Noam Chomsky

Interview by Dr Christopher Evans recorded at the Massacheusetts Institute of Technology, May 1973. Copyright Ferranti Ltd. 1973

from Audio Arts supplement 1981

Transcript

Given this concern for language, artists referred to and often appropriated from theoretical work in related disciplines – philosophy and linguistics being primary points of reference. In the first issue Cyril Barrett, Reader in Philosophy at the University of Warwick, talks about the relationships between philosophy and art, and in the following extract Dr Christopher Evans and Noam Chomsky discuss models for the description of language.

Christopher Evans: Many people assume that there are three models for the description of language: a ‘finite state’ grammar; what you call a ‘phrase structure’ grammar; and your rather dramatic development, the ‘transformational’ grammar. These are the kinds of things that many people find difficult to get clear in their minds. Could you say something about the three types of grammar: how they work, where they interact, and so on.

Noam Chomsky: Finite state grammar is the kind of system that would result from the most sophisticated psychological models of perception and production that one can imagine, on the basis of the concepts of habit, or habit structure, or stimulus and response. The idea is that the system, let’s say the mind, has a finite set of configurations, which is certainly true; that it accepts linguistic material one symbol at a time, which again is true, and as it accepts a particular linguistic symbol, a sentence is produced. Let’s say the first symbol and the second symbol, and so on, are accepted by this system. Each time that the system is activated by the new symbol which appears, it moves into a new state. Finally, if it ends up in some designated terminal state then it has accepted the sequence as a sentence and has understood it or perceived it.

There’s no doubt that one ought to be able to describe any perceptual model in these terms for a finite device. The question, however, was whether the actual knowledge of language that the mind possesses is entirely representable in terms of such a device. While there was a belief in the 1950s that this was true, it can be demonstrated that this is not true, and there is certainly no paradox or problem in that. So I think finite state models have virtually lost their interest as models for grammar. In fact, I know of no one who studies them.

Evans: You are basically a linguist, as well as a philosopher and mathematician, and I suppose one needs to be those three things – or at least a mathematician and philosopher – to be a good linguist, perhaps. And yet you have done some very productive work with psychologists, in particular George Miller. How did you first come to work with a psychologist at a time when linguists and psychologists were considered to be absolutely poles apart?

Chomsky: Well, I think George Miller and I found that our thoughts were moving along parallel tracks in the mid 1950s. As I recall, we must have begun talking seriously together and then working together by about 1956 or so. Miller had done very productive work on cognitive psychology, developing information theoretic models, and had begun to see their basic inadequacy, I believe. We were jointly interested in the question of the relationships among the various kinds of models that we were discussing earlier and the reasons, the basis, for the inadequacy of finite state models – the most abstract models that one could dream of achieving within the framework of even cognitive psychology, let alone behavioural psychology as then understood. We then went on to investigate jointly both the mathematical properties of some of these more complicated systems and also the possible implications for psycholinguistics (a field of which he is very much the grand master) of the hypothesis that these more abstract and complex models play a fundamental and central role in knowledge of language and hence in performance.

Evans: Several things that you were saying must have seemed absolutely anathema to the psychologists of the 1940s and 1950s. First, the notion of there being a universal grammar, which is one of your important hypotheses, a grammar that is constant for every shape of human being, and secondly, the idea that the basis of this grammar must be somehow inherited and passed over. In many ways these ideas don’t seem as shocking now as they did then, but I think it’s worth trying to find out exactly what you are saying about the carry over of, first of all, the universal grammar. Do you still feel that such a grammar is constant and present in all human beings?

Chomsky: Well, I think there can be no doubt among rational people that there is a very rich and complicated innate structure that permits us to acquire these highly articulated, very specific, extremely rich systems of knowledge, such as knowledge of language, on the basis of the highly degenerate information that is presented to us as language learners or learners of anything else for that matter. The only question is, what is the nature of this innate system? Here, I think, one can very roughly distinguish two major tendencies, two major beliefs which have a long tradition of controversy and conflict.

