Linguistic Society of America

Autonomy and Functionalist Linguistics

Author(s): William Croft
Source: Language, Vol. 71, No. 3 (Sep., 1995), pp. 490-532
Published by: Linguistic Society of America
Stable URL: http://www.jstor.org/stable/416218
Accessed: 11/09/2010 23:59
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=lsa.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.

http://www.jstor.org

AUTONOMY AND FUNCTIONALIST LINGUISTICS

WILLIAMCROFT

University of Manchester
Functional analyses of grammatical phenomena, and the functionalist approaches that
promote them, are appealing to those who believe that an integrated view of language
structure and language function is desirable. But functional analyses have been held to
founder on basic grammatical facts that are taken to support the autonomy of grammar.
The concept of autonomy is a complex one, and at least two different notions are found
in current linguistic theory: arbitrariness and self-containedness. These notions of autonomy apply either to the syntactic component of the grammar, or (a more recent claim)
to the grammar itself, with respect to change, use, and acquisition. The arbitrariness
of syntax must be accepted; and many functional analyses are compatible with selfcontainedness. However, mixed formal/functional analyses provide an argument against
the self-containedness of syntax, and in fact even many formal theories of syntax accept
non-self-containedness. The arbitrariness of grammatical knowledge must also be accepted; and many functional analyses of the dynamic process affecting grammar are
compatible with self-containedness. An argument against the self-containedness of grammar comes not from these functional analyses but from sociolinguistics.*

1. INTRODUCTION.
'Functional' analyses of grammar, though rather varied,

center on linguisticexplanationbased on language's function in a largercontext.' Whileit is commonlyagreedthat manyaspects of humanlinguisticbehavior can be explainedonly in termsof the functionof language,whatdistinguishes
functionalistapproachesis the hypothesis that at least some basic facts of syntax can be accountedfor in functionalterms as well. In this respect functionalism appears to contrast with formalist or more accurately structuralist
approaches, whose most prominentexample is generative grammar.
Unfortunately,there has been very little dialoguebetween structuralistsand
functionalists,and as a result there has been little serious comparisonof their
competinghypotheses aboutthe natureof grammaticalexplanation.The mutual
isolationof theoristsand theirapproachesis in contrastto the dialogue, acrimonious though it was, in the 'linguistic wars' between generative semanticists
and interpretivesemanticistsin the early 1970s(Harris1993).Newmeyer (1983,
1991, 1992)is virtuallythe only linguistto examine seriously the literatureon
both sides of this divide and present provocative challenges and proposals to
functionalist linguists (but see also Morgan 1982, Green 1982, Sadock 1984,
Nichols 1984).
This paper addresses one issue that is central to language yet problematic
for functional analyses: the notion of autonomy. The notion of autonomy
emerges from a undeniablefact of all languages, 'the curious lack of accord ...
* I would like to
thank Matthew Dryer, Mark Durie, Martin Haspelmath, Frederick J. Newmeyer,
and David Wilkins for reading and commenting extensively on earlier versions of this article; others
too numerous to mention for their advice and suggestions on particular points; and especially the
two referees, one of whom later revealed himself as Russell Tomlin. Their combined efforts vastly
improved the result. I accept responsibility for all flaws in the final product.
For general approaches to functionalism, see Dirven & Fried 1987, Givon 1979, Hickmann
1987, Langacker 1987, and Tomlin 1990.
490

AUTONOMY AND FUNCTIONALIST LINGUISTICS

491

between form and function' (Sapir 1921:89).One could say that the basic difference between structuralismand functionalismis the opposite direction each
goes in from this fact. The structuralisttakes the lack of accord to be evidence
for autonomy,while the functionalistinsteadfocuses on the 'unconsciousanalysis into individualconcepts whichis never entirelyabsentfromspeech, however
it may be complicated with or overlaid by the more irrationalfactors' (Sapir
1921:90).
Of course, there are many intermediatepositions that integratethe two realities Sapirdescribes. In fact the structuralist-functionalist
dichotomyis a continuum. This article discusses the issue of autonomy and how linguistic theories
must accommodatethe facts that are said to supportautonomy, while finding
a place for linguistic function. My goal here is to argue that structuralistand
functionalistapproachescan be compared,even if they cannotalways be reconciled; to raisethe fundamentalissues on which they differ;and to suggest points
at which differenttheorists are comparingsimilarphenomena.
The question of autonomyis not a simple one. Whatis claimed to be autonomous? In ?2.1, 1 arguethat the autonomythesis, and its functionalistcritiques,
have been applied to two very different linguistic systems. The autonomy debates of the 1970s largely centered around the status of syntax, relative to
semanticsor pragmatics(thatis, semiotic function)-all withinthe grammatical
competenceof the individual.Around 1980there were parallelconceptualshifts
both in generative grammarand in functionalist thinking. The debate shifted
to the status of the grammar,that is, the linguisticknowledgeof an individual,
relative to its larger socio-psychological context: language use, language
change, and languageacquisition. Because of this, the notion of function also
changed,from representingthe signifiedin the semiotic system of the grammar
to representingthe goals and purposesof the linguisticbehaviorof the individual
in the speech community(externalfunction).
In ?2.2, 1 arguethat the autonomythesis can be brokendown into two claims
about a linguistic system: a weaker claim, that it includes arbitrariness,and a
strongerclaim, that it is self-contained.Thus, for any given functionalanalysis
(henceforthFA), one must ask: which of these autonomyclaims does it actually
challenge?
Withrespect to an FA in syntax, thereare three logicallypossible conclusions
that might be drawn, with functionalist positions associated with each of the
conclusions:
(1) a. syntax is arbitrary and self-contained (AUTONOMIST FUNCTIONALISM)

b. syntax is arbitrary,but not self-contained (MIXED FORMAL/FUNCTIONALISM and TYPOLOGICAL FUNCTIONALISM)

c. syntax is not arbitraryor self-contained(EXTREME FUNCTIONALISM)
In ??3-6, I survey the four named functionalismsand the FAs that support
them. In ?3, it is shown that many FAs simply provide a semiotic function that
can explainconstraintson syntactic behaviorfor which syntactic analyses have
been offered; they do not challenge autonomy. In ?4, I argue that there are
valid FAs that demonstratethat syntax is not self-contained-and that there

492

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

seems to be a general consensus on this point at the present time. Section 5

describes typological functionalism, in which arbitrariness is given a much more
limited role than in the mixed approaches described in ?4. Section 6 describes
an untenable extreme position that was advanced briefly and which unfortunately has become a straw man for criticism of functionalist research.
I then turn to FAs of grammar. Again, there are three logically possible
conclusions that can be drawn, two of which characterize theoretical positions
currently held by linguists:
(2) a. grammar is arbitrary and self-contained

FORMAL(CONTEMPORARY

ISM and EXTERNAL FUNCTIONALISM)

b. grammar is arbitrary, but not self-contained

(INTEGRATIVEFUNC-

TIONALISM)

c. grammar is not arbitrary or self-contained (not attested)

In ?7, I briefly describe the positions under 2a, and point out that FAs of
language use, change, and acquisition do not in themselves challenge the autonomy of grammar. In ?8, I argue that arguments for the non-self-containedness of
grammar must emerge from an account of the interaction between the linguistic
cognition of an individual and the social context of language use in which the
individual is situated. An argument against the self-containedness of grammar
is developed based on variationist research (?8.1), and a series of hypotheses
and predictions that could be used to test the nonautonomist position are advanced (?8.2). A summary is provided in ?9.
2. BASIC ISSUES AND DEFINITIONS.The

structuralist-functionalist

debate

comes down to the question: to what extent can 'grammar' be considered a

system of formal structures that is autonomous from its function? In this section, this question is shown to be multiply ambiguous. Different linguists-and
sometimes the same linguist in the same paper-have employed different interpretations of the terms grammar, autonomy and function. Likewise, a variety
of functional analyses have been put forward as arguments against autonomy.
Subsequent sections will disentangle which FAs count as arguments against
which autonomy hypotheses.
2.1. SYNTAX, GRAMMARAND FUNCTION.As indicated in ? 1, 1 will use the term
grammar in this paper to refer to an individual's knowledge of their language.
Hence, grammar includes knowledge not only of syntax, but also knowledge
of the conventional semantic, pragmatic and discourse functions of the syntactic
forms. These represent

the SEMIOTICFUNCTIONof a syntactic

structure,

the

function actually encoded by the structure in the speaker's mind and used by
the speaker in producing and comprehending utterances. That is, the semiotic
function represents the signified of the linguistic sign, while syntax (and morphology and lexicon) represents the signifier. Some functional properties may
be statistically correlated with a particular structure in language use, but not
actually encoded by the mental grammar; semiotic function refers only to the
latter.
A grammar is a semiotic system including both syntax and semantics, as
even Chomsky notes (e.g. Chomsky 1977:44). The expression autonomy of

AUTONOMY AND FUNCTIONALIST LINGUISTICS

493

syntc.axrefers to the relationship between the syntactic component and the com-

ponents used to represent the semiotic functions. Functional analyses in syntax

argue for a certain degree of interpenetration of semiotic function and syntactic
form. In ?3-6 I evaluate particular kinds of FAs as to whether they constitute
valid arguments against the autonomy of syntax in any of its forms.
All parties to the 1970s debate agreed that semantics was a component of
the grammar. However, by the 1980s, other critics questioned the autonomy
of the grammatical system as a whole-from the forces or processes involved
in language change, language use, and language acquisition. The fundamental
relationship now highlighted is between the psychological linguistic system of
the individual-called grammar-and language as a social entity, that is, something resulting from the communicative interactions of individuals in social
contexts. For lack of a better term, I will use language3 to refer to this latter
concept.
This is a quite different relationship than that between syntax and semiotic
function, and the term function has been redefined to accommodate this new
perspective.

In this context,

function refers to EXTERNAL FUNCTION: the pur-

poses to which the grammatical knowledge of the speaker is put in social interactions, most importantly the communication of information but also other social
purposes. It is called external because it pertains to phenomena external to the
psychological grammar. Nevertheless, these external functions interact with
the grammar, the result being language use, language change, and language
acquisition.-

The question now is, how much interaction is there, and how much does it
affect an individual's grammar'? The expression autonomy
of graimnmar will
be used to describe this issue-the relationship (or lack thereof) between the
grammar of an individual and the external functions embodied in language as
a social phenomenon.3 This issue has importance beyond linguistics. The social
sciences have long struggled with the dichotomy between the social and the
psychological. Different subdisciplines of linguistics find themselves on either
side of the divide. A better understanding of the issues underlying the autonomy
of grammar may shed light on this very interdisciplinary problem. These issues
will be discussed in ??7-8.
So when the autonomy debate was resumed by Newmeyer in the 1980s (Newmeyer 1983), it was a different sort of autonomy that he was arguing for, and
2 There is another
type of function that is often referred to in the literature, which can be called

Some European structuralists argue that (systemic) 'functional' forces cause

languages to change in order to create or preserve symmetry in phonological, morphological, or
syntactic paradigms (e.g.. Martinet 1952). This is not an external function as defined here. That
is, preserving symmetry in formal paradigms in itself is not an interactive (communicative) goal.
Only if preserving symmetry can be empirically demonstrated to serve a communicative goal can
it be an external function. Some critiques of FAs (e.g. Labov 1994:547-68) are directed at systemic
FAs. We will not discuss systemic FAs further here.
3 The autonomy of syntax and autonomy of grammar hypotheses must not be confused with yet
another autonomy hypothesis, the autonomy of language from other human cognitive (and social)
abilities. The autonomy of language hypothesis is logically independent of the other two autonomy
hypotheses.
SYSTEMIC FUNCTION.

494

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

it was a different sort of functionalism that he considered to be autonomy's

opponent. Discussions about autonomy and function, however, still confuse
the two issues. Newmeyer 1992 does not clearly distinguish the two claims: in
his argument that generative grammar already accommodates functionalist
iconic principles, he argues that generative grammar already includes a semantic component, which does not tell us anything about the autonomy of grammar.
And in many functionalist accounts of syntax, evidence that a particular construction has a discourse function is taken to entail that that function is external
to the grammatical system, when it is in fact simply part of the signified for
that construction (see ??4, 9).
2.2. THE DEFINITI()NOF AUTONOMY.The term AUTONOMYhas been used am-

biguously in the literature to express different claims. 1 argue here that there
are three different claims associated with autonomy that are relevant to the
debates on the autonomy of syntax and the autonomy of grammar.
Since autonomy has been central to generative grammar, it is logical to begin
with the generative account as it was applied to the autonomy-of-syntax problem. The principle of autonomy of syntax is most extensively discussed by
Chomsky (1977, esp. pp. 37-44). A succinct formulation of this autonomy thesis
as generally understood is given by Newmeyer: 'According to the autonomy
thesis, there exists a set of nonsemantic and nondiscourse-derived grammatical
properties whose principles of combination make no reference to system-external factors' (Newmeyer 1992:783).
The hypothesis of autonomy of syntax is not a take-it-or-leave-it claim. It
can be broken down into three subclaims; the three need not be accepted or
rejected in toto. The three subclaims are:
(3) a. At least some elements of syntax are arbitrary (ARBITRARINESS);
b. The arbitrary elements participate in a system (SYSTEMATICITY);
c. That system is self-contained (SELF-CONTAINEDNESS).

The first and in some respects most fundamental property of autonomy is

the syntactic component contains elements and rules of combiARBITRARINESS:
nation that are not derivable from semantic or discourse categories and their
combination. By not derivable, it is meant that the syntactic element (say,
adjective) or rule of combination (say, passive) cannot be replaced with some
combination of semantic and/or discourse properties and produce the same
(correct) predictions of syntactic behavior.
For example, the claim is made that one could not replace the category adjective with a set of semantic properties and correctly predict the distribution
of adjectives in English (and every other language), or replace the syntactic
categories of the elements of the passive construction and rules of combination
for creating the passive with a set of semantic or discourse-functional category
definitions and rules of combination, and correctly predict the acceptable and
unacceptable passives of English (and every other language).
Arbitrariness is a property of the mapping between form and semiotic function; if correct, it means that formal syntactic structure must be posited independent of semantic structures. In fact, arbitrariness alone appears to be what

AUTONOMY AND FUNCTIONALIST LINGUISTICS

495

Chomsky (1977:37-40) refers to by the phrase 'independence of grammar'.

