Cartesian Linguistics

by Noam Chomsky

  A related point is found in a distinction Descartes made in his “Comments on a Certain Broadsheet” (CSM I, 303–4) where he explains his view of innate concepts (or ‘ideas’). There are two sorts, he argued, those that lie in the mind from birth such as TRIANGLE, and those that are “adventitious,” meaning that they require some kind of occasion or triggering data to come into operation. His example was the “common view” of the sun, offered to all of us by the innate but adventitious (common-sense) concept, SUN. These two classes of innate concepts are clearly distinct from another SUN-concept, one that is “made up” (created, manufactured) by the scientist who constructs a theory of the sun. Naturalistic theory-construction is clearly different from practical problem-solving, such as deciding whether to plant at noon in full sun, or limit your efforts to the early morning or late afternoon. The ‘common’ concept/idea of the sun serves practical problem-solving well, and these concepts are available to everyone. We use them all the time – when, for example, wondering whether to get up before the sun or linger for an hour. But the common concept of the sun is of no use to the scientist. In science, the sun does not rise, nor set, nor move across the sky. In science, common-sense concepts provide little guidance; one must follow instead what Descartes called “the light of nature,” plausibly understood as seeking simplicity in nature by making one’s theories simple, theories that are then tested in experiments that control for irrelevant factors. Similar points are made in Descartes’s skeptical reflections: if we want full explanations of phenomena, we cannot rely on the view of the world and the things in it that common sense gives us. We cannot assume that that piece of paper out there is yellow, or – as with Chomsky – that language is some kind of public institution, learned from parents and friends, described by appeal to rules for ‘correct usage’. Taking the route of seeking simplicity, one is led to producing formally explicit abstractions removed from everyday understanding, and hoping that these can be integrated with the findings of other scientists, and other sciences. The point is fundamental to making progress in any science. Like Galileo and Descartes, Chomsky often remarks on the need in the scientific study of language to idealize and construct theories. Only by doing so can one hope to get anywhere.

  Descartes helped initiate natural science, a project that people can undertake that at its most general level is a strategy for research, or a methodology. The scientist, whatever domain s/he investigates, seeks descriptive and explanatory adequacy in a theory of natural phenomena; s/he demands simplicity and, to get it, constructs formal and explicit theories that idealize the phenomena under investigation; s/he seeks objectivity and to get it, abandons the anthropocentrically oriented concepts of common sense that prove so useful in resolving practical problems but fail in attempts to construct objective theories. When Descartes followed these principles, he enjoyed for his time remarkable success. He provided a detailed account of optics, dealt in an interesting and still-current way with neurophysiology, offered a cosmological theory, presented and defended a contact mechanics that he tried to turn into a “theory of everything,” and even pointed plausibly in the direction of a computational theory of vision. Despite offering the rudiments of a computational theory of vision, however, he balked at applying these methodological desiderata to the mind to develop sciences of the mind.24

  III.2 Linguistic creativity

  One possible explanation for Descartes’s reluctance to venture into the mind by using the tools of science lies in Galileo’s experience with the church. Descartes might have been unwilling to appear to be offering a naturalistic account of the mind, or what the church authorities might have thought the soul. That motivation, if it was one, is of little interest to us. The other is relevant, and important. It is found in his effort to take the creative aspect of language use observations into account by using the tools that his sciences gave him, especially those found in his contact mechanics. We have seen already that linguistic creativity observations are important for the science of mind – at least, for those who adopt an RR strategy. Here I discuss them in more detail and describe Descartes’s inadequate attempt to contend with them. There are lessons in his mistakes, mistakes that were excusable at the time, but are no longer so.

  Descartes’s creativity observations appear in Part V of the Discourse after a lengthy effort to try to show that a contact mechanics could be used to deal with everything – cosmology, neurophysiology, optics, and so on. As he understood the situation, a contact mechanics proved sufficient for describing and explaining anything and everything having to do with “body.” The creativity observations indicated that some phenomena lay outside the scope of science as he understood it. They seem to be explicable only by something like a “creative principle,” and creativity is absurd from the standpoint of a deterministic mechanics. If science fails here, he reasoned, it must be because something non-bodily is at work. Giving it a name, he called it “mind.”

  Descartes assumed, essentially without argument, that a person knows in his or her own case that s/he has a mind (in his gloss, that s/he is a rational, thinking being). To decide whether others – humanoid organisms, animals, or machines – have such minds, he suggested observing their linguistic behavior when asked questions or otherwise prompted to speak.25 Observing the way they use language is, he thought, sufficient to conclude that one is dealing with a human, not a zombie, automaton, or animal. It suffices, he thought, because when humans use language – and he was careful to point out that he meant humans across the spectrum of intelligence and apparently without regard to education and social position – they display a form of linguistic creativity that is not duplicated in the behaviors and actions of any non-human organism that has been trained to produce linguistic sound or signs, nor machine that has been built or programmed (or ‘taught’ itself through exercising some kind of generalized learning procedure) to produce linguistic sounds or signs.26 He said,

  [No such animal or machine] could . . . use words, or put together other signs, as we do in order to declare our thoughts to others. For we can certainly conceive of a machine so constructed that it utters certain words, and even utters words which correspond to bodily actions causing a change in its organs. . . But it is not conceivable that such a machine should produce different arrangements of words so as to give an appropriately meaningful answer to whatever is said in its presence, as the dullest of men can do.