One approach – I’ll call it empiricist, because I think it does express the leading idea of traditional empiricism – is that what is innate to the mind is a system of processes and procedures and analytic mechanisms, a kind of data-processing procedure by which the data of sense is given a preliminary analysis in terms of innate structures of perception and is then further organised. So, for example, when Hume says that the part of our knowledge which comes to us from the original hand of nature is simply the animal instinct of induction and that everything else is simply a matter of forming associations, he is describing what we might think of as a primitive data-processing procedure, speaking of innate knowledge – incidentally I think quite properly. That is a clear expression of one general approach toward the problem of acquisition of knowledge. We have data of sense, we have a preliminary processing system which analyses it into properties and so on, and then we have certain procedures, like the procedure of forming associations, or of carrying out induction, or of developing habit structures of some sort, or carrying out constituent analysis. That’s what I meant by saying that Harris’s approach was the most sophisticated empiricist approach that existed, and our knowledge is simply the result of the application of these innate data-processing procedures to the presented data of sense.

That’s one approach. Another approach, which seems to me to express the leading ideas of traditional rationalism – and I’ll here call it a rationalist approach – is to say that the structure of our system of knowledge is what is predetermined. It’s the form of the knowledge that’s predetermined. We have a schematicism which determines the kinds of knowledge and the arrangement of principles and concepts that can be acquired on the basis of experience. This approach pays little attention to, and sees little interest in, such procedures as may exist to move from the data to the knowledge. If we think of the problem of learning as having three elements, namely input, operation and output, the system has a certain input of experience and carries out a certain operation and it has some sort of output as a result. That’s learning in a most general sense. The empiricist approach takes the position that it’s the procedure, it’s the operation that is innate and is of fundamental importance, and the output is anything that that procedure determines when it is applied to the data. So if you think of learning as a function mapping input into output, it’s the function that’s of interest to the empiricists. The rationalists, on the other hand, are saying the function isn’t of much interest and the output is very sharply constrained. It must be of a certain type, so a rationalist theory of perception would say that we perceive the data of sense in terms of schematicisms which are determined by, let’s say, geometrical notions of a regular triangle, innately: we see actual triangles as distorted variants of regular triangles, because the mind simply deals with those regular geometric figures. An empiricist might approach the same material by saying that we somehow form the notion of regular geometric figures by association or induction, or whatever. Here is a clear case for the rationalist approach which was developed in the seventeenth century; in fact, it is almost certainly correct if developed in a modern form.

Applying these perhaps overly sharp differentiations to the case of language, there is an approach which is very well represented by the best and the most sophisticated work in structural linguistics, which sought to express and develop precisely procedures of analysis which would yield grammatical structures, knowledge of language, if you like, by application to a corpus of data. That’s one approach. The rationalist approach, on the other hand, says the kinds of language that are accessible to the mind are a very narrowly restricted class, and the task that the child has is to discover which of the humanly accessible languages is the one to which he is being exposed. On empiricist grounds, what the child would be asking is this: he would say, ‘Well, I have this data of sense, now what kind of associations can I form and what kind of habits can I construct?’ And so on and so forth – not consciously, of course. On the rationalist approach the child is more or less asking, ‘I have this data of sense, which of the possible languages is this a specimen of? Is it a specimen of Japanese? No, because it doesn’t fit the principles of Japanese. Is it a specimen of English?’ And if that hypothesis is continued, he then uses the knowledge of English.

As these remarks indicate, I believe that the rationalist approach is correct, that the process of learning language is a process which we can explain as follows: the mind innately has a universal grammar, a very narrow restriction that permits certain systems and excludes other systems which would be equally useful for communication, but just don’t happen to fit the structure of mind, and the task of learning is to discover which of the possible systems is the one to which the child is being exposed. It is the task of taking this universal schematicism, this framework, and putting a little meat on the bones, that is, finding out how it is articulated specifically in this particular case [ … ]

It shouldn’t be a surprise to, let’s say, a physiological psychologist that our concepts of visual space are based on highly articulated, very specific schematicisms that are determined by the structure of the brain, by the structure of the visual cortex. I think it would be a rare physiological psychologist these days who would believe that our interpretation of visual space is developed by long procedures of association and conditioning. In fact, what he would expect to find and what we do find is very specific structures that are put into operation on the occasion of experience, but that have the character that they do because that’s the way our brain is. If we were a different organism we would do it differently, and I don’t see any reason to suppose that the case of language, or for that matter any other cognitive system, is any different [ … ]