The Science of Language

by Noam Chomsky

  The history of efforts to deal with the distinctiveness of human cognitive capacities is instructive. While few of them were as sharp in their observations as Descartes, many in both empiricist and rationalist schools followed him in the seventeenth and eighteenth centuries by attributing the differences seen in human cognitive powers to reason. Animals, it was thought, operate not by reason but by instinct. While there was agreement on this, however, the schools diverged in how they thought humans come by reason: rationalists assumed it was an innately endowed capacity, heavily dependent on the operation of several innate faculties. Descartes famously placed reason – the mind, or the mental – in a separate substance. Empiricists suggested that this supposedly distinctive feature of humans comes through exposure to the environment and – certainly by the time of Herder and the majority of the Romantics, and arguably before – to a large extent it comes by learning language and social practices from the community. The claim becomes in effect, then, that human invention, history, and culture make the difference between us and other creatures – that, and the supposed fact that must accompany this kind of explanation, that to a large extent the human mind is empty at birth, and that it has available a large area where some kind of generalized learning procedure operates, guided by training and experience, and shaped by learned habits and rules that are assumed to link some kind of input to some kind of output.

  Chomsky's work advances the rationalist cause considerably by making reason into not just a consequence rather than an apparently independent explanatory principle, but a consequence, largely, of treating language – its growth/development and its internal operations – as an ‘animal instinct’ introduced by mutation into the human species. (The use of language, however, remains well within the domain of freedom.) Unlike Descartes, he accepts that our minds and the combinatory mechanism of language especially are apt objects for natural scientific research.3 Moreover, unlike the empiricists, he maintains that what makes us human is not society, culture, and the training of a plastic mind, but the introduction of a special kind of instinct to our species. As an organ of the human body, language develops automatically and operates internally according to innate principles. And most of our commonsense concepts, at least, seem to be innate too – hence, the result of some kind of internal system or systems. The same can be said for the kinds of linguistic sounds that we can produce. However, by providing a way to put arbitrary concepts together in complex structures at arbitrary times in arbitrary circumstances, language surely provides humans with the essential tools for speculation, explanation, inference, and the like – certainly within the commonsense domain, at least. Making a contribution no doubt too to our capacity to create scientific theories – at the very least, assuming recursion came via Merge – it appears to yield the discrete infinities of natural numbers, so it has a role to play in the development of science. Chomsky remarks on that above and below in the text. In effect, then, the claim in its simplest form is that the introduction of recursion through the mutation that introduced Merge leads not just to the conceptually necessary operations of language (putting elements together, and moving them), but to what Jared Diamond called “the great leap forward,” the introduction to the species of the distinctive features of human cognitive capacity. That is not quite enough, as suggested above: human-unique concepts play a role too. But their distinctive nature may be the result of their being – at least in some measure – due to Merge or something like it too. The topic is taken up in the text and Appendix V.

