The Science of Language

by Noam Chomsky

  JM: If you give up the idea that you have to answer the question, what is it for . . .

  NC: It's not for anything . . .

  JM: But put it this way: don't you then have to give up also talk about interfaces, and talk about organs, because . . .

  NC: It has to relate to the interfaces, for otherwise it would die out. It would be a lethal mutation. But lethal mutations are no different from beneficial mutations from nature's point of view; they just die out. And in fact many of them remain. Why do we have an appendix?

  JM: You can't even say that it's for thought, then?

  NC: If it weren't adaptable to thought, it probably would just have died out. But functioning for something is a remote contingency; that was Haldane's point. If it's beneficial, it'll probably die out anyway, because statistically that's just what happens. But something may survive. And if it survives, it may be for physical reasons. The more that's being learned about evolution and development, the more it looks like most things happen because they have to; there's no other way. Speculations in the 1970s that suggested – at least for me – the principles and parameters approach to the study of language, such as [François] Jacob's speculations about the proliferation of organisms – well, they turned out to be pretty solid. The idea that basically there's one organism, that the difference – as he put it poetically – the difference between an elephant and a fly is just the rearrangement of the timing of some fixed regulatory mechanisms. It looks more and more like it. There's deep conservation; you find the same thing in bacteria that you find in humans. There's even a theory now that's taken seriously that there's a universal genome. Around the Cambrian explosion, that one genome developed and every organism's a modification of it.

  JM: Due to difference of timing in development, difference of gene position . . .

  NC: Yes, so it doesn't sound as crazy as it used to. They've found in the kinds of things that they've studied, like bacteria, that the way that evolutionary development takes place seems to be surprisingly uniform, fixed by physical law. If anything like that applies to language, you'd expect that the internal, unconscious system that is probably mapping linguistic expressions into thought systems at an interface ought to be close to perfect.

  JM: So language came about as a result of an accident – maybe some minor rearrangement of the human genome – and other creatures don't have it because they didn't have the same accident, at least in a form that survived . . .

  NC: In fact, the human line may have had the accident many times, and it just never took off. And the accident could have been – no one knows enough about the brain to say anything – but there was an explosion of brain size around a hundred thousand years ago which may have had something to do with it. It might be a consequence of some change in brain configuration about which people know nothing. And it's almost impossible to study it because there's no comparative evidence – other animals don't have it – and you can't do direct experimentation on humans in the way they used to do at McGill [University] . . .

  JM: To our shame . . . What happens then to the strong minimalist thesis?

  NC: Maybe it's even true. Of course, it would have to be pared down to apply just to the cognitive [conceptual-intentional, or SEM] interface, and the mapping to the sensory-motor interface may not even be a part – strictly speaking, may not even be a part of language in substantial respects – in this technical sense of language. It's just part of the effort to connect these two systems that have nothing to do with each other, and so it could be very messy, not meet any nice computational properties. It's very variable; the Norman invasion changes it radically, it changes from generation to generation so you get dialects and splits, and so on. And it's the kind of thing you have to learn; a child has to learn that stuff; when you study a language, you have to learn it. And a lot of it is probably pretty rigid. It's not that everything goes; there are certain constraints on the mapping. I think that there's a research project there, to try to figure out [just what they are]. That's what serious phonology and morphology ought to be – to find out the constraints in which this mapping operates and ask where they come from. Are they computational constraints? I think it opens up new questions. And the same for syntax. You can find some cases where you can give an argument that computational efficiency explains the principles, but . . .

  It's interesting that people have expectations for language that they never have in biology. I've been working on Universal Grammar for all these years; can anyone tell you precisely how it works [– how it develops into a specific language, not to mention how that language that develops is used]? It's hopelessly complicated. Can anyone tell you how an insect works? They've been working on a project at MIT for thirty years on nematodes. You know the very few [302] neurons; you know the wiring diagram. But how does the animal work? We don't know that.

