Pragmatic functions, semantic classes, and

lexical categories*
WILLIAM CROFT

Abstract

Smith o¤ers a critique of the theory of parts of speech in Croft (1991,

2001) inter alia. Smith tries to make a functionally-based universal-
typological theory of parts of speech provide an answer to the problem of
defining word classes and giving those classes the same names across lan-
guages (‘‘noun’’; ‘‘adjective’’); this is not possible and not what I intended.
Smith conflates semantic properties with pragmatic properties, and he con-
flates di¤erent pragmatic properties that cannot be conflated. There are
challenging issues in defining pragmatic functions and their linguistic re-
flexes, but Smith’s critique only briefly touches on them.

1. Introduction

Mark Smith criticizes approaches to major syntactic categories that uti-

lize pragmatic functions, in particular those by Hengeveld, Bhat, Baker
and myself (Croft 1991 [Syntactic categories and grammatical relations;
henceforth SCGR] and 2001 [Radical construction grammar; henceforth
RCG]). Smith argues that the pragmatic functions are too vague and gen-
eral to discriminate noun, verb and adjective, or where too specific, ex-
clude plausible members from those categories; and that absence of evi-
dence gives rise to disagreements about basic properties of pragmatic
functions. But disagreements among scholars does not necessarily entail
that all of them are wrong. I have myself argued against Hengeveld
(2001: 65–75) and Baker (Croft 2009), hence lumping my approach with
theirs necessarily leads to inconsistencies. In fact, there are significant
misrepresentations of my approach so that very few of Smith’s criticisms
apply to it.
In SCGR and RCG, I propose typologically valid universals of the en-
coding of linguistic function — semantics and pragmatics — in linguistic

Linguistics 48–3 (2010), 787–796 0024–3949/10/0048–0787

DOI 10.1515/LING.2010.024 6 Walter de Gruyter

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
788 W. Croft

form — morphosyntax. The universals have three components: the

semantic content of words, the pragmatic functions performed by cer-
tain constructions in which those lexical items are found, and the mor-
phosyntactic patterns that are produced, i.e., the structure of the relevant
grammatical constructions and the distribution of words in the relevant
roles of those constructions within and across languages. I begin here
with the morphosyntactic patterns, and then turn to the semantic and
pragmatic definitions.

2. Morphosyntactic patterns and lexical categories

Smith presents my analysis of the prototypes for noun, adjective and verb
as typologically unmarked combinations of pragmatic function and se-
mantic class (reference to objects, modification with properties, and pred-
ication by actions respectively; p. 720–721). Smith then argues that it
does not provide an adequate solution to the problem of lexical catego-
ries, particularly about ‘‘nonprototypical words’’ (actually, nonprototypi-
cal combinations of semantic word class and pragmatic function).
I have long argued that parts of speech cannot define word classes. The
beginning of Chapter 2 of RCG (the chapter devoted to parts of speech)
states: ‘‘Noun, verb, and adjective are not categories of particular lan-
guages; but noun, verb and adjective are language universals — that is,
there are typological prototypes which should be called noun, verb and
adjective’’ (Croft 2001: 63). This is opposed to the widely held view that
noun, verb and adjective are crosslinguistically comparable word classes
— a view held by Hengeveld, Baker and Bhat, who Smith rightly
criticizes in this regard. But the approach in RCG is the only crosslinguis-
tically tenable approach to parts of speech. This is true not just of the
so-called major syntactic categories but all grammatical categories, as I
show in RCG. Lexical categories are language-specific, defined by their
distributional patterns in the specific constructions of the language.
Labeling such categories as ‘‘noun’’, ‘‘verb’’ and ‘‘adjective’’ is fine only
if one acknowledges that these are language-specific word classes and
they do not have any crosslinguistic validity. And if one attempts to de-
fine such classes within a language, one generally does so based on their
distribution in an arbitrarily selected subset of constructions (I called this
methodological opportunism in RCG). If one uses all constructions to de-
fine word classes, then there would be a myriad of overlapping categories,
not a small set of large word classes, because almost every word has its
own unique distribution across all constructions (Croft 2001: 34–47).
Smith cites my critique of Hengeveld in this regard (p. 762), but does
not recognize that he himself falls prey to methodological opportunism.

