Noam Chomsky

Noam Chomsky
Norbert Hornstein
In Edward Craig (ed.), Routledge Encyclopedia of Philosophy, London, 1998

Fish swim, birds fly, people talk. The talents displayed by fish and birds rest on specific biological structures whose intricate detail is attributable to genetic endowment. Human linguistic capacity similarly rests on dedicated mental structures many of whose specific details are an innate biological endowment of the species. One of Chomsky’s central concerns has been to press this analogy and uncover its implications for theories of mind, meaning and knowledge.

This work has proceeded along two broad fronts.

First, Chomsky has fundamentally restructured grammatical research. Due to his work, the central object of study in linguistics is ‘the language faculty’, a postulated mental organ which is dedicated to acquiring linguistic knowledge and is involved in various aspects of language-use, including the production and understanding of utterances. The aim of linguistic theory is to describe the initial state of this faculty and how it changes with exposure to linguistic data. Chomsky (1981) characterizes the initial state of the language faculty as a set of principles and parameters. Language acquisition consists in setting these open parameter values on the basis of linguistic data available to a child. The initial state of the system is a Universal Grammar (UG): a super-recipe for concocting language-specific grammars. Grammars constitute the knowledge of particular languages that result when parametric values are fixed.

Linguistic theory, given these views, has a double mission. First, it aims to ‘adequately’ characterize the grammars (and hence the mental states) attained by native speakers. Theories are ‘descriptively adequate’ if they attain this goal. In addition, linguistic theory aims to explain how grammatical competence is attained. Theories are ‘explanatorily adequate’ if they show how descriptively adequate grammars can arise on the basis of exposure to ‘primary linguistic data’ (PLD): the data children are exposed to and use in attaining their native grammars. Explanatory adequacy rests on an articulated theory of UG, and in particular a detailed theory of the general principles and open parameters that characterize the initial state of the language faculty (that is, the biologically endowed mental structures).

Chomsky has also pursued a second set of concerns. He has vigorously criticized many philosophical nostrums from the perspective of this revitalized approach to linguistics. Three topics he has consistently returned to are:

· Knowledge of language and its general epistemological implications
· Indeterminacy and underdetermination in linguistic theory
· Person-specific ‘I-languages’ versus socially constituted ‘E-languages’ as the proper objects of scientific study.

1 The aims and principles of linguistic theory

There is an intimate relation between how a problem is conceived and the kinds of explanations one should offer. Chomsky proposes that we identify explanation in linguistics with a solution to the problem of how children can attain mastery of their native languages on the basis of a rather slender database. This is often referred to as ‘the logical problem of language acquisition’.

A natural language assigns meanings to an unbounded number of sentences. Humans typically come to master at least one such language in a surprisingly short time, without conscious effort, explicit instruction or apparent difficulty. How is this possible? There are significant constraints on any acceptable answer.

First, a human can acquire any language if placed in the appropriate speech community. Grow up in Boston and one grows up speaking English the way Bostonians do. However, the ‘primary linguistic data’ (PLD) available to the child are unable to guide the task unaided. There are four kinds of problems with the data that prevent it from shaping the outcome:,

(a) The set of sentences the child is exposed to is finite. However, the knowledge attained extends over an unbounded domain of sentences.
(b) The child is exposed not to sentences but to utterances of sentences. These are imperfect vehicles for the transmission of sentential information as they can be defective in various ways. Slurred speech, half sentences, slips of the tongue and mispronunciations are only a few of the ways that utterances can obscure sentence structure.
(c) Acquisition takes place without explicit guidance by the speech community. This is so for a variety of reasons. Children do not make many errors to begin with when one considers the range of logically possible mistakes. Moreover, adults do not engage in systematic corrections of errors that do occur and even when correction is offered children seem neither to notice nor to care. At any rate, children seem surprisingly immune to any form of adult linguistic intrusion (see Lightfoot 1982).
(d) Last of all, and most importantly, of the linguistic evidence theoretically available to the child, it is likely that only simple sentences are absorbed. The gap between input and intake is attributable to various cognitive limitations such as short attention span and limited memory. This implies that the acquisition process is primarily guided by the information available in well-formed simple sentences. Negative data (the information available in unacceptable ill-formed sentences) and complex data (the information yielded by complex constructions) are not among the PLD that guide the process of grammar acquisition. The child constructing its native grammar is limited to an informationally restricted subset of the relevant data. In contrast to the evidence that the linguist exploits in theory construction, the information the child uses in building its grammar is severely restricted. This suggests that whenever the linguistic properties of complex clauses diverge from simple ones, the acquisition of this knowledge cannot be driven by data. Induction is insufficient as the relevant information is simply unavailable in the PLD.