  (Descartes 1637/1985 (CSM II): 140; emphasis mine)

  Glossing these remarks, his observations note that people, unlike machines or animals, can put together any of an unbounded set of sentences and manage nevertheless to bring what they put together to bear in a way that seems appropriate and coherent (“rational”) for the discourse context (not local spatiotemporal context of the speaker or hearer) in question. Nothing in the environment causes the sentences produced: while a question or comment might prompt or incite someone’s linguistically expressed thoughts or utterances, they do not cause it. You can ask someone how best to get to Cambridge, no matter where you are located. If they have an idea about how to do this at all, they will no doubt come up with a set of sentences, each sentence differing from the others the person uses in the detailed description given, and each differing again from those that that person will produce at another time for another person asking the question, and differing too – obviously – from what others might produce then or at other times. And yet these clusters of sentences differing from one another and, within a cluster, differing from each other again, typically offer appropriate, coherent answers to the request. There is no upper limit on the set of sentences for performing this task, or other specific tasks where language might be employed. ‘Unboundedness’ seems to be a property of sentences produced in discourse contexts across the board, whether those contexts bear on requests for information, attempts to persuade, criticisms of actions performed, gossiping remarks, attempts to be humorous, efforts to excuse one’s actions, etc. – or, in the case of thought, per
haps no verbal or other prompting at all. Humans seem to be able to produce an unbounded number (“different arrangements of words”) of sentences without causal antecedents (that are externally and internally “stimulus free”) although perhaps prompted (in Descartes’s time, “occasioned”) by questions or other factors, while remaining appropriate and rational in what they produce. Clearly, their production cannot be the result of some deterministic mechanism that offers the answer it is caused to produce, nor even of a mechanism that, given a question and a context, provides one of a specifiable (and thus bounded) range of answers. If someone insists on the opposite, they are hereby challenged to come up with the mechanism, for no deterministic system can yield any of an unbounded set of sentences, for a specific discourse context, all of which are coherent or appropriate. If the conceptual difficulties involved in this do not convince, then lack of success coming up with a science of such a mechanism suggests that those who insist on a deterministic mechanism are on very thin grounds indeed.

  Descartes expresses his test in terms of conceivability, not observability. Obviously, though, his test is readily applied. In the case of animals such as the chimp Nim Chimpsky (the subject of a massive effort at Columbia University in the 1970s to get an ape to learn sign language) and other primates, it arguably has been, although – because no ape has managed to learn even the rudiments of syntax and morphology for human sign language – it is difficult to distinguish failures due to creativity from those attributable simply to lacking language. A closer approximation to human speech, it is thought, might be found with machines; they can be programmed to produce orthographically represented sentences on a terminal or printer, or even relatively realistic voices. Applying Descartes’s test to machines, a candidate speaking machine would have to satisfy one or more persons that what it produces in response to questions (keyed in letters and words, perhaps) are as appropriate as those of a human asked the same question. If a machine’s performance is judged no differently than a person’s, that might be a reason to say that the machine thinks. Alan Turing (in Turing 1950) re-invented Descartes’s test and predicted – too optimistically – that a computer program would be able to satisfy one or more persons that its responses to questions were as appropriate as a human’s. Adapting Turing’s test, a competition was created (the Loebner competition) which awards a machine program that can manage to convince a jury after a designated period of interactions with two terminals, one controlled by the program and the other by a person, that the machine’s responses to jury questions are no less appropriate than the responses one gets, or expects from, a person. No program has won the grand prize, which allows unrestricted questioning, without regard to subject or context – in effect, Descartes’s test, although with a limited time period.

  Notice that the test assumes that an arbitrary human (with competence in the relevant language) can understand what an animal, machine, or human says (or ‘says’), and manage to administer the test. That assumption raises the question of what resources humans have available to understand and judge the appropriateness of what is said. Obviously, no one relies on a causal theory of human linguistic behavior; if we had such a theory, we could predict what machine or person will say, or at least limit the set of answers that could be produced; we would be in the position of a god, and no doubt quite bored with humans and their pretense to novelty and creativity. That kind of fantasy is a philosopher’s indulgence, at best; there is nothing to recommend it – nothing in experience or science to take it at all seriously. And it has no bearing on the issue of what a human mind has available to interpret what another says, and judge its appropriateness. We know from discussion in Part II that we have many such resources, including common-sense ‘folk’ principles, common biologies, environments, interests, etc. – generally, shared biologies and discourse-context information plus anything one gets from familiarity with the speaker. It is hardly surprising that no program has passed the unrestricted Turing/Descartes test. Machines aren’t people. They simulate human behaviors in carrying out specific tasks – often besting them where dealing with restricted contexts and specific problems. Computers can win at chess. But they do not play chess the ways humans do.

