The Oxford Handbook of Cognitive Linguistics Part 61 pdf

a sentence was determined not by the fact of its occurrence in a corpus but on the basis of native speaker intuitions. The primitive elements of the rule system were symbols such as S (¼ ‘sentence’), N (¼ ‘Noun’), NP (¼ ‘Noun Phrase’), Aux (¼ ‘Auxiliary’); these, like the mathematician’s x and y or the logician’s p and q,were variables, contentless in themselves, but which could take on as values any contentful element of the appropriate category. The rules operated over strings of such symbols, expanding them, rearranging them, and performing other kinds of transformations. Like Bloomfieldian descriptivism, the ‘‘algebraic shift’’ left its mark on academic linguistics by introducing a new rigor into linguistic discourse. Linguists were re- quired to formulate rules with a degree of precision and explicitness which allowed the rules to be evaluated against counterexamples and alternative rule formulations. Within a relatively short span of time, all manner of ‘‘tests’’ were developed which could be applied in order to substantiate a given analysis. Many of these tests (for example, tests for confirming the status of a nominal as clausal subject, tests for diagnosing complements as opposed to adjuncts, tests for distinguishing control verbs from raising verbs) are now part of the arsenal of every practicing linguist. It did not take long for the algebraic enterprise to be ‘‘biologized.’’ This move was made explicit in Chomsky’s next major publication, Aspects of the Theory of Syntax (Chomsky 1965). The system of rules which constitute the grammar had to be evaluated, not only in terms of its descriptive adequacy, but as a hypothesis about the cognitive state of a speaker (47–59). Moreover, since human beings are not born with knowledge of any particular language, the rule system pertaining to a given language had to be acquired. Attention therefore moved toward those fea- tures of the human mind which support and enable the acquisition and mental representation of language. The shift signaled the emergence of linguistics as a cognitive science. To be sure, some linguists have ignored the ‘‘cognitive turn.’’ They have been concerned, as was Chomsky during the earlier stages of this career, with formu- lating syntactic rules with maximal accuracy and precision, with little regard for the cognitive reality of the rules. Generalized Phrase Structure Grammar (Gazdar et al. 1985) belongs to this tradition. (It comes as no surprise, therefore, that it is models such as Generalized Phrase Structure Grammar, rather than the latter-day versions of Chomskyan grammar, which have been exploited for Natural Language Pro- cessing.) On the whole, however, it would be probably true to say that the majority of today’s linguists would be comfortable with being described as ‘‘cognitive,’’ on a broad understanding of the term; they would, in other words, subscribe to the view that a linguistic analysis is a hypothesis about the mental representation of lan- guage in the minds of its speakers. 4 Neither would today’s linguists want to sac- rifice the argumentative rigor introduced by the Chomskyans, nor the basic de- scriptive tools bequeathed by the Bloomfieldians. In an important sense, therefore, we witness a progressive, incremental development of the discipline, from the tax- onomic phase, through the algebraic phase, to the cognitive phase. My sketch of this developmental trajectory, however, hides a number of contentious issues. It is to these that we need to turn in order to find the roots of Cognitive Linguistics, its 570 john r. taylor points of contrast with Chomskyan linguistics, and its distinctive way of imple- menting the cognitive agenda. Of particular relevance to our topic is the turmoil within the Chomskyan camp during the late 1960s and early 1970s. At around that time, a number of linguists closely associated with Chomsky pushed the generative enterprise in the direc- tion of what came to be known as ‘‘Generative Semantics.’’ As is well known, early Generative Grammar proposed a ‘‘deep’’ level of syntactic structure, which was converted into a ‘‘surface’’ structure by means of transformations. The generative semantic view was that all aspects of the meaning of a sentence, even its pragmatic force, had to be represented in the deep structure. Moreover, sentences which are (roughly) synonymous on the surface, even though they might differ in their wording, had to share the same deep structure. Regarding deep structure itself, it was usual for it to be represented in the format of predicate logic and to incor- porate what were supposed to be semantic primitives. An often-cited example was the representation of kill in terms of the abstract elements [cause] and [become not alive]. 