (thus a collocation may turn into a
complex, lexically loaded construction, while fusion such as hussy out of housewife
creates a new simplex word). As indicated above, his theory similarly views the change
affecting the whole construction as the only constant element, while the rise of a new
grammatical word or morpheme is optional (Lehmann 2002b:15).
Nevertheless, we think that our account has an advantage in its clearer separation of
constructionalization and delexicalization as two linked, but distinct types of develop-
ment. Lehmann repeatedly describes grammaticalization as moving ‘downward’—that
is, toward the inflectional corner of his diagram (Lehmann 2002b:15, 2004:168–69).
This suggests that he understands the role of constructionalization as ‘setting the stage’
for the standard (and still central) development that leads toward inflectional mor-
phemes. In contrast, our account focuses on the creation of conventionally secondary
and thus ancillary expressions (whether constructions, words, or morphemes) as the

19 Lehmann (2004:155), however, defines grammaticalization as involving a sign losing autonomy and be-

coming more subject to constraints of the linguistic system (cf. the arguments against boundness as a criterion
of grammatical status in §§2 and 7.1). Our theory can be viewed as a narrowed-down version of his: what we
claim is that only the loss of autonomy that follows from secondariness (‘second violins do not come alone’)
counts as a criterion of grammaticalization.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 37

defining feature of grammaticalization. Moreover, it suggests an additional function for

the whole construction: it provides the frame for the competition that the emerging
grammatical word or morpheme is in the process of losing—as when a complement
construction ends up as a construction involving an auxiliary plus a main verb.
10. THE PLACE OF OUR PROPOSAL IN THE GRAMMATICALIZATION LITERATURE. Grammat-
icalization has grown into a vast and sprawling area, and it is not possible in a journal
article to provide an account of how precisely our ‘ancillariness’ theory relates to all of
the literature. We try instead to give a bird’s eye view of where the theory belongs in the
field as a whole.
Beginning at the most general level, Rosenkvist (2006) has provided a useful
overview of approaches to grammaticalization, distinguishing between three main
trends. The first (GramE) is the broad tradition from Meillet and Kuryłowicz, which he
calls the ‘empirical’ approach because it does not do much more than point to the exis-
tence of the phenomenon and instantiations of it, without positing any theoretical as-
sumptions about it. Its central feature is that it describes a range of changes that involve
more or less of the features that were discussed in §7. The second is the modern func-
tionalist approach (GramT for ‘theory’), which focuses on grammaticalization as a way
of showing the dependence of grammatical expressions on meaning and discourse and
puts on the agenda a number of general claims including unidirectionality, the existence
of a limited number of crosslinguistically frequent paths, and suggested phases of prag-
matic strengthening alternating with semantic bleaching (e.g. Traugott 1989). The third
kind is a more recent development by formally oriented linguists (GramS for syntax),
including Roberts and Roussou (2003). A central interest is in showing the extent to
which grammatical change can be viewed as reflecting the pressure to simplify syntac-
tic structures.
Our proposal situates itself in continuation of the modern functionalist discussion
(GramT), both by taking its point of departure in usage phenomena and by addressing
the issue at a theoretical level. Within this tradition, one can roughly distinguish be-
tween two orientations, one with a focus on the basis in language use and one with a
focus on semantically oriented generalizations. The first of these, representing what has
been called ‘West Coast functionalism’, played a major role in putting grammaticaliza-
tion on the map as something that contradicts the theoretical chasm between structure
on the one hand and discourse on the other. Two major positions in this tradition are part
of the foundation we build on: Givón’s illustration of how ‘today’s syntax is yesterday’s
discourse pattern’ (1979:209) and Hopper’s (1991:33) emphasis on how ‘the same’
word may occur sometimes in lexical uses and sometimes in grammatical uses. When
we say that expressions may serve an ancillary function in discourse, and that ancillary
status sometimes becomes part of the structure of language, we are echoing central
claims of both Givón and Hopper.
The other focus, on semantic generalizations, is also part of our foundations. On one
point, however, it is at odds with our key theoretical claim, since we offer our proposal
as an alternative to theories that try to define grammatical status and grammaticaliza-
tion in terms of specific semantic characteristics. Once the case has been made for that
(from our point of view) crucial point, it is clear that there is a close affinity between the
relevant semantic characteristics and the ancillary status we argue for. As argued in
§7.3, semantic reduction can be motivated by the ancillary status of grammatical ex-
pressions—there is a natural relationship between losing prominence and losing seman-
tic substance. However, the two things can occur independently of each other, and it is
38 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