Chomsky makes his case for the independence of grammar (in our terms, the
arbitrariness of syntax) by showing that some interpretations of ambiguous
sentences are excluded by pragmatic or real-world factors, but others are excluded by syntactic constraints. The existence of the syntactic constraints demonstrates the arbitrariness of syntax.
Chomsky distinguishes the arbitrariness of syntax hypothesis from the 'thesis
of autonomy of formal grammar' (Chomsky 1977:42), or as it is called here,
the autonomy-of-syntax hypothesis. The latter hypothesis is of course the
stronger one. Autonomous syntax in the generative view is not simply a random
inventory of lexical items and grammatical constructions governed by enough
ad hoc constraints and rules of combination to be descriptively adequate for
the range of facts to be analyzed. It is not just enough that the mapping relation
between meaning and form is (at least partially) arbitrary; the form end of the
mapping must itself be a system. We will take the term SYSTEM here to mean
that there is a set of interlocking regularities that structure the phenomenon in
question (in this case, the syntactic units and constructions of a language) that
'holds together' overall-tout se tient, as the Saussurean saying goes (Koerner
1989:404-6).
The degree to which everything in the syntactic system must 'hold together'
is generally accepted to allow a bit of leeway for advocates of the systematicity
of syntax. Some irregularity is not incompatible with systematicity in syntax.
Irregularities and exceptional features have to be stipulated in an autonomous
syntactic system, although such stipulation is kept to a minimum. While this is
clearly a necessary and reasonable allowance-no theory would be empirically
adequate if such an allowance were not made-it does lead to some degree of
subjectivity in determining whether the data merit the establishment of a system
in this sense. One linguist's elegant and principled system is another linguist's
grab-bag of ad hoc generalizations; this is true of both structuralist and functionalist theories.
Our definition of system deliberately leaves out the question of whether the
regular, interlocking patterns in question include only the concepts of syntax
or not. If they include only the concepts of syntax, then we may say the system
is SELF-CONTAINED.
That is to say, the rules of the system interact with each
other but do not interact closely with the rules existing elsewhere. For the
autonomy of syntax, 'elsewhere' is generally taken to be semantics.
Again, there must be a loose interpretation of self-containedness in the autonomous syntactic system. Not all regularities in syntactic phenomena need be
attributed to the autonomous syntactic system; some may be governed by other
components of the grammar, specifically the semantic component (Chomsky
1977:45-48). Chomsky suggests for example that the parallel syntactic and semantic differences between restrictive and nonrestrictive relative clauses are
not a 'choice of grammar [syntax-WAC]' (Chomsky 1977:46). There must
also be interface rules between otherwise self-contained systems. For instance,
the semiotic relation between syntax and semantics requires at the least interface rules referring to both form and meaning. To the extent that interface rules

496

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

are allowed, a system is not self-contained in an absolute sense. However, it

is generally taken that syntax is self-contained if at least statements of categories
and rules of combination do not make reference to nonsyntactic elements (see
the Newmeyer quote at the beginning of this section, and see ?4 for further
discussion).
Given these caveats, the autonomy-of-syntax hypothesis is the hypothesis
that the syntactic component of the grammar is independent, (largely) systematic and (largely) self-contained with respect to the semantic, pragmatic and/
or discourse-functional components. But one need not buy the whole autonomy
package. Syntax may be arbitrary relative to its semantic counterpart, and
systematic also, yet not self-contained, the rules of syntax being interwoven
with those of semantics (or discourse, or other components of the grammar).
Arbitrariness and systematicity do not entail self-containedness. Arbitrariness
does not entail systematicity either; it is logically possible to believe that syntax
is arbitrary and not systematic as well as not self-contained, being merely a
residue of irregularities. In fact, those functionalists who accept arbitrariness
also accept systematicity (see n. 20 below). Hence, in the classification of functionalisms in this paper, acceptance of arbitrariness also includes acceptance
of systematicity. The remainder of this article will survey FAs and the corresponding functionalist positions, in order to see what sorts of arguments have
been, or can be, presented against which hypotheses of autonomy. Arguments
can be made against the self-containedness of both syntax and grammar. The
former argument is widely accepted in practice. The latter argument is much
newer, and many questions still remain. I begin with the autonomy of syntax,
proceeding from the fully autonomist position (?3) to various nonautonomist
positions (??4-6). Turning to the autonomy of grammar, we proceed in the same
fashion, from fully autonomist positions (?7) to the one plausible nonautonomist
position (?8). I will specify what substantive claims are being made on each
side, and-as much as can be done in a single article-evaluate those claims.
3. AUTONOMIST FUNCTIONALISM: COMPATIBLE WITH THE SELF-CONTAINEDNESS

OFSYNTAX.Many FAs of syntactic phenomena are in fact compatible with the

strongest version of autonomy, self-containedness. Specifically, an FA that
accounts for constraints on syntactic behavior in terms of the semiotic function
of a syntactic construction, but does not attempt to account for the structure
of that construction, is compatible with the self-containedness claim. I will call
such analyses AUTONOMIST
FAs, and linguists who make no stronger claims for
their analyses AUTONOMIST
these include Kuno, Prince, Ward
FUNCTIONALISTS;
and their associates and students (see Prince 1991:79 for a representative list).
Note that this does not mean that linguists holding stronger functionalist positions do not also offer autonomist FAs. However, autonomist FAs cannot be
used to argue against the self-containedness of syntax, let alone against weaker
claims of autonomy.
Autonomist functionalists are chiefly concerned with determining the discourse functions of the grammatical constructions sanctioned (or perhaps not
sanctioned) by the autonomous syntactic component, using naturally occurring

AUTONOMY AND FUNCTIONALIST LINGUISTICS

497

discourse data. No claim is made for a motivated (e.g. iconic) relationship

between the syntactic structure of the construction in question, such as English
it-clefts, and the discourse function that they perform-in fact, Prince explicitly
argues against iconicity in this particular example (Prince 1988:168-71). However, autonomist functionalists do argue for a conventional relation between
syntactic structure and discourse function, and moreover argue that 'many
sentences that are judged ungrammatical can be shown to be flawed for discourse reasons, with no need to invoke the (autonomous) grammar to account
for their infelicity' (Prince 1991:80-81). In other words, many constraints on the
behavior and/or distribution of sentences need not be provided by the syntactic
component. Nevertheless, the conventions governing the formation of the syntactic structures themselves are not provided by discourse (or conceptual structure, for that matter).
Thus, functional explanations for constraints on the distribution of sentence
types need not represent a direct attack on the autonomy of syntax. They may
reduce the number of principles found in autonomous syntax. And with the
increasing emphasis on principles governing constraints in generative syntactic
theory over the past two decades, any research program that attempts to replace
formal grammatical constraints with functional ones, if it is empirically successful, would narrow the scope of autonomous syntax considerably. But the sentence types themselves, that is the syntactic elements and the combinatoric
rules by means of which they are formed, are left to the syntactic component.
So there is no inherent contradiction in being a functionalist in this sense and
believing in autonomous grammar (cf. Morgan 1982:201).4
In fact, many FAs challenge purely syntactic approaches even for basic constraints in generative core grammar. The basic strategy in challenging a purely
syntactic analysis is to demonstrate that sentences excluded by a given syntactic
constraint are acceptable.5 This would imply that the given syntactic constraint
must at the least be weakened. At this point, a functional constraint may be
sought that accounts for the exceptions as well as the normal cases. If total
elimination of the syntactic constraint is possible, then the syntactic phenomenon can be shipped off to another autonomous component of the grammar. In
the remainder of this section, I briefly examine examples of FAs that purport
to provide a complete reduction: FAs of root transformations, the anaphoric
island constraint, and constraints on extraction phenomena.
Certain transformations were once argued to be restricted to the root S node
of a syntactic tree (Emonds 1976; examples from Green 1976:383):
4 One anonymous referee suggests that autonomist functionalists are not necessarily endorsing
autonomous syntax. Instead, they are providing descriptions of semiotic functions without making
any claim as to whether these functions in fact challenge or undermine the self-containedness of
syntax. However, at least for those Prince (1988:165-66) calls 'generative discourse analysts', it
appears that the self-containedness of syntax is assumed.
5 A possible response is to argue that a reanalysis occurs that changes the exceptionally acceptable sentences into a structure that is not subject to the syntactic constraint. But then the question
remains, how is reanalysis licensed in these cases but not in other structurally identical cases in
a noncircular fashion'?

498

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

(4) Never before have prices been so high.

(5) *Nixon regrets that never before have prices been so high.
There are many cases, however, in which root transformations are found in
subordinate clauses.
(6) 1 knewrthat never before had prices been so high. [Green 1976:383]
(7) I know that all too seldom does he bring her flowers.
I[Bolinger 1977:515]
(8) I'm gonna have breaukfastnowt, becauLseam 1 ever hungry!
[Lakoff 1984:474]
4-8
the
on
which
FAs
of
basic
converge is this:
story
Grossly simplifying,
root transformations

are all what Lakoff calls SPEECH ACT CONSTRUCTIONS, con-

structions which indicate some feature of an illocutionary act. Illocutionary

acts are normally performed in main clauses; hence the apparent restriction to
root Ss. But there are a number of contexts in which a speaker may perform
an illocutionary act in a subordinate clause, and use a speech act construction
in the process (see Hooper & Thompson 1973, Green 1976, Bolinger 1977,
Lakoff 1984 for the complex details). This argument appears to have been accepted; more precisely, root transformations per se are no longer a part of core
grammar and the phenomena analyzed in the references mentioned here have
dropped out of the autonomous-syntax literature.
Similarly, Ward et al. 1991 show that pronominal reference inside anaphoric
islands is possible, contrary to the anaphoric island constraint (Postal 1969):
(9) 1 think if I were a PERUVIAN I wouldn't want to live THEREffor the next
couple of years.
{Ward et al. 1991:470, from a conversation]
They argue that this phenomenon is governed by the same pragmatic principles
governing pronominal anaphora, and the syntactic anaphoric island constraint
should be dropped; and in fact it also has disappeared from the literature on
core grammar.
A similar tack has been taken on phenomena dearer to the heart of formal
syntacticians, the constraints asSociated with extraction phenomena (Kuno
1976, 1987; Erteschik-Shir & Lappin 1979, Grosu 1981, Lakoff 1986, Takami
1989, Deane 1991). This is a far more complex area of grammar than speech
act constructions and we can only outline an equally oversimplified version of
the proposed FAs. These constraints are manifested in a class of constructions
usually called extraction constructions (including wH-questions and relative
clauses). We may divide the extraction constructions into three parts: the extracted element, the matrix phrase from which the element was extracted, and
the intervening syntactic structure.
In essence, syntactic accounts argue that syntactic properties of the intervening structure render an extraction construction ungrammatical, while FAs point
to exceptional extraction as counterexamples to the syntactic accounts.
Who did you see a picture oJY
(10) a.
[Takami 1989:309, after the literature]
b. *Who did you destroy a picture of?
[ibid.]

AUTONOMY AND FUNCTIONALIST LINGUISTICS

(11) a.

499

did he grab and write to his Congressman?

[after Deane 1991:23]
b. Who did he grah his pen Cind write to?
[after Deane 1991:24]
c. Whaltdid he gralb clnd throw to his sister?
d. How manlycourses caInyol tClkefor credit, still realnIinscane, and
get all A 's in?
[Lakoff 1986:153]
Deane's FA, the most general proposed, adapts Anderson's (1983) spreading
activation model of memory and attention and argues that cognitive or discourse
properties of all three parts of an extraction construction jointly license or
prohibit extraction. Very roughly, the extracted element and the matrix phrase
must be in the center of attention while the rest must be backgrounded.6 Normally, the lexical head noun that appears to block extraction is informative
enough to attract the focus of attention (as in lOb), but in lOa it is evoked by
the semantic frame of see and thus permits extraction. In 1la-b, Deane argues
that the first event is a preparatory action for the second event in the coordinate
structure. Hence the first event is backgrounded (although the structure remains
coordinate; see Lakoff 1986); extraction is possible from the second clause
alone (but not the first clause alone) because it alone is the center of attention.
The normal case for coordination is for both events jointly to be the center of
attention, and so across-the-board extraction from all conjuncts is possible
(I Ic); but any combination of focused/nonfocused events is possible (lid).
All of these FAs postulate a semiotic function associated with the construction in question and deduce the syntactic behavior from the conventional function. Root transformations are speech act constructions; pronominal anaphora
denote a salient entity in the discourse; and the extraction constructions signify
a certain distribution of focus of attention. Likewise, Prince's accounts of various clefting and fronting constructions (Prince 1978, 1981) specify an informational status that they signify. This important point is often overlooked. Some
functionalists claim that these FAs provide evidence against the autonomy of
GRAMMAR,
because they involve semantic, pragmatic and discourse-functional
principles. But autonomist FAs make claims primarily about facts INthe grammar, namely that certain words or constructions have particular conventional
discourse functions or cognitive statuses-that is, semiotic functions. It is by
virtue of the semiotic functions of particular constructions that functional principles account for the syntactic constraints.7
4.

MIXED

* What

FORMAL/FUNCTIONALISM:

CHALLENGING

THE SELF-CONTAINEDNESS

OFSYNTAX.
As demonstrated in ?3, complete reductions of syntactic constraints
6 Kuno 1976 argues that the extracted element
must be topical in order to be extracted, while
Takami 1989 argues that the matrix must be the informational focus; Deane argues that both requirements must be met to license extraction, and that both requirements may be analyzed in terms of
focus of attention.
7 Of course, opinions may differ as to what needs to be specified as conventional and what is
derived from pragmatic principles; 'radical pragmatics' (Cole 1981) tries to maximize the latter
with respect to the former.

500

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

to functional constraints do not challenge the self-containedness of syntax,

though they may shrink the autonomous syntactic component significantly.
Partial reductions of syntactic constraints to functional principles, however,
pose a more serious problem for the self-containedness of syntax. If only a
partial reduction is possible, then we are left with a partially structural, partially
functional account, and the question can be raised as to how self-contained the
syntactic component is. I argue here that such analyses are quite common, and
in fact cannot reasonably be avoided. I will call such analyses MIXED FORMAL/
FUNCTIONAL analyses, and the position that accepts such analyses MIXED FORI will conclude
MALISM/FUNCTIONALISM.

this section by suggesting that indeed

many structuralist as well as functionalist theories are mixed, that is, they
accept mixed analyses.
Existing FAs of control theory and binding theory do not fully replace syntactic accounts by functional ones. Ladusaw & Dowty (1988) argue that thematic
control is determined by the semantic properties of the situation being described
rather than thematic roles, but assume that obligatory control is at least partially
syntactically determined. Farkas (1988) argues that the semantic relationship
of responsibility determines the difference between promise- and persnuade-type
obligatory control, but allows a stipulated but semantically motivated extension
of her principle to accommodate the multiple control possibilities (for some
speakers) in examples like 12 (Farkas 1988:47).
(12) THE PUPIL asked THE TEACHER to leave early.
The literature on binding, both syntactic and functional, is as voluminous
and the phenomenon is as complex as with constraints on extraction. Here
there is also some question as to whether binding is a unified phenomenon.
The contrast between anaphoric pronouns (Pronominals) and lexical NPs-the
Principle B/C contrast-has been argued to be purely pragmatic, based on
information structure and attentional phenomena (see inter alia Bolinger 1979,
Koster 1986, Ariel 1990, van Hoek 1992, Gundel et al. 1993). In fact, it is widely
accepted that most pronoun/NP choice is determined by pragmatic factors
(Reinhart 1983, Ward et al. 1991:440), and the dispute centers around whether a
syntactic condition such as c-command is an appropriate description for certain
intrasentential pronoun uses-for instance, Bolinger 1979 explicitly argues
against the syntactic description; Koster 1986 suggests it is purely pragmatic
but sensitive to syntactic criteria; and Gundel et al. 1993 propose a general
attentional account of pronoun/NP choice tested in texts in a variety of languages (though not with attention to the c-command cases). If these analysts are
correct, then an autonomist FA can be provided for the Principle B/C contrast.8
Levinson (1987, 1991), on the other hand, argues for only a partial pragmatic
x Incidentally, the relationship between functional accounts of pronoun/NP choice and constraints on extraction constructions in terms of levels of attention (Ariel 1990 for the former, Deane
1991 for the latter) suggests that even for functionalists, these two phenomena are related, while
NP-movement phenomena are distinct; cf. Levinson 1991:133, fn. 31. This is an interesting example
where otherwise conflicting formalist and functionalist analyses both converge on the same ideep'
relationships among syntactic phenomena.