5 Such an approach, as will be evident, necessitated a vast inventory of transformational rules, many of them entirely ad hoc, which added, deleted, re- arranged, and replaced material as the deep semantic structure got transformed into a surface syntactic structure. There was a time in the early 1970s when it seemed that Generative Semantics was going to conquer the field. After a particularly vi- cious and acrimonious period (narrated from different perspectives by Newmeyer 1980 and Harris 1993), the Generative Semantics movement collapsed, largely, it would seem, under the weight of the unconstrained transformations which it pos- tulated. Henceforth, ‘‘orthodox’’ Generative Grammar would severely restrict both the kinds of transformations that were admissible and the range of data that the theory was intended to account for. In the end, only one transformation came to be recognized, that of movement, and even this became an option of ‘‘last resort’’ (Chomsky 1995: 150). At the same time, the empirical scope of the theory came to be restricted to ‘‘core’’ grammatical phenomena, to the exclusion of pragmatics and all manner of ‘‘idiosyncratic’’ syntactic and lexical data. Although now largely defunct as a distinctive movement, 6 Generative Se- mantics is relevant to our topic because one of its main exponents, George Lakoff, was to become a leading figure in Cognitive Linguistics. Lakoff has since empha- sized the line of continuity between the two movements, even going so far as to describe Cognitive Linguistics as ‘‘an updated version of generative semantics’’ (Lakoff 1987: 582). 7 The link between the two movements lies in the importance attached to semantic structure and to the need to study surface phenomena from the point of view of the meanings that are being conveyed. But whereas Generative Semantics had represented semantic structure in model-theoretic terms, Cognitive Linguistics construes meaning more broadly to encompass topics such as meta- phorical mappings, image schemas, and mental models. And whereas Generative Semantics had proposed all manner of transformations intervening between deep and surface structure, the emerging Cognitive Linguistics proposed a more direct, symbolic relationship between aspects of form and aspects of meaning. cognitive linguistics and autonomous linguistics 571 4. Characteristics of Mainstream Generative Linguistics Returning now to the main line of the story, let us consider the development of ‘‘mainstream’’ Generative Grammar, which won out over Generative Semantics. Over the years, the Chomskyan enterprise has undergone many upheavals and transformations, from the ‘‘Standard Theory,’’ as the approach in Chomsky (1965) came to be known, through the ‘‘Extended Standard Theory’’ outlined in Chomsky (1970) but more fully articulated in Jackendoff (1972), to ‘‘Government and Bind- ing’’ (Chomsky 1981), ‘‘Principles and Parameters’’ (Chomsky 1986), and latterly ‘‘Minimalism’’ (Chomsky 1995). 8 The main features of the Chomskyan enterprise have, however, remained largely unchanged over the years. Some enduring char- acteristics are the following: a. The Centrality of Syntax. The central component of a grammar was, and remains, the syntax—Jackendoff (1997: 15) refers to this aspect of the theory as its ‘‘syntactocentrism.’’ Syntax here is construed as a computational system which operates over contentless symbols. The symbols receive pho- nological and semantic content only in the phonological and semantic components of the grammar after lexical material has been ‘‘inserted’’ into the syntactic strings. In this way, the set of grammatical sentences of a language is generated. b. Formalism. Given the centrality of syntax, which is viewed as a compu- tational mechanism, it is not surprising that Generative Linguistics has placed a high premium on formalism, that is, on precise statements of rules and the conditions under which they apply. The rules are typically presented in a quasi-mathematical format and make reference mostly to general cate- gories, not to actual linguistic items (words, meanings, pronunciations). c. Grammaticality. As a corollary of (a) and (b), the ‘‘grammaticality’’ of an expression comes to be defined in terms of whether the expression can be generated by the formal rules of the grammar. The likelihood of its occurrence in a corpus or even native speakers’ judgments on its accept- ability are secondary. d. Abstractness. Especially since the 1970s, Chomskyan linguistics has become increasingly abstract, in the sense that the entities with