The Science of Language

by Noam Chomsky

  This view is also attractive for another reason, I believe. It nicely places the study of linguistic meaning – of language's contribution to human ways of understanding – directly in syntax, and specifically in the syntax of the language faculty. In effect, as Nelson Goodman suggested several decades ago with regard to what he called linguistic “meanings,” to avoid being misled, do not say “concept of X,” but say instead “X-concept.” More specifically for language, for what appears at SEM, do not say “SEM of X,” but say “X-SEM.” This bears on what follows in the discussion in the main text.

  1 Its color-space volume is not a ‘map of the world at a time’. It is a form of representation, understood internalistically. It can be ‘used’ to deal with the organism's current environment. But ‘using’ it is not a matter of inspecting some internal scene and employing it to navigate. That way of thinking of its use is a variation on the sense datum theme.

  Commentaries

  Chapter 1

  Page 11, On uses of the word function

  The term “function” has several uses, including those in its everyday commonsense employment and its use in mathematics and natural science. There also appear to be some uses specialized for certain kinds of project in biology. An outline of some of the various uses of this term is found in Appendix II.

  Page 13, On creative options and their role

  These last remarks on the “creative options” that the introduction of language affords us (and us alone) reflect an important strain in Chomsky's thinking. I comment here briefly on the idea that human language use is ‘free’ and gives us humans cognitive advantages unavailable to other creatures that lack language.

  Unlike what appears to be the case with animal communication systems, human language production (in the head or externalized) does not depend in any discernible way on causal antecedents, can take unboundedly many forms, and yet generally remains appropriate to discourse circumstances. We can and do produce (usually in our heads alone) any number of sentences and fragments thereof, with no causal explanation to be found in environment or elsewhere in the head, and yet manage to be ‘rational’ in what we produce. As Descartes's follower Cordemoy and others put it, internal and external circumstances can incite or prompt us to say or think what we do, but they do not cause. No doubt when tortured or subjected to threats against life people can often be coerced to confess or utter all sorts of things. Nevertheless, they can and sometimes do choose not to – although at very considerable cost.

  Freedom from causal antecedence (“stimulus freedom”), unbounded flexibility in what is produced (“unboundedness,” “innovation”), and coherence and rationality (“appropriateness”) are the three elements in what Chomsky calls the “creative aspect of language use.” In work that goes back to at least his Cartesian Linguistics in 1966 (the roots were in place before that, including his 1959 review of B. F. Skinner's Verbal Behavior), Chomsky explores the implications of the fact that everyday language use can be and often is not only uncaused (by either current external or internal circumstance), but also virtually boundless (unbounded) in possible ‘output,’ and yet nevertheless (while uncaused and unbounded) generally appropriate to whatever the circumstances of conversation and thought might happen to be. Speech production or action seems to be a free yet reasonable – not random – form of action. Even in a court of law, where one is supposed to constrain what one says and provide an objective description of a single event, fifty witnesses will come up with fifty different descriptions utilizing many more than fifty different sentences to describe “the same thing” – and they would no doubt do so, no matter what they were asked to describe. All their sentences express the way they see the circumstance, and there appears to be no upper limit on what counts as appropriate for this one event, even where all try to converge in the interests of objectivity.

  The impact of the creative aspect of language use on the study of language and, more generally, cognitive science (or better, what that impact should be) is discussed at some length in my introduction to the third edition of Cartesian Linguistics (2009). For the moment, it is enough to point out that it was quite likely the introduction of language to the human species that led to what Jared Diamond called the “great leap forward” in human cognitive range, giving humans unique means to “solve problems.”

  Page 14, On what is distinctive about human nature (see also Appendix III)

  Appendix III takes up the issue of what is distinctive about human beings and discusses two prominent ways to try to explain it, the rationalist-biolinguistic approach endorsed by Chomsky and supported by his theory of language as a natural object and an empiricist approach adopted by the large majority of contemporary psychologists, philosophers, and social scientists.

  Page 15, How to get elementary mathematics from the operation Merge

  Here is one way to express how Merge restricted to an element that is merged again can yield the natural number system. Start with 0; if you merge it, you get {0} [in effect, 1]; merge it again and you get {0,{0}} [2]; merge it again and you get {0,{0,{0}}} [3], and so on.