The general picture that emerges from these considerations is that attaining linguistic competence involves the acquisition of a grammar, and that humans come equipped with a rich innate system that guides the process of grammar construction. This system is supple enough to allow for the acquisition of any natural language grammar, yet rigid enough to guide the process despite the degeneracy and deficiency of the PLD. Linguistic theorizing takes the above facts as boundary conditions and aims both at descriptive adequacy (that is, to characterize the knowledge that speakers have of their native grammars) and explanatory adequacy (that is, to adumbrate the fine structure of the innate capacity) (see Language, innateness of).

Issues of descriptive and explanatory adequacy have loomed large in Chomsky’s work since the beginning. Chomsky’s objection, for example, to ‘Markov models’ of human linguistic competence was that they were incapable of dealing with long distance dependencies exemplified by conditional constructions in English and hence could not be descriptively adequate. His argument in favour of a transformational approach to grammar rested on the claim that it allowed for the statement of crucial generalizations evident in the judgments of native speakers and so advanced the goal of descriptive adequacy (Chomsky 1957). Similarly, his influential critique (1959) of Skinner’s Verbal Behavior consisted in showing that the learning theory presented therein was explanatorily inadequate. It was either too vague to be of scientific value or clearly incorrect given even moderately precise notions of stimulus or reinforcement.

The shift from the early Syntactic Structures (1957) theory to the one in Aspects of a Theory of Syntax (1965) was also motivated by concerns of explanatory adequacy. In the earlier model the recursive application of transformations allows for the generation of more and more complex sentences from the sentences produced by the ‘phrase structure’ component of the grammar. In the Aspects theory, recursion is incorporated into the phrase structure component itself, and removed from the transformational part of the theory (see Syntax §3). The impetus for this was the observation that greater explanatory adequacy could be attained by grammars that had a level of ‘Deep Structure’ incorporating a recursive base component. In particular, Fillmore (1963) observed that the various optional transformations in a Syntactic Structures theory always applied in a particular order in any given derivation. This order is unexplained in a Syntactic Structures theory; in Aspects it is deduced. Thus, the move to an Aspects-style grammar is motivated on grounds of greater explanatory adequacy: introducing Deep Structure and moving recursion to the base allows for a more restricted theory of Universal Grammar. All things being equal, restricting UG is always desirable as it advances a central goal of grammatical theory; the more restricted the options innately available for grammar construction, the easier it is to explain how language acquisition is possible, despite the difficulties in the PLD noted above.

The same logic motivates various later additions to and shifts in grammatical theory. For example, a major move in the 1970s was radically to simplify transformational operations so as to make their acquisition easier. This involves eliminating any mention of construction-specific properties from transformational rules. For example, an Aspects rule for passive constructions looks like (1), the left-hand side being the Structural Description (SD) and the right hand side being the Structural Change (SC):

X-NP1-V-NP2-Y->(1)X - NP2 - be + en V - by + NP1 - Y x

This rule would explain, for instance the grammaticality of ‘the ball is kicked by John’ given that of ‘John kicks the ball’. Observe that the SC involves the constants ‘ ’ and ‘by’. The SD mentions three general expressions, ‘NP1’, ‘V’ and ‘NP2’ and treats these as part of the context for the application of the rule. In place of this, Chomsky proposed eliminating the passive rule and replacing it with a more general rule that moves NPs (Chomsky 1977, 1986). The passive rule in (1) involves two applications of the ‘Move NP’ rule, one moving the subject ‘NP1’ to the ‘by’ phrase, and another moving the object ‘NP2’ to the subject position. In effect, all the elements that make the passive rule in (1) specific to transitive constructions are deleted and a simpler rule (‘Move NP’) replaces it.

There is a potential empirical cost to simple rules, however. The simpler a transformation the more it generates unacceptable outputs. Thus, while a grammar with (1) would not derive ‘was jumped by John’ from ‘John jumped’, a grammar eschewing (1) and opting for the simpler ‘Move NP’ rule is not similarly restricted. To prevent overgeneration, therefore, the structure of UG must be enriched with general grammatical conditions that function to reign in the undesired overgeneration (Chomsky 1973, 1977, 1986). Chomsky has repeatedly emphasized the tension inherent in developing theories with both wide empirical coverage and reasonable levels of explanatory adequacy.