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
 Pragmatic functions, semantic classes, and lexical categories 789

For example, on p. 726 Smith writes, regarding a definition of predica-

tion, ‘‘such general definitions apply to nouns when they occur in the
predicative or modifying positions typically occupied by verbs and adjec-
tives’’ (e.g., John is a doctor). Smith presupposes what we are trying to ex-
plain: he assumes that doctor in the predicate nominal construction is a
noun, and therefore that predication is defined too broadly for parts of
speech. But what about languages in which ‘doctor’ inflects for subject in-
dexation, tense, aspect etc. when predicated, i.e., the same predication
construction is used for object words like ‘doctor’ as well as for action
words like ‘sing’? Likewise, on p. 731 Smith assumes that lucky in Gomez
is lucky is an adjective. Smith implicitly uses an English-specific definition
of word classes, apparently based on morphological distribution (occur-
rence with certain inflections), to criticize a language-universal character-
ization of parts of speech.
In John is a doctor and Gomez is lucky, words in the object semantic
class and property semantic class respectively are used in a predication
construction, which in English happens to di¤er from the action word
predication construction by the occurrence of a copula and the absence
of tense-indexation inflection on the head word. From a language-specific
point of view, one can define word classes by arbitrarily selected
language-specific constructions, such as the ‘‘English Copula construc-
tion’’ vs. the ‘‘English Tense-Indexation inflection construction’’, and
call the classes ‘‘English Noun’’ and ‘‘English Verb’’; but those definitions
of word classes do not carry over into any other language because the
constructions that define them are English-specific (as well as being arbi-
trarily chosen). One must take a radically di¤erent approach to develop a
crosslinguistically valid theory of parts of speech, which I do in SCGR
(see especially Croft 1991: 93–95) and more explicitly in RCG (Croft
2001: Ch. 1–2).
One can try to compare constructions across languages and their distri-
bution patterns, but how is that done? Because of the typological diversity
of grammatical structures, the only crosslinguistically valid way to do it is
by comparing functions, as has been done since the beginnings of typol-
ogy (see Croft 2003: 13–19 and references cited therein). For the distribu-
tion of words in constructions, one must compare equivalent semantic
classes, and for the constructions themselves, one must compare equiva-
lent pragmatic functions in this case. In fact, one must have a coherent
basis for selecting constructions to compare across languages, and for se-
lecting the semantic features for defining semantic classes of words to
compare across languages. The major propositional acts (reference, pred-
ication, modification) form a coherent paradigmatic functional set (Croft
2005: 439; 2007a: 421–22). The semantic features (see Section 2) are those