only the kind of bleaching that is associated with secondary status that is characteristic
of grammaticalization (cf. also Traugott 1988:408). The notion of ‘strengthening’, as in
‘pragmatic strengthening’ (cf. Traugott 1988), might at first glance appear to be at odds
with our central notion of discursively secondary status, but this is not the case for the
kind of meanings Traugott is interested in, which involve a ‘situating’ relation to the
text or to the speaker. While the meaning of an expression becomes enriched when
these previously purely situational interpretations become conventionalized, this kind
of meaning is not a candidate for primary status. When Old English hwilum (cf. Trau-
gott 1988:407) turns into the temporal connective while, thereby strengthening its tex-
tual meaning, the new meaning is by convention secondary. This is not to say that
pragmatically inferred meanings can never be lexical (cf. the development whereby
corn comes to mean ‘maize’ in America)—it is the ‘situating’ meaning that is naturally
secondary to the meaning that it situates. Traugott’s theory entails that bleaching, where
it occurs, is a later stage than pragmatic strengthening—and this would fit into our ac-
count in the following way: once an inherently ancillary meaning (such as the connec-
tive meaning of while) has become conventional, the next stage may be that this
meaning takes over from the potentially primary meaning as the raison d’etre for using
the word. When the potentially primary meaning disappears, the word goes from being
a part-time secondary to being a full-time secondary, as it were, thus becoming fully
grammaticalized (cf. §5.1).
This can also throw light on the sense in which our theory is congenial with key ele-
ments of theories that in our view are not sufficient in themselves, including Diewald’s
(2010) theory that grammatical items are characteristic by being ‘weakly deictic’, and
Nicolle’s (1998) theory that grammatical elements are procedural. Like bleached and
pragmatically recruited meanings, deictic and procedural meanings are obvious candi-
dates for being ancillary. Deictic (indexical) meanings have the natural function (cf.
Deacon 2003) of providing a situational anchor for the symbolic meaning that serves as
the focal point—the only exception to this is STRONG deictic meaning, which may con-
stitute the focal point, when it is the situational location that is the issue (he’s here!).
(Purely) procedural meanings are almost by definition ancillary, since (in relevance the-
ory) their purpose is to pave the way for the propositional content (the ‘explicature’).
It should be emphasized that other authors have introduced notions closely related to
what we mean by discursively secondary and discursively primary. The list includes My-
hill (1988:261), who sets up a distinction between morphemes that function as ‘nucleus’,
which conveys ‘information of central importance’, and morphemes that are ‘satellites’,
which convey peripheral information; Talmy (2007:267; cf. Talmy 2000:76–84), who
speaks of ‘attentional backgrounding’ of grammatical expressions; Croft (2001:259),
who speaks of lexical heads as ‘primary information-bearing units’; and Dixon (2006:
11), who distinguishes between primary and secondary concepts. The intuition on which
we build is thus far from new, but previous accounts all tend to see their categories as two
inherently different classes of expressions (whether in terms of semantic content or dis-
course function). We believe we have supplied an account of the usage-based pathway
from primary to secondary status, and a theory that covers both the diachronic relation
and the synchronic difference of status between lexical and grammatical expressions.
Our theory includes many of the standard assumptions about grammaticalization and
language change, including the frequent claim that all semantic changes can occur
equally in lexical change and grammaticalization (including metaphor, metonymy, ex-
tension, and narrowing). We suggest the conventional ancillary relation, and the devel-
opment of this relation, not as something that replaces the points made by all the other
accounts, but as something that makes them fall into place.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 39