  The empiricist explanation has not changed substantially, nor has it advanced appreciably since its beginnings – ignoring in this regard contemporary redefinitions of empiricism as efforts to seek the best explanations. No doubt connectionism and the like provide pictures of what the area of the mind might look like, if something like the empiricist picture were correct. In this way, it is an advance over Locke's “blank slate” (assuming he actually believed that tale, which is less than obvious). But the rest of the story – and especially the dependence on some kind of training or learning – is unchanged. We differ from animals by having weak instincts, or perhaps lacking instincts altogether with regard to our higher cognitive operations. For these to become available, we need recurring experiences, acculturation, and (typically) training in order to bring shape to and constitute these operations and the conceptual materials on which they operate.4 We must have large areas of our minds that allow for this. And reason continues to be celebrated as what makes us distinctive. Consider, for example, the work of Wilfrid Sellars and his followers – a group that includes several contemporary philosophical stars. In Sellars's essay “Philosophy and the Scientific Image of Man” in (1963a), he presents a sweeping picture – not unlike Hegel's – of how humans come to be what they are now, with advanced cultures, sciences, and institutions. In this picture, humans are portrayed as gradually being weaned from a framework in which they placed themselves at the center of the universe and had no inkling of science to our current epistemically advanced states and our sciences by continuous refinement in their ability to reason. Their ability to reason comes to be modified through greater and greater sophistication of common sense (Sellars's “original” and “manifest” images of world and humans in them), which eventually comes to be scientific understanding in its modern form. Improvement and greater sophistication are throughout treated in terms of ways of offering better and better ways to describe and explain the world and humans in it – that is, as improvements in our abilities to reason, provided by better and better ‘theories.’ And concepts are characterized by their contributions to reasoning – by their roles in reasoning about the world and ourselves. Concepts – like language itself – are treated as normatively governed, and the largely epistemic norms that do the governing are seen as inferential rules that Sellars calls “practices.” These rules are in turn portrayed as the rules of reason and language, conceiving of language in the way that Wittgenstein did as a game or set of games we ‘play.’ Crucially, unlike rationalist accounts of primitive cognitive systems, including for Chomsky language, we must learn to reason. To learn how to reason is largely a matter of learning how to infer (or play the game); and learning that is learning how to speak in accord with the rest of the specific community in which we find ourselves. We learn how to speak by being trained by our communities; our communities, in fact, are repositories of, and in a way constitute, the standards of correct reasoning. Communities train their children to produce the right words in the right circumstances. Once a child sufficiently meets community standards, he or she “knows a language” (a form of know-how), or masters it, and can teach it to others, because he or she has the relevant discriminative capacity to recognize divergence and conformity. The view of learning is basically behavioristic; indeed, Sellars acknowledged his behaviorism, and in fact celebrated a slightly sophisticated form of it as the proper and only science of mind. The slightly sophisticated form of it appears in essays in which Sellars treats the brain as a neural net that is modified by experience and training to emulate the inferential connections he thought constituted the rules of languages. These essays offer an early form of a connectionist model of the brain, one according to which the brain's pathways leading from sensory input to behavioral output are modified as training proceeds to produce the right outputs, given specific inputs. To have a language is to have a brain that yields what a community takes to be appropriate behavior (epistemically correct, etc.), given circumstances. And to have a concept is to have a node in the brain that yields the right outputs, given a specific input. It is remarkable that this Sellarsian picture of language, mind, brain, and reason continues to dominate philosophical (and psychological, etc.) study of language and mind. In philosophy, it is found in both the ‘analytic’ school and (although with less emphasis on learning, science, and the brain), the ‘continental.’ Both are empiricist in their assumptions, and differ largely in style and emphasis.

  Chomsky's rationalist altern
ative to this empiricist account treats reason – exemplified in problem-solving in both science and common sense – as (at least for common sense) heavily dependent not on training and acculturation, but on our having the innate instincts we do. This very different perspective on the matter is found in earlier rationalists Chomsky discussed in his Cartesian Linguistics, such as Herbert of Cherbury, one of the Cambridge Platonists of the seventeenth century. Herbert noted that we have to have (innate) “common notions” in place in order to come to reason at all – that is, to describe and explain at all. These common notions are, essentially, the commonsense concepts typically expressed in our natural language use. Chomsky adds a further, and crucial, instinct or innate contribution to Cherbury's picture, a language organ and its combinatory powers. Given this, reason that is divorced from circumstances – and flexible cultures, human institutions, and individual styles, etc. – becomes possible.

  So it turns out that the distinctive difference between humans and other primates lies largely in the fact that we have an instinct they do not: language. Introducing language allows reason to develop. We are also invited to rethink our view of animals, of course: they need not be conceived as deterministic natural machines, as Descartes thought – although incorrectly, because as Newton showed, there are no such things. No doubt many animals also have concepts, minds, freedom, intention, etc. However, they don't have what we have, language and recursion.

  That said, keep in mind that it is quite possible, even likely, that it is not just the fact that we have language that accounts for the differences in human cognitive capacities (in “reason”), but that our concepts are just different from animals’. There is more on this in Appendix V. First, though, Chomsky clarifies what Merge amounts to, how it operates, and what it gives the human species. He begins with an account of the relationship between language and another cognitive benefit of recursion apparently unique to the human species – mathematics, especially the natural numbers. And he offers a very interesting explanation of one respect in which natural languages differ from another form in which humans employ ‘symbols.’ We produce and use invented, formal systems such as those found in advanced forms of mathematics and in the natural sciences. No other creature does, of course. And no doubt our capacity to invent these symbol systems depends in part on our having language. These formal systems differ from natural language not only in the fact that they are invented, that they are artifacts, where our languages and our concepts (and linguistic sounds) are not. They differ also in that at least some of them depend heavily on natural languages, at least with regard to making them learnable. Arithmetic is plausibly an exception, as hinted in the discussion in the main text. For arithmetic is as suggested there a product of internal Merge operating over a lexicon with a single element, and it is a very impoverished natural language. I thank Chomsky for pointing this out to me.