A high point of this research agenda is Chomsky’s Lectures on Government and Binding (1981). Here the transformational component is reduced to the extremely simple rule ‘Move a’ - that is, move anything anywhere. To ensure that this transformational liberty does not result in generative chaos, various additions to the grammar are incorporated, many conditions on grammatical operations and outputs are proposed, and many earlier proposals (by both Chomsky and others) are refined. Among these are trace theory, the binding theory, bounding theory, case theory, theta theory and the Empty Category Principle. The picture of the grammar that Chomsky’s Lectures presents is that of a highly modular series of interacting subsystems which in concert restrict the operation of very general and very simple grammatical rules. In contrast to earlier traditional approaches to grammar, Lectures witnesses the virtual elimination of grammatical constructions as theoretical constructs. Thus, in Government Binding (GB)-style theories there are no rules of Passive, Raising, Relativization or Question Formation as there were in earlier theories. Within GB, language variation is not a matter of different grammars having different rules. Rather, the phenomena attested in different languages are deduced by variously setting the parameters of Universal Grammar. Given the interaction of the grammatical modules, a few parametric changes can result in what appear on the surface to be very different linguistic configurations. In contrast to earlier approaches to language, variation consists not in employing different kinds of rules, but in having set the parameters of an otherwise fixed system in somewhat different ways (see Chomsky 1983).

The GB research programme has proven to be quite successful in both its descriptive range and its explanatory appeal. Despite this, Chomsky has urged a yet more ambitious avenue of research. He has embarked on the development of a rationalist approach to grammar that goes under the name of ‘Minimalism’ (Chomsky 1995). The theory is ‘rationalist’ both in that it is grounded on very simple and perspicuous first principles, and in that it makes use only of notions required by ‘virtual conceptual necessity’. Chomsky hopes to make do with concepts that no approach to grammar can conceivably do without and remain true to the most obvious features of linguistic competence. For example, every theory of grammar treats sentences as pairings of sounds and meanings. Thus, any theory will require that every sentence have a phonological and an interpretative structure. In GB theories, these sorts of information are encoded in the PF (Phonetic Form) and LF (Logical Form) phrase markers respectively. In addition, GB theories recognize two other distinctive grammatical levels: S-structure and D-structure. A minimal theory, Chomsky argues, should dispense with everything but LF and PF. It will be based on natural ‘economy’ principles and indispensable primitives. Chomsky has suggested reanalysing many of the restrictions that GB theories impose in terms of ‘least effort’ notions such as ‘shortest move’ and ‘last resort movement’. For example, he proposes that the unacceptability of sentences such as ‘John is expected will win’, are ultimately due to the fact that the moved NP ‘John’ need not have moved from the embedded subject position (between ‘expected’ and ‘will’) as it fulfils no grammatical requirement by so moving. This work is still in its infancy, but it has already prompted significant revisions of earlier conclusions. For example, with the elimination of D-structure, the recursive engine of the grammar has once again become the province of generalized transformations. Whatever its ultimate success, however, Minimalism continues the pursuit of the broad goals of descriptive and explanatory adequacy enunciated in Chomsky’s earliest work.

2 Knowledge of language

According to Chomsky, the three fundamental epistemological questions in the domain of language are ‘What constitutes knowledge of language?’, ‘How is knowledge of language acquired?’ and ‘How is this knowledge put to use?’. The answer to the first question is given by a particular generative grammar. Harold’s knowledge of English is identified with Harold’s being in a particular mental/brain state. A descriptively adequate grammar characterizes this part of Harold’s mental/brain make-up. An answer to the second question is provided by a specification of UG and the principles that take the initial state of the language faculty to the knowledgeable state on exposure to PLD. Harold knows English in virtue of being genetically endowed with a language faculty and having been normally brought up in an English-speaking community. Beyond this, further issues of grounding are unnecessary. Issues of epistemological justification and grounding in the data are replaced by questions concerning the fine structure of the initial state of the language faculty and how its open parameters are set on the basis of PLD. The third question is answered by outlining how linguistic knowledge interacts with other cognitive capacities and abilities to issue in various linguistic acts such as expressing one’s thoughts, parsing incoming speech and so on (see Chomsky 1986).

How much does the language case tell us about epistemological issues in other domains? In other words, should knowledge of quantum mechanics be analysed in a similar vein, that is, being in a particular mental state, grounded in specific innate capacities and so on. Chomsky only makes sparse comments on this general issue, but those he advances suggest that he believes that knowledge in these domains should be approached in much the same way they are approached in the domain of language. This suggests that humans have an innate science-forming capacity that underlies our success in the few domains of inquiry in which there has indeed been scientific success. As in the domain of language, this capacity is focused and modular rather than being a general all-purpose tool and this, Chomsky speculates, might well underlie the patchiness of our successes. Where we have the right biological propensities, we develop rich insightful theories that far outpace the data from which they are projected. Where this mind/brain structure is lacking, mysteries abound that seem recalcitrant to systematic inquiry. Stressing our cognitive limits is a staple of Chomsky’s general epistemological reflections. If humans are part of the natural world we should expect there to be problems that fall within our cognitive grasp and mysteries that lie outside it. The rich theoretical insights allowed in the natural sciences are the result of a chance convergence between properties of the natural world and properties of the human mind/brain (see Chomsky 1975).

