The Science of Language

by Noam Chomsky

  As for the soundness of selectional adaptationist claims, several have pointed out that evidence is too often question-begging (just-so stories), or does not exist and cannot be found, making an adaptationist claim empty. Focusing on cognitive systems, Richard Lewontin (1998) raised serious doubts about the evidence claimed for selection of several cognitive capacities, including some that seem to be unique to humans, such as language. At the end of the article, he remarks in response to the volume's editors’ suggestion that he soften his criticism, “We should reserve the term ‘hypothesis’ for assertions that can be tested.” And at the very end of his remarks, he emphasizes his main point, that history provides virtually no evidence for change and adaptation in cognitive capacities. One of Lewontin's primary targets was Pinker and Bloom's view that selection alone can explain the introduction of a “complex system” such as language. His criticism of Pinker and Bloom resurfaces in modified and more precise form in Hauser, Chomsky and Fitch (henceforth HCF) (2002). Some of the main points of the HCF article are raised elsewhere in this volume. HCF point out that while we might find at least some evidence for selectional adaptation of aspects of language that are outside the core in what HCF call “FLB” (for “faculty of language, broad”) as opposed to FLN (“faculty of language, narrow”), claims about the adaptive evolution of the core computational system of language FLN with its recursive character and capacity to link sounds and meanings over an infinite range have nothing to back them up. For discussion, see HCF (2002) and critical response from Pinker and Jackendoff (2005) followed in the same volume by Fitch, Hauser, and Chomsky's (2005) and Jackendoff and Pinker's further effort (2005). On the relevance of the Minimalist Program to this discussion, see the online Chomsky, Hauser, and Fitch (2004), available on Hauser's website. See also the discussion in the main text, and especially Chomsky's remarks on perfection and design in our 2009 discussion (pp. 50ff.).

  Discussions of function, adaptation, and evolution are complicated by the fact that it is not clear what ought to figure in evolutionary theory. Is Turing's work on morphogenesis (which he thought is best located in chemistry or physics) in or out? What about Gould and Lewontin's spandrels? Should one admit Chomsky's saltational view (explained below) of how language came to be introduced into “evolutionary theory”? These decisions can be left to those working in the field, as can the explanatory status of natural selection. Nevertheless, there is good reason to believe that the claim that adaptationist accounts can explain everything in the way of biological structure, development, evolution, and speciation is at an end – supposing, of course, that responsible scientists actually believed that they did in the first place.

  While we can leave the issue of the precise explanatory status of selection, adaptation, and adaptive function to biologists, there is a simplistic view of selection and evolution that needs no expertise to reject out of hand. Perhaps in an effort to popularize evolutionary theory and/or present to lay audiences a readily-understood alternative to a theological explanation of why biological species are the way they are, some have exploited a connection between evolution and commonsense understanding's apparently default view of learning, a view found in behaviorism's and empiricism's view that we learn by coming to respond to input in the ‘right’ way by being trained to form the ‘right’ dispositions to respond. We shape our responses to the environment by having the ‘right’ responses rewarded, and the ‘wrong’ ones punished. Relying on this picture, many have treated evolution as an historicized version of learning so conceived: we and other creatures are the ways we are because each species has adapted in its structure and development to have an optimal strategy for its ecological niche; Skinner himself supposed this, revealing how little he understood of what evolution involves. Exploiting a connection to this view of learning is misguided strategy; one ends up defending something for which there is no warrant except that it appears to be all-encompassing. There is no modesty to this view – no recognition that selection's role is limited, if even that. Nothing is said about how evolution and phenotype development and growth must take place within the constraints set by physics, chemistry, biology, perhaps some form of information theory. There is no mention of the fact that many genes are conserved over species and clades. Epigenetic factors are ignored. ‘Happy accidents’ of the sort found in what Lewontin and Gould called “spandrels” are not mentioned. Too often – especially in the cognitive domain – there is only a minimal effort to find evidence for claims: just-so stories are common. Virtually all features within a species and their structures and behaviors are treated as though they were ‘selected’ over a long period, chosen by virtue of adapting to the environment in which a species is found. While there are gestures toward a role for the genome, nothing is said about its precise structure and how it works and one finds no mention of epigenetic factors. The genome's role on this naïve view of evolution – one too often popularized by those who should know better – is just to transmit from generation to generation successful ways for a species and its members to ‘solve problems’ in dealing with an environment. There is little or no disentangling of the notion of the interests and actions of agents from the ‘interests’ and ‘actions’ of genes. Compounding the error, genes are often presented as having some direct control over an organism's behavior, rather than providing an organism with various systems that it can use to deal with its environment, and – certainly in the case of humans – do a great deal more too. Like ‘learning’ for the behaviorist, the naïve but very popular view of ‘selection’ comes to be treated as an explanation of virtually every trait of an organism. And this kind of naïve simplification comes to assume the trappings of a metaphysical principle. Rather than God, one finds appeals to evolution (so conceived) alone as the explanation of why we are the way we are and act in the ways we do. Confusion and ideology are served, not clarity and reason – nor science.