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
790 W. Croft

that interact with the propositional act functions to produce the typologi-
cal prototype patterns.
When one compares functionally equivalent constructions and func-
tionally similar word classes across languages, one finds variation in their
morphosyntactic expression. But the crosslinguistic morphosyntactic vari-
ation is constrained by the universal typological prototypes of semantic
class and pragmatic function referred to above. What are typological
prototypes? They are typologically unmarked combinations of values of
functional properties, in this case the combinations 3object, reference4,
3property, modification4 and 3predication, action4. Smith misconstrues
typological markedness on p. 760. He assumes that the fact that in
English, a ‘‘noun’’ like university can modify another ‘‘noun’’ like housing
without additional morphemes is a counterexample to typological
markedness, and writes that in RCG I propose a ‘‘revised version’’ of
typological markedness that requires nonprototypical combinations to
have only at least as many morphemes as the prototypical combinations
(in this case, object modification has the same number of morphemes
as object reference, i.e., zero). He also argues in the same passage that
English -er nominals (reader, teacher etc.) are prototypical nouns (object
reference) but contain additional morphemes.
Typological markedness (Greenberg 1966) di¤ers from Prague school
and generative markedness theory (Croft 2003: 87–91). It describes cross-
linguistically valid implicational universals, one of which (structural cod-
ing) states that ‘the marked value of a grammatical category will be ex-
pressed by at least as many morphemes as the unmarked value of that
category’ (Croft 2003: 92). First, the morpheme must encode the function
in question; I called this function-indicating morphosyntax in SCGR
(Croft 1991: 58). English -er nominals are not function-indicating mor-
phosyntax: they change the semantic type of the stem from an action to
an object. They are irrelevant to typological markedness. The formula-
tion of structural coding in terms of at least as many morphemes (con-
trary to the Prague school and generative definitions) is the only typolog-
ically empirically valid one, and has been used since Greenberg; it is not
a recently ‘‘revised’’ version to accommodate ‘‘counterexamples’’. Smith
attacks a straw man here. It is also extremely robust across many concep-
tual categories, not just the functional prototypes for parts of speech (see
Croft 2003: 156–157 for a list of categories with typological markedness
asymmetries), and is ultimately explained via token frequency e¤ects (By-
bee 1985; Croft 1991: 87–93; Croft 2003: 110–117).
The variation of morphosyntactic expression accommodated in typo-
logical markedness does not allow the universal-typological theory of
parts of speech to be used to define language-specific word classes. But

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
 Pragmatic functions, semantic classes, and lexical categories 791

that is not the intention of the universal-typological theory of parts of

speech. Linguists (including Smith) should give up trying to find a cross-
linguistic theory of word classes, and instead focus on the genuine univer-
sals of language that traditional grammarians were aiming at with their
analyses of parts of speech. In other words, Smith is wrong to conclude
that a theory of parts of speech utilizing pragmatic functions is no better
than a theory based on morphosyntactic criteria: the latter is impossible,
while the former has revealed empirically valid typological universals.

3. Semantic classes and pragmatic functions

A genuinely universal theory of parts of speech is based on the semantic

classes of words and the pragmatic functions of constructions: these two
interact to yield the universal typological prototypes described above.
This requires good definitions of semantic classes of words and the prag-
matic functions. Both are provided in SCGR, and in the case of the prag-
matic functions, elaborated in other publications (Croft 1990, 2007b).
Smith, however, confuses the semantic and pragmatic definitions.
In SCGR, I use four semantic features to define the semantic word
classes of objects, properties and actions: relationality (called valency in
SCGR), stativity (state/process), persistence (persistent/transitory) and
gradability (Croft 1991: 62–66). Smith appears to think that I define pred-
ication in terms of persistence (pp. 731–732) and that relationality is a
pragmatic function (p. 757). Both of these assertions are incorrect. The
first one leads Smith to conclude that predication of a persistent property
(e.g., God exists) is a counterexample to my theory. It is not. The funda-
mental fact about linguistic communication is that any semantic class can
be expressed via any pragmatic function (Croft 2007b: 367). So any se-
mantic type will occur as a predication. But some combinations of seman-
tic class and pragmatic function are (typologically) prototypical, while
others are not. It is those prototypes that give rise to distinct parts of
speech (but do not allow us to form a theory of lexical categories). Hence
it is important to clearly distinguish semantic properties of concepts, ex-
pressed by words, from the pragmatic functions those concepts have, as
encoded by constructions.
I now turn to the pragmatic functions themselves, the chief target of
Smith’s critique. SCGR devotes a long section to definitions of reference,
predication and modification (Croft 1991: 101–126); detailed definitions
of the pragmatic functions given on p. 123. These are more elaborate
than the definitions discussed by Smith. Space limitations allow me only
to outline some major points.

