Of Minds and Language (68 page)

Thomas Hobbes argued that part of the core properties for looking at the world was characterizing things in terms of their origins, so you identified a river by its origin, or a constitution by its origin. Locke, far from being a caricature of an empiricist, assumed extremely rich cognoscitive powers. Relevant to us, his most significant contribution to this domain is, I think, his analysis of persons. Our concept of person is based critically on psychic continuity; it is continuity of the mind that individuates persons, it's not anything a physicist can find. And even what we would call science-fiction experiments – two minds in the same physical body and that sort of thing – go as far back as Locke. Some of the basic issues go back to Aristotle. He identified a house, let's say, as a combination, in his terms, of matter and form. Matter is sticks and bricks and so on, but there's also form: design, function, and standard use. It's a combination of the two. It's important to note that Aristotle was giving a metaphysical definition; he was not defining the word “house,” he was defining houses. And that leads to hopeless conundrums that go right through the history of philosophy: one which came up was the ship of Theseus, a modern version of
which is Kripke's puzzle.
⁵
And if you give a metaphysical interpretation to these things you run right off into impossible conundrums.

What began to happen in the seventeenth century was that these problems were restated as being essentially epistemological or cognitive rather than metaphysical, and it just turns out that our concepts don't apply in many cases. So the Ship of Theseus is simply a case where our concepts just don't give an answer.
⁶
And why should they? They're not supposed to answer every possible problem that comes up. The thing is still amusing to look at, but it is no longer a paradox or a conundrum. They didn't actually draw that conclusion but they should have. It ends up that the investigation of cognoscitive powers – which is quite a rich theory of meaning and still remains unexplored (and going back to Jim's lesson, it ought to be explored) – led finally to a quote of Hume's: the objects that we talk about are really objects of thought which are constructed by mental operations, and there is no peculiar physical nature belonging to them.
⁷
You can't identify them by some identifiable, mentally independent property. As in Locke's example, psychic continuity is not a mentally independent property. I don't know if this has been studied but we all know that infants have no problem with this. In fact, children's literature is based on these notions. In the standard fairy tale the handsome prince is turned into a frog by the wicked witch, and he is to all extents and purposes a frog until the beautiful princess kisses him and suddenly he's a prince again. The infant knows that he was the prince all along and it didn't matter if he looked like a frog. Locke's notion was much too narrow: it's not persons, it's almost anything.

My grandchildren have a favorite story about a baby donkey named Sylvester who is turned into a rock. For most of the story Sylvester the rock is trying to convince his parents that he is their baby donkey. And since children's stories always end happily – that's a law of nature – something happens at the end of the story and he turns back into the baby donkey and everybody's happy. However, the children know that the rock, which may be a rock by any physical test, is actually Sylvester because there's a psychic continuity running through it.

So Locke's distinction between person and man doesn't work; it goes to maybe anything organic, maybe beyond. But it's the typical case of some
semantic or conceptual property that is impossible to identify in material terms. And Hume's conclusion is, I think, plausible. When you look at case after case you find more and more that that's exactly the way it is. And it does mean (and I'll come back to this) that there simply is no notion of reference in natural language. There is in other language-like systems, but natural language is a biological object and we can't stipulate its properties, and one of its properties seems to be that it doesn't have reference. I'll come back to that.

Alongside the core domains, to get back to Rochel Gelman, there's also what she called HoW – “hell on wheels” (see page 226). So there's another kind of domain that has none of the properties of the core domains: you have to really work on it, people's talents differentiate, it's slow, its understanding is developed over generations, it's transmitted, and so on. It's analogous in the domain of physical abilities to, say, walking versus pole-vaulting. When you go to the Olympics there's a pole-vaulting championship but there's no championship for walking across the room. And that's because everybody can do it; it may seem very easy but trying to figure out what's going on might be very hard. Or, say, reaching for something; it's extremely hard to figure out what's going on but there's no competition for it because that's a core domain. Pole-vaulting on the other hand is for freaks; very few people can do it and ability is spread all over the place, so that's why it's a game or sport. In fact, games and sports are precisely those things that people are no good at. That's one of the reasons that I've always felt that the cognitive scientists and the artificial intelligence people were barking up the wrong tree when they started to study chess, because that's something that's for freaks – like Jim Higginbotham (who notoriously is a very good chess player), but not normal people. Normal people can figure out the moves, but if you want to have championship abilities spread enormously then it's like pole-vaulting. If you want to understand an organism, you look at the core domains not the freaky things at the edges. So chess is the wrong thing to look at, just as pole-vaulting would be the wrong thing to look at if you were trying to figure out the organization of motor skills.