which it deals have no overt manifestation in actual linguistic expressions. Abstract entities in this sense include traces, empty categories, pro and PRO, as well as movement operations of various kinds (which may themselves operate over ‘‘empty’’ categories), not to mention tree structures and the structural relations which these define, such as the c-command relation. e. Modularity. The syntactic component is regarded as an encapsulated mod- ule which functions independently not only of phonology and semantics 572 john r. taylor but also of more general cognitive capacities, such as perception and cat- egorization, memory and learning, and interpersonal and rhetorical skills. The semantic and phonological components ‘‘interpret’’ syntactic struc- tures (after lexical insertion) and ‘‘interface’’ with nonlinguistic domains, such as conceptual knowledge and processes of speech production and perception. Actual linguistic performance results from the interaction of strictly linguistic competence with nonlinguistic cognitive abilities. Not only is syntax regarded as an autonomous module, the syntactic module itself is ascribed a modular structure whose ‘‘submodules’’ have included the X-bar principle, the Theta principle, and the Case principle. f. The Neglect of Semantics. Given the syntactocentrism of Generative Gram- mar, it is not surprising that semantic issues, especially lexical semantics, have been largely ignored. Indeed, it is taken as axiomatic that the syntax is structured in accordance with its own principles, which cannot be ‘‘re- duced’’ to semantic principles nor ‘‘explained’’ in semantic terms. In- terest in semantics has been restricted, in the main, to those topics which have an obvious reflex in syntactic organization—in particular, matters of ‘‘logical form,’’ such as argument structure, thematic relations, quantifier scope, anaphors, reciprocals, and the like. 9 g. The Core and the Periphery. The aim of the generative enterprise has been from the very start the search for high-level generalizations. In this process, the idiosyncratic, the idiomatic, and the exceptional have been sidelined. The high-level generalizations define the ‘‘core’’ of the language system, while the idiomatic and the peculiarities of individual constructions and lexical items were relegated to the periphery; not being amenable to high- level generalizations, the periphery was of little theoretical interest and was therefore accorded little attention. Symptomatic is the fact that the gen- erative literature has tended to address only a very limited range of phe- nomena. Wh-movement, extraction, raising, anaphors, and reciprocals have been favorite and recurring topics. h. Universal Grammar and Language Acquisition. The generative enterprise, as characterized above, brings with it a rather specific view on the men- tal representation of language. It is important here to bear in mind that Chomskyan linguistics developed under its own theory-internal momen- tum; it was not driven by independently known facts about the mind. On the contrary, it was Generative Linguistics which imposed conditions on the structure of a more general theory of cognition. The linguistic theory required a theory of cognition in which syntax could be allocated to an encapsulated computational module. Moreover, given the abstractness of the syntactic representations and the implausibility that the syntactic rules could ever be induced through exposure to linguistic data, it was neces- sary that the general architecture of the syntactic component, with its var- ious interacting submodules, be genetically inherited. Acquisition thus became a matter of the ‘‘setting’’ of ‘‘parameters’’ provided by Universal cognitive linguistics and autonomous linguistics 573 Grammar, something which, it was assumed, would be possible on only minimal exposure to data. The downside was that the parameters of Uni- versal Grammar accounted only for the core of the language (defined, somewhat circularly, as those features of a language that could be handled by parameter setting). Everything else in a language (including such mat- ters as the meanings and idiosyncratic behavior of individual lexical items) belonged to the periphery, which had to be learned by generalization fol- lowing extensive exposure to data. 5. The Cognitive Linguistics Reaction Already in the early 1970s, a number of linguists were feeling increasingly unhappy with the direction being taken by Chomsky and his followers. Some were dissat- isfied with the increasing abstractness of Generative Grammar and sought to de- velop more surface-oriented models, such as Lexical-Functional Grammar (Bresnan 1978), Generalized