  Rejecting this notion of selection is not, of course, rejecting the explanatory power and scope of evolution, properly understood. But as indicated, it is not exactly clear what is to be included in the theory of evolution.

  II.4 Biology: function-of-an-organ

  With that caveat in place, and leaving aside the to-be-determined final status of the explanatory reach of adaptation and adaptive function in evolutionary theory, it is time to acknowledge that biology does seem to provide a place for functional explanation that several centuries of scientific research have convincingly excluded from most forms of physical science. By way of background, Ernst Mayr (2004) argues against reducing biological explanation to physical because, he suggests, biological theories are committed to speaking of function where physical theories are not. It is not clear what Mayr has in mind by the physical, nor – in a related vein – by biology offering an ‘autonomous’ mode of explanation. Granted, biology is not physics, but it – like linguistics, if Chomsky is right – is a naturalistic and thus in an important sense physical science. Ignoring, however, the issue of what is or is not physical, and focusing on the most plausible kinds of cases where functional explanation makes good contributions, let us look at a “system-within-an-organism” or “organ-within-an-organism” notion of function. It may not be quite what Mayr had in mind, but it looks reasonably close to it.

  An “organ-within-an-organism” notion of function is interesting for our purposes for two reasons. One is that it allows for clearly stated and falsifiable hypotheses, unlike far too many claims made in dealing with issues of function in biology. And second, on the face of it, it suits Chomsky's conception of a computational theory of language, where language is thought of as an ‘organ’ – which is how Chomsky construes the language faculty – or at least, how he conceives its computational core and its specified ‘contributions’ to other mental systems. ‘Doing a specific job’ (fulfilling a function within the organism by interacting with specific other systems in specific ways over interfaces) seems to be built into the idea that a computational theory of language is a
formal science of a system that “interfaces” with other systems by providing them information that they can use, and apparently even doing so in an optimal way, so that the design of the language faculty can be conceived as perfect – or at least, a lot closer to perfection than was thought not long ago. Language – using Chomsky's informal term – “instructs” production/perception and conceptual/intentional systems and, in order to do so, must provide them with ‘instructions’ (relevant forms of information) that they can ‘use’ or put to use – that is, come into operation in the way(s) characteristic of the relevant system(s). Note that ‘use’ here is not “use by creature” or “use by organism as a whole,” and especially not “use by agent.” It is “employment by other (designated) systems in specific ways,” which is what places this notion of function in the ranks of empirical investigation.

  Plausibly, in fact, we could entirely abandon the term “function” in dealing with specific system-internal system transactions and thereby avoid the obscurity that this term seems to invite. That is probably the wisest strategy – but not one, apparently, that people are fully willing to adopt. It is not clear why, although I suspect one factor is that its use in what appears to be a reasonably well understood context attracts attention. A possible parallel is found in the following case, and numerous others. A recent article in the Proceedings of the National Academy of Sciences uses words such as choose and move in speaking of the entirely chemical responses to environmental food sources for slime molds. The cost is confusion, but the reward is attention to an ‘amazing’ discovery.