11. CONCLUSION: THE MISSING LINK? The theory that we have presented does not rad-
ically change everything we know about grammar and grammaticalization. What it
does is to incorporate core elements of existing assumptions into a unified understand-
ing that has until now not been available.
The critical factor in the theory we offer is the functional link between conventional
discursively secondary status and structural differentiation. The prevailing focus on in-
dividual expressions and on development away from full lexical status casts the devel-
opment of grammatical status in the role of a form of loss. If instead we focus on the
new relation that comes into being whereby one linguistic expression assumes the role
of structural sidekick for the other, it becomes clear that the output is functionally en-
riched. The point is to look not at the single expression but at the combination of the
grammaticalized expression and the expression to which it is attached.
If we assume that this combination is what makes grammatical expressions ‘weaker’
than lexical expressions, while it makes the combination of lexical and grammatical ex-
pressions more powerful than isolated lexical expressions, we may perhaps venture to
speculate why such a unified account has not been reached before, in spite of the related
proposals that we have cited in the discussion. We believe the crux of the matter lies in
the polarization of the issue of structure. Two dominant positions have divided the field
in such a way that the most salient positions either separated structure radically from
other aspects of language, or, on the contrary, sought to challenge all dividing lines that
appeared to give structure an unwarranted special status. We have tried to show that the
development whereby elements acquire grammatical status has to be understood by
combining discourse and structure, recognizing at the same time that structure emerges
from usage and that once emerged, structural facts are not reducible to usage facts.
We see our proposal, therefore, as providing the missing link in the familiar, tradi-
tional picture. It is by no means all there is to say about grammatical status and gram-
maticalization—but all there ALWAYS is to say.

REFERENCES
AARTS, BAS. 2007. Syntactic gradience: The nature of grammatical indeterminacy. Oxford:
Oxford University Press.
ALGEO, JOHN. 1995. Having a look at the expanded predicate. The verb in contemporary En-
glish, ed. by Bas Aarts and Charles F. Meyer, 203–17. Cambridge: Cambridge Univer-
sity Press.
ANDERSEN, HENNING. 2006. Grammation, regrammation, and degrammation: Tense loss in
Russian. Diachronica 23.231–58.
ANDERSEN, HENNING. 2008. Grammaticalization in a speaker-oriented theory of change.
Grammatical change and linguistic theory, ed. by Thórhallur Eythórsson, 11–44. Am-
sterdam: John Benjamins.
ANDERSEN, HENNING. 2010. From morphologization to demorphologization. The Continuum
companion to historical linguistics, ed. by Silvia Luraghi and Vit Bubenik, 117–46.
London: Continuum.
BERGS, ALEXANDER, and GABRIELE DIEWALD. 2008. Contexts and constructions. Construc-
tions and language change, ed. by Alexander Bergs and Gabriele Diewald, 1–14.
Berlin: Mouton de Gruyter.
BOHNEMEYER, JÜRGEN. 2009. Temporal anaphora in a tenseless language. The expression of
time, ed. by Wolfgang Klein and Ping Li, 83–128. Berlin: Mouton de Gruyter.
BOLINGER, DWIGHT. 1968. Aspects of language. New York: Harcourt.
BOLINGER, DWIGHT. 1980. Wanna and the gradience of auxiliaries. Wege zur Universalien-
forschung: Sprachwissenschaftliche Beiträge zum 60. Geburtstag von Hansjakob Seile,
ed. by Gunter Brettschneider and Christian Lehmann, 292–99. Tübingen: Gunter Narr.
BOYE, KASPER. 2010. Raising verbs and auxiliaries in a functional theory of grammatical
status. Language usage and language structure, ed. by Kasper Boye and Elisabeth
Engberg-Pedersen, 73–104. Berlin: Mouton de Gruyter.
40 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

