Return to Human Knowledge and Human Nature contents  

 

6. THE CASE FOR INNATE MENTAL STRUCTURE

 

Having considered the empiricist case against nativism and found it wanting, I shall now begin to assess the evidence which supports nativism.

 

Principles of Learning

All empiricists have allowed that the mind is innately structured into distinct faculties, including for example thought, perception and memory. And all have allowed that the basic mechanisms involved in the acquisition of our beliefs - such as the recording of experiences in memory, and the laws of association amongst ideas - are innately given. (One cannot, from nothing, learn how to learn.) But they have denied that the mind divides into faculties having innate constitutent structures which are distinct from one another, embodying information about the domains which they concern. On the contrary, they have insisted that all belief-acquisition principles are general ones, operating in similar ways for all domains of knowledge. Much of the evidence which has emerged over recent decades, on the other hand, suggests that there is innate information embodied in the language faculty, for example, which is not present in the faculty of vision or vice versa. (I shall hold over until later, discussion of the question whether such information may be described appropriately as 'knowledge'.)

            In fact there might appear to be two different issues here. First, there is the question whether the various mental faculties employ distinct innate belief-acquisition principles. Then secondly, there is the question whether our innate mechanisms of belief-acquisition can be said to embody information about the world. It is easy enough to see why empiricists should have denied the latter possibility. For even if such information could not itself be counted as genuine knowledge, it would plainly contribute to our knowledge, and might thus be expected to fall within the scope of the empiricist rejection of nativism. However, it is not nearly so obvious why empiricists should have insisted that all belief-acquisition principles are general ones, the same for all domains of knowledge.

            One possible explanation lies in the conception of science endorsed by the early empiricists.¹ For the assumption within the physical sciences at the time, was that all physical phenomena should be explicable in terms of laws of nature which are entirely general, applying equally to all interactions and combinations of matter. In effect, it was assumed that all laws of nature which concern the physical realm should be reducible to the laws of physics. Now as well as being epistemologists, the early empiricists certainly saw themselves as attempting to develop a science of the mind (more on this in chapter 9). So quite probably they would have imported the above assumption into the domain of psychology, thus insisting that all mental processes should be explicable in terms of the same general psychological mechanisms, such as laws of association amongst ideas.

            Few thinkers today would endorse the reductionism inherent in early conceptions of science. Few would now insist that all sciences which deal with physical phenomena, including chemistry, biology and physiology (and since most scientists today are materialists, they would also add psychology), must in principle be reducible to physics. On the contrary, most would allow that the special sciences can contain laws which, while they may (indeed must) be consistent with the laws of physics, cannot be deduced from them.² Then if the laws of the special sciences are autonomous, there is no reason to insist, either, that within any given science, such as psychology, all phenomena must be explicable from a single set of laws. We therefore seem to have no reason, today, for insisting that the psychological mechanisms operative in the various faculties of the mind should be the same throughout.

            But in fact there is also a more direct explanation for the empiricist insistence that learning mechanisms should be the same for all domains of knowledge. For if those mechanisms were to differ in different cases, this could only mean that they embodied information about the domains which they concern. For example, suppose that the mechanisms through which we learn the grammar of our native language were to differ from the mechanisms though which we acquire our common-sense beliefs about material objects. It is then hard to see how this could be so - to see why one could not, for example, employ the one set of learning principles to acquire knowledge of the other domain - unless the mechanisms in question contained, at least tacitly, information specific to the domains which they concern. In which case the insistence on general learning principles may be seen to derive from the empiricist view that mechanisms of learning should not in any way embody information about the world.

            We now need to consider whether empiricism is correct in this regard. I shall focus first on the case for an innately structured language faculty, considering in the final section the case for an innately structured visual faculty.

 

Chomsky on Language

Noam Chomsky has been prominent in arguing that human beings possess a distinct language-faculty, which is involved in the acquisition and use of natural language. The principles of operation of this faculty are said to be distinct from those of other psychological faculties, containing a good deal of information about natural languages.³ In fact Chomsky's view is that much of our linguistic knowledge is innate, being embodied in the structure of the language-faculty. When a child acquires its first language, this is not so much a matter of learning, as of the language-faculty being triggered into spontaneous growth. Chomsky concedes, of course, that some exposure to English is necessary if the child is to grow up speaking that language rather than French or Japanese. But he denies that this is all a matter of learning - the child only has to learn the words of the language (the lexicon), not the grammar. He is therefore committed to the theory of local triggering of innate knowledge, which we distinguished from other varieties of nativism in chapter 4.

