by Noam Chomsky
Now of course these people are like Helen Keller. [Her condition] was not from birth, and nobody really knows what the effects were of that early – say year and a half – of experience. [It is true that] they have never found a successful case of someone blind and deaf from birth. So a lot is obviously going on in the first year even though nothing is being exhibited. [Nevertheless,] it can be done[; language can appear in modalities other than sound and sight]. Helen Keller did it – and she was a terrific writer.
JM: To go back to another matter we discussed last time, I asked whether canalization could be expressed in terms of parameters and the possible channels or paths for development that they provide. I take it that canalization for language would involve not just contributions by what we might call the innate biological endowment that gives us language, but also by other, non-linguistic systems.
NC: Waddington's point was that there must be architectural constraints and developmental constraints that are independent of the organism, and they function to channel the growth of the organism in particular directions. So, for example, if locality conditions or other efficient computation conditions contributed to the outcome of language – probably it doesn't have anything to do with language, or even humans, perhaps even biological organisms. That's the idea. I don't think that biologists doubt very much that something like that is going on. But how much is hard to determine.
JM: But in terms of your three sets of factors . . .
NC: That's the third factor. The choice of parameters is either the first [genetic] factor or the third [other constraint] factor; but the setting of them has to be the second factor.[C]
JM: OK.
7 Development, master/control genes, etc.
JM: Who was the person who did the interesting work on the eye and the PAX-6 gene; I forgot.
NC: Walter Gehring.
JM: Gehring in Switzerland. That kind of work might throw quite a different kind of light on the question of how a system that had Merge built into it . . .[C]
NC: His work is extremely interesting; and basically, what he shows – I don't have any expert judgment, but it seems to be pretty well accepted – is that all visual systems (maybe even phototropic plants) seem to begin with some stochastic event that got a particular class of molecules into a cell – the rhodopsin molecules that happen to have the property that they transmit light energy in the form of chemical energy. So you have the basis for reacting to light. And after that comes a series of developments which apparently are very restrictive. There's a regulatory gene that seems to show up all over the place, and the further developments, according to his account, are highly restricted by the possibilities of inserting genes into a collection of genes, which probably has only certain physical possibilities . . .
JM: the third factor . . .
NC: . . . yes, the third factor, which gives you the variety of eyes. That's very suggestive; it's quite different from the traditional view.
JM: Does it have any bearing on language?
NC: Only that it suggests that there is another system that seems to have powerful third factor effects.
JM: I've sometimes wondered about – well, take people working on Williams Syndrome children. Their brains have different morphologies – they're really quite extraordinarily different. And yet they have this amazing capacity to . . .
NC: Well, some of Eric Lenneberg's discoveries are even more dramatic, like the work he did on nanocephalic dwarfs, which is really dramatic. They have almost no cortex at all, yet almost perfect language ability.
JM: Well, it certainly throws monkey wrenches into the idea that language must be localized . . .
NC: And it just shows how crude our understanding is. But that's not too surprising. Language is the last thing we should expect to understand, because it's the one system that – for ethical reasons – you cannot directly investigate. Every other system you can investigate in other animals. Since there are no homologous structures for language, there's no comparative work. The only comparative work is on the precursors – like the sensory-motor system.
And the same is true on the conceptual side. I just don't see how you can – with our current understanding, at least – hope to get any possible insight into the evolution of the elementary concepts with their strange internalist properties. [Again] they're universal – if you go to a New Guinea native, he or she's going to have basically the same concept RIVER that we have. But we have no idea how it got that way.
JM: There are lots of just-so stories to the effect that it has something to do with evolution in the sense of selection giving advantages.
NC: But what's advantageous about having a concept RIVER that has the features we seem to be sensitive to that could have no discernible bearing on survival or selection? We can make up thought experiments about RIVER which you couldn't even imagine if you're a New Guinea native. Imagine a small phase change that turns the Charles River into a hard substance, which is apparently possible. And then you paint a line on it, and you start driving trucks on both sides of the line, so it becomes a highway and not a river. You can't explain that to a New Guinea native; none of the other notions you need to entertain the thought of a river undergoing a phase change and becoming a highway are around; so how could selection have played a role in leading us to acquire the features RIVER has that come into play when we engage in thought experiments like these, ones that lead us to declare that a river has become a highway?
In fact, the native has the same concept; if he or she grows up here or there, he or she's going to have the concept RIVER. So he or she's got it. But how could it possibly be selected? What function does it have in human life, for that matter? And since that's true of every elementary concept – take, say, Paul Pietroski's example in his recent paper about France being hexagonal and a republic. Why should we have that notion of France? It can't have any selectional role . . .