Phrase Structure Grammar (Gazdar et al. 1985), and Word Gram- mar (Hudson 1984). Cognitive Linguistics shares with these approaches a focus on surface phenomena and a general skepticism toward constructs that are not im- minent in primary linguistic data. Others deplored the encapsulation of the lan- guage system from the many uses to which language is put and argued that structural aspects of a language have been shaped by the functions it needs to perform. An early statement of the functionalist approach was Givo ´ n(1979). Finally, the self-styled cognitive linguists proposed to take seriously the claim that linguistic knowledge is a cognitive phenomenon, which needs to be studied as an integral aspect of human cognition. In practice, this has entailed framing linguistic analyses in terms of more general (i.e., not exclusively linguistic) cognitive abilities. Some, such as Lamb—who, incidentally, claims to have been the first to have used the term ‘‘cognitive linguistics’’ in print (1999: 381)—have gone further, insisting that lin- guistic theory should be consistent with what is currently known about the neuronal structure of the brain. (For this reason, Lamb prefers the designation ‘‘neurocog- nitive linguistics.’’) The neurocognitive approach has characterized Lakoff’s most recent work (see the Neural Theory of Language (NLT) Web site: http://www.icsi .berkeley.edu/NTL/). A number of chapters in this Handbook deal with cognitive abilities which have proved to be of crucial importance in the study of language, but which are not re- stricted to the domain of language. One of these is categorization, including cate- gorization by prototype and by family resemblance (Lewandowska-Tomaszczyk, chapter 6). Categorization is most obviously relevant to lexical semantics, as well as 574 john r. taylor to the structure of morphological, syntactic, and phonological categories (Taylor 1989). Categorization is also crucially involved in any act of linguistic performance, in that the unique features—phonological, structural, and semantic—of every speech act need to be assessed with a view to their categorization by existing schemas. Several general cognitive abilities can be brought under the broad title of ‘‘construal operations’’ (see Verhagen, this volume, chapter 3). At issue is the fact that linguistic expressions do not, and cannot, designate a state of affairs as it ‘‘ob- jectively’’ is; rather, the scene must be processed and conceptualized by the human mind. Construal operations include attentional processes (chapter 11), force- dynamic construals (chapter 12), metaphor, image schemas, and conceptual blend- ing (chapters 8, 9,and15). A further general ability concerns entrenchment, that is, the degree to which a cognitive representation can be strengthened through repeated activation (chapter 5). In neurocognitive terms, entrenchment is an instance of Hebb’s postulate: ‘‘Cells that fire together, wire together’’ (Harnish 2002: 73). Highly entrenched representations can be more easily accessed and can be accessed as units, without attention being paid to their internal composition. Langacker’s rejection of the rule/list fallacy, referred to earlier, was based on the view that complex expres- sions, even though their internal structure might be unexceptional, may nevertheless be stored and activated as wholes. Evidence that this is the case comes from the fact that frequently occurring, and therefore highly entrenched units, tend to acquire properties that are different from those of less entrenched units (Bybee 1995, 2001; Bybee and Hopper 2001;Taylor2002: 307–18). Perhaps the most fundamental cognitive ability, however—and the one which lends human language its distinctive properties vis-a ` -vis animal communication systems—is our capacity for symbolic thought (Noble and Davidson 1996; Deacon 1997). In contrast to indexical signs, which stand in a causal relation (or which are perceived to stand in a causal relation) to their designatum, the distinctive feature of a symbol is that it stands for a conceptualization and is independent of external stimuli. 