  Page 16, On external and internal Merge

  For a linguistic example of external Merge, with the lexical items eat and apples you make {eat, apples}, often represented as

  Internal Merge arises in (and helps constitute) linguistic derivations/computations. Consider a derived question: what (John) eat what. Here we find internal Merge yielding what used to be called “movement” or “displacement” of the ‘what’ inside a derived structure to its ‘edge’; the ‘earlier’ merged ‘what’ that appears at the right does not actually move; nor is it ‘erased’ from the position it originally occupied as a result of an external merge. It remains ‘there’ and is “copied” at the ‘edge’ of the constructed set but – for reasons consistent with the Minimalist Program that would require considerable additional explication – it is not pronounced, the copy is. We do not hear it (or see it, if dealing with sign), although ‘the mind’ does. So internal Merge – Move in an earlier instantiation of the Minimalist Program – is Chomsky's Minimalist Program substitute for a transformation. It does the jobs that transformations in the earliest and middle days of Chomsky's work for the last fifty years were asked to do, but with very much less additional – and now it appears, gratuitous – machinery required. The gain in economy of computational machinery is part of the reason for the label ‘minimalist’ for the new way of proceeding Chomsky outlines (2005), the focus of a large part of our discussion. Internal Merge not only – as suggested – plays a role in making sense of why the derived structure being used in this illustration is ‘sounded out’ in the form What did John eat, it also helps explain why the “what” on the left side is treated as a quantifier (what thing) which binds a variable x in such a way that the variable x (the residual and unpronounced “what”) is interpreted (understood) as the direct object of the verb eat. So while one hears What did John eat? one's mind ‘sees’ or understands John as eating something (x), the first-merged “what.” Representation of the relevant structure and the ‘movement’ of the element often takes this form in linguistics:

  As indicated, there is no actual movement.

  There is considerable discussion of internal Merge in the linguistic literature. Some argue that it is faulty and should be abandoned, some that it undermines other assumptions, some that still other forms of Merge should be introduced, including Chomsky (2001), who wants “pair merge” to deal with what linguists call “adjunction.” I do not go into details; there are several technical matters and often quite deep issues in the background, worth exploring if the reader has some understanding of the technical matters. A search for ‘internal merge: linguistics’ on the internet will give an idea of what various individuals think is at stake. For current purposes, the emphasis is on seeing that for Chomsky, internal Merge and thus transformations have come to be seen as parts of the ‘conceptually necessary’ machinery of linguistic computation. Internal M
erge makes movement and transformations less of a mystery than they had earlier appeared. And if Merge in its various forms is all that is needed for language, assuming that one is given lexical items (‘elements’ in the above) with at least semantic features, perhaps not phonological, it makes it easy to understand how language and complex thought could have been introduced by a single mutation. It also contributes to the idea that language is a ‘perfect’ solution to the ‘design problem’ that can be conceived of as confronting a biological engineer – that of taking lexical items as sound and meaning features (or perhaps meaning features alone, if we assume that the original mutation did not involve any linkage to production and perception [‘sound’ or ‘sign’] systems) and combining them to yield a potential discrete infinity of sentences (or at the very least, structured sets of semantic features) each of which can be understood.

  There is risk in speaking of a biological engineer solving a design problem; it can invite creationist claims. No claims of that sort are supported by Chomsky's proposals. In essence, he holds that the core computational system of language, found in Merge in its available forms, is the result of a biologically transmissible mutation in an individual somewhere between fifty thousand and a hundred thousand years ago, a mutation that proved advantageous to the mutant and his or her progeny, to the extent that the progeny soon (in evolutionary time) dominated. For Merge provided the means to engage in all-important complex thought and, where/when linked to sound/sign, it offered means to communicate, plan, and organize to carry out non-kin projects.

  Page 18, Galilean science and simplicity

  For a good discussion of how Chomsky employs idealization in his work in linguistics, see Norbert Hornstein's new introduction to the second edition of Chomsky's Rules and Representations (1980/2005).

  Page 19, On simplification, idealization, and explanation

  Chomsky's points concerning the need for simplifying and idealizing are central to naturalistic scientific research of the sort that postulates ‘hidden’ entities in order to (among other things) explain observed phenomena. This topic comes up several times, including the pages where our discussion turns to Chomsky's relationship to a mentor, Nelson Goodman. Goodman emphasized the tight connection between science and simplicity, and Chomsky's efforts to respect idealization and simplification are hallmarks of his approach. Chomsky's view of natural science is discussed in Appendix IV.

  Chapter 2

  Page 20, On biology as more than selectional evolution

  Kauffman, D’Arcy Thompson, and Turing (in his work on morphogenesis) all emphasize that there is a lot more to evolution and development than can be explained by Darwinian (or neo-Darwinian) selection. (In fact, Darwin himself acknowledged as much, although this is often forgotten.) Each uses mathematics in studying biological systems in different ways. Some of Kauffman's more surprising suggestions concern self-organizing systems and the use of statistical modeling in trying to get a grip on how timing of gene protein expression can influence cell specialization during growth.

  Page 22, On Plato's Problem and its explanation

  The term “I-language” is explained – along with “I-belief” and “I-concept” – in Appendix I. For discussion, see Chomsky (1986) and (2000).