  JM: OK. But now what happens to parameters? I guess you're pretty much committed to saying that all of the research on them should shift to focus on the mapping to the sensory-motor interface, PHON.

  NC: I guess that most of the parameters, maybe all, have to do with the mappings [to the sensory-motor interface]. It might even turn out that there isn't a finite number of parameters, if there are lots of ways of solving this mapping problem. In the field, people try to distinguish roughly between macroparameters and microparameters. So you get Janet Fodor's serious work on this. You get these kinds of things that Mark Baker is talking about – head-final, polysynthesis [which Baker suggests are among the best candidates for macroparameters]. It's probable that there's some small store that just may go back to computational issues [hence, mapping to the SEM interface]. But then you get into the microparameters. When you really try to study a language, any two speakers are different. You get into a massive proliferation of parametric differences – the kinds of stuff that Richard Kayne does when you study dialects really seriously. Very small changes sometimes have big effects. Well, that could turn out to be one of the ways of solving the cognitive problem of how to connect these unrelated systems. And they vary; they could change easily.

  JM: So you think that with the possible exception of the head and polysynthesis parameter, they're all going to have to be shifted over to the PHON mapping?

  NC: Well, but take the head parameter – it looks like the most solid of the macroparameters (reinterpreted, if Kayne is right, in terms of options for raising), although it's not really solid because while there are languages like English and Japanese where it works, a lot of languages mix them up and one thing works for noun phrases and something else with verb phrases, and so on – but even that, that is a linearization parameter, and linearization is probably in the externalization system. There's no reason why internal computation should involve linearization; that seems to be related to a property of the sensory-motor system, which has to deal with sequencing through time. So it could be that that too is an externalization parameter. The same is true of polysynthesis, Mark Baker's core parameter. It has to do with whether a sentence's arguments – subject, object, and so on – are internal to the syntactic structure, or are marked in the syntactic structure just like markers, kind of like pronouns, and they hang around on the outside. But that is also a kind of linearization problem.

  So it may turn out that there aren't any [computation-]internal parameters; it's just one fixed system.

  JM: What happens then to parameter setting?

  NC: That's the problem of language acquisition, and a lot of it happens extremely early . . .

  JM: As Jacques Mehler's work indicated . . .

  NC: All the phonetic stuff, a lot is going on before the child even speaks.

  JM: Familiarization with the native tongue . . .

  NC: It's known that Japanese kids lose the R/L distinction before they can even speak. So some kind of stuff is going on there that is fine-tuning the sensory apparatus. The sensory apparatus does get fine-tuned very early, in other areas too.

  JM: So in principle, it's possibl
e that you don't have to set (‘learn’) any parameters – that it all happens automatically and at an extremely early age, even before the child speaks.

  NC: Certainly no kid is conscious of what is going on in his or her head. And then you get a three- or four-year-old child who is speaking the language mostly of their peers.

  JM: OK. How does early and automatic acquisition fit with the kind of data that Charles Yang comes up with, data that suggest that when children ‘grow’ a language, they go through a stage at around two and a half where they exhibit a kind of parameter-setting experimentation: their minds ‘try out’ computational patterns available in other languages that are extinguished as they develop a pattern characteristic of, say, English . . .

  NC: There is interaction, but it's not so obvious that the feedback makes much difference, because most of the interaction is with children.

  I don't know about you, but my dialect is [that of] a little corner of Philadelphia where I grew up, not my parents’ – which is totally different. What about you?

  JM: I grew up speaking both English and Tamil.

  NC: How's that?

  JM: I was born in the southern part of India.

  NC: Did your parents know Tamil?

  JM: My father did; he learned it by squatting with kids on the floors of their schools.

  NC: Did they speak Tamil at home?

  JM: No, they didn't. But some of my friends spoke Tamil.