10 Symbolic thought appears to be underpinned by other uniquely human cognitive abilities, such as our ability to empathize (Lieberman 1991) and our ca- pacity for joint attention (Tomasello 1999), both of which follow from our reali- zation that other human beings have minds and a conceptual life not unlike our own (Taylor 2002: 67–68). Cognitive Linguistics is absolutely committed to a symbolic view of language (Lakoff 1987: 583; Langacker 1987: 11; Taylor 2002: 38–58). It is this feature which distinguishes Cognitive Linguistics not only from Chomskyan Generative Linguis- tics but from several other approaches which emerged in reaction to the Chom- skyan agenda, such as Lexical-Functional Grammar and Phrase Structure Gram- mar. Thus, in Langacker’s Cognitive Grammar there are only three objects of study: phonological representations (language in its ‘‘overt,’’ perceptible form), semantic representations, and symbolic relations between phonological and semantic rep- resentations. Importantly, patterns for the combination of smaller units into larger configurations (traditionally, the province of morphology and syntax) are them- selves regarded as symbolic units, albeit schematic ones, which are abstracted on cognitive linguistics and autonomous linguistics 575 the basis of encounters with their instances. As befits their symbolic nature, these ‘‘constructional schemas’’ or ‘‘constructions’’ for short, are characterized both in their formal aspects (typically, the formal characterization consists of a sequence of slots, which are able to be filled by items of the appropriate category) and with respect to their semantics (see Croft, this volume, chapter 18). Generative Linguistics had sought to eliminate constructions from the gram- mar. Constructions were ‘‘epiphenomena’’ which emerged through the interaction of principles of Universal Grammar (Chomsky 1995: 170). In Cognitive Linguistics, on the other hand, constructions are central and perform the work that, in other theories, is handled by the syntactic component. Evidence that constructions can- not be reduced to more general syntactic principles comes from the fact that a construction may be able to contribute its schematic meaning to the meaning of an expression which instantiates it; the schematic meaning may go beyond, or even override, the meanings contributed by the component units. Thus, to take an often- cited example, it is the construction in which the word occurs which gives sneeze a caused-motion reading in sneeze the napkin off the table (Goldberg 1995: 9). 5.1. Methodological Consequences A commitment to a symbolic view of language and attempts to ground the study of language in more general cognitive abilities have profoundly influenced the Cog- nitive Linguistics research program. Thus, a distinctive feature of Cognitive Linguistics research has been a focus on what in Chomskyan linguistics would be dismissed as the theoretically uninter- esting periphery. There have been a wealth of studies on the properties of indi- vidual lexical items, especially highly polysemous ones such as the prepositions, as well as on quirky grammatical constructions which have properties that cannot be predicted from general principles; see, for example, Tuggy (1996) on the double-be construction and Lambrecht (1990) on the incredulity response construction. Along with an interest in the peripheral and the particular has been a general lack of enthusiasm for mathematical formalism. Rejection of formalism has not, however, meant that cognitive linguists have ignored the need for precise and de- tailed characterizations of the phenomena under discussion. What we witness, in fact, has been a search for alternative, especially visual, modes of representa- tion. Pictographic representations of semantic structures are a distinctive and well- known feature of Langacker’s writings. Diagrammatic representations have also been employed by Talmy (1988) in his seminal paper on force and were also an important aspect of early treatments of prepositions, such as Brugman (1981) and Lindner (1981). Network models of polysemy have also lent themselves to visual representation (e.g., Tuggy 2003), as has the notion of a construction as a linking of semantic, syntactic, and lexical specifications (Goldberg 1995). A further notable feature of Cognitive Linguistics has been attempts to offer conceptual motivations for syntactic and morphological structures as well as for 576 john r. taylor the entities over which syntax and morphology operate, such as word classes. Recall that in autonomous linguistics the syntax was a computational device which op- erated over contentless symbols, blind to their phonological and semantic proper- ties. This procedure was deemed to be necessary since it was assumed that the major categories over which syntax operates—categories such as Noun, Verb, and so on—were resistant to a coherent semantic characterization, as were certain structural relations such as that of a clausal subject. Jackendoff, for whom a reg- ular relationship between syntactic and semantic categories was indeed a ‘‘working assumption’’ (a position which Cognitive Linguists would certainly endorse), specifically mentions the case of nouns and