            It might be objected that there can be no way of testing Chomsky's hypothesis. For once we allow that innate knowledge may be locally triggered by experience of the domain which it concerns, how are we to tell, in any particular instance, whether the knowledge has been learned or triggered? But in fact, the general point about a mechanism of learning is that the final state of the system should be a direct product of the initial experience, covarying with it. In contrast, a mechanism of local triggering will malfunction in the face of initial experiences for which its operative assumptions fail to hold. So if our knowledge of the grammatical basis of natural languages is locally triggered, then we may predict that a child exposed only to an artificially constructed language will either fail to learn it at all, or will only do so extremely slowly by comparison with normal language acquisition. However, as this example makes clear, direct testing of the issue may be ruled out on moral grounds. In settling the question whether some item of knowledge is learned or locally triggered, we shall often have to rely upon indirect arguments.

            However, this need not mean that we cannot sometimes form reasonable beliefs on the matter. In particular, wherever it is hard to see how our knowledge of some domain could have been acquired from our experience via any combination of memory, inductive extrapolation, and inference to the best explanation, then we shall have reason to believe that local triggering has occurred. As we shall see in the next section, this is just what Chomsky claims. He insists that a child's mature knowledge of its native language will go well beyond anything that they could have learned from their limited exposure to it.

            Crucial in the development of Chomsky's ideas has been the discovery of what he calls 'linguistic universals'. These are abstract features of syntax which have been found to be common to all natural languages.⁴ For example, all known languages have a syntax which is dependent upon phrase-structure rather than, for example, linear word-order. And the basic form of sentence-structure within all natural languages is subject-phrase/verb-phrase (where the object of the verb is incorporated into the verb-phrase), rather than the form subject-phrase/verb-phrase/object-phrase employed in the artificial languages of logic. Chomsky's view is that these universals reflect our common genetic endowment, given in the structure of the language-faculty.

            Moreover, even where there are manifest syntactic differences between languages, Chomsky thinks that these can be seen as merely different settings in a single underlying innate parameter.⁵ In his view, acquiring such a feature of syntax will be a matter, not of learning, but of switching the parameter to the appropriate setting. Of course Chomsky concedes that the lexicon (the individual words) of one's language need to be learned, together with some specific rules of grammar. But he maintains that this is largely a matter of learning the labels for a set of innate concepts, most concepts themselves being unlearned. I propose to set aside his views on concepts to the next chapter, concentrating here on his nativism about syntax.

            One argument for nativism concerns the explanation of the existence of linguistic universals themselves. For why should all natural languages have features in common, unless this reflects the innate structure of a language-faculty which we all possess? Certainly there does not appear to be any general feature of our psychology, nor any general learning-principle, which could explain why the syntax of all languages is phrase-structure-dependent. However, an equally plausible explanation, alternative to nativism, exists in the hypothesis of a common origin. Let us suppose, as is likely, that all humankind are descended from a common stock. Then there will probably have been a time in the distant past when all living human beings spoke the same language. As different groups then dispersed around the globe, their language would have begun to change and develop, in perhaps radically different directions; but always, it may be supposed, retaining certain general features in common - today's linguistic universals.

 

Plato's Problem

In fact Chomsky himself places greater emphasis on the argument from what he calls 'Plato's Problem' (so-named after the slave-boy example in Plato's Meno). The problem is essentially this: how, in the absence of innateness, would children manage to learn so much, so fast, and on the basis of such meagre exposure to the language? The issue of quantity learned is of course impossible to measure precisely, but some sense of it can be gained by glancing at any contemporary work in linguistics. Natural languages seem to be governed by a great many different syntactic rules of very considerable complexity. While these rules are not consciously known by a native speaker of the language, they seem nevertheless to implicity govern linguistic behavior. For example, they emerge in the speaker's judgements concerning what constructions are or are not permissible in particular cases. The question then is: how does the child manage to discover all of these rules within the space of just a few years, employing only learning mechanisms, when even many years of cooperative labour on the part of linguists still leaves much of the grammar of natural languages controversial?