  So ignoring the matter of whether biology is “autonomous” in the way Mayr seems to think, and focusing on reasonably well-specified cases in sciences that deal with specific ‘organs’ (or at least their ‘cores’), and ignoring too the advice of the paragraph above, it might make no sense to speak of a biological ‘organ’ of this sort without introducing the relevant notion of a function – of what the organ ‘does’ for and with the systems with which it interacts. And if Chomsky accepts such a notion of function and – as indicated – this notion of function is not thought of in terms of adaptations to an environment, nor in terms of what an organism ‘does’ with what a system provides, one can make a plausible case for the language system serving a biological function in something at least reasonably close to what Mayr may have had in mind.

  It is not altogether obvious, however, that Chomsky does accept this as an autonomous notion of biological function. Some reasons for doubt appear in the discussion of the skeleton example in the main text (see also Lewontin 2001); not only are there many plausible claims about functions for the skeleton (which is also an ‘organ’ of sorts in the body interacting with immune, circulatory, motor . . . systems), but some descriptions of its functions look very informal (giving structure to the body), with little prospect for making them precise within a naturalistic theory. Others appear in our 2009 discussions of the design and perfection of the language faculty on pp. 50–55. Likely the most significant of all appears two paragraphs above: for the relevant scientific purposes in question, why not abandon the term “function” entirely and instead speak in a well-understood way of an organ's ‘interfaces’? Ignoring all this, however, there is a case to the effect that Chomsky does acknowledge some notion of inter-organic function (cf. Epstein 2007).

  Footnotes

  1 Some (e.g. Carruthers [2006]) have tried to make the language system itself a – or even the – central coordinating system, the device that gives humans cognitive flexibility by taking contributions from various systems and integrating them. That move makes some sense because of Elizabeth Spelke's work (2003, 2004). And I do not doubt (McGilvray 2005b) that language contributes to human cognitive flexibility. I confess, though, that I find Carruthers's very complex claims difficult to take very seriously. For one thing, like some psychologists and others (including some of his critics), where Carruthers believes some mental ‘job’ needs to be done (with jobs specified, apparently, by assuming that humans are agents with minds that serve their actions, for that seems to be what he has in mind when he treats humans as “complex functional systems”), he introduces a module to perform it. The result is a boxological picture of the mind, with little or nothing offered concerning what is in the boxes. I admire his courage in tackling the issues he wants to deal with – describing and in some sense explaining how a ‘massively modular’ mind manages to integrate its components and produce coordinated actions, with special focus on the role of language. However, for reasons that go back to the creative aspect of language use, I doubt that these can be addressed in a scientific way, pace Fodor and his “representational theory of mind” and similar efforts on the parts of others. Commonsense folk psychology and belief-desire stories deal with the matter in some sense, no doubt, but they do not provide a science of mind and action; they depend on commonsense concepts and on conceiving of humans as essentially practical problem-solvers.

  2 Some – such as Fodor – believe that they can salvage a representationalist theory of the mind by introducing a view of reference/denotation claimed to depend on natural laws. For discussion of such efforts, see the main text and references in it and my (2010).

  3 One can to a degree ‘determine’ a referent with respect to a context by including in an account of ‘what is said’ a set of indices that specify time of speech, person speaking, context of speech, purpose of speech, and so on. Or rather, one can hope that something like this would succeed, for it surely cannot: the indices would have to be endless in number to cover all possible cases, and there would be no discourse-independent means of fixing even a single set of assignments, assuming – counterfactually – that that would suffice. Nothing like this appearance of progress can aid the efforts of those who want to construct an honest theory of language that has some degree of objectivity. For in the final analysis, one would have to have an index for “object of current concern” in a specific discourse context (or the like), and fixing that would require being in the circumstances in question employing whatever cognitive resources appear to be required and making then a reasonable guess about what someone intends. There are no ‘boundary conditions’ on what is or could be involved in coming to such a ‘situational’ judgment. In his “A Nice Derangement of Epitaphs,” Donald Davidson concluded from this that there can be no theory of language. What he should have concluded is that there can be no theory of language as use – basically, Wittgenstein's point, made some forty years earlier.