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
792 W. Croft

Pragmatics covers all aspects of language use (Clark 1996). The prag-
matic functions appealed to in analyses of parts of speech are only a small
part of pragmatics. The pragmatic functions are intended to explain how
conceptual content is structured in an utterance for presentation to the
hearer by the speaker. This goes under the name of information structure,
and is what Clark calls the formulation or presentation process in his
analysis of language use (Clark 1996: 21, 151–153). I also use the term
propositional act (Searle 1969), since the formulation of information is
represented by propositions in philosophical semantics.
Propositional acts are encoded by propositional act constructions:
referring expressions, attributive constructions and predication con-
structions. This answers the question that Smith raises about what sort
of linguistic structures encode pragmatic functions; it is explicitly asserted
in RCG (Croft 2001: 88; see also Croft 2005, 2007a). The close relation-
ship between propositional acts and lexical classes is that the relevant
roles in the propositional act constructions are the heads of those con-
structions, under the semantic definition of ‘‘head’’ argued for in RCG
(Croft 2001: Ch. 7): a head is the element in a construction whose deno-
tation is the closest to the denotation of the construction as a whole. This
is the sense in which object, property and action words may be associated
with propositional act functions.
Information structure involves the structuring of all information in the
utterance, not just the major content words. In Croft (1990), I develop a
comprehensive framework for the information structuring functions of
function words and grammatical inflections, not just the major content
words described in SCGR. I call these the minor propositional acts, sup-
plementing the major propositional acts of reference, predication and
modification. Smith describes this as an indefinite multiplication of prag-
matic functions that undermines the pragmatic enterprise (pp. 767–768).
In fact it is an attempt to develop a full-fledged theory of information
structure in language, essential to understanding how all information is
structured in discourse. Croft (2007b) provides an explanatory model of
all of these propositional act functions in terms of an extended version of
Chafe’s (1977a, 1977b) theory of the verbalization of experience.
Information structure at the clausal (propositional) level is a whole ge-
stalt, not independent parts. Predication is the predication of a referent;
reference and predication are two distinct roles in the information struc-
ture of a clause. Lambrecht (1994: 51–52) describes this interrelation of
functions in his analysis of clause-level information structure. Lambrecht
defines three clause-level information structures: topic-comment (also
called categorical by Sasse [1987], following the philosophical tradition),
event-reporting (thetic), and identificational (Lambrecht 1994: 120–127).

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
 Pragmatic functions, semantic classes, and lexical categories 793

Reference and predication are the major roles defined in the topic-
comment clausal information structure. Reference (topic) can only be de-
fined relative to predication and predication (comment) only relative to
reference: a topic is what a proposition is about, and a comment is what
is asserted about the topic (Lambrecht 1994: 118–119, 121–122). Lam-
brecht also distinguishes clause-level information structure, as outlined in
the preceding sentences, from the information status of discourse refer-
ents, defined via the properties of identifiability and activation (Lam-
brecht 1994: Ch. 3).
Smith’s discussion of pragmatic functions paints too simplistic a picture
of pragmatic theories. Smith argues that predication and modification
cannot be di¤erentiated, since they are both assertions about a referent.
However, I define modification as a secondary function to reference and
predication (Croft 1991: 111, 130–133; 2001: 97). Modification serves a
secondary function of establishing reference (restrictive modification) or
adding a predication (nonrestrictive modification). Modification of refer-
ents (the adjective prototype) can be distinguished further by its construal
as enriching the nominal category with a value on a dimension (see
Wierzbicka 1986 and the discussion in Croft 1991: 101–102).
Smith’s discussion of reference confuses four separate theoretical con-
structs: reference as a major propositional act; denotation — the relation-
ship between form and meaning; the minor propositional act of particula-
rizing (Croft 2007b, described as selecting and situating in Croft 1990);
and the discourse status of identifiability. Some of this confusion is found
in the semantic and pragmatic literature, in which refer is often used to
mean ‘denote’ and referential is used for an identifiability status; but the
distinctions have been made clearly by others.
Denotation is simply the symbolic relation between a linguistic form
such as a content word and the concept it symbolizes or signifies. In The
car exploded and The explosion shook the building (Smith, pp. 738–739),
both exploded and explosion denote an action, possibly the same action;
but in the first, the action is predicated of the referent the car and in the
second the action is itself being referred to, and shake is predicated of it
(and the building). Again, reference and predication play distinct roles in
the structuring of the information communicated in the two utterances.
Denotation of course applies to all content words, but that does not un-
dermine the notion of reference as a distinct propositional act.
Smith confuses reference and denotation on pp. 738–740, but notes
that pragmaticists distinguish the concepts (p. 740). However, he then
confuses reference (or denotation) with particularizing, that is, the fact
that words denote types but are used to denote tokens on most occasions
of use. Again, as pointed out in Croft (1990, 2007b; see also Langacker