JM: That seems pretty obvious to me. Let me get back to Laura Petitto just for a moment and what she had to suggest about the way in which the STG (Superior Temporal Gyrus) looks for certain kinds of patterns. Her idea was that, at least in part, the reason we are bimodal – that we could develop without difficulty in either or both of two ways – was because you're using the same system in both cases.
NC: I suspect that there are other domains in which it could happen. Maybe you could do it in dance. I don't know, but I presume that infants would be capable of externalizing their language in dance motions – with their legs, let's say. Or perhaps any movements of your head, or eye blinking . . . In fact, people with severe paralysis.
JM: But if they did it with dance, say, they'd still require the visual system and certain kinds of patterns . . .
NC: We can't do it with smell, because we're not developed enough; we can't use taste, because we don't have the sensory range – maybe dogs could, but we can't. So you're stuck with vision and hearing. Those are the only adequate sensory capacities that we have. So everything is going to use vision and hearing, and some kind of action that we can carry out with our bodies. That's just given. OK, that leaves certain possibilities. But perhaps any possibility that makes use of those capacities will work for externalization. And she could be right; it's all going to have to fall into certain subcategories, because that's all that our brains can process. So whenever externalization comes along, as an aspect of language, it's going to have to make use of these facts about our nature. If dogs suddenly underwent a mutation in which they got Merge, maybe they'd use the sense of smell.
JM: Another fascinating aspect of her work is that she suggests that rhesus, macaques, and several other species have what seem to be perfectly homologous parts of the brain – the STG – but they just do not have this capacity to develop even the rudiments of speech or sign. Is that because they lack a language faculty in addition? Or why? Do you attribute it to some specific feature of the human STG, or . . .
NC: You could make up different stories. It could be that our hominid an
cestors lacked these brain structures, developed Merge, and then developed the brain structures. But there isn't enough time for that. The brain structures had to have been there for a long period before anything like this explosion took place. And we do know that there hasn't been anything since, because of the essential identity of people all over the world. So you've got an upper bound and a lower bound, and they seem to be awful close. So unless something entirely new comes along, the only plausible story seems to be that the apparatus was in place, for whatever reason. And maybe special adaptations like these were used for grunts; after all, you can have polysyllabic lexical items, and maybe polysyllabic lexical items were used, and maybe with the complex characteristics of human concepts, for some unexplained and unintelligible reason. But it still requires the ability to have infinite generative capacity, which apparently comes along in a flash, giving everything else.
JM: If, whatever it is, whatever that gene is that introduced Merge and carries it, if it acted something like a control gene along the Gehring PAX-6 line, it might . . .
NC: . . . it might affect the development of other things. We don't know enough about neurology to tell. So maybe some regulatory gene emerged which both gave Merge and permitted [neural systems to embody it].
JM: Fascinating speculations . . .
NC: . . . so little is known about the evolution of the brain that no one can tell.
JM: Does anyone speculate about these kinds of things . . .
NC: I don't think so, because the overwhelming assumption is that language evolved slowly through natural selection. Yet that doesn't seem at all consistent with even the most basic facts. If you look at the literature on the evolution of language, it's all about how language could have evolved from gesture, or from throwing, or something like chewing, or whatever. None of which makes any sense.
8 Perfection and design (interview 20 January 2009)
JM: I want to ask some questions about the ‘perfection’ of the language faculty. First, a background matter: if you speak of perfection and in particular perfection in design of the language faculty – or at least, the mapping to the SEM interface – you seem to be invited to answer the question, “design for what?”
NC: I think that's misleading. That's because of connotations of the word design. Design suggests a designer, and a function of the designed thing or operation. But in biology, ‘design’ just means the way it is.
JM: The structure, whatever it is . . .