AUTONOMY AND FUNCTIONALIST LINGUISTICS

5O1

account of the Principle A/B contrast-the distribution and interpretation of

reflexives and reciprocals (Anaphors) vs. Pronominals. He argues against specifying (non)coreference as part of the meaning of the Anaphor or Pronominal,
depending on the status of Anaphors in the language. In B-FIRSTlanguages, with
no Anaphors or weakly grammaticalized Anaphors, Principle B is grammatical,
while Principle A is pragmatic; in A-FIRSTlanguages, with grammaticalized Anaphors, it is the other way around.
These mixed analyses are controversial. However, apparently mixed syntactic phenomena are quite commonplace in syntactic descriptions. I will use two
simple examples to illustrate the plausibility of the mixed analysis and argue
that an attempt to reanalyze the phenomenon as purely syntactic smuggles
semantic properties into the self-contained syntactic component.
Babungo is a Grassfields Bantu language (Schaub 1985). Its noun class system
is typical of Bantu languages, in that nouns fall into a range of noun classes,
and the noun classes are quite arbitrary in their class membership. That is, there
is no way one could completely predict noun class membership on semantic
principles, although Schaub notes some correlations, e.g. class 1/2 with humans
(1985:174; the paired numbers refer to singular/plural forms). These agreement
patterns indicate that noun classes are not only arbitrary but participate in a
system of syntactic rules. Modifiers agree with their head nouns in class, and
verbs agree with their core arguments in class. Anaphoric pronouns agree with
their antecedents in noun class in the same way. There is one exception to this
generalization: anaphoric pronouns that refer to humans, regardless of the noun
class of the human noun, take class 1/2 agreement (Schaub 1985:193). Anaphoric pronouns that refer to nonhumans agree in the noun class of their nominal
antecedent.
The facts as just described argue against the self-containedness of syntax. If
one ships out the human-anaphoric agreement generalization to the semantic
component, then a very unsystematic hole has been punched into the syntax
of agreement in Babungo, and hence into the system of a large range of syntactic
dependencies. If one tries to ship out a more natural class of phenomena, say
the phenomena of anaphoric reference, then one loses an important linguistic
generalization also: the anaphoric agreement patterns with respect to nouns
referring to nonhumans are the same as those for modifiers and verbs, which
remain in the syntactic component. The simplest way to save systematicity and
the linguistic generalization (applying Ockham's razor) is to say that there is a
single grammatical system here, but syntactic and semantic primitives coexist
peacefully in the statement of grammatical rules (using graminatical
in the sense
of grammar as defined in ?2.1).
Phenomena of the Babungo type are quite common in the world's languages: a
particular syntactic phenomenon is sometimes constrained by (at least partially
arbitrary) syntactic factors, sometimes by purely semantic or discourse-functional ones, and they are interconnected to a high degree. Many field linguists
write mixed grammatical descriptions, including some who call themselves
functionalists (e.g. Haiman 1980a, McGregor 1990). These functionalists assume the independence and systematicity of syntax. But in practice they deny

502

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

the self-containednessof syntax: they make reference to functionalcategories

and features in their syntactic descriptions. Although these functionalists do
not generally work in one of the officially named theoreticalframeworks,it is
a logically possible and empiricallyplausible intermediateposition, and something like it has been developed in Dik's FunctionalGrammar(Dik 1981)and
in Role and Reference Grammar(Foley & Van Valin 1984, Van Valin 1993).
All of these approachesfit the definitionof mixed formal/functionalismat the
beginningof this section.9
There is a solution to the Babungo problem that appears to save the selfcontainednessof syntax. One could provide an additionalformalfeature to the
lexical entries of nouns, let's call it [ + H], which is [+ H] for nouns referring
to humansand I - H] for nouns referringto nonhumans.Rules for agreement
other than anaphoricpronominalagreementwill requirematchingof the class
value with the class of the noun agreed with. The rule for anaphoricreference
will yield [class 1/2]for IL
+ H] nouns and [class a] for I-H] nouns (a referring
to the class feature for the L-Hi nouns). This may appear to be an ad hoc
descriptionof the anaphoricagreementsystem. But the mixed syntactic-semantic account also has to specify when the syntactic classes dictate agreement
and when the semantic property of humanness does; so both accounts are
equally ad hoc in this respect (but see ?5.1). And the purely syntactic account
preserves the self-containednessof syntax.
This strategy is in fact an instance of a general technique for reanalyzing
putative mixed syntactic-semanticrules. In the mixed formulation,there is a
grammaticalrule that makes reference to both syntactic and semantic primitives. The purely syntactic formulationshifts the semantic part of the mixed
formulationto a mappingrule between syntax and semantics, namely the mapping rule that specifies that the syntactic feature [? H] is mapped into the semantic feature [? HUMAN]. In general, this strategy can be employed so as
to restrict reference to semantic primitives to mappingrules between syntax
and semantics.
Nevertheless, an importantgeneralizationhas been lost in the purelysyntactic account. There is a precise match between the distributionof the [?H]
featureandthe semanticpropertyof humannessof referents.Thatis, the feature
? H] CAN be reduced to a nonsyntactic primitive(cf. Newmeyer's definition
of autonomyat the beginningof ?2.2). The self-containednessof syntax can only
be maintainedby essentially replicatinga semantic primitivein the syntactic
component. Semantics is smuggledinto the syntax via a syntactic feature that
duplicates a semantic one. If the employment of this sort of reformulationof
syntactic rules is taken as proof of the self-containednessof syntax, then the
self-containednessof syntax is an unfalsifiableproposition.'0
9 Van Valin refers to Role and Reference Grammar as a 'structural/functionalist' theory (Van
Valin 1993:2).
'0This last statement must be qualified. The self-containedness of syntax, while not falsifiable
in this sense, may be proven to be true if every case in which a grammatical rule is described to
be sensitive to a semantic, pragmatic or discourse-functional property is empirically demonstrated
to be false. This is unlikely and as we will see shortly, is hardly an issue any more. I am indebted

AUTONOMY AND FUNCTIONALIST LINGUISTICS

503

One might consider the Babungo example as an instance in which the price
to pay for semantic smuggling is relatively small, since the agreement rules are
by and large governed by syntactic factors. However, one finds examples of
all proportions of mixture of syntactic and semantic or other functional rules
in the grammars of the world's languages. Obviation in Cree is an example of
the opposite balance between syntax and semiotic function. Cree (and other
Algonquian languages) possesses a nominal inflection whose values are termed
PROXIMATEand OBVIATIVE.
In general, the rules for the employment of the
and
obviative
are
proximate
exclusively a matter of discourse pragmatics (see
for example the brief remarks in Wolfart & Carroll 1981:25-27). That is, there
is a mapping rule between the noun forms and the values [proximate, obviative]
and discourse-functional properties, and the distribution of proximate and obviative noun phrases is dictated by the discourse component of the grammar.
There are a few cases, however, in which choice of form is determined by
the syntax. Specifically, in a possessive construction the possessed item is
always obviative, though the possessor may be proximate or obviative (Wolfart
& Carroll 1981:47-49). In some sense, this is worse for the self-containedness
of syntax than the Babungo case. One could provide a wholly syntactic account
only by importing all of obviation into the syntax. In this case, the syntactic
component would dictate the distribution of proximate vs. obviative in possessive constructions. But it would only be able to specify that either value is
licensed in virtually every other syntactic context, and leave it to the discourse
component to filter out the discourse-functionally inappropriate utterances. In
contrast, a mixed formal-functional account would license only the syntactically
and discourse-functionally acceptable utterances.
As a matter of fact, it may be that semantic (or discourse) smuggling is no
longer a black market activity. Consider the grammatical structures of Headdriven Phrase Structure Grammar (HPSG; Pollard & Sag 1987, 1994) and Fillmore & Kay's version of Construction Grammar (1993). Both theories use typed
feature structures to represent grammatical structures, including elements and
rules of combination. Each typed feature structure contains both syntactic and
semantic information, as values for the (syn) and (sem) features respectively.
They are both semiotic theories in the Saussurean sense. Neither theory, as
far as I can tell, explicitly excludes reference to either (syn) or (sem) values or
subvalues in the formulation of grammatical generalizations, though particular
generalizations may involve only (syn) subvalues. Both of these theories are
mixed theories, by our definition. It is worth quoting the proponents of the
supposedly more formalist of these two theories:"1

to Matthew Dryer and Russell Tomlin for providing essential parts of the semantic smuggling
argument.
" At this point, it is also worth pointing out another use of the term formal in describing linguistic
theories, namely that a formal theory is one in which descriptions of grammatical structures and
generalizations are formulated in a rigorous mathematical metalanguage. In principle, it is quite
possible to characterize any formalist or functionalist theory in a rigorous mathematical metalanguage, although in practice rejection of formalism has often included rejection of mathematical
formalization.

5()4

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

HPSG differs from all the syntactic theories which have influenced its development, for it is
not at heart a theory of syntax. Rather, it is concerned with the interaction among all the
forms of information that bear upon the linguistic meaning relation, including (inter alia) both
the syntactic information borne by signs as well as their semantic content ... syntactic and
semantic aspects of grammatical theory are built up in an integrated way from the start, under
the assumption that neither can be well understood in isolation from the other. (Pollard & Sag
1987:16-17)

Finally, it is a serious question as to whether even recent versions of Chomskyan generative grammar are autonomous in the sense of a self-contained
syntactic component. For example, the theta-criterion (Chomsky 1981:36, 335)
imposes a restriction on syntactic structure depending on whether a theta-role is
assigned to a particular position, but theta-roles are semantic entities (Chomsky
1981:35). This appears to be a constraint on syntactic structures in which semantic structure plays a direct role. Matthews makes this point more generally,
discussing different passages in Chomsky's recent work (Chomsky 1981, 1986)
which appear to equivocate on whether certain components of the grammar (in
particular, LF) are syntactic or semantic.
In citing these passages I am not trying to suggest that Chomsky was confused or inconsistent,
but rather to illustrate what now seemed to be true, that in the 'principles-and-parameters'
model' to call one level 'syntactic' and another 'semantic' no longer made any real sense
...
Insofar as [the levels] were like those envisaged before what Chomsky calls his second 'conceptual shift', we could perhaps continue to use the old names. But if I have understood the logic
correctly, they were mere names. (Matthews 1993:245)

LF may have begun as a semantic level; when syntactic processes such as

movement were sanctioned in LF, it appears to have become a syntactic level;
but Matthews suggests that LF hasn't fully lost its semantic character in the
process. Properties internal to the grammatical system, some of which were
once called semantic, now appear to be part of the grammar; and what really
matters is whether they are a manifestation of Universal Grammar. More recently, Jackendoff writes, 'we ought now to be willing to consider mixed alternatives when the facts push us that way' (1992:30).
Thus, it appears that in fact even contemporary Chomskyan generative grammar is mixed. Instead, it appears that Chomsky's attention has now turned to
the autonomy of grammar. Chomsky describes 'real semantics' as 'the study
of the relation between language or language use and the world' (1981:324).
The minimalist program takes X (formerly LF) as the interface between the
grammar and the 'conceptual-intentional system' (Chomsky 1993:2), not as an
interface between a syntactic and a semantic component of the grammar. These
issues will be taken up again in ?7, after a survey of two stronger positions that
have been taken against the autonomy of syntax.
5. TYPOLOGICAL FUNCTIONALISM: LIMITING ARBITRARINESS IN SYNTAX. Some

typologists have argued for grammatical analyses that integrate form and function more profoundly than other mixed models. These accept that the grammar
includes arbitrariness, and they provide mixed analyses. However, the mixed
analyses are such that the arbitrary parts of the analysis are language-specific,
while the universal properties instantiated in the analysis are functional. These

AUTONOMY AND FUNCTIONALIST LINGUISTICS

5()5

FAs are more functional than mixed analyses with regard to universals, not
with regard to autonomy. We will call such analyses TYPOLOGICALFAs, and
their proponents TYPOLOGICAL
FUNCTIONALISTS.
The basic strategy for constructing a typological FA is to examine a correlation between syntax and semantics (or perhaps discourse function), seek a
functional prototype that is found across languages, and construct implicational
universals (particularly implicational hierarchies) holding between nonprototypical semantic types and the prototypical ones. The universal hierarchies and
prototypes are the crosslinguistic manifestation of particular kinds of relationships among semantic and/or discourse elements in a speaker's mind. These
functional elements and relationships partake in the grammatical knowledge of
an individual. Hence the grammatical system is mixed, but the mixture is of a
specific type: it is made up of functionally defined universal elements and relations as well as arbitrary language-specific elements and relations.
We illustrate this analytical strategy first with a typological FA of basic syntactic category membership-a fundamental syntactic primitive if there ever
was one. As with the FAs in ?3-4, this illustration is highly simplified (the
general analysis is based on Dixon 1977 and Croft 1991:36-148; other studies
are cited below).
The basic syntactic categories Noun, Verb and Adjective are generally defined in terms of morphological properties, mainly inflectional categories (tenseaspect, number, case, etc.), and syntactic distributional criteria (e.g. occurrence
with articles, auxiliaries, etc.). These properties are language-specific, but we
may compare the major syntactic categories of languages by comparing across
languages the function of the inflections and co-occurring words used to define
category membership and the semantics of the word classes that combine with
them. In particular, the functions of the usual inflectional and syntactic criteria
used to define the major syntactic categories are associated with the pragmatic
functions of reference, predication and modification (Searle 1969; Croft 1991:
101-26 integrates Searle's pragmatic analysis with conceptual-semantic and
discourse-functional analyses of the same concepts).
Dixon (1977) observes that in comparing adjective classes across languages,
using comparable syntactic-semantic criteria, certain semantic classes of property concepts form the prototype of the adjective category. Wetzer (1992)
slightly modifies Dixon's analysis and argues that even in languages where
adjectives appear to be a subset of nouns or verbs on the usual morphosyntactic
criteria, peculiarities in morphosyntactic behavior distinguish the adjectival semantic classes (properties) from true verbal semantic classes (actions or processes).
Croft (1991) extends this typological FA to nominal and verbal categories.
Certain semantic classes of words consistently display the full range of morphosyntactic behavior and the most simply encoded syntactic structure for the
pragmatic functions of reference and predication-that is, they are unmarked
in the typological sense (Croft 1990:64-154). Words denoting objects are the
least marked referring expressions, and words denoting actions are the least
marked predications. Consider the following English examples:

(06

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

(13) The BIRDflew away.