  1 In light of some matters discussed later and in the main text at pp. 26–28, I should emphasize that this story presupposes that human concepts were in place before this. As to their origin – assuming that they are in large measure unique to humans – that is something that is likely to remain a mystery. As Lewontin (1998) reminds us, there is little (that is, nothing) that one can offer in favor of or against any specific hypothesis.

  2 In fact, if it turns out (for discussion, see pp. 11–15) that communication is at best something that language allows for, rather than being central to it, their argument is completely irrelevant.

  3 Chomsky suggested to me that it is interesting to speculate about what Descartes would have agreed to, had he had available and held a competence/performance distinction. If he had had this available, he might have agreed that the computational operations of language are inherent in what he would have called “body.” In that regard, look at what he says about vision; I speculate on the matter in a section called “Descartes's Contribution” in my introduction to the third edition of Chomsky's Cartesian Linguistics. Keep in mind, though, that when Newton came along, Descartes's notion of body as a scientific concept had to be rejected.

  4 Hume, who often appealed to instinct (while insisting that its operations were and would remain obscure), is a partial exception. Like Descartes again, it is interesting to speculate about what he would have maintained, if he had had available a competence/performance distinction. His view of himself as a scientist of human nature makes the speculation even more interesting.

  Appendix IV: Chomsky on natural science

  Chomsky's discussions of Merge illustrate the fruitfulness of idealizing and simplifying in order to construct a successful science. The point is discussed further in several places in our interchange and very usefully in Norbert Hornstein's new preface to Chomsky's Rules and Representations (1980/2005), so I will not pursue it here. Instead, I want to make some further remarks about Chomsky's view of natural science as I understand it.

  As suggested before, although humans constructing sciences get some help from whatever innate resources lead to our capacity to engage in what Peirce called “abduction” (see above and below), the sciences themselves – the explicit formal symbol systems that constitute theories – are largely artifacts. They are products of human ingenuity and effort, typically of people working together. And they advance, often, over centuries, leaving still many unanswered questions. They are artifacts put together to do a job because, in effect, science is a project. It is an attempt by humans to construct theories of various domains. In the case of the natural sciences, that job is reasonably well understood, and there is agreement, usually implicit, on the goals. There are, of course, differences between sciences in subject matters (in the ‘entities’ the sciences investigate), in specific research techniques and experimental devices, and – of course – in the laws and principles of the theories. Yet natural scientists generally aim toward a uniform goal. Their practices reflect what Chomsky calls “methodological monism” (2000). Abstracting from differences in experimental techniques, etc., there is sufficient uniformity in the goals of natural scientists, no matter what the science, that the term “monism” (implying a single approach to a domain) is called for. The goal – the “goal of science,” the project the scientist tries to carry out – is to produce a theory of a domain that offers descriptive and explanatory adequacy, that provides formalized (explicit, mathematical) statement, that is simple (in some rather hard to define sense), that aims towards objectivity, and that allows for accommodation to other sciences. Progress – and progress is necessary, for it is the sign of success – is measured by improvements in one or more of these desiderata.

  The best theory at a time by these measures offers what counts as a true theory, and it can tentatively be assumed that the subject matter that the theory focuses on is correctly so described and explained. From this perspective, it is plausible to speak of the project of science as an effort to “seek the truth” about the natural world. Thinking of it as a project constrained by the tools that demonstrably yield success suggests that attempts to construct theories of domains where one or more of the desiderata cannot be satisfied – for example, attempts to construct theories of human action – or where one encounters continued lack of progress, should be abandoned. Failure in certain domains – perhaps those that are too complex, among others – should be no surprise, Chomsky notes. We are biophysical creatures, and there is no reason to expect that our cognitive powers are anything but limited, just as are those of other creatures. As emphasized in Chomsky (1988), that fact is of course to our advantage, for without limitation, there would be no growth, no knowledge . . .

  Given the points about theories and truth, and given the shape of the theories that we have managed to construct, perhaps we can speculate to an extent about what the world ‘in itself’ is like. For one thing, with the otherwise-explainable exception of biological organisms that develop over time, the world (as we can understand it) seems to have entities and systems that remain quite stable, and to have (within limits) predic
table states and consequences. Perhaps this stability depends on maintaining stable structures. For another, these states seem to stand in what for our mathematically and formally endowed minds appear to be simple relations to one another. The claims made by these speculations may, of course, be nothing more than artifacts of our theories. But the surprising – almost miraculous – success of our formal theory-construction techniques does suggest that our theories track how things are. Remarkably, moreover, the theory-construction goals that have yielded improving sciences in the ‘physical’ domain seem to be equally successful in the ‘mental’ domain. Examples include computational theories of language and vision. This point is emphasized again in later discussion.