NC: How is the galaxy designed? Because the laws of physics say that that's the way it's designed. It's not for anything, and nobody did it. It's just what happens under certain physical circumstances. I wish there were a better word to use, because it does carry these unfortunate connotations. In a sense – a negative sense – there's a function. If the structure were dysfunctional, it wouldn't survive. And OK, in that sense, it's designed for something. It doesn't mean it's well designed for survival. So take language and communication. Language is poorly designed for communication, but we get by with it, so it's not dysfunctional enough to disappear [or at least, disappear with regard to its use for communication, which isn't its only use, by any means]. Take, for example, trace erasure [or in the more recent terminology of copies, non-pronunciation of copies]. It's good for efficiency of structure, but it's very bad for communication. Anyone who tries to write a parsing program [encounters it] . . . most of the program is about how to find the gaps. Where are the gaps, and what's in them? If you just repeated – if you spelled out [or pronounced or otherwise exhibited] copies – the problem would be gone. But from a computational point of view, that would be poor design, because it's extra computation, so there's no point in it. So you cut it out. And there's case after case like that. So take garden path sentences and islands, for example. Islands prevent you from saying things you would like to say. You can't say, “who did you wonder why visited yesterday.” It's a thought; you know what it means. But the design of language on computational grounds doesn't allow it. To the extent that we understand them, at least, these things follow from efficient computational structure. But computational structure has no function. It's like cells breaking up into spheres instead of cubes: it just works; but if it broke up into cubes, it would work too, it just can't [because of third factor constraints on possible shapes – in this case, physical ones]. Here too I think that what you find more and more is just efficient design from a computational point of view independent of any use you might want to put it to. And I think that from an evolutionary point of view, that is exactly what should be expected. That's what these papers are about that I probably forgot to send you.
We know almost nothing about the evolution of language, which is why people fill libraries with speculation about it. But we do know something. You can roughly fix the time span. You can argue fifty thousand years more or less, but that doesn't matter; it's basically instantaneous [from an evolutionary point of view]. Something suddenly happened, and then there's this huge explosion of artifacts and everything else. Well, what happened? The only thing that could have happened – it's hard to think of an alternative – is that suddenly the capacity for recursive enumeration developed. That allows you to take whatever simple thoughts are that a chimpanzee may have, like act or action or something and turn it into an infinite array of thoughts. Well, that carries advantages. But even that is not so trivial, because Haldane, I think it was, proved – eighty years or so ago now, I guess – that beneficial mutations almost never survive. The probability of a beneficial mutation surviving is almost minuscule. It does, of course, happen sometimes, so you get some changes. But that suggests that whatever it was that gave this may have happened many times and just died out. But at some point, by some accident, the beneficial mutation survived. But it survived in an individual; mutation doesn't take place in a group. So the individual that had this property – which does carry advantages: you can talk to yourself, at least, and you can plan, you can imagine, things like that. That partially gets transmitted to offspring. By enough accidents, it could dominate a small breeding group. And at that point, there becomes some reason to communicate. And so you develop ancillary systems. You know, morphology, phonology, and all the externalization systems. And they are messy. There's no reason for them to be computationally good. You're taking two completely independent systems. The sensory-motor system has apparently been around for hundreds of thousands of years. It doesn't seem to have adapted to language, or only marginally. So it's just sitting there. You've got this other system – whatever developed internally – and there's every reason to expect that it might be close to computationally perfect, for there are no forces acting on it. So it would be like cell division. So then, when you're going to map them together, it's going to be a mess.
JM: But wait, when I think to myself, I think to myself . . .
NC: In English, yes. But that's when you think to yourself consciously. And of course, we don't know what's going on unconsciously. So consciously, yes, because that is our mode of externalization, and we reinternalize it. Here, I think, is where a lot of the experimentation going on is very misleading. There's a lot of work recently that's showing that before people make a decision, something is going on in the brain that is related to it. So if it's a decision to pick up a cup, something is going on in the motor areas before you make the decision. I think it's misinterpretation. It's before the decision becomes conscious. But lots of things are going on unconsciously. There's this philosophical dogma that everything has to be accessible to consciousness. That's just religious belief. Take mice. I don't know whether they're conscious or not, but I assume that they make decisions that are unconscious. So when we talk to ourselves, the part that is reaching consciousness is reconstructed in terms of the form of externalization that we use. But I don't think that tells you much about the internal use of language. It's evidence for it, just like speech is evidence for it.[C]
Anyhow, whatever this first person was who had the mutation, maybe the mutation j
ust gave Merge. That's the simplest assumption. If that happened, that person would not be conscious of thinking; he or she would just be doing it. He or she would be able to make decisions on the basis of internal planning, observations and expectations, and whatever. Now if enough people in the community had the same mutation, there would come a point where someone had the bright idea of externalizing it, so that they could contact somebody else. This may not have involved any evolutionary step at all. It may have [just been a matter of] using other cognitive faculties to figure out a hard problem. If you look at language – one of the things that we know about it is that most of the complexity is in the externalization. It is in phonology and morphology, and they're a mess. They don't work by simple rules. Almost everything that's been studied for thousands of years is externalization. When you teach a language, you mostly teach the externalization. Whatever is going on internally, it's not something that we're conscious of. And it's probably very simple. It almost has to be, given the evolutionary conditions.