[object, reference]
[action, predication]
(14) The bird FLEWaway.
(15) That WASA BIRD.
[object, predication]
The word bird in 15 is considered syntactically to be a noun; it shares nounlike
morphosyntactic behavior with bird in 13, and in contrast it is unable to take
verbal inflections of the sort acceptable with flew in 14. However, bird in 15
is also characterized by the fixed use of a (or bare plural for plural subjects),
in contrast with ordinary nouns (as in 13) and the obligatory presence of the
copula in predication, in contrast with ordinary verbal predications (as in 14).
In other words, the English predicate nominal does not display the full range
of morphosyntactic behavior of a noun in a referring expression. Also, unlike
verbal predications, the predicate nominal requires a copula. And even though
the copula expresses some verbal categories such as tense, others such as the
progressive are unacceptable:
(16) *That was being a bird.
Moreover, in other languages, the predicate nominal displays even fewer
noun-like properties, in fact almost none at all, as in Classical Nahuatl (Andrews
1975:146):
(17) n- oqiiich -tli
1ST- man
-sG
'I am a man.
Predication of object concepts varies considerably across languages, but are
consistent in lacking some morphosyntactic nominal criteria in comparison to
reference with object concepts, and in lacking some verbal criteria in comparison to predication of action concepts. In other words, predication of object
concepts is nonprototypical in comparison to reference to objects and predication of actions, both of which are consistent in their coding and behavior across
languages.
If one brings adjectives (more precisely, word classes denoting properties),
into the picture, one can construct more precise crosslinguistic generalizations
about the predication of various sorts of semantic word classes. The (non)use
of a copula conforms to a hierarchy of predication:
(18) Action word (verb) < Property word (adjective) < Object word (noun)
For instance, if a copula form is required for any member of the hierarchy in
a specific language, then it is required for any member to the right of the hierarchy (Croft 1991:130). The same hierarchy is found for the possibility that a
word will be inflected like a main predicate (Stassen 1992). The (non)use of a
copula and inflectability with the inflectional categories of predication indicate
the prototypicality of action words as predicates and the increasing nonprototypicality of object words as predicates-whether predicate nominals look more
nounlike (as in English) or verblike (as in Classical Nahuatl).
The hierarchy of predication in 18 is universal, that is, there is substantial
empirical evidence that it holds for all languages. As with universals postulated
by structuralist theories, it is hypothesized that some cognitive structure is
responsible for the crosslinguistic universal, namely an ordering of semantic

AUTONOMY AND FUNCTIONALIST LINGUISTICS

507

classes with respect to the pragmatic function. This is a property of the semantic
and pragmatic components of the grammar. If we return now to the description
of a specific grammar such as that of a speaker of English, then the grammatical
representation of the predicate nominal can make reference to this ordering of
semantic classes with respect to the encoding of predication.
In fact, reference to the universal ordering relation simplifies the statement
of the rules for the specific grammar. For instance, all that needs to be specified
as a language-specific fact of English is that predicate adjectives require a copula; the fact that predicate nominals also require a copula follows from the
universal ordering. In this way, the statement of the rule is mixed, as with the
theories described in ?4. But (ideally, at any rate), the primitives that are taken
to be language universal are defined functionally, while the arbitrary statements
of how function is encoded into form are language specific. The goal of typological functionalism is to provide a functional characterization of those aspects
of grammatical structure that are empirically demonstrated to be universal. The
fact that language-specific structures may be arbitrary, that is, not derivable
from function, is not of particular concern; in fact, it is one of the chief ways
in which grammars vary.
The typological analysis of basic syntactic categories just sketched out proposes a typological functional analysis of that core area of syntax. Another
core area of syntax that has been explored in great detail is the expression of
predicate-argument relations. The encoding of predicate-argument relations is
extremely complex, but much order is found in the typological variation. In
the typological functionalist account, the complexity is due to a large number
of semantic and discourse-functional factors, including animacy, referentiality,
topicality, event-related properties of participant roles, and a hierarchy of syntactic constructions with respect to ergativity, which interact to determine
crosslinguistic patterns of encoding grammatical relations, including case and
adpositions, agreement, transitivity and voice. Of course, such accounts can
be compared to theories of linking in generative grammar; however, that is far
beyond the scope of this paper.12
Typological FAs can also be applied to arbitrary syntactic facts that have
been used to support autonomist, or at best mixed formal/functional, syntactic
analyses. The result is still a mixed analysis of course, but one in which a
functionally motivated universal can be extracted leaving a language-specific,
often arbitrary, residue. I will illustrate this with two examples from Newmeyer
1992 and the Babungo example used in ?4.
Newmeyer discusses the tendency for morphologically incorporated nouns
to 'have less conceptual independence, as measured by their ability to have
independent reference or to be focussed or stressed, than do nonincorporated
nouns' (Newmeyer 1992:763). He mentions counterexamples in a footnote on
12 For various

typological perspectives on predicate-argument encoding, see Comrie 1978, 1989;

Croft 1991; DeLancey 1981; Dixon 1994; Foley and Van Valin 1984; Hopper & Thompson 1980;
Kazenin 1994; Mithun 1991; Moravcsik 1978; Silverstein 1976. 1993; Van Valin 1993; and references
cited in those works.

508

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

the same page, citing Sadock (1980, 1986). But the generalization (Mithun 1984,
1986), is not a tendency but an implicational hierarchy of incorporability.
demoted
externally
backgrounded
(19) nonreferential
< modifiable NP
< NP
< NP
noun
(Mithun's Type I)
(Type II)
(Type III)
(Type IV)
This hierarchy provides a universal constraint on language types such that
if a language allows incorporation of a semantic type at a certain point on the
hierarchy, then it allows incorporation of all types to the left of it on the hierarchy. Sadock's counterexample languages, in which highly referential nouns
can be incorporated, fall on the extreme end of that scale (Type IV), and even
within one of those languages (Southern Tiwa), Sadock finds an incorporability
hierarchy among the different NP types in it (Sadock 1985). The hierarchy
points to the least conceptually independent nouns being the first to be incorporated crosslinguistically. Of course, the cutoff point for how conceptually independent a noun must be before it can no longer be formally incorporated must
be specified in a grammar, but that arbitrary fact is language specific.
Another example of Newmeyer's pertains to the interpretation of the temporal order of events in Chinese serial verb constructions. Here the principle
invoked by typological functionalists is tense iconicity (Haiman 1980b): the
order of verbs or clauses reflects the actual order of events. Newmeyer notes
that although the tense iconic order is an entailment in the serial verb construction, it is only a (defeasible) implicature in the conjoined sentence construction
(analogous in structure and implicature to English conjoined sentences; 1992
776-77). The difference in behavior between coordinate and serial constructions
can be partially accounted for by grammaticalization. Grammaticalization Theory predicts that the more grammaticalized a construction is, the more conventionalized its semantic interpretation will be. Grammaticalization Theory also
provides phonological and syntactic criteria, as well as semantic criteria, for
determining the relative degree of grammaticalization of constructions (see for
example Heine & Reh 1984, Lehmann 1985, Hopper & Traugott 1993). The
Chinese serial verb construction is more grammaticalized than the coordinate
construction-there is crosslinguistic evidence that the serial construction is
related to the coordinate construction (Noonan 1985:140, Schiller 1990:
38)'3-and its semantic interpretation is more fixed than the coordinate construction. Again, Grammaticalization Theory provides predictions only about
degrees of formal and functional grammaticalization; the exact specification of
the point on the grammaticalization scale where an implicature becomes an
entailment must be provided for particular language grammars.
We now return to the Babungo noun class example. Although the Babungo
agreement rule makes reference to a semantic feature, humanness, it does so
'3 Newmeyer also gives an example of anti-iconic order in a sentence with a preposed before
subordinate clause as a further counterexample. However, tense-iconicity is only claimed to apply
to coordinate constructions and related structures, not adverbial subordination. The relationship
between clauses in adverbial subordination signifies a Gestalt figure-ground pattern (Reinhart 1984,
Talmy 1978). a conventionalized semantic relation that can override an implicature of tense iconicity
(see also Haiman 1983:812-13).

AUTONOMY AND FUNCTIONALIST LINGUISTICS

09

in only one seemingly completely arbitrary part of the rule. Why humans rather
than some other semantic class? Why anaphora rather than noun modifier agreement or verb agreement? A typological perspective provides answers to these
questions. Crosslinguistically, the semantic class that is least arbitrary in its
agreement patterns is that of humans: it is either semantically completely consistent (as with Babungo anaphora), or the most semantically consistent of noun
classes (as with many European sex gender systems). And Corbett (1979) has
identified an agreement hierarchy in which agreement with semantic rather than
syntactic features is more likely with syntactically more distant controllers; the
antecedent of an anaphoric pronoun is more distant from the pronoun than a
modifier is from its head or a predicate from its argument. Thus, even quite
arbitrary-looking facts such as Babungo human anaphoric agreement turn out
to participate in typological universal patterns that are part of human linguistic
competence in the typological view.
These examples show that typological functionalism posits an even tighter
integration of (arbitrary) form and function than the mixed approaches described in ?4. In the grammar of a particular language the functionally-defined
universals cannot be disentangled from the statement of language-specific arbitrary conventions such as the cutoff point for how conceptually independent a
noun must be for incorporation to be prohibited, or how typological prototype
categories are extended to nonprototypical functions. That is, the grammar of
a particular language includes both universal functional factors and languagespecific arbitrary conventions of the language. The arbitrary facts of the language do not make up a self-contained system. In fact, they cannot even be
understood outside the context of the functionally motivated universal patterns
of variation, one possibility of which they are instantiating. Universal grammar,
in the typological functionalist view, is in large part the full set of functional
categories, patterns and motivations, including their interactions and conflicts
(see ?7.2).
6. EXTREME FUNCTIONALISM: DENYING ARBITRARINESS IN SYNTAX. In 1979 two

papers were published that rejected even the arbitrariness of syntax. Garcia
(1979) denies the existence of arbitrariness in language beyond the arbitrariness
of the form-function link between single words. Since Garcia accepts the arbitrariness of word meanings, she does not reject the independence of grammar.
In her view, grammar exists, but it consists only of a lexicon; syntax is derived
from discourse, which is outside the grammatical system. A similar argument
is proposed by Kalmar (1979), specifically for predicate-argument relations in
(following Nichols
Inuktitut. We will call this position EXTREMEFUNCTIONALISM

1984).
The evidence of the languages of the world is overwhelmingly against extreme
functionalism, which probably explains why it has hardly ever been advocated.
Beyond the arbitrariness of the lexicon, there is ample evidence of relations
between syntactic structure and its function that are arbitrary in every human
language. Yet there are several reasons to bring up this functionalist position
here.
First, it is a logical possibility that must be considered and rejected; we may

510

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

conclude that the most radical version of functionalism is invalid. Second, all
other functionalists accept the existence of arbitrariness, a fact which has not
always been noted by their critics. Thompson discusses this explicitly in her
response to Newmeyer 1991: 'he claims that "... in the functionalist view,
grammatical patterning mirrors discourse function in a direct way" [Newmeyer
1991:4] ... I am not aware of any functionalist who would take such a position.
It is obvious to any grammarian that there is much in grammar that is arbitrary'
(Thompson 1991:93).
Finally, the few direct critiques by formalists of anti-autonomy claims (Newmeyer 1983, 1992; Morgan 1982; Sadock 1984) equate autonomy with arbitrariness. With respect to the question of whether basic syntactic categories have
a functional basis, Newmeyer writes: 'However, the question we must ask is
whether they [basic syntactic categories] can be derived INTHEGRAMMATICAL
SYSTEM by some algorithm from their extragrammatical function' (Newmeyer
1992:785); and he answers this question in the negative. The lack of an algorithm
is due to some arbitrariness in the relationship between syntactic form and
meaning or function: if the relation were not arbitrary, there would exist an
algorithm deriving the former from the latter. The same points are made by
Sadock (1984) in a critique of Kalmar's (1979) analysis of case relations in
Inuktitut.
These criticisms attack a straw man and defend only the weakest version
of autonomy. None of Newmeyer's adversaries disavows the arbitrariness of
language. Admittedly, Morgan and Sadock were attacking the extreme functionalist view of Garcia and Kalmar. But Newmeyer assumes that arbitrariness
is sufficient to demonstrate autonomy: 'these principles [of combinations of
phonological segments] cannot be formulated in purely phonetic terms. That
is, autonomous phonological rules exist ... [T]he principles of combination governing [syntactic] categories cannot be stated in semantic or discourse terms.
That is, autonomous syntactic rules exist' (Newmeyer 1992:786). Yet arbitrariness is only part of the autonomy story, one that all parties to the debate accept.
The more interesting issues, for which there are plausible alternative views, are
the self-containedness of syntax (see ??4-5 above) and the self-containedness of
grammar, to which we now turn.
7. CONTEMPORARY FORMALISM AND EXTERNAL FUNCTIONALISM: COMPATIBLE

OF GRAMMAR.
The autonomy of grammar inWITHTHESELF-CONTAINEDNESS
volves the relationship between the grammar and language, that is, the individual cognitive system and the social context of language use, including language
acquisition and language change. Again, we may ask if there are FAs that are
compatible with the strongest sense of autonomy of grammar, self-containedness, and if there are FAs that challenge self-containedness, as well as
systematicity or arbitrariness. In this section, I note that FAs of grammar operating in language use, language change, and language acquisition are compatible
with the autonomy of grammar. First, however, a brief clarification of the contemporary formalist position is necessary.

AUTONOMY AND FUNCTIONALIST LINGUISTICS

FORMALISM.To my knowledge,
7.1. CONTEMPORARY

511

there do not exist any

extreme formalists. No formalist denies the possibility that functional forces

may be involved in language change, language acquisition, and language use:
In short, the autonomy thesis of generative grammar maintains that, as a synchronic system,
grammatical principles are not replaceable by discourse principles, although the latter may
indeed exert an effect on the former that could lead in the direction of a new (autonomous)
synchronic system ... IT]here is no necessity that some factor internal to the autonomous
grammar is responsible for predicting any given instance of grammatical change. (Newmeyer
1992:785)

The consequence of this position is that some, perhaps many, aspects of language change, language acquisition, and language use are beyond the reach of
generative grammar: 'My feeling is that in truth generative grammar has little
to offer in the way of an explanation of phenomena such as these [functional
factors in grammaticalization] ... a [functional account] falls outside the explanatory ___domain of generative grammar' (Newmeyer 1992:777).
Likewise, generative grammarians do not deny the possibility that some systematic aspects of the grammar itself may not be due to the expression of the
innate UG in a particular individual's core grammar. Chomsky describes those
aspects of grammar that are not attributable to UG as peripheral. Chomsky
admits the likelihood that the periphery will 'contain independent structure as
well' (Chomsky 1981:8),'4 and cites the accessibility hierarchy, a typological
universal, as an example.
Hence, the scope of the claims of an innate UG as an explanatory factor for
grammars is limited. Not all forces dynamically affecting grammars are to be
accounted for by UG principles, or by the possible settings of UG parameters.
Nor are all systematic aspects of a grammar to be accounted for by the principles
and parameters of UG. All that formalists are claiming is that certain systematic
aspects of grammar are attributable to UG, and certain dynamic processes
affecting grammar are determined by settings of UG parameters. In other
words, contemporary generative grammar is not intended to be a complete
theory of grammar, nor is it intended to be a complete theory of language (in
our sense of that term; cf. Tomlin 1990:156). '5
The grammar itself, however, is self-contained from factors operating in the
social context. To be sure, there must be some interaction, just as there must
be interaction between syntactic form and semiotic function in the grammar.
This interaction, between the grammar and its use in social contexts, occurs
in language processing. But the processing mechanism, or any external (functional) factors, cannot alter UG in the formalist view. The only exception to this
14 By independent Chomsky means independent of the innate principles and parameters of UG.
'5 The rhetoric is less modest. The part of grammar that UG purports to account for is called 'core

grammar' (Chomsky 1981:7); the 'periphery' is described as consisting of 'idiosyncratic factors, as

contrasted with the more significant reality of UG' (Chomsky 1981:8). Chomsky suggests that if
complexity due to functional utility or evolutionary accident turns out to be largely true of language,
then UG would not be interesting (Chomsky 1981:14); he does not consider the possibility that a
universal grammar built around functional principles may be interesting in itself.