            The problem becomes even more pressing when one considers the nature of the data available to the child. For adult linguists, who may themselves be native speakers, will have available to them as data in the construction of their grammars a potential infinity of sentences of the language. They can call up sentences at will, and judge for themselves whether they are permissible on the one hand, or ill-formed on the other. The child, in contrast, has to arrive at its knowledge of the grammar of its native language on the basis of the very limited sample of sentences to which it happens to have been exposed. Moreover, it will have, in general, only positive data to go on. No one explicitly tells the child that certain constructions are impermissible. Nor, when the child itself speaks, are its errors often corrected. In general people simply speak to children, without attempting to tell them how to speak; and they allow children to speak for themselves, only very rarely correcting their mistakes. Even more strikingly, indeed, much of the data available to the child is actually corrupt (that is, ungrammatical). To see this, try taping part of an ordinary conversation, transcribe it on to paper, and then study the result. What you will find is that much of the conversation will consist of sentences which are never finished, sentences which start out in one direction and then conclude in another, and so on. Yet on this sort of basis the child acquires a mastery of the grammatical rules of the language which still eludes complete description by linguists.

            This argument of Chomsky's appears extremely powerful. It is impossible to see how the child could construct a nonconscious model of the grammar of the language, using only general learning principles (such as analogy, inductive extrapolation, inference to the best explanation, and so on), on the basis of such fragmentary and corrupt data. Nor is it easy to see how there could be any learning mechanism specific to the language faculty which could generate knowledge of grammar from such data without presupposing anything about the structures of natural languages. It appears that Plato's Problem in the case of language can only be solved if we suppose that much of the child's grammatical knowledge is already innate. The child would then only have to learn the lexicon of its language and a few grammatical rules which are language-specific, provided that it has sufficient exposure to instances of sentences of the language to set the various parameters within its innate language-faculty. This looks like a manageable task.

            There is a particular development of the above argument which Chomsky also employs. This is that children simply do not make the sorts of mistake which one would expect them to, if they were arriving at their knowledge of the language through general learning principles. For example, consider how the child learns to construct questions from the corresponding declarative sentences. It first has experience of simple sentences such as 'The man is at home' and 'The man is happy', noticing that the corresponding questions are 'Is the man at home?' and 'Is the man happy?' respectively. At this stage the simplest, most obvious, candidate for the rule of question-formation, is to remove the first occurrence of the word 'is' (or its cognates) to the front of the sentence, leaving the sentence otherwise unchanged. One might then expect that when faced with the task of constructing the question corresponding to the sentence 'The man, who is happy, is at home', a child might try 'Is the man, who happy, is at home?'. But no child ever says such a thing! The next most obvious candidate for the rule, would be to remove the final occurrence of the word 'is' to the front of the sentence, leaving the sentence otherwise unchanged. But applying this rule to the sentence 'The man is at home, who is happy' would produce 'Is the man is at home, who happy?' But no child ever makes this sort of mistake either!

            Of course the correct rule, as Chomsky points out, is not linear, as these are, but phrase-structure-dependent. The rule is that one should remove the occurrence of 'is' from the main verb (the verb of the main verb-phrase) to the front of the sentence. For in reality the sentence 'The man, who is happy, is at home' has the structure '[[The man] [who is happy]] is at home', where the outside brackets enclose the subject-phrase. Similarly, the sentence 'The man is at home, who is happy' can be represented as having the form '[The man] is at home [who is happy]', where the brackets collectively enclose the subject-phrase. Since these structures are amongst the last that one might have expected, had one been employing only general learning principles, the information that natural languages are phrase-structure-dependent (as well as much else besides) must be supposed to be innate to the language-faculty.