  NC: So you picked up Tamil from your friends, from other kids. That's normal. No one knows why, but children almost always pick up the language of their peers. And they're not getting any feedback – certainly not teaching. The parents may be trying hard to teach you something, but all they do is teach you artificialities [irregularities]. So it looks like a tuning problem. It works in other things too. There are styles of walking. If you go to Finland – Carol and I noticed as soon as we were there – they just walk differently. These older women carrying shopping bags racing down the streets; we could barely keep up. It's just the way they walk. People just pick that up.

  I remember once when Carol and I were walking down the streets in Wellfleet [Massachusetts] one summer and Howard Zinn was walking in front of us, and right next to him was his son, Jeff Zinn. And the two of them had exactly the same posture. Children just pick these things up. If people really studied things like styles of walking, I'm sure that they'd find something like dialect variation. Think about it: you can identify somebody who grew up in England just by mannerisms.

  JM: Assume so, then what gets put into the lexicon in the way of phonological features?

  NC: Well, as we both agree, a lot of what ends up in the lexicon comes from inside. Nobody's conscious of it, nor can be conscious of it. It's not in the dictionary . . .

  JM: We hope it's accessible to some theory, surely.

  NC: It has to be; there has to be some kind of theory about it, if you're going to understand it at all. As far as I know, we can't go much beyond the seventeenth century on this. It looks like they found a considerable amount of what we can be aware of. [But of course, that has nothing to do with what scientific theory can reveal.]

  So it [that is, the question of what ends up in the lexicon] is a topic, but it's not going to be investigated until people understand that the externalist story [about language and its sounds and meanings] just doesn't get anywhere. Until people understand that it's a problem, it'll just not be investigated.[C]

  Some of the stuff that is coming out in the literature is just mind-boggling. Do you look at Mind and Language?

  JM: Yes . . .

  NC: The last issue has an article – I never thought that I would see this – you know this crazy theory of Michael Dummett's, that people don't know their own language, etc? This guy is defending it.

  JM: Terje Lohndal [a graduate student in linguistics at the University of Maryland] – he and Hiroki Narita [a linguistics graduate student at Harvard] – wrote a response to it. I think it's good; I don't know if it will be published. I hope so. [See Lohndal & Hiroki 2009.]

  Is there anything you want to add about design?

  NC: Well, the main thing is, we've got to find another term, because it's just too misleading. And it's true for biology altogether. In biology, people aren't usually misled by it, even though the connotations are there. Well, maybe some of them are misled. So if you read, for example, Marc Hauser's book on the evolution of communication – which is a very good book, and he's one of the most sophisticated people working in biology – well, if you read through the chapters, there's almost nothing about evolution there. The chapters are discussions of how perfectly adapted organisms are to their ecological niche. A bat can pick out a random mosquito far away and go right after it. And that shows that animals fit their ecological niche. The assumption [in the background] is, of course, that that's because of natural selection; that they evolved [to fit their niche]. [But the book] doesn't say anything about evolution [– about how it took place in these specific cases]. [So far as the discussion of the book is concerned,] a creationist could accept it: God designed bats to be able to catch mosquitoes. But that move is very fast. To try to demonstrate anything about evolution is extremely hard. Richard Lewontin has a paper coming out on this, about how difficult – just on the basis of population genetics – about what it would take for natural selection actually to have worked. The way it looks, it seems to be a really remote possibility.

  JM: Jerry Fodor is against selection too . . .

  NC: But he's against it for other reasons concerned with intentionality, and that kind of stuff about what something is for. His instincts are right, but I think that's the wrong line to take. You don't ask whether a polar bear is white for surviving or for mating, or something like that. It just is, and because it fits the environment, it survives. That's why people like Philip Kitcher and others go after him.

  Do you think that there's anything else to say about design?

  JM: No, although I'm sure that discussion of the topic will not end there.

  9 Universal Grammar and simplicity

  JM: OK, now I'd like to get clear about the current status of Universal Grammar (UG). When you begin to focus in the account of acquisition on the notion of biological development, it seems to throw into the study of language a lot more – or at least different – issues than had been anticipated before. There are not only the questions of the structure of the particular faculty that we happen to have, and whatever kinds of states it can assume, but also the study of how that particular faculty developed . . .