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
794 W. Croft

1991: 31–35), particularizing structures all types of content information,

not just object words, or words used in reference. This point is not always
appreciated by the philosophically-influenced literature, hence Smith’s
criticism of certain other scholars is valid. But Smith does not recognize
the di¤erence between reference and the process of getting from a type
(category) to a token that particularizes the type.
Smith then confuses reference and identifiability. Identifiability is a
property of discourse referents, having to do with the ability of the hearer
(and speaker) to identify and situate a particular referent in the real world
or in a possible world (mental space; Fauconnier 1985; Croft 1990). But a
discourse referent is being referred to, no matter what its identifiability
status is. Hence ‘‘nonreferential’’ (type identifiable; Gundel et al. 1993)
referents are still referents, that is, the targets of the referring proposi-
tional act. In fact, identifiability appears to be part of the particularizing
function (it is part of the situating minor propositional act), and so should
not be restricted to reference anyway.
An important discourse motivation for the reference/predication con-
trast is that referents generally recur across clauses in a discourse, while
predications are generally asserted just for the one proposition (Chafe
1977a, 1977b; Croft 2007b). It is not impossible for expression of a pred-
ication to recur, as Smith notes (pp. 749 and 753), but he does not note
the di¤erences between tracking of a referent and tracking of a predica-
tion. Referent anaphora indicate token identity: in I called the director
and told her I was coming, her picks out the same individual as the direc-
tor. So-called predicate anaphora such as do so/the same and VP ellipsis
indicate type identity: in Mary tore up her application form, and John did
the same (Example 28), the tearing-up events are di¤erent tokens. In He
was standing and he fell. He fell and he kneeled. He kneeled and he bled.
He bled and he died (Example 45), the same event token is predicated
more than once, but only with separate full predications, not predicate
anaphora. In I heard the same explosion that you heard (Example 31),
there is token identity of the event, but the event is being referred to, not
predicated. In other words, there does not appear to be reference tracking
of tokens of predicated concepts, at least in English.

4. Conclusion

Smith’s critique of pragmatics in parts of speech largely does not apply to

the approach I advocate. Smith tries to make a functionally based univer-
sal-typological theory of parts of speech provide an answer to the hope-
less task of defining word classes and giving those classes the same names

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
 Pragmatic functions, semantic classes, and lexical categories 795

across languages (‘‘noun’’; ‘‘adjective’’); this is not what I am doing and it

is doomed to failure anyway. Smith treats some semantic properties as if
they were part of the pragmatic functions, and therefore mistakenly ar-
gues that some pragmatic functions are too vaguely defined. Finally,
Smith conflates many pragmatic categories, especially in discussing refer-
ence, that scholars have carefully distinguished in order to develop a com-
prehensive and usable theory of information structure.
This is not to say that there are no challenges in pragmatic analysis, or
that Smith’s critique completely misses the mark. Smith describes the
pragmatic functions as intangible (p. 770). The reason they seem so is
that they structure information in discourse, rather than just symbolize
conceptual content. Hence they are best analyzed in terms of discourse
and cognitive processes, whose analysis is harder to operationalize empir-
ically than conceptual content. Unfortunately Smith does not address this
genuine challenge.
Also, there is not a perfect one-to-one mapping between pragmatic
function and propositional act construction: the head of a predication
construction is not always the primary predication, and the head of a re-
ferring construction is not always a referent (Croft 1991:125–126). Smith
gives examples of the former — adjective-light noun predications like He
was a mournful man — and of the latter — light verb constructions like
Sue had a chat with Mike (pp. 763–764). In these cases, a new construc-
tion violates the mapping; but these will eventually grammaticalize so
that the light verb becomes an auxiliary and the light noun becomes a
predication or indexation marker. Another challenging example is the
pragmatic status of nonsubject referents (p. 757): some analyze them as
part of the predication (comment), others not, at least not always (Lam-
brecht 1994: 147–150).
Smith also gives the example of NP coordination (pp. 756–757), which
creates a composite referring expression out of component expressions
(Wierzbicka 1980: 223–285). Some nesting of pragmatic functions needs
to be allowed in a theory of information structure.
Many interesting problems in the pragmatic analysis of parts of speech re-
main, but the analysis of information structure has already provided in-
sights into parts of speech, and further research will certainly provide more.