  Descartes ‘explained’ creativity by attributing it to reason; he even went so far as to claim (see CL’s text) that reason is a universal instrument, able to solve any problem. The self-contradiction is obvious: it cannot be a universal instrument if it cannot deal with linguistic creativity. We do much better by attributing creativity to humans with the biologies they have, biologies that give them the cognitive capacities that they have, including a capacity to produce boundless numbers of sentences and understand and interpret them. And we can assess the appropriateness of sentences with respect to discourse contexts that we can appreciate. We can appreciate them by being, like the speaker we interpret, humans with the innate resources we have.

  Apparently, our sciences cannot deal with creativity and – given the reasonable assumption that language and the cognitive resources it offers will shape and play a constitutive role in most of what we understand about ourselves and the world, and in how we deal with the multiple problems that everyday life poses – our sciences are inadequate to deal with human action and behavior in that very wide range of cases in which sententially expressed concepts play a constitutive role. We can, nevertheless, develop sciences of the mind by focusing on specific internal systems. Chomsky’s science of language, as we have seen, is a theory of such a system; his aim is to produce a “generative” theory that individuates a language by appeal to a set of principles or laws that takes arbitrary selections of specified sets of lexical items and returns a sound-meaning pair, a sentence; else the derivation crashes. This is a deterministic theory. It cannot explain how we manage to use language creatively; it does not explain linguistic behavior/action. But it can and does contribute to making some sense of how that readily observable kind of creativity is possible, and why it is available only to humans.

  Summarizing, ‘ordinary’ linguistic creativity is uncaused, unbounded/innovative, and appropriate/coherent. Explaining why it is available only to humans is easy: so far as is known, only humans have language faculties, and these are organic systems in a brain/mind that has multiple systems, some of which the language system ‘communicates’ with.

  Explaining how creativity is possible is not scientific explanation, but a matter of taking what is known in the science of mind – and especially language – now, and trying to make sense of linguistic creativity’s readily observed features. Beginning with being uncaused, it is plausible to suggest that it must have something to do with the fact that the mind is modular, and that language in particular is not only modular, but neither an input nor an output system. To say it is modular is at least to say that it operates in accordance with unique principles, taking system-specific inputs and providing system-specific (to the faculty) outputs. To say that it is neither an input nor an output system is to say that it is not tied closely, as sensory or output systems are, to dealing with signals and other forms of input from outside the head, nor to the direct production of bodily movements. These characteristics of the language system do not provide a fully satisfactory account of the apparently great degree of autonomy the language system displays. But they make a plausible contribution, and may be the best that we can offer.

  Unboundedness or innovation plausibly has something to do with the language faculty’s ‘productivity’.27 A generative theory of language, if correct, suggests that the language faculty can provide at the “conceptual-intentional” interface(s), in principle, an unlimited number of “perspectives” (to use Chomsky’s term for them). These perspectives can be thought of as complex forms of concept – essentially, sententially ‘expressed’ concepts. The unlimited set of possible outputs is a discrete set: each complex item (sentence/expression/sentential ‘concept’) is in its structure and character distinct from any other. Intuitively, the theory explains why an ‘und
erstanding’ of one sentence is distinct from an understanding of another or, to use semi-technical vocabulary that avoids the direct use of charged words like ‘understand’, it indicates that the internal or intrinsic content of any one sentence that a person’s I-language can ‘generate’ is distinct from that of any other. It is so because language relies essentially on a recursive procedure that yields hierarchically structured expressions, and each expression consists of two complexes of features at language’s “interfaces,” the phonetic interface and the semantic. Expressions have these characteristics because of Merge, an operation that combines arbitrary ‘concepts’ (meanings of lexical items) to yield through phonetic features ‘sound’ instructions to articulatory and perceptual systems, and through semantic features, ‘information’ for the conceptual–intentional systems with which language ‘communicates’. Various constraints limit the total understandable perspectives available to persons, the total of perspectives available for their use. These include memory constraints, ‘parsing’ constraints, some forms of embedded construction, and so on. But by any measure, what remains is massive expressive power – by reasonable estimates, more comprehensible sentences for people with moderate vocabularies than they could produce or listen to in their lifetimes. So far as making sense of the creative aspect of language use is concerned, this is plenty. All that is asked is that for any given linguistic ‘task’ (suggesting, describing, querying, scolding, gossiping. . .) and any given discourse context (and its “immediate focus of interest,” to recall Strawson’s phrase (in Strawson 1950)), there is no way to set an upper limit on the set of sentences that can be understood and produced by a speaker, even though every such sentence remains appropriate.