Cartesian Linguistics

by Noam Chomsky

  4 Keep in mind here and throughout that the labels ‘rationalist’ and ‘empiricist’ are labels for specific views of the mind and of the best way to study the mind scientifically. Rationalists when they study the mind assume nativism and adopt internalism; empiricists are anti-nativist and assume that the study of mind cannot proceed apart from study of how the environment in which a mind is found ‘shapes’ and gives specific content to the mind.

  Historical and contemporary candidates for these labels can sometimes be difficult to place in one camp as opposed to another. While Hume, for example (usually classed as an empiricist), held that we get our higher-level concepts and language through a process of learning that amounts to some form of associative grouping stabilized over time by repetition of similar experiences or “impressions” (clearly a general-purpose learning mechanism), he also held that at least some of the operations of our mind seem to be performed automatically and in a remarkably rule-like manner by “secret springs and principles.” He noted, for example, that humans seem to be able to comprehend any novel action and still make what look to be rule-governed judgments about its permissibility. One way of reading the claim about secret springs and principles (and his refusal to try to say what they are) is to construe him as at heart a nativist (at least in certain domains), but one whose skepticism prevents him from undertaking a research project to figure out just what those internal springs and principles are. Jerry Fodor, on the other hand, is a self-declared rationalist, and is indeed like other rationalists a nativist. But he adopts a non-internalist view of mind and its study, as one would expect of someone who wants to think of the study of mind as the study of a representational (denoting, referring) system. Granted, he does not rely on a generalized learning procedure; but his RR credentials – so far as my labeling conventions go – are forfeit. Chomsky, though, is clearly in the RR camp: he is both nativist and rationalist, and – with language and its use – an opponent of the “representational theory of mind” that Fodor says is “the only game in town” if you want to be a cognitive scientist. Chomsky is also an even more vociferous opponent of empiricist “dogma” – of their anti-nativist and externalist view of the mind and their research strategy. Behaviorists – among them, many philosophers still (although they might shun the label) – are clearly in the empiricist camp. So too are many of the psychologists (and philosophers) who call themselves “connectionists.” Connectionism as it is practiced is primarily devoted to pressing the idea that with language and concepts (the complex ones, presumably) the mind gains its language-producing shape and its conceptual contents by the operations of a generalized learning procedure. Their claim that the mind is made up of “neural nets” is innocuous; it is their claim about the initial state of the net (undifferentiated, approximating Locke’s “blank slate”) and their view about how this net gets its “content” (by training, learning) that place them firmly in the empiricist camp. It should surprise no one that a lot of connectionist work is directed to trying to show that some rule or another that is part of a RR theory might be “learned” by a neural net shaped by a generalized learning procedure involving “training.”

  5 People (using what their minds provide) also put together clusters of sentences called “stories” or “tales” or “descriptions”, etc. – depending on the job the cluster is supposed to perform. These are not under consideration here. Stories clearly are not innate. Nor does one learn them. One has to have a language and construct sentences (complexes of concepts of sentential form) to be able to construct stories. The basic issue is how one acquires concepts and the combinatory principles involved in putting sentences together.

  6 The RR theorist has no qualms about naturalistically determinate causal relations in the world-head direction. Relations of this sort figure in an account of acquisition, not use.

  7 Chomsky emphasized this quotation from Strawson (1950: 336) in a talk outlining and defending an internalist approach to semantics (meaning) at Harvard on October 30, 2007.

  8 One can, of course, introduce a technical notion, proper noun, and stipulate that a proper noun denotes a single entity. Probably you will have to introduce a technical notion entity too. But you should not expect to convince an RR theorist aiming to construct a naturalistic theory of language and its meanings that s/he should take your technical terms seriously.

  9 Keep in mind that talks and papers by philosophers and others provide contexts to a large extent under the control of the author or speaker and, unsurprisingly, those writing or speaking tend to construct contexts that suit what they want to defend.

  10 Von Humboldt (discussed earlier in CL and again at its end) makes the same points – as did Descartes, Cudworth, and others in the RR tradition.

  11 Perhaps there is “hi?” But there is also “hey there,” “hello,” “greetings,” “welcome,” not to mention a current favorite among some, “dude.” And none of these or any of the other possibilities for greeting terms – or the few other cases where one can find some degree of regularity in use – will sustain anything like Lewis’s and Sellars’s claims about conventions and practices, not to mention their semantic efforts.

  12 Animal communication often displays something like relatively fixed uses of what their communication systems provide. But that fact – if it is so – has little or nothing to do with human use of natural languages, not to mention human language.

  13 Chomsky and Marr call their theories “representational,” but certainly in Chomsky’s case – and, I would argue, Marr’s too – this has nothing to do with the Fodorian notion of representation, which amounts to something like ‘re-presentation’. Granted, we use vision often to navigate and otherwise deal with our immediate environments. That is only an occasional and insignificant aspect of our use of language. Yet neither theory is ‘about’ the world; each concerns what goes on in the head.