subjects as evidence that syntax still needs to be accorded a certain autonomy vis-a ` -vis semantics (Jackendoff 1983: 14). As a matter of fact, there is now a substantial body of literature which has addressed the functional and conceptual basis of the major word classes (Givo ´ n 1984; Wierzbicka 1986; Langacker 1987; Croft 1991; Taylor 1996). A more radical view has recently been expressed by Croft (1999, 2001). Turning on its head the genera- tive view that constructions are mere epiphenomena, Croft argues that construc- tions are basic and that word classes are epiphenomenal, since they need to be de- fined in terms of the constructions in which they are eligible to occur. The notion of subject has also come in for intense scrutiny by Langacker (1993, 1999). Recognizing that the identification of ‘‘subject’’ with the semantic role ‘‘Agent’’ is bound to be inadequate, he proposes a highly schematic characterization in terms of the ‘‘pri- mary figure’’ at the clausal level. Especially challenging for this approach are so- called expletive subjects, as in It seems that a mistake has been made and There seems to have been a mistake. On generative approaches, it and there are mere place- holders, inserted by the syntax in order to satisfy the requirement that finite clauses in English must have a grammatical subject. A symbolic view of language entails that even the so-called expletive subjects have semantic content, by virtue of which they are able to function as the primary figure. Attempts to offer a conceptual motivation for structural aspects of a lan- guage were evident from the earliest days of Cognitive Linguistics. The earliest full- fledged presentation of Cognitive Grammar was Langacker’s (1982) analysis of the English passive. His strategy was to examine each of the constituents of a passive clause—the verb be, the participle, the optional by-phrase—and to identify their contribution to a passive clause in relation to the values which these elements have elsewhere in the language. It turned out that a passive clause was not simply a compositional function of the standard values of its constituents—passive be, for example, has a processual value unique to the passive. Nor could a passive be re- garded as the consequence of an algebraic transformation of an active. The passive turned out to be a construction in its own right whose global properties were mo- tivated by the properties of its parts. Motivation was also a key concept in Lakoff’s analysis of English there -constructions (1987: 462–585). Lakoff proposed a ‘‘cen- tral’’ construction in which there has a deictic, referential meaning and, radiating out from this, a large number of secondary constructions which are based on the central deictic. cognitive linguistics and autonomous linguistics 577 Especially significant from a methodological point of view have been a number of cognitive linguistic studies which have tackled head-on the kinds of data that have been central concerns in autonomous linguistics and which are often cited as evidence, precisely, for the correctness of the autonomous approach. These include anaphoric reference (van Hoek 1997), raising (Langacker 1995), and ex- traction (Deane 1992). Since anaphors are discussed elsewhere in this Handbook (van Hoek, chapter 34), I will restrict myself to a few comments on raising. Raising has been a stock-in-trade of autonomous syntax. It was claimed that a sentence such as Don is likely to leave exhibits a divergence of semantic and syntactic structure. Only events are likely (to happen), not individuals. The deep structure therefore has ‘be likely’ predicated of the proposition ‘(for) Don to leave’. The divergence be- tween the presumed semantic structure and the attested surface structure results from the ‘‘raising’’ of the subject of the embedded clause to be the subject of the main clause. Langacker (1995) argued against this, claiming that ‘be likely’ can indeed be predicated of a nominal, with the to-clause designating the process with respect to which the likelihood is being assessed. Statements such as A war is likely are fully acceptable, since the process whose likelihood is being assessed can be inferred from the semantic structure of the subject nominal (‘A war is likely to occur’). In brief, Langacker claims that the overt, surface structure of the raised expression corresponds rather directly to its semantic structure; raising, as a process of rearranging elements of a presumed deep structure, is superfluous. 