  4 Reproductive capacity may be the most generally accepted view of what ‘drives’ selection, but it would be very difficult to exclude other options.

  Appendix III: On what is distinctive about human nature (and how to deal with the distinction)

  There are at least two important points made here in the main text. One is remarked on elsewhere: that the evolution of language into its modern form could well have consisted entirely in a single mutation in a single individual – a mutation that allowed a human to construct complex thoughts.1 The key is the introduction of a single operation, Merge, which gives a creature that has this operation the capacity to engage recursion to an in-principle unbounded extent. There is no need, as there is with the Pinker-Bloom story about language and its evolution, and other appeals to the selectional advantages of increased capacity to communicate, to suppose that language developed slowly over many millennia.2 A single step would suffice, assuming that Merge provides the means to join concepts together to make complexes and to move them, and that conceptual (thought) systems and – when and where externalization begins to figure, which need not be at the start – articulatory and perceptual systems were in place. There is more to language and the conditions under which it can be acquired and operate than that, and more needs to be done to make sense of how language could be innate yet take so many forms. For more on these matters, see the discussion below of parameters and of what Chomsky calls the “third factor” in language development at both the species and i
ndividual levels. However, the basic point seems to remain: the hypothesis that Merge alone is sufficient for the recursive feature of language remains plausible, and its saltational introduction makes sense of how language came to be introduced into the species. (For a dissenting view, see Pinker & Jackendoff 2005 and for a response to them, Fitch, Hauser & Chomsky 2005.)

  The other point is found in Chomsky's remark that our concepts and thought systems do not operate the way animals’ do. Part of the reason is that the concepts are just different; I return to that below. But another aspect of the matter recalls the point made above about stimulus freedom and other features of the creative use of language. Descartes in his 1637 Discourse, and a large number of others following him, were struck by the difference in ways in which animals use their conceptual/cognitive tools and the ways in which humans, able to combine concepts in complex forms of ‘perspective,’ employ theirs. (Arguably, it is this observation, no doubt only obscurely understood under rubrics such as “knowledge” or “reason,” that lie at the root of the myth of special creation for humans: we have language and flexible ways to combine concepts, other creatures do not.) At least two factors play a role in bringing about this difference. One is that our concepts are just different; there is more on that below. The other has to do with what a language faculty provides. It can in its use operate – so far as we can determine – autonomously, thereby supporting our ability to speculate and wonder about anything whatsoever, without regard to circumstances, external or internal. Stimulus freedom, already noted, is one factor in this. Another is “unboundedness”: by means of the operations of the language faculty, arbitrarily chosen concepts can be joined with others to form an endless number of complexes (‘expressions’). Elizabeth Spelke (2003) – among others – makes much of the combinatory capacity of language and what it affords humans alone; she could add: combinatory to an unbounded degree. Chomsky (2000) puts it this way: with language, we can produce novel cognitive “perspectives” that allow us to conceive in ways that are clearly unavailable to creatures that lack recursion. Still another factor – almost certainly related – is that human actions seem to be free. Perhaps that is an illusion, but it is not likely one we can overcome, or for which there is any evidence that we should overcome. Given the number of systems in our heads and the fact that the production of actions requires the coordination of at least several contributing mental, motor, and input systems (with different ones at different times) operating cooperatively over an interval, analyzing the causes of an action would be like trying to solve a massive multi-body problem where there are few if any constraints on what contributes when, and how. As extensive discussion of the n-body problem indicates, that is very likely beyond what we humans are capable of managing. So we might as well acknowledge now that we, with our scientific tools, are very unlikely to be able to produce a deterministic account of the much more complex phenomena of linguistic behavior, especially one that can be generalized to all individuals. However, instead of bemoaning the fact, we should recognize that – as Descartes emphasized – we have the evidence of experience that we feel free. Lacking also any evidence of science against ‘free will,’ and acknowledging the evidence of experience for it, perhaps we should celebrate our freedom instead.