Received 27 October 2009 University of New Mexico

Note
* Correspondence address: MSC03 2130, 1 University of New Mexico, Albuquerque,
NM 87131-0001, USA. E-mail: [email protected].

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM
796 W. Croft

References

Bybee, Joan L. 1985. Morphology: A study into the relation between meaning and form.
Amsterdam: John Benjamins.
Chafe, Wallace. 1977a. Creativity in verbalization and its implications for the nature of
stored knowledge. In Roy Freedle (ed.) Discourse production and comprehension, 41–55.
Norwood, NJ: Ablex.
Chafe, Wallace. 1977b. The recall and verbalization of past experience. In Peter Cole (ed.),
Current issues in linguistic theory, 215–246. Bloomington, IN: Indiana University Press.
Clark, Herbert H. 1996. Using language. Cambridge: Cambridge University Press.
Croft, William. 1990. A conceptual framework for grammatical categories (or, a taxonomy
of propositional acts). Journal of Semantics 7. 245–279.
Croft, William. 1991. Syntactic categories and grammatical relations: The cognitive organiza-
tion of information. Chicago: University of Chicago Press.
Croft, William. 2001. Radical construction grammar: Syntactic theory in typological perspec-
tive. Oxford: Oxford University Press.
Croft, William. 2003. Typology and universals, 2nd edn. Cambridge: Cambridge University
Press.
Croft, William. 2005. Word classes, parts of speech and syntactic argumentation. Linguistic
Typology 9. 431–441.
Croft, William. 2007a. Beyond Aristotle and gradience: A reply to Aarts. Studies in Lan-
guage 31. 409–430.
Croft, William. 2007b. The origins of grammar in the verbalization of experience. Cognitive
Linguistics 18. 339–382.
Croft, William. 2009. Methods for finding language universals in syntax. In Sergio Scalise,
Elisabetta Magni & Antonietta Bisetto (eds.), Universals of language today, 145–164.
Berlin: Springer.
Fauconnier, Gilles. 1985. Mental spaces. Cambridge, MA: MIT Press.
Greenberg, Joseph H. 1966. Language universals, with special reference to feature hierarchies
(Janua Linguarum Series Minor 59.) The Hague: Mouton.
Gundel, Jeannette K., Nancy Hedberg & Ron Zacharski. 1993. Cognitive status and the
form of referring expressions in discourse. Language 69. 274–307.
Lambrecht, Knud. 1994. Information structure and sentence form: topic, focus and the mental
representations of discourse referents. Cambridge: Cambridge University Press.
Langacker, Ronald W. 1991. Foundations of cognitive grammar, vol. II: Descriptive applica-
tion. Stanford, CA: Stanford University Press.
Sasse, Hans-Jürgen. 1987. The thetic-categorical distinction revisited. Linguistics 25. 511–
580.
Searle, John R. 1969. Speech acts: An essay in the philosophy of language. Cambridge: Cam-
bridge University Press.
Wierzbicka, Anna. 1980. Lingua mentalis: The semantics of natural language. New York:
Academic Press.
Wierzbicka, Anna. 1986. What’s in a noun? (or: how do nouns di¤er in meaning from adjec-
tives?). Studies in Language 10. 353–389.

Brought to you by | Brown University Rockefeller Library (Brown University Rockefeller L

Authenticated | 172.16.1.226
Download Date | 7/16/12 9:36 PM