BOYE, KASPER, and PETER HARDER. 2007. Complement-taking predicates: Usage and lin-
guistic structure. Studies in Language 31.569–606.
BOYE, KASPER, and PETER HARDER. 2009. Evidentiality: Linguistic categories and grammat-
icalization. Functions of Language 16.9–43.
BRINTON, LAUREL J. 2008. ‘Where grammar and lexis meet’: Composite predicates in En-
glish. Theoretical and empirical issues in grammaticalization, ed. by Elena Seoane and
María Jóse López-Couso, 33–53. Amsterdam: John Benjamins.
BRINTON, LAUREL J., and ELIZABETH CLOSS TRAUGOTT. 2005. Lexicalization and language
change. Cambridge: Cambridge University Press.
BRUGMANN, KARL, and BERTHOLD DELBRÜCK. 1911. Grundriss der vergleichenden Gram-
matik der indogermanischen Sprachen. Strassburg: Teubner.
BÜHLER, KARL. 1934. Sprachtheorie. Jena: Fischer.
BUNDESEN, CLAUS. 1990. A theory of visual attention. Psychological Review 97.523–47.
BYBEE, JOAN L. 1985. On the nature of grammatical categories. Eastern States Conference
on Linguistics (ESCOL) 2.17–34.
BYBEE, JOAN L. 1994. The grammaticalization of zero: Asymmetries in tense and aspect sys-
tems. Perspectives on grammaticalization, ed. by William Pagliuca, 235–54. Amster-
dam: John Benjamins.
BYBEE, JOAN; REVERE PERKINS; and WILLIAM PAGLIUCA. 1994. The evolution of grammar:
Tense, aspect, and modality in the languages of the world. Chicago: University of
Chicago Press.
CAMPBELL, LYLE. 1991. Some grammaticalization changes in Estonian. In Traugott &
Heine, vol. 1:285–99.
CAMPBELL, LYLE. 2001. What’s wrong with grammaticalization? Language Sciences 23.
113–61.
CLARK, HERBERT H. 1996. Using language. Cambridge: Cambridge University Press.
COMRIE, BERNARD. 1976. Aspect. Cambridge: Cambridge University Press.
CROFT, WILLIAM A. 2000. Explaining language change: An evolutionary approach. London:
Longman.
CROFT, WILLIAM A. 2001. Radical construction grammar: Syntactic theory in typological
perspective. Oxford: Oxford University Press.
CROFT, WILLIAM A. 2003. Typology and universals. 2nd edn. Cambridge: Cambridge Uni-
versity Press.
CRYSTAL, DAVID. 2008. A dictionary of linguistics and phonetics. 2nd edn. Oxford: Black-
well.
DAHL, ÖSTEN. 2004. The growth and maintenance of linguistic complexity. Amsterdam:
John Benjamins.
DAVIDSEN-NIELSEN, NIELS. 1990. Tense and mood in English: A comparison with Danish.
Berlin: Mouton de Gruyter.
DEACON, TERRENCE. 2003. Universal grammar and semiotic constraints. Language evolu-
tion, ed. by Morten H. Christiansen and Simon Kirby, 111–39. Oxford: Oxford Univer-
sity Press.
DIDERICHSEN, PAUL. 1946. Elementær dansk grammatik. Copenhagen: Gyldendal.
DIESSEL, HOLGER. 1999. Demonstratives: Form, function, and grammaticalization. Amster-
dam: John Benjamins.
DIESSEL, HOLGER. 2006. Demonstratives, joint attention, and the emergence of grammar.
Cognitive Linguistics 17.463–89.
DIEWALD, GABRIELE. 2006. Context types in grammaticalization as constructions. Construc-
tions all over: Case studies and theoretical implications (Special issue of Constructions
1), ed. by Doris Schönefeld, 1–29.
DIEWALD, GABRIELE. 2010. On some problem areas in grammaticalization theory. Grammat-
icalization: Current views and issues, ed. by Katerina Stathi, Elke Gehweiler, and
Ekkehard König, 17–50. Amsterdam: John Benjamins.
DIK, SIMON. 1997. The theory of functional grammar, vols. 1–2. Berlin: Mouton de Gruyter.
DIXON, R. M. W. 2006. Complement clauses and complementation strategies in typological
perspective. Complementation, ed. by R. M. W. Dixon and Alexandra Y. Aikhenvald,
1–48. Oxford: Oxford University Press.
DUCROT, OSWALD. 1972. Dire et ne pas dire: Principes de semantique linguistique. Paris:
Hermann.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 41