5.2. Acquisition The Cognitive Linguistics enterprise offers a distinct perspective on language ac- quisition (Tomasello, this volume, chapter 41). There can, to be sure, be no question about humans’ unique ability to acquire and to use language. To this extent, lan- guage builds on uniquely human and therefore innate, that is, genetically inherited, abilities. These abilities, however, may not be peculiar to language. I have already drawn attention to the hypothesis that language rests on the capacity for symbolic thought rather than on an innate algebraic syntax. Not to be overlooked, too, is the possibility that a number of basic grammatical categories may be grounded in cog- nitive abilities that emerge during the early years of life. Langacker, in this con- nection, speaks of ‘‘conceptual archetypes’’ (1999: 41). Among these are the concepts of a spatially bounded physical object and of a spatially distributed substance. These concepts underlie the emergence of the noun category and the distinction, crucial to the morphosyntax of many languages, between count and mass nouns (Soja, Carey, and Spelke 1991; Imai and Gentner 1997). Of special significance to the cognitive linguistic view of syntax has been research on the acquisition of constructions (Tomasello 2000). On the genera- tive view, the acquisition of a structural configuration is essentially complete once the appropriate parameters of Universal Grammar have been set. It appears to be the case, however, that constructions tend to be acquired gradually on a word- 578 john r. taylor by-word basis. The two-year-old who comes out with I kick ball may simply have acquired an idiosyncratic property of the verb kick (Tomasello and Brooks 1998). Only when a number of lexically specified transitive clauses have been learned does the child apply the transitive construction productively to new verbs. A fur- ther finding (Tomasello 2000) has been that by far the greatest part of what the young child learner says consists of a collage of previously used utterances— utterances which, in the child’s previous linguistic experience, have acquired the status of entrenched units. ‘‘Creativity,’’ in the Chomskyan, generative sense, hardly features at all. 6. Autonomy I now address the concept of autonomy and the extent to which the notion is relevant to the polemics between Cognitive and Generative Linguistics. An object of study may be said to be autonomous if it cannot be ‘‘reduced’’ or explicated by reference to other objects of study. There are a number of ways in which the notion of autonomy can be applied to the study of language (Croft 1995; Newmeyer 1998). a. First, we can enquire whether linguistics, that is, the academic study of language, constitutes an autonomous field of study. It is amusing to re- call that as recently as 1986, Newmeyer discussed ‘‘opposition to autono- mous linguistics’’ in precisely these terms. He defended the autonomy of linguistics as an academic discipline against three charges. These came from humanist scholars, for whom the study of language should be subsumed under the study of literary texts; from Marxist scholars, for whom language was a reflex of economic processes; and from political activists, for whom the study of language should be just one facet of the study of oppression. A concern to establish linguistics as a legitimate and ‘‘autonomous’’ aca- demic discipline goes back at least to Saussure ([1916] 1967), who recog- nized that many disciplines have a legitimate interest in language. He argued, however, that if linguistics was to have a place in the academic curriculum, there had to be a distinctly linguistic way of studying language, and this, for Saussure, lay in recognizing the ‘‘linguistic sign’’ (comparable to the ‘‘symbolic unit’’ in Langacker’s theory) as the central object of enquiry. b. We can enquire whether knowledge of a language constitutes an autono- mous object of study or whether language knowledge can be reduced to more general conceptual knowledge. While Cognitive Linguistics certainly subscribes to the view that language may be grounded in more general cognitive abilities, there is no suggestion that the study of these general cognitive linguistics and autonomous linguistics 579 . Aspects of the Theory of Syntax (Chomsky 1965). The system of rules which constitute the grammar had to be evaluated, not only in terms of its descriptive adequacy, but as a hypothesis about the cognitive. corollary of (a) and (b), the ‘‘grammaticality’’ of an expression comes to be defined in terms of whether the expression can be generated by the formal rules of the grammar. The likelihood of its occurrence. about the mind. On the contrary, it was Generative Linguistics which imposed conditions on the structure of a more general theory of cognition. The linguistic theory required a theory of cognition