512

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

is the setting of parameters in language acquisition-and language acquisition is

treated as a purely cognitive process. Once the language is acquired (natively),
with the parameters set, no external influences can affect the grammar; it is a
fixed, self-contained system (cf. Newmeyer 1992:784-85).
7.2. EXTERNALFUNCTIONALISM.
Contemporary

formalism,

including the hy-

pothesis of the autonomy of grammar, is logically compatible with the existence

of functional forces operating external to the grammar, although formalists believe that the latter cannot account for all (or perhaps any) of the interesting
properties of language (in our sense). FAs of language acquisition, language
use, and language change challenge or even deny the existence of an innate
UG. They are nevertheless compatible with the self-containedness of grammar.
We call such FAs EXTERNAL
FAs, and the position associated with them EXTERNAL FUNCTIONALISM.

An example of external functionalism is Bates & MacWhinney's Competition

Model (Bates & MacWhinney 1987, 1989). Bates & MacWhinney state that the
competition model is offered as an account of language change, performance
and acquisition, but that it 'is not offered as a formal model of LINGUISTIC
COMPETENCE'
(Bates & MacWhinney 1987:160; cf. Bates & MacWhinney 1989:
3).16 Their competition model takes for granted the existence of a semiotic
grammatical system of some sort, but assumes that speakers manipulate that
system in acquisition, use, and change in accordance with functional principles.
The competition model uses cue validity and cue strength and principles of
competition derived from connectionist models in order to test hypotheses of
comprehension of utterances, without suggesting that those principles shape
or alter the grammatical structures they manipulate. Bates & MacWhinney's
external FAs of language acquisition are functional accounts of dynamic processes that presuppose the existence of a grammar. External functional theories
of language use, such as the competition model, presuppose an existing selfcontained grammar.'7
'6 Bates and MacWhinney (1982:178-90, 1987:160) define four levels of functionalism: diachronic
change, performance, acquisition, and competence. Their competition model is intended to cover
only the first three levels, i.e. those levels external to grammatical knowledge. 1 see no reason to
distinguish the three external situations in terms of levels. Also, it is not clear that Bates and
MacWhinney distinguish the autonomy of syntax from the autonomy of grammar. In their 1982
paper they describe the fourth level thus: 'the grammar or system of representation that mediates
the interaction between form and function can be fully described in terms of natural functional
categories and performance constraints' (Bates & MacWhinney 1982:187). This appears to be a
statement of extreme functionalism; but there are several varieties of functionalism that are less
extreme yet assume a role for function in the grammar. In their 1987 paper they veer in the opposite
direction: 'rules in the grammar make direct reference to semantic and pragmatic symbols' (Bates
& MacWhinney 1987:160). This seems to be a description of the fairly uncontroversial mixed
formal/functionalism. It should also be pointed out that Bates & MacWhinney are provisional
external functionalists; if a functionalist (nonautonomist) model of competence existed, they would
probably endorse it (Bates & MacWhinney 1989:32).
17Actually, Bates and MacWhinney's competition model is a model of comprehension only; it
is not clear how well it will extend to production (I am grateful to an anonymous referee for pointing
this out).

AUTONOMY AND FUNCTIONALIST LINGUISTICS

513

Another sort of external functionalist model emerges from typological research. Since these linguists (unlike Bates & MacWhinney) do argue against
the self-containedness of grammar, we must examine their external FAs in
somewhat more detail. I argue here that while these external FAs appear to
provide useful models of language change and the resulting typological variation, they do not in themselves challenge the autonomy of grammar.
These models, like Bates and MacWhinney's, are competition models: factors determining the expression of grammatical form compete with one another,
thereby causing a language change. The classic example of competition is found
between phonological processes and morphological processes in language
change. An example of this sort of competition is found in Slavic noun declensions (Greenberg 1969:185-89). The elimination of the reduced vowels (jers)
of Slavic, a natural sound change that is presumably phonetically motivated,
led to a change in the nominal declension contrary to typological markedness
patterns. Typological markedness predicts that a more frequent (unmarked)
category is expressed by no more morphemes than the marked category, in
fact often by zero (Croft 1990:73). But in Slavic, the genitive plural of certain
stems (particularly feminine a-stems), an infrequent form, is expressed by zero,
while the very frequent nominative singular is expressed by a nonzero suffix.
Diachronically, there has been a subsequent change such that nonzero genitive
plural endings of other minor declensions have spread at the expense of the
zero genitive plural in almost all Slavic languages. (There has also been a less
widespread tendency for zero nominative singular endings to spread at the
expense of the nonzero feminine a-stems.) The opposite change-spread of
zero genitive plural-has not occurred, and is in fact predicted not to occur in
the diachronic competition model.
Typological functionalists argue that there is competition among universals
determining form-(semiotic) function relations as well, and that these determine
crosslinguistic patterns of variation. For example, Greenberg's original explanation for word order variation (Greenberg 1990 [1963]:60-64) involves competition between DOMINANCE (a preference for one order over its opposite) and
HARMONY (a preference for two word orders to be aligned). The occurrence of
any given word order in a language is due to dominance or harmony; if neither
principle holds, the word order pattern is predicted not to be found. Greenberg
did not offer a functional account of his competing factors, but Hawkins (1983)
attributed harmony to head-dependent relations (possibly a semantic factor,
but see Hawkins 1990 for a processing account) and a factor likening dominance
to heaviness (a processing explanation).
The main advocate of typological competition models is Haiman (1983, 1985).
Haiman posits two primary explanatory factors for form-function relations:
economy (the expression of more commonly used concepts in shorter forms)
and iconicity (the employment of a syntactic structure that is a simple one-toone mapping from the encoded conceptual structure in certain respects). Haiman offers iconicity and economy as deeper explanations for particular typological universals. For instance, iconicity is argued to account for the isomorphism
between conceptual distance and linguistic distance, which itself is argued to

514

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

account for universals such as 'In no language will the linguistic distance between X and Y be greater in signaling inalienable possession, in expressions
like 'X's Y', than it is in signaling alienable possession' (Haiman 1983:793).
Typological markedness patterns, as in the Slavic example given above, are
generally attributed to economy (Croft 1990:156-60).
Haiman notes that particular form-function mappings, iconically or economically motivated, may be in conflict in particular constructions, and typological
variation may be attributed to whichever principle applies in a particular grammar. For instance, a normally transitive expression, particularly one with an
indefinite or generic object, is often rendered intransitive crosslinguistically in
two apparently opposite ways (Haiman 1983:814). There can be demotion of
the object to an oblique, in accordance with the principle of linguistic distance
reflecting conceptual distance (for additional evidence, see Haiman 1983:
790-92). Or the object is incorporated-the opposite linguistic distance, but
incorporation is iconically motivated by lack of conceptual independence of
the object from the event (a transitive event requires two highly individuated
participants; cf. Hopper & Thompson 1980).8
These external FAs have some appeal as explanations for language variation
and change. The notion that certain properties of the form-function relation
inherently conflict provides a plausible explanation for why there is variation
in the syntax of the world's languages: the variation represents different nonoptimal grammatical coding strategies conforming to one or another competing
motivation. Competing motivation accounts, and unidirectional processes such
as grammaticalization (Hopper & Traugott 1993), also provide an explanation
for language change that does not require an appeal to language contact. Alternative form-function mappings may make their way into the grammar, or a
functionally motivated change in another part of the grammar will create an
unmotivated structure. Also, competing motivation models provide constraints
on the directionality of change or more generally on the directionality of deviation from the conventional rules of the grammar in any dynamic aspect of
language use, including performance 'errors' (see ?8.2.3) and language acquisition.
Technically, though, only one of two competing motivations is manifested
WITHIN a grammar. Iconic or economic principles are part of mixed typological
FAs. (In fact, functional explanations are treated as universals of form-function
relations within the grammar in Croft 1990:155-97.) But only one competing
motivation can be found in a (nonvariable) grammar (cf. Newmeyer 1994:72).
functional or otherl8 Explanations based on competition between principles-phonological,
wise-may appear to be unfalsifiable. This is not the case for nonvacuous competing motivation
models (Croft 1990:193). A nonvacuous model requires that there are logically possible language
types that do not satisfy any of the relevant competing motivations, and hence are predicted not
to exist; this makes the model falsifiable. lt also requires that not all competing motivations may
be satisfied at once, or else all languages would be predicted to be of the same perfect type. More
interestingly, competing motivation models may allow one to make quantitative predictions about
the frequency distribution of deviations from the rules of the grammar in terms of relative optimality
of language types; see also ?8.5.

AUTONOMY AND FUNCTIONALIST LINGUISTICS

515

A grammatical rule or construction represents a choice between two competing

motivations-for instance, an indefinite object is either demoted (distance) or
incorporated (independence); it can't be both at once. Haiman's principles may
offer a deeper FA of particular grammatical constructions, but they do not
challenge the autonomy of grammar.
However,

for a competition

model of LANGUAGE CHANGE to work, both com-

peting motivations must be present in the minds of speakers, and this, it might
be argued, challenges the autonomy of grammar. One motivation is part of the
grammar (that is, part of the typological FA) while the other is latent, potentially
available to trigger the novel form or analogical change. Economic motivation
for a nonzero genitive plural exists in the minds of speakers of Slavic even
when the genitive plural is zero and so the economic motivation cannot be part
of the grammatical representation. The latent motivation manifests itself in the
subsequent language changes, away from the zero genitive plural. Likewise,
there is a potential for grammaticalization of particular semantic categories
even when the grammar exhibits little or no grammaticalization (as yet) of the
category in question. In this way, functional factors external to the grammar
can affect the grammar and hence the grammar is not self-contained.
There are two problems with this argument. The first is that it has not been
demonstrated that the competing motivations postulated by typologists and
historical linguists are external functions, that is, are a result of communicative
interactions among individual speakers. As pointed out in ?7, these motivations
must be justified in terms of language use, that is, comprehending and producing
utterances in context. That is, the competing motivation that is not part of the
grammar does not obviously emerge from language use and its social context.
The principles that actually compete in particular cases are relatively specific-vowel
reduction and markedness, conceptual distance and conceptual
independence. Economy and iconicity provide the functional explanation for
these principles; but they still make reference only to (semiotic) function, a
conceptual structure. Some typologists have recognized that processing accounts must underlie economy and iconicity in order to link competing-motivation models to external function (Givon 1985:198, Croft 1990:252-55). There are
intuitively natural processing explanations-economy
represents the efficient
of
the
most-used
expression
concepts (Haiman 1985:11), and iconicity represents the simplest mapping of meaning onto form, namely a mathematical isomorphism (Givon 1985:189)-but no psycholinguistic evidence has been
adduced to support these analyses. An alternative external account might be
Gricean: iconicity and economy are essentially Horn's Q-based and R-based
principles respectively (Horn 1989:192-94), applied to form-function relations
within the grammar. But this connection has not been made in the functionalist
literature (cf. Newmeyer 1992:774, n. 24). Finally, a comprehensive theory of
motivations based on external functions (communicative, phatic, poetic, etc.)
is needed to define more precisely which motivations are applicable to which
syntactic constructions. All of these would strengthen external FAs of dynamic
linguistic processes.
The second problem with the argument is that the time scale of the dynamic

516

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

processes frequently appears to be much greater than the lifetime of an individual speaker. The changes in Slavic declensions described by Greenberg have
taken place over centuries, and the process is far from over (in the sense of
restoring an economically motivated system in all the daughter languages). Typological variation is also the result of language changes over very large time
scales. Even if typological external functionalists are successful in constructing
a model of external functions to motivate these language changes, one must
still bring it down to the scale of an individual speaker. One must explain how
the latent functional factors interact with the grammar in the speaker's mind,
and why seemingly unmotivated arbitrariness is at least as strong a motivation
for an individual speaker as the latent functional forces described here. Only
this would challenge the self-containedness of grammar.
8.

INTEGRATIVE

FUNCTIONALISM:

CHALLENGING

THE SELF-CONTAINEDNESS

OF

GRAMMAR.
In ? 1, I pointed out that with respect to the nonautonomy of grammar, only one position is currently held, the one challenging self-containedness
(but see also ?6). Here I briefly consider what stronger nonautonomist positions
would constitute and why nobody has advocated them, before turning to the
weaker position of non-self-containedness.
What would denial of even arbitrariness, that is, the independence of grammar, constitute? Essentially that would mean that all of the grammatical knowledge of an individual can be derived from principles of communication and
social interaction, and hence there is no mental grammar (except perhaps for
lexical knowledge). This position, like extreme functionalism, is immediately
falsified by the existence of any grammatical regularity with any degree of
arbitrariness. The denial of systematicity would argue that linguistic behavior
is not rule-governed, whether those rules were external (social-pragmatic) or
internal (the system of the grammar). This position is also immediately falsified
by the high degree of regularity in both cognition and social behavior.
This leaves the denial of only self-containedness. This would lead to the
counterpart of mixed or typological functionalism at the level of the grammar.
That is, linguistic phenomena may be systematic, and may be (partly) arbitrary,
but they would involve such a close interaction of cognitive and external social
factors that one could not reasonably describe the internal cognitive system as
self-contained, contrary to the positions described in ?7. The position denying
FUNCTIONALonly the self-containedness of grammar I will call INTEGRATIVE
ISM.19 Integrative functionalists include Givon, Bybee, Hopper, Thompson, and
Haiman.20

'9 Integrative functionalism is a bit of a misnomer. It is logically possible that the external factors
are not functional (communicative, phatic, poetic, etc.). What really matters here is the integration
of the psychological grammar with its social context. In fact, the linguists who argue against the
self-containedness of grammar assume these factors are (external) functional. Having issued this
warning about these linguists' assumptions, for the rest of this paper we will accept their assumption
that the external factors are indeed functional.
20 is not obvious from some of the rhetoric
of these linguists that they are denying only the
It
self-containedness of grammar (for representative passages, see Givon 1982:112. Bybee et al. 1994:

AUTONOMY AND FUNCTIONALIST LINGUISTICS

517

In fact, few if any arguments for integrative functionalism exist in the functionalist literature. This is because FAs in themselves do not provide such
arguments. Autonomist, mixed, or typological FAs of grammatical phenomena
may prove to be superior to syntactic analyses, and external FAs of language
use, change, and acquisition may prove to be superior to UG analyses, but they
cannot in themselves decide the issue of the self-containedness of grammar. In
the next two sections, we present an argument that does challenge the selfcontainedness of grammar.
8.1.