  14 This needs qualification for science, as opposed to common sense. Science’s concepts seem to be inventions of human beings who construct theories. The point about ‘the world’ conforming to concepts remains, but in doing science, we hope that the concepts we construct with our theories offer a more objective (less anthropocentric) way to understand.

  15 This is obvious with philosophers who think that what is in the head is the product of history, acculturation, or the like (Foucault, for example). And it is obvious in Sellars, Putnam, and even Quine. In case it is not obvious with the more experimentally inclined connectionists, consider the following excerpts from an article by Morris et al. (2000) describing the aim of their effort to get ‘neural nets’ (computer models of what they take neural nets to be) to “learn language,” where languages are understood as communal forms of behavior, or “usage.” They explain that children “learn grammatical relations over time, and in the process accommodate to whatever language-specific behaviors . . . [their] target language exhibits.” Further: “From beginning to end this is a usage-based acquisition system. It starts with rote acquisition of verb-argument structures, and by finding commonalities, it slowly builds levels of abstraction. Through this bottom-up process, it accommodates to the target language.”

  16 Classifying Herder and Foucault among the empiricists will surprise some readers. Keep in mind that ‘empiricism’ as used here is a label for a research strategy concerning the human mind, one based on assumptions about the mind’s contents and how they got there. Clearly, both Herder and Foucault are anti-nativist and externalist in their views of language and how to study it. They are, then, empiricists, differing from others largely in a tendency to deny that natural science offers objective descriptions and explanations of world and at least some aspects of mind.

  17 I assume a distinction between common sense and science – or at the very least, the advanced mathematical sciences. The distinction goes back to Descartes, perhaps before. In his Discourse he contrasts the kind of study he is interested in (what we would now call natural science) and its methodology (which he helped clari
fy) to what one finds in “bon sens,” sometimes (and plausibly) translated as “common sense.” Chomsky adopts the distinction; it appears in much of his work. See his (1975a, 1988a, 1995a, 2000). Motivations for the distinction include the fact that children do not readily acquire scientific concepts and theories (although they may use sounds like “lepton”) nor do they – or any but those very familiar with the sciences – routinely display scientific creativity. It is relevant too that scientists and mathematicians try to regulate their uses of technical terms when communicating with others in the field. Their uses of technical terms are much closer to what the empiricist seems to believe is the case with common-sense concepts and their everyday use in natural languages.

  18 That is perhaps true for physics, but not for study of the mind. Quine’s “naturalized epistemology” holds that there are “causal” relations between sensory impingements and the beliefs and knowledge people develop about the world, where psychology is supposed to cash out these causal relations. But the story he tells about psychology in his naturalized epistemology material (Quine, 1969) seems to be little different from the view he develops of concepts, language, and world in his neglected 1974 The Roots of Reference. In that work, one finds standard empiricist claims about how psychological “causal” relations come to be established: with the exception of some “saliencies” found in sensory systems, causal relationships are under exogenous control. This is not ‘naturalizing’ the study of mind – it is not treating the mind as a natural object that grows according to a biophysical agenda and using the tools of naturalistic research to understand it. As for Sellars, his (1960, among others) simply assumes that the science of mind (psychology, I presume) is behaviorism, and he explicitly adopts an early version of the connectionists’ view of the brain and its ‘learning’. He does gesture in the direction of evolution for bee languages – a naïve version of evolution, at least. But he shows no inclination to extend what he says there to human language. That would have the effect of separating language from “reason,” which he takes to be ours by virtue of learning language. It would detach what he took to be the epistemic norms of reason from their ‘home’ in the linguistic community. Chomsky’s idea of studying language (including its meanings) apart from the use of language did not occur to him, or – obviously – to many other philosophers and cognitive scientists.

  19 The term is due to Elman, one of the more famous and productive connectionists.

  20 The preface Chomsky added to a 1967 reprint of his 1959 review of the behaviorist B. F. Skinner’s Verbal Behavior remains apropos. “I had intended this review not specifically as a criticism of Skinner’s speculations regarding language, but rather as a more general critique of behaviorist (I would now prefer to say ‘empiricist’) speculations as to the nature of higher mental processes. My reason for discussing Skinner’s book in such detail was that it was the most careful and thoroughgoing presentation of such speculation. . . Therefore, if the conclusions I attempted to substantiate in the review are correct, as I believe they are, then Skinner’s work can be regarded as, in effect, a reductio ad absurdum of behaviorist [and empiricist] assumptions. My personal view is that it is a definite merit, not a defect, of Skinner’s work that it can be used for this purpose, and it was for this reason that I tried to deal with it fairly exhaustively. I do not see how his proposals can be improved upon, aside from occasional details and oversights, with the framework of the general assumptions that he accepts. I do not, in other words, see any way in which his proposals can be substantially improved within the general framework of behaviorist or neobehaviorist, or, more generally, empiricist ideas that has dominated much of modern linguistics, psychology, and philosophy. The conclusion that I hoped to establish in the review, by discussing these speculations in their most explicit and detailed form, was that the general point of view is largely mythology, and that its widespread acceptance is not the result of empirical support, persuasive reasoning, or the absence of a plausible alternative.” (1959/1967: 142)

  21 The scare quotes are justified. The connectionists’ neural nets are computers made to simulate not what one actually finds in the case of actual neural nets – very complex forms of interconnected neurons that develop under genetic control to provide the organism with cognitive and other systems that are more or less the same across the human population. They are made to simulate Locke’s blank slate.