3 Indeterminacy and underdetermination

Knowledge of language, Chomsky has argued, presents a strong argument in favour of traditional rationalist approaches to mind and against traditional empiricist approaches (see Learning §1; Rationalism). In particular, ‘learning’ is treated as more akin to growth and the course of acquisition is seen more as the unfolding of innate propensities under the trigger of experiential input than as the result of the shaping effects of the environment. This rationalist perspective is now quite common and this is largely due to Chomsky’s efforts. Chomsky has consistently warned against empiricist prejudices in philosophy, and in no instance more strongly than in his critique of Quine’s methodological remarks on linguistics (for example, see Quine 1960).

Chomsky takes Quine to be arguing that linguistic investigations are beset with problems greater than those endemic to inquiry in general. Whereas empirical investigation in general suffers from underdetermination of theory by evidence, linguistic study is beset with the added problem of indeterminacy (see Radical translation and radical interpretation §§2-3). Indeterminacy differs from standard inductive underdetermination (see Underdetermination) in that where there is indeterminacy ‘there is no real question of right choice’ among competing proposals. Chomsky interprets Quine as arguing that ‘determining truth in the study of language differs from the problem of determining truth in the study of physics’ (Chomsky 1975: 182-3).

In reply, Chomsky (1969) argues that Quine’s thesis rests on classical empiricist assumptions about how languages are acquired. Quine, he argues, supposes that humans have ‘an innate quality space with a built-in distance measure’ tuned to certain ’simple physical correlates’. In addition, certain kinds of induction in this space are permitted. Beyond this, however, ‘language-learning is a matter of association of sentences to one another and to certain stimuli through conditioning’. Further, one cannot ‘make significant generalizations about language or common-sense theories, and the child has no concept of language or of "common-sense" prior to this training’ (Chomsky 1969: 54-5, 63).

Chomsky notes that Quine provides no evidence to support these assumptions. Nor can there be any good evidence to support them if the nature of the learning problem in the domain of language is characterized as Chomsky has argued it must be. Chomsky concludes that ‘Quine’s thesis of the indeterminacy of translation amounts to an implausible and quite unsubstantiated empirical claim about what the mind brings to the problem of acquisition of language (or of knowledge in general) as an innate property’ (Chomsky 1969: 66). Stripped of these tendentious empirical assumptions, Quine fails to show that indeterminacy is anything other than the familiar problem of underdetermination of theory by evidence as applied to linguistics. Chomsky (1996) has since argued that the ultimate source of many critiques of the mental sciences in general and linguistics in particular (including Quine’s indeterminacy thesis) is a kind of methodological dualism that takes humans to be separate from the natural world. This dualism is manifest in the a priori constraints that philosophers place on explanations in the mental sciences, which would be regarded as inappropriate if applied to the physical sciences.

In this vein Chomsky asks, for example, why access to consciousness is so often taken to be crucial in substantiating the claim that humans have I-language or follow rules. Suppose, he asks, we had a theory that perfectly described what happens when sound waves hit the ear, stimulating the performance system to access the cognitive system and construct a logical form that interacts with other cognitive systems to yield comprehension, in so far as the language faculty enters into this process. What more could be desired? The insistence that this entire process be accessible to consciousness in order for the account to be credible, he argues, is a demand beyond naturalism, a form of methodological dualism of dubious standing that would be summarily rejected if raised elsewhere.

Or consider the oft-voiced suspicions concerning mentalist approaches in psychology. Many philosophers are ready to accept these as perhaps temporarily necessary but ultimately, the view seems to be, mentalist theories must reduce to physical ones to be truly legitimate. Chomsky argues that this sentiment is another manifestation of methodological dualism and should be rejected. First, it presupposes that there is a tenable distinction between the mental and the physical. However, Chomsky argues that since Newton undermined the Cartesian theory of body by showing that more ‘occult’ forces were required in an adequate physics, mind-body dualism has lost all grounding. Second, even if reduction were possible, reduction comes in many varieties and there is little reason to believe that the contours of the reducing physical theory would be left unaffected by the process. Since Newton, Chomsky notes, ‘physical’ has been an honorific term that signifies those areas in which we have some nontrivial degree of theoretical understanding. The relevant scientific question is whether some theory or other offers interesting descriptions and explanations. The further insistence that its primitives be couched in physical vocabulary is either vacuous (because ‘physical’ has no general connotation) or illegitimate (another instance of methodological dualism).