  “Plato's Problem” labels an issue that any linguist constructing a science of language must speak to: saying (by offering a theory that constrains language growth and thus explains the relevant poverty of the stimulus phenomena) how any child given minimal input can acquire a natural language (or several) quickly, going through approximately the same developmental stages as other children acquiring a language, and without apparent training or “negative evidence.” It is called “Plato's Problem” because it is a bit like that faced by Plato/Socrates in Plato's dialogue Meno: a slave boy without training and given only prompting (not being told what the answers are) manages to come up in short order with the basic principles of the Pythagorean Theorem. He does not state the formula, of course; he does not have the tools. However, with what he has, he can go on to produce correct solutions to the length of the hypotenuse of various right triangles. So too the child with language: the child gets no training and is exposed to a limited data set, but has no difficulty displaying adult linguistic competence by around 4. The child cannot state the principles by which he or she speaks either, of course; that would require the child to have a science of language available. But lack of an articulate way to say what he or she knows by no means gives a reason to hold that therefore language must be a kind of know-how, gained by intensive training and familiarization.

  Plato's ‘explanation’ of the slave boy's capacity consisted in appealing to a myth of recollection, one in which the ‘soul’ had received knowledge before birth, and this knowledge came to be recollected as a result of experiential input of the relevant sort. Chomsky's explanation is, of course, a quite different naturalistic one that appeals to efforts to understand how automatic and ‘channeled’ development proceeds.

  Page 24, On recursion and its forms

  Marcus Tomalin correctly points out (2007) that there are various formal ways (lambda calculus, computability, general recursion, Peano-Dedekind [arithmetical] induction . . .) to define recursion, and it is important to say precisely what kind, for not all functions are definable in each. He argues that the preferable one is induction, for it yields what is needed for what is claimed in Chomsky's work, it is the least tied to a specific formalism, and it emphasizes a connection between linguistic recursion and mathematics. So he recommends speaking of (mathematical) induction, not recursion. While his point would be a reasonable one if the issue were that of defining recursion, that is not the case. The issue when dealing with Merge and its introduction to the human species focuses on an extremely simple subset of recursive functions that appear in the inductive and all other approaches to offering a general definition of recursive functions. One would need empirical evidence that indicates that more complex approaches to recursion such as those Tomalin considers are relevant.1

  Keep in mind that the issue of how to define recursion of the sort needed in the study of language (and perhaps ‘native’ parts of mathematics) is a matter for empirical research, not convenience or stipulation.

  Page 24, On human concepts

  There are several central points here. One is that our concepts (RIVER, HOUSE, PERSON . . .) do not seem to have – as Locke might have put it – “resemblances” in the natures of things in the world. They seem to be constructed – not by persons, but by their minds. Locke, an empiricist, had little of value to contribute to saying how the construction took place. A better answer, offered by a rationalist working at the same time, came from Ralph Cudworth, who said that the concepts we have available are provided by an “innate cognoscitive power” that takes ‘occasions’ of input to put together concepts that happen to serve our cognitive interests. While he offered no specific proposal for a mechanism, he did at least recognize that concept-acquisition is virtually instantaneous and cannot be explained in the empiricist manner, so that it demands a devoted internal “power.” In effect, our minds yield concepts, and they in turn can shape the ways we experience and understand, without being shaped by them.

  Another is that our concepts do not seem to be tied by some kind of one-to-one correspondence to events or states of the world outside, nor to inner states. Gallistel calls these correspondences “functioning homomorphisms,” which are partial ‘samenesses’ in abstract mathematical form; environmental maps employed in insect navigation are examples. See Gallistel (1990) and (2008).

  A third is that they (along with a capacity to produce sounds/signs) might well have been – and probably were – in place before the introduction of language, at least where language is conceived of as a capacity to take associations of sound and concepts (‘words’ or lexical items) and, starting with a finite cluster of these, combine/concatenate them with others to yield, in principle, an unlimited number of those forms that we call
“sentences,” and that Chomsky calls “expressions” consisting of paired sound and meaning instruction sets.

  A warning is in order for those who have read Hauser, Chomsky, and Fitch's “The Faculty of Language” in Science (2002) and who might have developed the impression from it that Chomsky entirely endorses the idea that human conceptual resources are quite like those found in other organisms – probably primates, in particular. Apparently, the difference was discussed in earlier versions of the paper, but a lot was cut out by the editor, including the discussion of concepts and almost everything about linguistics. In comments on the MS for this volume, Chomsky remarks: “About all that remained [on concepts in the article] was ‘Even for the simplest words, there is typically no straightforward word–thing relationship, if “thing” is to be understood in mind-independent terms.’”