EGAN, THOMAS. 2010. The ‘fail to’ construction in Late Modern and Present-Day English.
English historical linguistics 2008, vol. 1: The history of English verbal and nominal
constructions, ed. by Ursula Lenker, Judith Huber, and Robert Mailhammer, 123–41.
Amsterdam: John Benjamins.
ENGBERG-PEDERSEN, ELISABETH; MICHAEL FORTESCUE; PETER HARDER; LISBETH F. JAKOB-
SEN; and LARS HELTOFT. 1996. Content, expression and structure: Studies in Danish
functional grammar. Amsterdam: John Benjamins.
ERTESCHIK-SHIR, NOMI. 2007. Information structure. Oxford: Oxford University Press.
EVANS, VYVYAN. 2009. Semantic representation in LCCM theory. New directions in cogni-
tive linguistics, ed. by Vyvyan Evans and Stephanie Pourcel, 27–55. Amsterdam: John
Benjamins.
EVANS, VYVYAN, and MELANIE GREEN. 2006. Cognitive linguistics: An introduction. Edin-
burgh: Edinburgh University Press.
FISCHER, OLGA. 2007. Morphosyntactic change: Functional and formal perspectives. Ox-
ford: Oxford University Press.
FRAUENFELDER, ULI H., and ROBERT SCHREUDER. 1992. Constraining psycholinguistic mod-
els of morphological processing and representation: The role of productivity. Yearbook
of Morphology 1991.165–83.
FRONEK, JOSEF. 1982. Thing as a function word. Linguistics 20.633–54.
GIVÓN, TALMY. 1979. On understanding grammar. New York: Academic Press.
GOLDBERG, ADELE E. 2006. Constructions at work: The nature of generalization in lan-
guage. Oxford: Oxford University Press.
GREENBERG, JOSEPH H. 1991. The last stages of grammatical elements: Contractive and ex-
pansive desemanticization. In Traugott & Heine, vol. 1:301–14.
HARDER, PETER. 2007. Shaping the interactive flow: Language as input, process and prod-
uct. Acta Linguistica Hafniensia 39.7–36.
HARDER, PETER, and OLE TOGEBY. 1993. Pragmatics, cognitive science and connectionism.
Journal of Pragmatics 20.467–92.
HASPELMATH, MARTIN. 2004. On directionality in language change with particular reference
to grammaticalization. Up and down the cline: The nature of grammaticalization, ed. by
Olga Fischer, Muriel Norde, and Harry Perridon, 17–44. Amsterdam: John Benjamins.
HEINE, BERND; ULRIKE CLAUDI; and FRIEDERIKE HÜNNEMEIER. 1991. Grammaticalization: A
conceptual framework. Chicago: University of Chicago Press.
HEINE, BERND, and TANIA KUTEVA. 2002. World lexicon of grammaticalization. Cambridge:
Cambridge University Press.
HEINE, BERND, and TANIA KUTEVA. 2007. The genesis of grammar: A reconstruction. Ox-
ford: Oxford University Press.
HEINE, BERND, and MECHTHILD REH. 1984. Grammaticalization and reanalysis in African
languages. Hamburg: Buske.
HENGEVELD, KEES, and J. LACHLAN MACKENZIE. 2008. Functional discourse grammar: A
typologically-based theory of language structure. Oxford: Oxford University Press.
HOPPER, PAUL J. 1991. On some principles of grammaticalization. In Traugott & Heine, vol.
1:17–35.
HOPPER, PAUL J., and ELIZABETH CLOSS TRAUGOTT. 1993. Grammaticalization. Cambridge:
Cambridge University Press.
HOPPER, PAUL J., and ELIZABETH CLOSS TRAUGOTT. 2003. Grammaticalization. 2nd edn.
Cambridge: Cambridge University Press.
HUDDLESTON, RODNEY. 1976. Some theoretical issues in the description of the English verb.
Lingua 40.331–83.
HUDDLESTON, RODNEY, and GEOFFREY PULLUM. 2002. The Cambridge grammar of the En-
glish language. Cambridge: Cambridge University Press.
JANDA, RICHARD D. 2001. Beyond ‘pathways’ and ‘unidirectionality’: On the discontinuity
of language transmission and the counterability of grammaticalization. Language Sci-
ences 23.265–340.
JOSEPH, BRIAN D. 2001. Is there such a thing as ‘grammaticalization’? Language Sciences
23.163–86.
KIPARSKY, PAUL. 2005. Grammaticalization as optimization. Stanford, CA: Stanford Univer-
sity, MS. Online: http://www.stanford.edu/~kiparsky/Papers/yalegrammaticalization
.pdf.
42 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