VARIATION: AN ARGUMENT AGAINST SELF-CONTAINEDNESS. The autonomy

of grammar pertains to the relationship between the grammar as an individual

speaker's cognitive representation and the language as something emerging
from communicative interaction of individuals, and evolving over time (using
grammar and language as defined in ?2.1). The self-containedness question can
be recast as follows: Can the dynamic processes in language be conceived
of as the interaction among self-contained grammars of individuals? Or is the
interaction such that the grammars of individuals are more intimately interconnected with those dynamic functional forces-that is, can an individual adult
speaker's grammar be influenced by those forces? In this section, I present an
argument adapted from variationist theory that suggests that a grammar is not
self-contained. In ?8.2, I address the consequences of the non-self-containedness argument for the representation of an individual's grammar.
Arguments against the self-containedness of the grammatical system must
demonstrate that the cognitive processes that cause changes to the grammar
are still (potentially) operative in the adult grammar. This involves demonstrating that the adult grammatical system is inherently variable and dynamic, that
is, it may still instigate change or be affected by external changes. If this is the
case, then the simplest representation of grammatical knowledge will make
reference to those forces.
There is a theoretical anomaly in the self-containedness view. Language
change is not really external to grammatical systems. Many language changes
such as grammaticalization are internal; that is, they are not due to language
contact. Even those language changes that are due to contact must manifest
themselves in changes in a speaker's grammatical system in order to constitute a
change. We cannot say that a change has occurred until an individual speaker's
grammar has undergone a change: 'languages don't change: people change language tin our terms, grammar-WAC]' (Croft 1990:257). In this context, it
2. Hopper 1987:145, Haiman 1985:259; see also Greenberg 1968:177-96 and Jakobson 1971 [1961]:
574, 576 for earlier statements along the same lines). In fact, all of them carefully hedge their
claims. Regarding arbitrariness, Hopper speaks of 'formulas' (Hopper 1987:145). The functionalist
literature is full of reference to 'routinization' or grammaticalization of discourse functions (e.g.
Thompson 1991:96). Nor is systematicity denied: Bybee et al. refer to a 'logic of grammar' (Bybee
et al. 1994:2), Haiman admits that a grammar exists (Haiman 1985:259), and Hopper writes that
his intention is 'not to abolish [grammar]' (Hopper 1987:148). Instead, all of these argue in the
passages cited in the preceding sentence that the 'true regularity' (Greenberg 1969:186) is found
in the interplay between grammar and the external dynamic-and presumably functional-forces.

518

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

seems odd to assume that the forces that effect language change turn off at
some point, and then turn on again at a time that language change occurs.
Generative (and other) advocates of the autonomy of grammar avoid this
anomaly by arguing that language change-or at least, change in core grammar-occurs during child language acquisition.2' If a child sets different parameters from the parent generation, language change results (Lightfoot 1979, 1982:
161-63, 1991; cf. Newmeyer 1992:777, 784). The adult language system that
the child eventually arrives at is static, and does not participate in language
change: either it hasn't changed at all in comparison with the earlier generation,
or it has already changed in the course of language acquisition. Functional
factors may operate in language acquisition (e.g. semantic bootstrapping [Pinker
1984]); but once the child has acquired the syntactic category, the functional
factors are no longer part of the representation of the adult grammar, nor can
they alter the grammar in the course of language use.
However, sociolinguistic research has demonstrated overwhelmingly that
language change occurs in shifts in variable patterns of use in the adult speech
community (Weinreich et al. 1968, Labov 1972, Milroy 1987, inter alia). That
is, language change occurs in the dynamics of interaction of adult speakers.
Even creolization, considered a paradigm case of language in the making, can
occur in an adult speech community, and is rarely solely due to child language
acquisition (Muhlhausler 1986:51-95, esp. 85-86). Much quantitative variation
in the speech community involves language change in progress. These changes
can progress through the community within a generation, not only between
generations (see below).22
The sociohistorical account of language change is based on variation in the
speech community. The existence of variation also argues against the self-containedness of grammar. Advocates of autonomy abstract away from this variation to an ideal speaker-hearer in a homogeneous speech community (Chomsky
1965:3-4; cf. Chomsky 1981:8). Variation occurs in the speech of single individuals, and hence must be a part of a speaker's knowledge of the language. This
variation is not simply performance errors, as implied in Chomsky's famous
passage (cf. Weinreich et al. 1968:125, and see ?8.5). Variation in the speech
community, though largely socially determined, is not due to different grammars possessed by different speakers. The (socio)linguistic variable-the system of variants-is a property of individuals' grammars. Most speakers have
most variants as part of their grammatical competence, and employ them differently depending on their social position in the speech community and the circumstances of the conversational interaction. In other words, the systematicity
of an individual's variable grammar is tied to the individual's interactions in
different social settings; in this sense the grammar is not self-contained.
2!

In this, they were preceded by Hermann Paul (Weinreich et al. 1968:108-9).

22 Also, the
population of a speech community does not divide neatly into generations (Saussure

1959:72). It may appear to do so within a family, but even there the parents' language does not
determine the structure of the children's language-for instance, children of nonnative speaker
parents learn the language natively (Weinreich et al. 1968:145, n. 36).

AUTONOMY AND FUNCTIONALIST LINGUISTICS

519

One analytical strategy to preserve homogeneity (and hence self-containedness) in an individual's linguistic system is to posit SOCIOLECTS,that is,
multiple grammars for an individual corresponding to the appropriate values
for variables in particular social dimensions of the speech community. Each
sociolect is homogeneous, and a social account of language change would be the
shift in use from one sociolect to another. The argument against this approach is
exactly parallel to the argument against the self-containedness of syntax in ?4.
The problem with variation in a grammar is the same as the problem with
the mixing of syntactic and semantic/discourse properties in the description of
grammatical rules. Positing multiple sociolects in an individual is analogous to
positing a pure syntactic level separate from a semantic level, and has the same
result. Homogeneity of each sociolectal grammar is preserved, but one loses
the generalizations over the identical parts of each sociolect, leading to largescale redundancy in grammatical representation. Also, the most important regularity in variation is the regularity of the relationship BETWEEN variants. To
express these regularities in a multiple-sociolect model, one would have to posit
a supergrammar that is made up of all of the sociolects. A simpler solution
that preserves the generalizations and discards the redundancies is to abandon
homogeneity.23
Sociolinguistic research has linked language change and language use; less
attention has been paid to the role of language acquisition in this model. Language acquisition-both first and second-is a social as well as psychological
phenomenon: it is the process of joining the speech community via particular
social networks. Language use in conversational interaction put the grammatical system in the speaker's memory in the first place and influences its organization, including variation, relative frequencies (see ?8.3), etc. It still exercises
its influence in the adult, as the speaker moves through social networks and
contributes and absorbs variants in the language of the speech community.
The preceding paragraphs state commonplaces in sociolinguistics, almost all
of which were expressed by Weinreich, Labov and Herzog and have been tested
repeatedly since then. Unfortunately, Weinreich et al. still speak in terms of
the 'grammar of the speech community' (1968:188). This description tears apart
the social and the psychological. There is no grammar of the speech community
(cf. Lightfoot 1991:162); the speech community is a collective entity and linguistic knowledge resides in individuals. There are only the grammars of individuals
and the mutual influence of those grammars in communicative interaction.
Variation in the speech community, language change, and typological diversity
23 Not all cases wherea

speakeris multilectalshouldbe reducedto STRUCTURED

HETEROGENEITY

(as Weinreich et al. call it). In general, structured heterogeneity is found particularly when the two
varieties are very close in lexical form and grammatical structure, as in postcreole continua, diglossic continua, and dialect contact. Multilectalism is usually found when the two varieties are very
different-e.g. two different languages, in the structural sense of language. This is exactly what
integrative functionalism would predict: the more overlap, the more likely that the speaker would
collapse two distinct grammars into one. Nevertheless, even a multilectal speaker organizes his/
her lects-which are themselves heterogeneous-according
to their complementary social functions, and creates intermediate variants through interference.

520

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

are all ultimately properties of individuals' grammars ANDtheir mutual interactions. Part of the value of sociolinguistic research is making the link between
(for instance) language change as an obviously supraindividual phenomenon
and processes that can be situated with respect to an individual's grammar
(though sociolinguists have rarely made the final, cognitive step).
In fact, more recent work in sociolinguistics has shifted to networks of interacting individuals (e.g., Milroy 1987). Research in linguistic variation in social
networks has demonstrated that differences in patterns of interaction among
individuals lead to differences in the employment of variants in the language.
For example, the rapidity of diffusion of a language change-that is, changes
in the frequency of variants of a linguistic variable-is iensitive to the structure
of social networks (Milroy & Milroy 1985). A variable internal grammar can
thus evolve within a speaker's lifetime, as the speaker's social networks and
sense of social identification change over time. These shifts writ large are the
dynamic processes that are external to the grammar; but in the representation
of a variable grammar, the dynamic processes are already present internal to
it.
Variationist work in sociolinguistics provides the strongest empirical arguments against the self-containedness of the grammatical knowledge of the individual: grammars must incorporate variation; this variation changes through
use, in ways that may lead to the propagation of a change in the speech community that will be established as such in the mental representations of speakers'
(variable) grammars. Sociolinguists and integrative functionalists, however,
have largely focused on the social, interactional aspects of their theories and
not on their consequences for the mental representation of the grammar. The
next section explores these consequences in somewhat more detail.
8.2.

PROBLEMS AND PREDICTIONS OF INTEGRATIVE FUNCTIONALISM. One of the

important missing elements of integrative functionalism is an integrated model

of all of the dynamic aspects of language, including the (variable) grammar.
Typological functionalists discover crosslinguistic universals and propose as
explanations for them principles of cognition and discourse that are based in
the human mind and its interaction with its environment. But the mechanisms
that lead from the internalized cognitive-social principles to the diversity of
linguistic form in the world's languages have hardly been explicated. The variationist research described in ?8.1 indicates an intimate relationship between
the dynamic processes in language and the representation of an individual's
grammar. Ultimately, then, the dynamic processes must be represented cognitively; this is the issue that sociolinguists and integrative functionalists have
rarely addressed. The gaps and problems with the integrative model are as
follows:
(20) a. Integrative functionalism must provide a system of grammatical
representation that can model a variable grammar and its acquisition and use.
b. Integrative functionalism must account for stable as well as dynamic characteristics of the grammatical system.

AUTONOMY ANDI)FUNCTIONALIST LINGUISTICS

521

c. The role of functional (that is, cognitive and discourse) factors

must be integrated with the role of social factors.
I address each of these problems in the remainder of this section.
Problem 20a is to establish
REPRESENTATION.
8.3. GRAMMATICAL

a link be-

tween the psychological model of grammatical knowledge and the sociolinguistic model of a variable grammar. Bybee's (1985) model of morphology is perhaps
the only example of an integrative functionalist model for which substantial
empirical evidence has been adduced (see also Andersen 1980).
Bybee presents evidence from typological generalizations, diachronic processes, language acquisition, and adult performance that converges on certain
patterns determining the organization of verbal paradigms. She argues that they
result from the interaction of semantic relationships among word forms and the
frequency of those forms. Bybee proposes a spreading activation network
model for grammatical representation to account for these facts.24 Word forms
are stored in memory in a network whose relations reflect semantic relationships among word forms. Individual word forms are entrenched to a greater or
lesser degree largely as a function of their frequency in use.2-5Linguistic units
and schemas (rules) are learned inductively through routinization of the symbolic relation between form and function. The psycholinguistic evidence for
routinization comes from linguistic phenomena that correlate with the token
frequency of linguistic units in language use (including acquisition and language
change). Lower-frequency word forms may not be entrenched-they are generated by the schemas when necessary.
Bybee's model allows for the variable entrenchment of word forms in the
grammar and for adjustments to degree of entrenchment via language use (i.e.
social interaction). Entrenchment is gradient, allowing for continued sensitivity
to variations in input and possible resulting changes (innovations) in morphological form, either in acquisition or during adulthood. Bybee's model thus has
the necessary prerequisites for representing changes in a variable grammar as
a result of changes in patterns of social interaction in adults as well as in children
acquiring the language of the community.
A major challenge for integrative functionalism is to apply the same or a
similar model to syntax. Such a model can be described by substituting syntactic
construction for word form in the description of Bybee's model in the two
preceding paragraphs. A number of models of construction grammar have been
proposed (Lakoff 1987, Langacker 1987, Fillmore & Kay 1993). These models
offer a grammatical representation of syntactic constructions as independent
syntactic units, or rather semiotic units including their semiotic functions. De24
Spreading activation networks are not the same as connectionist models. Connectionist models
are (or claim to be) nonsymbolic models of human knowledge; activation networks usually assume
some sort of symbolic representation, e.g. of word forms and their meanings, and use activation
levels and network organization to model other cognitive phenomena associated with conceptual
structures (entrenchment, attention, etc.).

25

Bybee uses the term AUTONOMOUS;I borrow the term ENTRENCHEDfrom Langacker'svery

similar cognitive model (Langacker 1987:59-60).

522

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

tailed descriptions of constructional polysemy and function have been offered

(e.g. Lakoff 1987:462-585). However, the properties of a construction grammar
that make contact with the representational issues raised here-activation patterns, degrees of entrenchment, dynamic changes resulting from the network
structure-have not been developed in these theories.26 Empirical evidence
from typology, diachrony, acquisition and performance are also necessary to
support integrative functionalist analyses of specific syntactic problems. Such
evidence would be required to confirm or disconfirm the non-self-containedness
of grammar.
In general, integrative functionalists have devoted more pages to describing
the role of functional factors in the characterization of form than with the status
of form itself. Integrative functionalists must make more explicit proposals
regarding the organization and behavior of linguistic form than many of them
have in fact done-in other words, the integrative functionalists should not
leave form to the formalists. This in turn will help to make integrative FAs
more directly comparable to other analyses. Until such FAs are forthcoming,
we cannot assess this aspect of integrative functionalist models with respect
to less radical models.
FORSTABILITYIN GRAMMAR.We have assumed
8.4. EXTERNALMOTIVATIONS

so far that arbitrariness in the grammar, which all linguists must accept, is a
purely internal phenomenon. Its often stable existence in adult grammars-our
problem 20b-is one of the main empirical problems faced by integrative functionalism. The question is: What about language when it's NOTchanging, when
arbitrariness in the signifier (to the extent that it is arbitrary), is not being pushed
around by external functional forces such as economy and iconicity? Isn't this
a basic fact of language that an integrative functionalist must be able to account
for? In fact, however, the stability of arbitrariness itself has an external
function.
There are two arguments for the functional motivation of stable arbitrariness.
The first is psychological, and is based on the concept of entrenchment employed by Bybee and discussed above as a model of variable grammatical
knowledge. The phenomenon of entrenchment (routinization) also provides a
psychological motivation for the stability of arbitrariness. It is hypothesized
that by adulthood mental routinization of most constructions is so strong that
it is largely a force for stasis, and hence it supports the maintenance of arbitrariness. This phenomenon is attributed to efficiency of communication: it's easier
to express a meaning using the already entrenched form for doing so.
The social motivation for stasis can be found in a model of social conventions
or norms. The employment of an arbitrary link between form and function, to
whatever extent it is arbitrary, is a matter of convention in the speech community, and its maintenance in the face of competing functional motivations is a
social question. Conventions themselves have a functional utility: communica26
Langacker makes reference to spreading-activation and interactive-activation networks in his
model of grammatical organization (Langacker 1987:385, n. 12, 429-33), but does not offer analyses
of quantitative relationships between syntactic constructions.