            Someone might be inclined to object against Chomsky that grammar (the grammar of English, for example) is explictly taught, often in schools. Indeed, are not the letter columns of newspapers full of complaints that grammar is inadequately taught? How, then, can Chomsky claim that children acquire a near-perfect grasp of it by an early age? But this is a misuderstanding. In the view of Chomsky, as of most contemporary linguists, the various dialects of English constitute autonomous languages, with their own rules of sentence construction. It is only for political reasons that we select one of these dialects as primary, and try to impose it uniformly across all sections of the community. So Chomsky's reply to the objection would be that children do indeed have a near-perfect grasp of the grammar of their native English dialect at an early age; what they are then required to master in school is the grammar of another dialect, which they have to learn almost as they learn a foreign language.

 

Are Grammars Known?

We may agree that in some sense or other a mature speaker of a language has succeeded in internalising the grammar of that language. For such a speaker has the ability to tell, of any candidate sentence of the language, whether or not it is syntactically well-formed. There must then be an internal structure which explains this ability, in some way corresponding to the rules constructed by linguists. But is it appropriate to describe the ordinary speaker as knowing the grammar of their language? Should we say that the innate structure of the language-faculty provides us with locally triggered innate knowledge of the syntactical rules governing our native language? Chomsky argues at some length that we should, insisting that we are here dealing with a genuine species of innate - although nonconscious - knowledge.⁶

            It is worth noticing that if we accept the position argued for in the previous section, then in any case we shall be provided with some locally triggered innate knowledge. For example, your knowledge in the particular case that 'Is the man, who happy, is at home?' is not a permissible question-form, would constitute just such knowledge. For you know this without being told, and without having learned it from your experience of the language. This is not at issue. What is at issue is whether the underlying systems of rules, which generate this particular item of knowledge, may be said to be known in their turn.

            Chomsky contrasts his claim that we have knowledge of grammatical rules with the idea that our grasp of them constitutes, rather, a practical capacity, or skill. He presents a number of arguments against the latter hypothesis. In the first place he points out that two people can have exactly the same knowledge of English, while differing greatly in their ability to use it. For example, one may be an accomplished and creative public speaker, whereas the speech of the other is pedestrian and banal. Yet they each have the same knowledge of the grammar of English, for example agreeing as to which sentences are or are not well-formed. But here Chomsky has picked on a different ability to the one intended. The skill which Chomsky's opponent contrasts with our supposed knowledge of the grammar of English, is not the ability to use English well, or creatively. It is simply the capacity to produce, and to recognise, sentences of English which are grammatical. And in this respect there may be no difference whatever between the creative and the dull speaker.

            Chomsky also cites the case of people who suffer complete aphasia after a head-wound, losing all ability to speak or to understand their native language. But then as the effects of the injury recede, they may recover their ability to understand without any relevant instruction or experience. What this shows, Chomsky thinks, is that since people can retain their knowledge of the language while losing their ability to use it, our grasp of grammatical rules cannot simply be a matter of ability, but is a genuine instance of knowledge. What he fails to notice, however, is that one can retain the categorical basis of a capacity, while circumstances prevent that capacity from being exercised. (To say that something has a capacity, is to say that it will behave in a certain way if certain other things happen. The categorical basis of a capacity is the positive - non-hypothetical - fact which explains why the thing would behave in that way in those hypothetical circumstances.)

            For example, consider the brittleness of a glass, which is a matter of its capacity to be broken easily (it will break if you hit it). This capacity has a categorical basis, presumably in the molecular structure of the glass, which explains why even modestly severe impact will shatter it. Now suppose that the glass has been tightly packed in cotton wool. Of course it retains its brittleness (its molecular structure remains unchanged). But as a matter of fact it is no longer easy to break it. Something similar may also be true in the case of temporary aphasia. It may be that the categorical basis in the brain of the person's ability to construct and to recognise grammatical sentences remains undamaged, but that the mechanisms which allow this ability to be exercised are no longer operative.

            In addition to the arguments above, Chomsky seems to assume that anyone holding the thesis that our grasp of grammar is a practical capacity will also be an opponent of nativism. He apparently thinks that such a thesis must go together with a picture of language-learning as a mere matter of habituation, of developing a skill through repetitive activity. But there is no reason why these ideas should be connected. We could agree with Chomsky's arguments for the innateness of the language-faculty (as indeed I have). We could also agree that the categorical basis of a mature speaker's grasp of their language must be structured in a way which reflects the systems of rules developed by linguists. For how otherwise are we to explain the productivity of their ability (that is, their ability to produce, and to recognise as grammatical, sentences previously unencountered)? Yet we can combine all this with a denial of Chomsky's claim that this categorical basis - structured though it may be - is appropriately described as a system of knowledge.