KLIMA, GYULA. 2005. Syncategoremata. Encyclopedia of language and linguistics, 2nd

edn., ed. by E. Keith Brown, 353–56. Oxford: Elsevier.
KURYŁOWICZ, JERZY. 1975 [1965]. The evolution of grammatical categories. Esquisses lin-
guistiques, vol. 2, 38–54. Munich: Wilhelm Fink. [Originally published in Diogenes
51.55–71, 1965.]
LADD, D. ROBERT. 1980. The structure of intonational meaning: Evidence from English.
Bloomington: Indiana University Press.
LADD, D. ROBERT. 2008. Intonational phonology. 2nd edn. Cambridge: Cambridge Univer-
sity Press.
LAMBRECHT, KNUD. 1994. Information structure and sentence form: Topic, focus, and the
mental representations of discourse referents. Cambridge: Cambridge University Press.
LANGACKER, RONALD W. 2008. Cognitive grammar: A basic introduction. Oxford: Oxford
University Press.
LEHMANN, CHRISTIAN. 1982. Thoughts on grammaticalization: A programmatic sketch, vol.
1. Köln: Universität zu Köln.
LEHMANN, CHRISTIAN. 1985. Grammaticalization: Synchronic variation and diachronic
change. Lingua e stile 20.303–18.
LEHMANN, CHRISTIAN. 1992. Word order change by grammaticalization. Internal and exter-
nal factors in syntactic change, ed. by Marinel Garritsen and Dieter Stein, 395–416.
Berlin: Mouton de Gruyter.
LEHMANN, CHRISTIAN. 2002a. Thoughts on grammaticalization. 2nd edn. Erfurt: Seminar für
Sprachwissenschaft der Universität.
LEHMANN, CHRISTIAN. 2002b. New reflections on grammaticalization and lexicalization.
New reflections on grammaticalization, ed. by Ilse Wischer and Gabriele Diewald,
1–18. Amsterdam: John Benjamins.
LEHMANN, CHRISTIAN. 2004. Theory and method in grammaticalization. Zeitschrift für ger-
manistische Linguistik 32.152–87.
LEHMANN, CHRISTIAN. 2008. Information structure and grammaticalization. Theoretical and
empirical issues in grammaticalization, ed. by Elena Seoane and María Jóse López-
Couso, 207–29. Amsterdam: John Benjamins.
LEWIS, DAVID. 1969. Convention: A philosophical study. Cambridge, MA: Harvard Univer-
sity Press.
LOCKWOOD, WILLIAM B. 1968. Historical German syntax. Oxford: Clarendon.
LYONS, JOHN. 1968. Introduction to theoretical linguistics. Cambridge: Cambridge Univer-
sity Press.
MATTHEWS, PETER H. 2007. Oxford concise dictionary of linguistics. 2nd edn. Oxford: Ox-
ford University Press.
MEILLET, ANTOINE. 1958 [1912]. L’évolution des formes grammaticales. Linguistique his-
torique et linguistique générale, 130–48. Paris: Champion. [Originally published in
Scientia 12.384–400, 1912.]
MYHILL, JOHN. 1988. Categoriality and clustering. Studies in Language 12.261–97.
NEWMEYER, FREDERICK J. 2001. Deconstructing grammaticalization. Language Sciences
23.187–229.
NICOLLE, STEVE. 1998. A relevance theory perspective on grammaticalization. Cognitive
Linguistics 9.1–35.
NORDE, MURIEL. 2001. Deflexion as a counterdirectional factor in grammatical change. Lan-
guage Sciences 23.231–64.
NORDE, MURIEL. 2009. Degrammaticalization. Oxford: Oxford University Press.
PALMER, FRANK R. 1986. Mood and modality. Cambridge: Cambridge University Press.
POSNER, ROLAND. 1972. Theorie des Kommentierens: Eine Grundlagenstudie zur Semantik
und Pragmatik. Frankfurt am Main: Athenäum.
PUTNAM, HILARY. 2001. Enlightenment and pragmatism. Amsterdam: Koninklijke van
Gorcum.
ROBERTS, IAN, and ANNA ROUSSOU. 2003. Syntactic change: A minimalist approach to gram-
maticalization. Cambridge: Cambridge University Press.
ROOTH, MATS. 1992. A theory of focus interpretation. Natural Language Semantics 1.75–
116.
 A USAGE-BASED THEORY OF GRAMMATICAL STATUS AND GRAMMATICALIZATION 43