AUTONOMY AND FUNCTIONALIST LINGUISTICS

523

tion of meaning, social identity, status, and so forth would be difficult without
an agreed-upon convention. Hence, social convention can also be a force for
stasis and the maintenance of arbitrariness (Saussure 1959:71-74).
There is evidence from sociolinguistic research that the rigidity of these conventions depends on the cohesiveness and centralization of the speech community. LePage and Tabouret-Keller (1985:115-16, 182-86) describe DIFFUSE
speech communities, such as (until recently) the Belize City vernacular, in
which this cohesiveness and centralization are lacking, and the linguistic conventions are therefore in much greater flux than for most middle-class American
speakers of English. That is, the internalized grammars are much more variable
and flexible for members of the Belizean speech community than they are for the
much more homogeneous middle-class American English speech community.
Hopper's emergent grammar (Hopper 1987) is a useful description of highly
diffuse speech communities. However, Hopper's description is probably too
strong for what LePage & Tabouret-Keller call FOCUSED
communities, where
social network ties are much tighter. In focused speech communities, grammatical conventions are much more fixed and do not vary dramatically from one
face-to-face interaction to the next. The variability and fluidity of grammar is
a matter of degree that is dependent on the social characteristics of the speech
community, and middle-class America and Belize are close to the opposite
extremes in this continuum.
If the stability of arbitrariness is functionally motivated, then those motivations should interact with the functional motivations based on efficiency of
the form-meaning mapping that are usually invoked (iconicity, etc.). Hence
integrative functionalism should make the following empirical prediction:
(21) Those components of the grammar where there are more variants in
the expression of a semiotic function in the grammar, or where there
is a lower frequency of occurrence, will be more susceptible to
change due to the functional motivations (e.g. iconicity and economy) than those components where there is less variation and a
higher frequency of occurrence.
If stable arbitrariness is not motivated, then degree of variation and token frequency should not have a discriminatory effect (other things being equal). Bybee's research confirms 21, at least for the morphological categories that she
examines. For example, less frequent members of a paradigm are more susceptible to levelling on analogy with semantically closely related, more frequent
members (Bybee 1985:49-79, Bybee & Slobin 1982).
8.5.

THE INTERACTION OF SOCIAL AND FUNCTIONAL FACTORS IN LANGUAGE.

Problem 20c is a serious one for integrating functionalist accounts of language

with sociolinguistic models of a nonautonomous grammar. Typological variation and diachronic change are supposed to reflect interacting factors that are
determined by external function. But the evidence in sociolinguistics suggests
that change is diffused through the speech community via social factors such
as prestige, social identity, and social network contacts. Functional forces appear to play little if any role in the spread of a variant through a speech commu-

524

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

nity (see e.g., Milroy 1987:186, Weinreich et al. 1968:1 11). Can these two types
of factors be reconciled?
I propose a solution to this problem which provides another empirical prediction for integrative functionalism. The problem in 20c is related to the 'riddle
of actuation' (Weinreich et al. 1968:102): Why does a novel variant occur,
and why is it adopted in one community but not in another? The (external)
functionalist model provides an answer to the first part of the riddle of actuation:
competing motivations bring about the novel variants. A grammar consists of
a balance between multiple competing functional forces. From an external functional point of view, the system is inherently unstable. While the external motivations for stability preserve the balance most of the time (see ?8.4), alternative
resolutions of competing motivations are occasionally produced by individuals.
These include speech errors; however, they may also include deliberate creations as well. This linguistic creativity is not the creativity of producing neverbefore-heard utterances from existing grammatical rules, but the creativity of
stretching or breaking existing grammatical rules (consciously or unconsciously).

We will use the neutral term INNOVATIONS to cover both errors and

creations. The competing motivations model of functionalism provides a builtin mechanism for generating innovations.
We may thus make the following hypotheses and prediction for integrative
functionalism:
(22) a. Hypothesis: Innovation will occur as the result of functional factors, specifically novel resolutions of competing functional motivations (stretching of the conventional rules).27
b. Hypothesiss: Propagation of innovations is determined by social
factors.
c. Prediction: The frequency distribution of different innovations will
be a function of the relative functional utility of the innovation
relative to the existing conventional variant(s), as inferred from
crosslinguistic distribution.
Hypothesis 22a is the proposed solution to the relevant part of the riddle of
actuation. Hypothesis 22b appears to be the consensus of current sociolinguistic
research. Assuming 22a and 22b are correct, then prediction 22c must hold.
Social factors influencing the spread of change are independent of functional
utility; thus, they select innovated variants randomly. The random effects of
socially determined diffusion of variants should even out across a large enough
sample. Hence, if the crosslinguistic distribution of grammatical types is determined by functional factors that operate in individuals' grammars, then it should
match the frequency distribution of individual innovations; this is 22c.
Prediction 22c represents the claim that external functional factors determine
both individual linguistic behavior (innovations) and crosslinguistic/diachronic
patterns that are obviously not part of an individual's grammar. Since these
27
Novel variants may also occur in contact, as some sort of compromise between the alternative
forms (Trudgill 1986). Presumably such intermediate variants must also fit within the space of
possibilities provided by a functionalist account. Prediction 22a refers to changes of internal origin.

AUTONOMY AND FUNCTIONALIST LINGUISTICS

525

are independent sources of evidence, one can develop hypotheses of functional

factors on the basis of one source of evidence and test them in the other. For
example, Bybee and Pardo (1981) tested their hypothesis that morphological
paradigms have a partly semantically governed structure, a hypothesis based
on crosslinguistic and historical evidence, in a nonce-probe task testing the
organization of paradigms by Spanish speakers.
Integrative functionalism as a theory of grammar derives its theoretical
models and empirical support from highly disjoint fields of research. On the
one hand, it uses a competing functional motivations model which has been
defended with typological and diachronic evidence. On the other hand, its nonautonomist approach to grammatical representation owes its empirical support
to variationist and interactional sociolinguistics, and its theoretical model (to
the extent that it exists) to the cognitively oriented research of Bybee and
others. The disadvantages of this are obvious: different sorts of data and different theoretical models must be reconciled in order to establish a sound and
testable theory. But if it succeeds, it will reintegrate the field of linguistics,
which has become highly fragmented. It will reintegrate the social and the
psychological and thereby make a general contribution to the social sciences.
That, at least, is the promise of integrative functionalism.28
This paper has examined the status of autonomy in contem9. CONCLUSION.
porary theories of grammar, specifically with respect to functional analyses
that are said to challenge autonomy. We have identified different meanings of
autonomy and the specific claims about the nature of grammar that are associated with each. This allows us to evaluate just what an FA of a particular
phenomenon proves about autonomy.
If a syntactic structure is shown to have a consistent function, that shows that
even syntactic constructions are linguistic signs; but no more. If the behavior of
that syntactic structure can be predicted from its function, this shows that the
scope of an autonomous syntax has been narrowed (autonomist functionalism).
If a syntactic phenomenon is demonstrated to be better described partially in
functional terms, then it challenges the self-containedness of syntax (mixed
formal/functionalism). If the mixed formulation involves reference to functionally based typological universals so that the arbitrary aspect of the rule is only
language-specific, then the typological functionalist position is supported. If
the functional properties of a grammar can be reanalyzed as having (external)
28

It is worth pointing out parallels between the integrative functionalist model of grammar presented here and Ohala's model for phonology, where external phonetic factors explain phonological
patterns (Ohala 1981, 1983, 1989, 1992). Ohala argues that one must account for why sound systems
do NOT change much of the time (e.g., 1992:309); cf. ?8.4. Ohala also argues that the phonetic
(system-external) factors that he offers for explanations are responsible only for innovation, not
propagation, of novel variants (1989:175); cf. ?8.5. He also argues that the phenomena that we find
in the resulting language diversity (e.g., crosslinguistic and dialectal differences) are parallel to the
results of his psycholinguistic investigations, just as I have suggested that typological variation
should be parallel to variation caused by innovations that in principle could also be elicited in
psycholinguistic experiments (such as those done by Bybee and Pardo).

526~

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

functional value, and a functional account of the dynamic processes of language

is superior to an innate UG parameter-setting account, then the external functionalist position is supported. Only if a model of the mental representation of
grammatical knowledge is posited that accounts for the variable nature of an
adult grammar (as well as the dynamic forces influencing that representation),
and that model satisfies the criteria in ??7.2 and 8.2, can we say that the integrative functionalist analysis is supported.
In the course of this analysis, a near-continuum of structuralist to functionalist theories, and their central points of agreement and disagreement, have been
identified. Structuralist and functionalist theories share some important assumptions. One is the acceptance of the independence (arbitrariness) and systematicity of syntax within the grammar, and of the grammar with respect to
external factors. Another is the acknowledgement of a role for semiotic function
in the description of syntax, and a role for external functional forces in the
dynamics of language.
On many issues there are disagreements, between structuralists and functionalists and between different kinds of functionalists. My analysis of these disagreements demonstrates that these theories are comparable in ways that can be
empirically tested, and reveals some of the fundamental issues in grammatical
theory. There are many differences of opinion as to the degree of interaction
or interpenetration of properties of the signifier and the signified in the composition of the grammar. A major distinction is found between syntactic behavior
and syntactic structure. With respect to the external factors that make grammars the way they are, the belief in an innate UG vs. functional motivations
is a central issue for debate.
Two issues of deeper theoretical importance stand out. Perhaps the deepest
issue is the relationship between internal and external in language. This difference allows us to compare two views of language that have generally been quite
separate from each other, generative grammar and sociolinguistics. It remains
to be seen whether integrative functionalism can join the concern with the
mental knowledge of the individual found in the former with the concern for
the dynamic processes in society found in the latter.
Finally, there is an important difference hiding in the consensus that language
(and grammar) forms a system. The formal linguistic system is usually modelled
as a logical calculus, with noncontradictory axioms and discrete qualitative
differences. The system offered by some functionalist theories is quite different:
a spreading activation network, with competing motivations and quantitatively
variable representations. These two models make very different predictions
about the nature of mental representations and linguistic behavior.
In the course of this article, I have discussed a number of examples and made
reference to empirical studies in various subareas of linguistics to illustrate
various points of contact between the different approaches to grammar. I especially hope that this article will encourage more empirical research in potential
areas of contact between the various approaches to language described herein,
since that, ultimately, is what linguistics should be about.

AUTONOMYAND FUNCTIONALISTLINGUISTICS

527

REFERENCES
ANDERSEN,HENNING. 1980. Russian conjugation: acquisition and evolutive change.
Papers from the 4th International Conference on Historical Linguistics, ed. by Elizabeth Closs Traugott et al., 285-302. Amsterdam: John Benjamins.
ANDERSON, JOHN. 1983. The architecture of cognition. Cambridge, MA: Harvard University Press.
ANDREWS,

J. RICHARD. 1975. Introduction

to Classical

Austin: University

Nahuatl.

of

Texas Press.
ARIEL,MIRA. 1990. Accessing noun phrase antecedents. New York: Routledge.
1982. Functionalist

BATES, ELIZABETH, and BRIAN MACWHINNEY.

approaches

to gram-

mar. Language acquisition: The state of the art, ed. by Eric Wanner and Lila R.
Gleitman, 173-218. Cambridge: Cambridge University Press.
1987. Competition, variation and language learning. Mechanisms of language
--,--.
acquisition, ed. by Brian MacWhinney, 157-93. Hillsdale, NJ: Lawrence Erlbaum.
1989. Functionalism and the competition model. The crosslinguistic study
--, --.
of sentence processing, ed. by Brian MacWhinney and Elizabeth Bates, 3-73. Cambridge: Cambridge University Press.
BOLINGER,

DWIGHT.

1977. Another

glance

at main clause

phenomena.

Language

53.511-519.
. 1979. Pronouns in discourse. Discourse and syntax, ed. by Talmy Givon, 289-310.
New York: Academic Press.
BYBEE,JOANL. 1985. Morphology: a study of the relation between meaning and form.
Amsterdam: John Benjamins.
, and ELLYPARDO.1981. On lexical and morphological conditioning of alternations:
a nonce-probe experiment with Spanish verbs. Linguistics 19.937-68.
R;REVERE D. PERKINS; and WILLIAM PAGLIUCA. 1994. The evolution

of grammar:

Tense, aspect and modality in the languages of the world. Chicago: University of
Chicago Press.
, and DAN I. SLOBIN1982. Rules and schemas in the development and use of the
English past tense. Language 58.265-89.
CHOMSKY, NOAM. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.
1977 [1975]. Questions of form and interpretation. In his essays on form and interpretation, 25-59. Amsterdam: North-Holland. [Originally published in Linguistic
Analysis 1.75-109.]
. 1981. Lectures on government and binding. Dordrecht: Foris.
. 1986. Knowledge of language: Its nature, origin, and use. New York: Praeger.
. 1993. A minimalist program for linguistic theory. The view from Building 20, ed.
by Kenneth Hale & Samuel Jay Keyser, 1-52. Cambridge, MA: MIT Press.
COLE,PETER(ed.). 1981. Radical pragmatics. New York: Academic Press.
COMRIE, BERNARD. 1978. Ergativity.

Syntactic

typology,

ed. by Winfred Lehmann,

329-94. Texas: University of Texas Press.

-. 1989. Language universals and linguistic typology. 2nd edn. Chicago: University
of Chicago Press.
CORBETT, GREVILLE. 1979. The agreement hierarchy. Journal of Linguistics
CROFT, WILLIAM. 1990. Typology and universals. Cambridge: Cambridge

15.203-24.
University

Press.
--. 1991. Syntactic categories and grammatical relations: The cognitive organization
of information. Chicago: University of Chicago Press.
DEANE,PAUL. 1991. Limits to attention: A cognitive theory of island phenomena. Cognitive Linguistics 2.1-63.
DELANCEY,SCOTT.1981. An interpretation of split ergativity and related patterns. Language. 57.626-57.
DIK, SIMON.1981. Functional Grammar, third revised edition. (Publications in Language
Sciences, 7.) Dordrecht: Foris.

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

528

DIRVEN,RENEand V. FRIED(eds.) 1987. Functionalism in linguistics. Amsterdam: John

Benjamins.
DIXON, R. M. W. 1977. Where have all the adjectives gone? Studies in Language
1.19-80.
-- . 1994. Ergativity. Cambridge: Cambridge University Press.
EMONDS,JOSEPH. 1976. A transformational approach to English syntax. New York:
Academic Press.
ERTESCHIK-SHIR,

NOMI, and SHALOM LAPPIN.