            The motive for such a denial is as follows. Recall that knowledge implies belief. So Chomsky's claim requires speakers to be credited with nonconscious beliefs concerning the grammatical rules which govern their native language. Now there is no objection in principle to the idea of nonconscious beliefs. For example, consider a bicyclist who successfully negotiates obstacles in the road while their conscious attention is wholly directed elsewhere. Thus I might yesterday have been thinking so hard about the content of this chapter that I was unaware of what I was seeing while I cycled, so that if you had asked me even a moment later I should have been unable to say what I had been doing on the road. Yet an observer would say that I must have seen the vehicle parked by the side of the road, since I turned the handlebars to avoid it.

            In a case such as this it seems right to say that I had a nonconscious belief that there was a vehicle there. But notice that this is because the explanation of my behaviour fits neatly into the practical-reasoning-model. It is because I wanted to reach my destination safely, and believed that there was a vehicle in the way, and believed that unless I turned the handlebars I should hit it, that I acted as I did. It seems essential to the very notion of belief (whether conscious or nonconscious), that beliefs are states which are apt to interact in a distinctive way with other states of desire and belief, in such a way as to cause action. Yet it is this pattern of interaction which is apparently missing in the case of our supposed belief in the rules of grammar.

            Another way of putting the point, is that the structures which underlie our grasp of grammar do not have the sort of flexibility of behavioural effect which would be necessary for them to qualify as beliefs. Any genuine belief of yours can lead you to do one thing, given that you have one desire, or to do something quite different, given that you have another. If I believe that the ice on a nearby pond is thin, this may lead me to put away my skates, if my desire is safety; or to put them on, if I want to court danger; or to say to my wife 'That is a good place to skate' if I want to collect the insurance on her life. There is nothing analogous to this in connection with our mature grasp of the grammar of our native language. True enough, if we want to achieve one effect we will say one thing, and if we want to achieve something different we will say something different. But the relevant beliefs here will not be concerned with the rules of grammar, but rather that to say such-and-such in these circumstances will achieve such-and-such an effect. Our grasp of grammaticality will be equally implicated in any case, underlying in the same way whatever it is that we choose to say.

            If we really did have beliefs about the detailed grammar of our language, underlying our ability to produce grammatical sentences, then one would expect that they could be deployed equally in the service of ungrammaticality. For compare: your beliefs about the layout of a particular building (if true) are what enable you to find your way around it. But they will be equally serviceable if you wish to give the appearance of being lost, enabling you to wander from one corridor to another without ever reaching your destination. In contrast, we do not have a finely-tuned ability to produce ungrammatical sentences to order. If asked to do so, we mostly have to resort to very crude methods, such as constructing a random sequence of nouns ('Lion tiger stag'). This looks much more like the difficulty a practised cyclist encounters if asked to ride their bicycle badly. The best that most of us can do here, is to wabble a bit and then fall off. (It takes training to become a clown.)

            It is not easy for those who possess a skill to behave as if it were absent, whereas it generally is easy for those who have a belief to behave as if they lack it. In this respect our grasp of the rules of grammar seems much more like a practical ability than a set of beliefs. I therefore conclude that, while Chomsky is correct to claim that much of our grasp of language is innate, he is wrong when he says that the mental structures which constitute that grasp should be counted as innate knowledge. (Indeed, he may even be wrong in thinking that grammatical rules are explicitly represented in the mature structure of the language faculty. Rather, they may be implicit in its mode of organisation, in such a way that the system behaves in accordance with those rules, but without actually consulting them.)⁷

 

Structure in Vision

There is a case for thinking that many other faculties besides language have innate constituent structures, embodying information about the world. Here I can only gesture at some of a rich body of evidence concerning the faculty of vision in particular.