  22 ‘Generative’ can mean (and often does mean in CL, and certainly did for the Port-Royalists) ‘productive’, generally understood to require some recursive principles to allow for infinite competence, given human finite means. In more technical formal work, it means (or perhaps means also) ‘explicit’ or ‘formal’.

  23 Note that this measure of ‘better than’ explicitly relies on a technical sense of simplicity. Simplicity, as mentioned before, is another desideratum of science. The fate of the notion of simplicity in Chomsky’s work – its beginnings in the work of his teacher Nelson Goodman, its varieties in Chomsky’s work, and its extraordinarily important role in “minimalism” – is a fascinating study in itself, but unfortunately it is beyond the scope of this introduction.

  24 It is possible he dealt with such matters in the final volume of Le Monde. He destroyed that work when he heard of Galileo’s fate, however, so we will never know.

  25 Descartes held that animals are machines – that their actions are determined by external stimulus and inner state, and can be understood by using a deterministic contact mechanics (which includes a mechanical interpretation of neurophysiological function). We know he was wrong to think of them this way, but his test applies in any case.

  26 The test itself has obvious limitations. There are cases where, due to trauma or disease, people who demonstrated normal linguistic competence at one time become (perhaps temporarily) incapable of expressing language at all. An obvious if trivial example is a normal speaking person who has had a throat operation and cannot speak. There are other kinds of cases too, some of them extremely interesting from a scientist’s point of view because they reveal novel features of human language and mind. Still, Descartes’s test for mind is the best easily applied test anyone has come up with, and it takes no special skills or knowledge to apply. In only a slightly different form, Alan Turing reinvented Descartes’s test and suggested trying it out on programmable machines. He optimistically and incorrectly predicted that it would be possible to program a computer to pass it by the year 2000. Some of his other points are worthwhile, however, and his 1950 Mind paper “Computing Machinery and Intelligence” bears closer reading than it often receives. One of the more important of those insights, usually ignored, is that if a machine does pass the test, no fact of the matter has been determined; no scientific issue is resolved. The test offers no evidence in favor of a specific science of mind, and it does not show that the mind works the way the computer that passes the test does (any more than Big Blue’s win at chess shows that the human chess player’s mind works that way). All that is claimed to follow is that success might offer a reason to decide whether to say that a machine thinks – to decide whether to change one’s use of language and say that machines think (now often done anyway, without satisfying anything as strict as Turing’s test). For the naturalistic scientist of mind, this is about as interesting a question as whether excavators really do dig, or whether submarines swim. Usage – as linguistic creativity reveals – can vary, but with no consequence for (in this case) the naturalistic science of fish or the hydraulic systems of excavators.

  The significance of the test for Chomsky is quite different and more in line with another of Turing’s aims: “investigating the intellectual capacities of man.” The test (or rather, failure to pass it in the unrestricted case contemplated by Descartes) provides a reason to strongly suspect that some problems – here, constructing a science of the creative aspect of language use – are beyond the reach of human intelligence. And it underscores a basic assumption of RR research strategy: one shoul
d proceed on the assumption that there might well be limits to our science’s capacity to deal with human behavior – behavior that to all appearances seems to be free. The biolinguist expects limitations on human problem-solving capacities (both common-sense and scientific). We are, after all, natural biological organisms. See Chomsky 1988, ch. 5.

  Note that the significance of the test for Descartes, Cordemoy, and others in the seventeenth century was rather different; it was then clearly relevant to the principles of physical science as understood at the time and it suggested what seemed then to be a plausible hypothesis concerning how to deal with it: creativity must be due to a different principle, lodged in a non-bodily “substance.” Introducing mental substance to lodge a principle of creativity was for them then just doing what we would think of now as ‘normal science’. It was proposing a scientific solution to a scientific problem.

  Nothing came of Descartes’s proposed solution to the problem, obviously. Apparently, science still cannot deal with the matter, even though science – including the science of mind – has changed very considerably since Descartes’s time. Perhaps we should reintroduce a distinction Chomsky introduced (in Chomsky 1975a): if there is no known or even remotely plausible solution to a problem, in spite of many efforts to contend with it, it is not a problem but a mystery.