1979. Dominance

and the functional

ex-

planation of island phenomena. Theoretical Linguistics 6.41-86.

FARKAS,DONKA. 1988. On obligatory control. Linguistics & Philosophy 11.27-58.
FILLMORE, CHARLES, and PAUL KAY. 1993. Construction

Grammar coursebook,

chap-

ters 1-11. Berkeley: University of California, MS.

FOLEY, WILLIAM A., and ROBERT D. VAN VALIN, JR. 1984. Functional

syntax

and

universal grammar. Cambridge: Cambridge University Press.

GARCIA,ERICA. 1979. Discourse without syntax. Discourse and syntax. ed. by Talmy
Givon, 23-49. New York: Academic Press.
GIVON,TALMY.1979. On understanding grammar. New York: Academic Press.
--. 1982. Logic vs. pragmatics, with natural language as the referee: Towards an
empirically viable epistemology. Journal of Pragmatics 6.81-133.
. 1985. Iconicity, isomorphism and nonarbitrary coding in syntax. Iconicity in syntax, ed. by John Haiman, 187-220. Amsterdam: John Benjamins.
GREEN, GEORGIA. 1976. Main clause

phenomena

in subordinate

clauses.

Language

52.382-97.
. 1982. Review of Discourse and syntax, ed. by Talmy Givon. Language 58.672-80.
GREENBERG, JOSEPH. 1968. Anthropological

linguistics.

New York: Random House.

. 1969. Some methods of dynamic comparison in linguistics. Substance and structure

of language, ed. by Jan Puhvel, 147-203. Berkeley and Los Angeles: University of
California Press.
. 1990 [1963]. Some universals of grammar with particular reference to the order
of meaningful elements. On language: Selected writings of Joseph H. Greenberg,
ed. by Keith Denning and Suzanne Kemmer, 40-70. Stanford: Stanford University
Press.
GROSU, ALEXANDER. 1981. Approaches

to island phenomena.

Amsterdam:

North-

Holland.
GUNDEL, JEANNETTE; NANCY HEDBERG; and RON ZACHARSKI. 1993. Cognitive

status

and the form of referring expressions in discourse. Language 69.274-307.

HAIMAN,JOHN. 1980a. Hua: A Papuan language of the Eastern Highlands of New
Guinea. Amsterdam: John Benjamins.
1980b. The iconicity of grammar: Isomorphism and motivation. Language
54.565-89.
. 1983. Iconic and economic motivation. Lg. 59.781-819.
. 1985. Natural syntax. Cambridge: Cambridge University Press.
HARRIS, RANDY ALLAN. 1993. The linguistics wars. Oxford: Oxford University
HAWKINS, JOHN. 1983. Word order universals. New York: Academic Press.

Press.

. 1990. A parsing theory of word order universals. Linguistic Inquiry 21.223-61.

HEINE, BERND, and MECHTHILDREH. 1984. Grammaticalization

and reanalysis

in Afri-

can languages. Hamburg: Helmut Buske.

HICKMANN,M. (ed.) 1987. Social and functional approaches to language and thought.
New York: Academic Press.
HOOPER, JOAN B., and SANDRA A. THOMPSON. 1973. On the applicability

of root trans-

formations. Linguistic Inquiry 4.465-97.

HOPPER,PAUL. 1987. Emergent grammar. Berkeley Linguistics Society 13.139-157.
, and SANDRA A. THOMPSON. 1980. Transitivity

guage 56.251-299.

in grammar and discourse.

Lan-

AUTONOMYAND FUNCTIONALISTLINGUISTICS

529

-- , and ELIZABETH
CLOSSTRAUGOTT.1993. Grammaticalization. Cambridge: Cambridge University Press.
HORN,LAURENCE. 1989. A natural history of negation. Chicago: University of Chicago
Press.
JACKENDOFF, RAY. 1992. Mme. Tussaud

meets the binding theory.

Natural Language

and Linguistic Theory 10.1-31.

JAKOBSON, ROMAN. 1971 [1961]. Linguistics

and communication

theory. Selected

Writ-

ings 2.570-79. [Originally published in Proceedings of Symposia in Applied Mathematics, XII.]

KALMAR, IVAN. 1979. Case and context in Inuktitut (Eskimo). (National Museum of
Man Mercury Series, No. 49.) Ottawa: National Museums of Canada.
KAZENIN,

KONSTANTIN

I. 1994. Split syntactic ergativity: Toward an implicational

hier-

archy. Sprachtypologie und Universalienforschung 47.78-98.

KEFER, MICHEL,

and JOHAN VAN DER AUWERA.

1992. Meaning and grammar: Crosslin-

guistic perspectives. Berlin: Mouton de Gruyter.

KOERNER, KONRAD. 1989. Antoine

Meillet,

Saussure

et la linguistique

generale:

Une

question d' 'influence'. In his practicing linguistic historiography, 401-15. Amsterdam: John Benjamins.
KOSTER,JAN. 1986. Domains and dynasties: the radical autonomy of syntax. Dordrecht:
Foris.
KUNO,SUSUMO.1976. Subject, theme and speaker's empathy: A reexamination of relativization phenomena. Subject and topic, ed. by Charles Li, 417-44. New York:
Academic Press.
. 1987. Functional syntax: anaphora, discourse and empathy. Chicago: University
of Chicago Press.
LABOV,WILLIAM.1972. Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.
. 1994. Principles of linguistic change: Internal variation. Oxford: Basil Blackwell.
LADUSAW, WILLIAM, and DAVID DOWTY. 1988. Toward a nongrammatical

account

of

thematic roles. Syntax and semantics 21: Thematic relations, ed. by Wendy Wilkins,
62-74. New York: Academic Press.
LAKOFF,GEORGE.1984. Performative subordinate clauses. Berkeley Linguistics Society
10.472-80.
. 1986. Frame semantic control of the coordinate structure constraint. Chicago
Linguistic Society, 22.152-67.
. 1987. Women, fire and dangerous things: What categories reveal about the mind.
Chicago: University of Chicago Press.
LANGACKER, RONALD W. 1987. Foundations

of Cognitive

Grammar, vol 1: Theoretical

prerequisites. Stanford: Stanford University Press.

LEHMANN, CHRISTIAN. 1985. Grammaticalization:

Synchronic

variation and diachronic

change. Lingua e Stile 20.303-18.

LEPAGE, R. B., and ANDREE TABOURET-KELLER. 1985. Acts of identity.

Cambridge:

Cambridge University Press.

LEVINSON, STEPHEN. 1987. Pragmatics and the grammar of anaphora: A partial pragmatic reduction of Binding and Control phenomena.
Journal of Linguistics

23.379-434.
--. 1991. Pragmatic reduction of the Binding Conditions revisited. Journal of Linguistics 27. 107-61.
LIGHTFOOT,DAVID. 1979. Principles of diachronic

syntax. Cambridge: Cambridge Uni-

versity Press.
. 1982. The language lottery: Toward a biology of grammars. Cambridge, MA.:
MIT Press.
. 1991. How to set parameters: Arguments from language change. Cambridge, MA.:
MIT Press.

530

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

MARTINET,ANDRE. 1952. Function, structure and sound change. Word 8.1-32.

MATTHEWS, P. H. 1993. Grammatical theory in the United States from Bloomfield to
Chomsky. Cambridge: Cambridge University Press.
MCGREGOR, WILLIAM. 1990. A functional grammar of Gooniyandi. Amsterdam: John
Benjamins.
MILROY,JAMES,and LESLEYMILROY.1985. Linguistic change, social network and
speaker innovation. Journal of Linguistics 21.339-84.
MILROY, LESLEY. 1987. Language and social networks. 2nd edn. Oxford: Basil
Blackwell.
MITHUN,MARIANNE.1984. The evolution of noun incorporation. Language 60.847-94.
1986. On the nature of noun incorporation. Language 62.32-37.
--.
--. 1991. Active/agentive case marking and its motivations. Language 67.510-46.
EDITHA. 1978. On the distribution of ergative and accusative patterns.
MORAVCSIK,
Lingua 45.223-79.
MORGAN, J. L. 1982. Discourse theory and the independence of sentence grammar.
Analyzing discourse: Text and talk, ed. by Deborah Tannen, 196-204. Washington:
Georgetown University Press.
PETER.1986. Pidgin and creole linguistics. Oxford: Basil Blackwell.
MUHLHAUSLER,
NEWMEYER, FREDERICK J. 1983. Grammatical theory: Its limits and its possibilities.
Chicago: University of Chicago Press.
. 1991. Functional explanation in linguistics and the origin of language. Language
& Communication 1/2.3-28.
. 1992. Iconicity and generative grammar. Language 68.756-96.
. 1994. Competing motivations and synchronic analysis. Sprachtypologie und Universalienforschung 47.67-77.
NICHOLS,JOHANNA. 1984. Functional theories of grammar. Annual Review of Anthropology 13.97-1 17.
NOONAN, MICHAEL. 1985. Complementation. Language typology and syntactic description, vol. 2: Complex constructions, ed. by Timothy Shopen, 42-140. Cambridge:
Cambridge University Press.
OHALA,JOHN. 1981. Articulatory constraints on the cognitive representation of speech.
The cognitive representation of speech, ed. by T. Myers, J. Laver, and J. Anderson,
111-22. Amsterdam: North-Holland.
. 1983. The origin of sound patterns in vocal tract constraints. The production of
speech, ed. by Peter F. MacNeilage, 189-216. New York: Springer.
. 1989. Sound change is drawn from a pool of synchronic variation. Language
change: Contributions to the study of its causes, ed. by Leiv Egil Breivik and Ernst
Hakon Jahr, 173-98. Berlin: Mouton de Gruyter.
. 1992. What's cognitive, what's not, in sound change. Diachrony within synchrony:
Language history and cognition, ed. by Gunter Kellermann and Michael D. Morrissey, 309-55. Frankfurt: Peter Lang.
PINKER, STEVEN. 1984. Language learnability and language development. Cambridge,
MA: Harvard University Press.
POLLARD,CARL,and IVANA. SAG. 1987. Information-based syntax and semantics, vol.
1: Fundamentals. (CSLI Lecture Notes, 13.) Stanford: Center for the Study of
Language and Information.
-, --. 1994. Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press.
POSTAL, PAUL. 1969. Anaphoric islands. Chicago Linguistic Society 5.205-39.
PRINCE, ELLEN F. 1978. A comparison of wH-clefts and it-clefts in discourse. Language
54.883-906.
. 1981. Topicalization, Focus-movement and Yiddish-movement: A pragmatic differentiation. Berkeley Linguistics Society 7.249-64.
. 1988. Discourse analysis: A part of the study of linguistic competence. Linguistics:

531

AUTONOMYAND FUNCTIONALISTLINGUISTICS

The Cambridge survey, vol. 2, Linguistic theory: Extensions and implications, ed.
by Frederick J. Newmeyer, 164-82. Cambridge: Cambridge University Press.
--. 1991. On 'functional explanation in linguistics and the origins of language'. Language & Communication 1/2.79-82.
REINHART,TANYA. 1983. Anaphora and semantic interpretation. Chicago: University
of Chicago Press.
. 1984. Principles of gestalt perception in the temporal organization of narrative
texts. Linguistics 22.779-809.
SADOCK, JERROLD. 1980. Noun incorporation

in West Greenlandic:

A case of syntactic

word formation. Language 56.300-319.

. 1984. Whither radical pragmatics? Meaning, form and use in context: Linguistic
applications, ed. Deborah Schiffrin, 139-49. Washington: Georgetown University
Press.
. 1985. The Southern Tiwa incorporability hierarchy. International Journal of American Linguistics 51.568-71.
. 1986. Some notes on noun incorporation. Language 62.19-31.
SAPIR, EDWARD. 1921. Language. New York: Harcourt, Brace & World.
SAUSSURE, FERDINAND. 1959. Course in general linguistics, trans. by Wade Baskin.

(Translation of Cours de linguistique generale, ed. by Ch. Bally and A. Sechehaye,

Paris, 1916.) New York: McGraw-Hill.
SCHAUB,WILLI. 1985. Babungo. (Croom Helm Descriptive Grammars.) Dover, NH:
Croom Helm.
SCHILLER, ERIC. 1990. An autolexical account of subordinating serial constructions.
Chicago: University of Chicago dissertation.
SEARLE, JOHN. 1969. Speech acts. Cambridge: Cambridge University Press.
MICHAEL.1976. Hierarchy of features and ergativity. Grammatical cateSILVERSTEIN,
gories in Australian languages, ed. R. M. W. Dixon, 112-171. Canberra: Australian
Institute of Aboriginal Studies.
. 1993. Of nominatives and datives: Universal grammar from the bottom up. Advances in Role and Reference Grammar, ed. Robert D. Van Valin, Jr., 465-98.
Amsterdam: John Benjamins.
STASSEN,LEON. 1992. A hierarchy of main predicate encoding. In Kefer & van der
Auwera 1992, 179-201.
TAKAMI, KEN-ICHL. 1989. Preposition

stranding: Arguments

against syntactic

analyses

and an alternative functional explanation. Lingua 76.299-335.

TALMY, LEONARD.1978. Figure and ground in complex sentences. Universals of human
language, vol. 4, Syntax, ed. by Joseph H. Greenberg et al., 625-52. Stanford:
Stanford University Press.
THOMPSON, SANDRA A. 1991. On addressing

functional

explanation

in linguistics.

Lan-

guage & Communication 1/2.93-96.

TOMLIN,RUSSELL S. 1990. Functionalism in second language acquisition. Studies in
Second Language Acquisition 12.155-77.
TRUDGILL, PETER. 1986. Dialects

in contact.

Oxford: Basil Blackwell.

VANHOEK,KARENA. 1992. Paths through conceptual structure: Constraints on pronominal anaphora. San Diego: University of California dissertation.
VAN VAHIN, ROBERTD. JR. 1993. A synopsis of Role and Reference Grammar. Advances

in Role and Reference Grammar, ed. by Robert D. Van Valin, Jr., 1-164. Amsterdam: John Benjamins.
WARD, GREGORY; RICHARD SPROAT; and GAIL McKoON.

1991. A pragmatic analysis

of so-called anaphoric islands. Language 67.439-74.

WEINREICH, URIEL; WILLIAM LABOV; and MARVIN I. HERZOG. 1968. Empirical founda-

tions for a theory of language change. Directions for historical linguistics, ed. by
Winfred P. Lehmann and Yakov Malkiel, 95-195. Austin: University of Texas
Press.

532

LANGUAGE, VOLUME 71, NUMBER 3 (1995)

WETZER,HARRIE.1992. "Nouny" and "verby" adjectivals: A typology of predicative

adjectival constructions. In Kefer & van der Auwera 1992, 223-62.
WOLFART,H. CHRISTOPH, and JANET F. CARROLL.1981. Meet Cree: A guide to the
Cree Language. 2nd edn. Lincoln: University of Nebraska Press.
Departmentof Linguistics
[Received July 26, 1994;
Universityof Manchester
accepted January 16, 1995;
OxfordRoad
final version received April 29, 1995.1
Manchester,England
w.croft(uman.ac.uk