The general conclusion Chomsky draws is that whatever problems linguistic theory encounters, it is no more methodologically problematic than theories in other domains. He attributes the qualms of philosophers to lingering empiricist dogma or an indefensible epistemological dualism.

4 I-language versus E-language

Given the aims of Chomskian linguistic theory, the proper objects of study are the I-languages internalized by native speakers, rather than public E(xternal)-languages used by populations. Chomsky denies that public E-languages are interesting objects of scientific study. Indeed he denies that E-languages can be coherently specified as they simply do not exist. The proper objects of inquiry are I-languages; ‘I’ standing for intensional, internal and individual. An I-language is individual in that each speaker has one. This focus turns the common wisdom on its head. E-languages like English, Swahili and so forth are (at best) radical idealizations for Chomsky, or (at worst) incoherent pseudo-objects. At best, E-languages are the intersection of the common properties of various I-languages. Thus, for example, it is not that speakers communicate because they have a language in common; rather wherever I-languages overlap communication is possible.

An I-language is internal in the sense of being part of a speaker’s individual mental make-up. It is neither a Platonic object nor a social construct. Also, an I-language is intensional, not extensional. Comprised as it is of an unbounded number of sentences, a language cannot be ‘given’ except via a specification of the function that generates them, that is a grammar for that language. Thus, it is languages in intension, languages dressed in all of their grammatical robes, not simple concatenations of words, that are the proper objects of scientific interest. One consequence of this is that weak generative capacity (that is, the extensional equivalence of languages generated by different grammars) is of dubious interest. In short, the shift from E-language to I-language turns many long-standing questions around, raising some to prominence that were considered secondary and relegating many that previously were considered crucial to the status of pseudo-questions.

Many philosophers have found Chomsky’s focus on I-language problematic. To illustrate, we will consider an important philosophical critique and Chomsky’s reply.

Dummett (1986) argues against internalist approaches to language that they fail to provide an account of notions like ‘language of a community’ or ‘community norms’ in the sense presupposed by virtually all work in the philosophy of language and philosophical semantics. These notions, Dummett claims, are required to provide a notion of a common public language which ‘exists independently of any particular speakers’ and of which native speakers have a ‘partial, and partially erroneous, grasp’ (see Language, social nature of §2).

The naturalistic study of language, Chomsky counters, has no place for a Platonistic notion of language, a notion of language outside the mind/brain that is common to various speakers and to which each speaker stands in some cognitive relation. The reason is that this Platonistic reification rests on notions like ‘language’ and ‘community’ that are hopelessly under-specified. Asking if two people speak the same language is, in Chomsky’s opinion, to ask a highly context-dependent question - much like asking whether Boston is near New York. What counts as a community depends on shifting expectations of individuals and groups. Human society is not neatly divided into communities with languages and their norms. Thus, what counts as a community is too under-specified to be useful for theoretical purposes. Therefore, it is not a defect of linguistic theory that these notions play no role within it.

From Chomsky’s perspective E-languages are epiphenomenal objects, if coherent at all. I-language in its universal aspects is part of the human genotype and specifies one aspect of the human mind/brain. Under the triggering effects of experience a particular grammar arises in the mind/brain of an individual. From this perspective, universal grammar and the steady-state grammars that arise from them are real objects. They will be physically realized in the genetic code and the adult brain. E-language, in contrast, has a murky ontological status. Chomsky (1980) argues that the priority of I-language cannot be reasonably doubted once we observe that languages involve an infinite pairing of sounds and meanings. Given that language is infinite, it cannot be specified except in so far as some finite characterization - a function in intension - is provided. It might be possible to give some characterization to the notion ‘a language used by a population’ but only indirectly via a grammatical specification of the language. But this concedes the priority of I-language as the claim unpacks into something like: each person in the relevant population has a grammar in their mind/brain that determines the E-language. Thus, at best, an E-language is that object which the I-language specifies. However, even this might be giving too much reality to E-languages, for there is nothing in the notion I-language that requires that what they specify corresponds to languages as commonly construed, that is, things like French, English and so on. It is consistent with Chomsky’s viewpoint that I-language never specifies any object that we might pre-theoretically call a language. Whether this is indeed the case, the key point is to realize that the move from grammar to language is a step away from real mechanisms to objects of a higher degree of abstraction. I-language is epistemologically and ontologically hardier than E-language, much philosophical opinion to the contrary.