            First, it appears to be innately determined that we should perceive the world three-dimensionally. (This may be why even phenomenalists, who have become convinced that there in fact exists no independent reality, continue to see objects as distributed in three-dimensional space around them.) One argument for this conclusion, is that it is hard to imagine how we might learn, employing only general learning principles, to interpret the patterns in a two-dimensional visual field as representing objects in three-dimensional space. There are of course cues of various sorts, of the kind exploited by landscape-painters, such as perspective and shading. But it is hard to see how a child who as yet had no idea of distance could be supposed to deduce that these were indeed cues to three-dimensionality.

            It might be objected that this argument is too swift. For it could be maintained that a child will learn to interpret two-dimensional images in three-dimensional terms by moving around in the world, and noticing how the two-dimensional image changes.⁸ In fact, however, three-dimensional vision seems to be in place before human babies are capable of independent motion. At least, they will attempt to move away from a visual cliff as soon as they are old enough to control their movements. Moreover, it has been shown that three-dimensional vision is innate in at least some other species. In particular, new-born chicks will stay away from a visual cliff, despite having had no previous visual experience. Considerations of simplicity then suggest that three-dimensionality is innate to human vision also.

            Additionally, the development of computational theories of vision has made it seem increasingly likely that seeing in three-dimensions (as well as much else in our visual faculty) is innately determined.⁹ For as scientists have begun to construct computer models of how the visual faculty can work out the details of the visual scene from the two-dimensional information embodied in the pattern of activation of the rods and cones in the retina, it has became obvious that these computations simply could not be carried out realistically without a rich background of innate assumptions. Amongst these assumptions is that of three-dimensionality. Other suggested assumptions, whose innateness would make the operation of our visual faculties a good deal more intelligible, are that most moving bodies are rigid, and that most perceived change (once the movements of our own head and eyes have been allowed for) results from objects moving from one place to another. These have then received dramatic confirmation in experimental findings.¹⁰

            For example, suppose that you are looking at a television screen on which a circle is displayed. The circle then collapses, through narrowing ellipses, into a straight line, and then expands up to a circle again. What you will in fact see is a circle rotating in three-dimensional space (the straight line being the case where you see the circle end-on). Your visual faculty seems to embody the information that most changes of this sort do indeed result from the motion of rigid bodies, rather than from the (equally possible) collapse and expansion of bodies which are flexible.

            In another clear demonstration of the importance of a rigidity-assumption, experimenters attached tiny points of light to the main human joints in an otherwise completely darkened room. Observers were nevertheless able to interpret the movement of the lights as a person walking. If the lights were positioned in the middle of the rigid portions of the limbs, on the other hand, then subjects could make no sense of the resulting movements. Apparently the visual system is structured in such a way as to assume that the lights are linked by rigid bodies (as they are in the first experiment, but not in the second).

            There have also been a number of dramatic experiments conducted involving apparent motion. If a light is displayed at one point on a television screen, which is then extinguished and replaced by a light at a different point, what you will in fact see (if the timing of the change is correct) is the first light moving to the position of the second. Indeed, if a bar is positioned between the two lights, what you will see is the first light moving around the bar to the position of the second. And if the first light is red and the second blue, what you will see is the light change colour while it moves. What seems to be happening is that your visual faculty automatically assumes that if changes take place in what you see, then this will normally be the product of real motion, interpreting perceived change in just such a manner if possible.

            It is easy to understand from an evolutionary perspective why the basic structure of our visual faculty should be innate. For all human beings, at whatever times and in whatever places, have inhabited a world which is basically the same, at least in so far as it consists of a range of differently shaped, middle-sized, mostly rigid, moveable objects against an immobile three-dimensional background. There are then great advantages to the individual, both in terms of time and reliability, if the appropriate representational system is innately structured. For in that case the individual does not have to waste time learning such a system, and there is no risk of the inevitable mistakes which would occur in the course of such learning. Both of these facts would appear to have considerable survival-value, making the innateness hypothesis likely.

            In conclusion, it would seem that classical empiricists were radically mistaken in denying that our various mental faculties embody innate information about the world. Not only does the language-faculty contain information about human languages, but the visual faculty contains information about the objects in space around us. And something similar is probably true in connection with other faculties as well. Yet the innate structures of these faculties, while giving rise to innate knowledge of a rather particularised sort (such as your knowledge that a given sentence is ill-formed) should not in themselves be counted as knowledge.

 

On to chapter 7.