  NC: How it evolved? Or how it develops in the individual? Genetically, or developmentally?

  JM: Well, certainly genetically in the sense of how it came about biologically, but also the notion of development in a particular individual, where you have to take into account – as you make very clear in your recent work – the contributions of this third factor that you have been emphasizing. I wonder if that doesn't bring into question the nature of modularity [of language] – it's an issue that used to be discussed with a set of assumptions that amounted to thinking that one could look at a particular part of the brain and ignore the rest of it.

  NC: I never believed that. Way back about fifty years ago, when we were starting to talk about it, I don't think anyone assumed that that had to be true. Eric Lenneberg was interested in – we were all interested in – whatever is known about localization, which does tell us something about what the faculty is. But if it was distributed all over the brain, so be it . . .

  JM: It's not so much the matter of localization that is of interest to me, but rather the matter of what you have to take into account in producing an account of development. And that seems to have grown in recent years.

  NC: Well, the third factor was always in the background. It's just that it was out of reach. And the reason it was out of reach, as I tried to explain in the LSA paper (2005a), was that as long as the concept of Universal Grammar, or linguistic theory, is understood as a format and
an evaluation procedure, then you're almost compelled to assume it is highly language-specific and very highly articulated and restricted, or else you can't deal with the acquisition problem. That makes it almost impossible to understand how it could follow any general principles. It's not like a logical contradiction, but the two efforts are tending in opposite directions. If you're trying to get Universal Grammar to be articulated and restricted enough so that an evaluation procedure will only have to look at a few examples, given data, because that's all that's permitted, then it's going to be very specific to language, and there aren't going to be general principles at work. It really wasn't until the principles and parameters conception came along that you could really see a way in which this could be divorced. If there's anything that's right about that, then the format for grammar is completely divorced from acquisition; acquisition will only be a matter of parameter setting. That leaves lots of questions open about what the parameters are; but it means that whatever is left are the properties of language. There is no conceptual reason any more why they have to be highly articulated and very specific and restricted. A conceptual barrier has been removed to the attempt to see if the third factor actually does something. It took a long time before you could get anywhere with that.

  JM: But as the properties of language become more and more focused on Merge and, say, parameters, the issue of development in the particular individual seems to be becoming more and more difficult, because it seems to involve appeals to other kinds of scientific enterprise that linguists have never in fact touched on before. And I wonder if you think that the study of linguistics is going to have to encompass those other areas.

  NC: To the extent that notions such as efficient computation play a role in determining how the language develops in an individual, that ought to be a general biological, or maybe even a general physical, phenomenon. So if you get any evidence for it from some other domain, well and good. That's why when Hauser and Fitch and I were writing (Hauser, Chomsky & Fitch 2002), we mentioned optimal foraging strategies. It's why in recent papers I've mentioned things like Christopher Cherniak's work [on non-biological innateness (2005) and on brain wiring (Cherniak, Mikhtarzada, Rodriguez-Esteban & Changizi 2004)], which is suggestive. You're pretty sure that that kind of result will show up in biology all over the place, but it's not much studied in biology. You can see the reasons. The intuition that biologists have is basically Jacob's, that simplicity is the last thing you'd look for in a biological organism, which makes some sense if you have a long evolutionary history with lots of accidents, and this and that happens. Then you're going to get a lot of jerry-rigging; and it appears, at least superficially, that when you look at an animal, it's going to be jerry-rigged. So it's tinkering, as Jacob says. And maybe that's true, and maybe it isn't – maybe it looks true because you don't understand enough. When you don't understand anything, it looks like a pile of gears, levers, and so on. If you understood enough, maybe you'd find there's more to it. But at least the logic makes some sense. On the other hand, the logic wouldn't hold if language is a case of pretty sudden emergence. And that's what the archeological evidence seems to suggest. You have a time span that's pretty narrow.