ROSENKVIST, HENRIK. 2006. Fyra vyer av begreppet grammatikalisering. Vetenskapssoci-

eteten i Lund årsbok 2006.1–17.
ROSS, JOHN ROBERT. 1969. Auxiliaries as main verbs. Studies in philosophical linguistics,
ed. by William Todd, 77–102. Evanston, IL: Great Expectations.
SEARLE, JOHN R. 1995. The construction of social reality. Harmondsworth: Penguin.
SCHMERLING, SUSAN F. 1976. Aspects of English sentence stress. Austin: University of Texas
Press.
SLOBIN, DAN I. 1997. The origins of grammaticizable notions: Beyond the individual mind.
The crosslinguistic study of language acquisition, vol. 5: Expanding the contexts, ed. by
Dan I. Slobin, 265–323. Mahwah, NJ: Lawrence Erlbaum.
SPERBER, DAN, and DEIRDRE WILSON. 1995. Relevance: Communication and cognition. 2nd
edn. Oxford: Blackwell.
TALMY, LEONARD. 2000. Toward a cognitive semantics, vols. 1–2. Cambridge, MA: MIT
Press.
TALMY, LEONARD. 2007. Attention phenomena. Handbook of cognitive linguistics, ed. by
Dirk Geeraerts and Hubert Cuyckens, 264–93. Oxford: Oxford University Press.
THOMPSON, SANDRA A., and ANTHONY MULAC. 1991. A quantitative perspective on the gram-
maticalization of epistemic parentheticals in English. In Traugott & Heine, vol.
2:313–29.
TRASK, ROBERT L. 1993. A dictionary of grammatical terms in linguistics. London: Rout-
ledge.
TRAUGOTT, ELIZABETH CLOSS. 1980. Meaning-change in the development of grammatical
markers. Language Sciences 2.44–61.
TRAUGOTT, ELIZABETH CLOSS. 1982. From propositional to textual to expressive meanings:
Some semantic-pragmatic aspects of grammaticalization. Perspectives in historical lin-
guistics, ed. by Winfred P. Lehmann and Yakov Malkiel, 245–71. Amsterdam: John
Benjamins.
TRAUGOTT, ELIZABETH CLOSS. 1988. Pragmatic strengthening and grammaticalization.
Berkeley Linguistics Society 14.406–14.
TRAUGOTT, ELIZABETH CLOSS. 1989. On the rise of epistemic meanings in English: An ex-
ample of subjectification. Language 65.31–55.
TRAUGOTT, ELIZABETH CLOSS. 2003. Constructions in grammaticalization. The handbook of
historical linguistics, ed. by Brian D. Joseph and Richard D. Janda, 624–47. Oxford:
Blackwell.
TRAUGOTT, ELIZABETH CLOSS. 2008. The grammaticalization of NP of NP patterns. Con-
structions and language change, ed. by Alexander Bergs and Gabriele Diewald, 23–45.
Berlin: Mouton de Gruyter.
TRAUGOTT, ELISABETH CLOSS, and BERND HEINE (eds.) 1991. Approaches to grammatical-
ization, vols. 1–2. Amsterdam: John Benjamins.
TROUSDALE, GRAEME. 2008. Constructions in grammaticalization and lexicalization: Evi-
dence from the history of a composite predicate construction in English. Constructional
approaches to English grammar, ed. by Graeme Trousdale and Nicholas Gisborne,
33–67. Berlin: Mouton de Gruyter.
VAN GELDEREN, ELLY. 2005. Function words. Encyclopedia of linguistics, vol. 1, ed. by
Philipp Strazny, 362–64. New York: Fitzroy Dearborn.
WILLIS, DAVID. 2007. Syntactic lexicalization as a new type of degrammaticalization. Lin-
guistics 45.271–310.
WIWEL, H. G. 1901. Synspunkter for dansk sproglære. Copenhagen: Det Nordiske.
WOODWORTH, NANCY L. 1991. Sound symbolism in proximal and distal forms. Linguistics
29.273–99.

Boye [Received 24 April 2009;

LANCHART revision invited 10 November 2009;
University of Copenhagen revision received 24 February 2010;
Njalsgade 136 revision invited 31 July 2010;
2300 Copenhagen S, Denmark revision received 15 March 2011;
[[email protected]] accepted 13 September 2011]
44 LANGUAGE, VOLUME 88, NUMBER 1 (2012)

Harder
Department of English, Germanic and Romance Studies
University of Copenhagen
Njalsgade 128
2300 Copenhagen S, Denmark
[[email protected]]