A first approximation to the structure of the cognitive system, and it seems to me a reasonable one, is the core domain versus “hell on wheels.” We investigate these topics using capacities which allow us to carry out considered reflection on the nature of the world. It is given various names in different cultures. It's called myth, or magic, or in modern times you call it science. And they're all different but they're all sort of like that: there's some considered reflection on what's going on. It's very hard and there are all the other problems that we know about, but that's a first break.

If you look at something like Marc Hauser's father – you may remember his description of how he gave the wrong result (that is, the right result) on one of
the trolley problems (see page 325) – Marc Hauser's father was, I presume, not working in the core domain where you give your intuitive reaction to what you would do. Instead, he was thinking about it and saying that in a sense here is what you
ought
to do, which is called the wrong reaction. This suggests something about that whole topic – in fact any topic like it – which is reminiscent of things like the classic case of garden path sentences (see
Chapter 22
). When somebody is presented with a garden path sentence, they instantly get the wrong interpretation and they may say it's not a sentence because they're doing it the wrong way. They can't tell you the processes that they're using; they are unconscious, inaccessible to consciousness (there's that same inaccessibility to consciousness). It's in a certain sense wrong, whatever that means, to get the right answer with considered reflection. If you set up a certain context and the usual things, then they'll say “Yeah, there's that other interpretation.” It's a kind of performance–competence distinction, and I think all of these topics ought to be looked at that way, including the morality cases. Maybe Marc's father – for one of the cases but not the others – was in the sort of state which Jack Rawls,
⁸
who started this stuff off, called reflective equilibrium: not your immediate reaction but the interpretation you give when you think about it, interact with others, you figure out what your understanding and your ideas really are, and so forth. That's a distinction to keep in mind.

One of the core domains, as we get narrower, must be language. It has all the properties that Rochel Gelman talked about. To quote her again, “mental structures actively engage the world”:
⁹
it's reflexive; it just happens without effort; it happens in exactly the same way in everyone – pathology aside; and as some evidence brought up here shows, within two days the infant has picked some of the data in the world and has decided that this is linguistic data. And as far as I know that's a pretty tricky operation. I don't know if anyone's tried a computational theory to figure out how that's done, but I suspect it wouldn't be easy to figure out how with all this mess of stuff going on in the world you decide that's linguistic data and some of it clearly isn't. But apparently this happens by two days, and even more has happened, since, as you've heard, the child is differentiating different types of linguistic data.

That's the quick development of a complex Umwelt. And one reason that no non-human animal is ever going to have anything like language is that it can't get over that first step. It's just data for your parrot or your songbird or ape or whatever. It's just not making that distinction so nothing can happen after that. What we mostly study is what happens after that – after you've taken
that first step, what do you do with it? And that's the rest of the core domain. The first step is tricky enough, and I think that's one reason why ape language studies are pointless from the start. There are a lot of other reasons why they're pointless, but even from the first moment they're pointless.

After that comes the growth of the system. It's called acquisition. If we go back to our discussions here, when the terrain was laid out, Lila Gleitman talked about word meaning, Janet Fodor talked about the computational system, and Núria Sebastián-Gallés talked about parameters. These are three big issues that come up in the nature of the growth of the system, which is called language acquisition.

Starting in reverse order, why parameters? The question came up again and again in our discussions here. Mark Baker, as you heard, had a proposal (see page 95). I was telling Massimo, I thought Baker was joking, frankly, when he gave that proposal, but maybe not. It's kind of amusing but it can't be right, for the reasons that Massimo said. It is true that there's evidence from cultural anthropology that groups distinguish themselves from other groups by arbitrary practices. A famous example is Jews and Muslims not eating pork. Apparently the only reason for that, that anybody can figure out, is that they do so to differentiate themselves from other people around. There are no health reasons or anything like that. And there are plenty of examples like that, so the idea that that could happen is possible, but Baker's hypothesis is logically incoherent for the reasons that Massimo mentioned – you have to have the parameters before you can use them to differentiate yourselves from others.

A second plausible proposal is one that Donata Vercelli and Massimo mentioned: there is some kind of a minimax operation going on (see page 101). This would take you back to third-factor properties – optimization properties – and the intuition is that if you take a parameter and you genetically fix the value, it becomes a principle, it moves from the domain of parameters to principles. To spell this out is not so simple, but from a certain point of view, when you add the value, you're adding genetic information. Try to work that out, it's not so trivial. There's a way of thinking about it in which it gives more information if you give the value than if you leave it open. So adding parameters is reducing genetic information from this point of view. On the other hand, it's making language acquisition harder because you have to find the value of the parameter. So you can at least imagine that there's a nice theorem out there waiting for somebody to prove which says that the choice of parameters maximized for both of these contradictory tendencies does the best possible job – it's a mini-max problem. So those of you who are looking for Ph.D. dissertations or maybe Nobel prizes might try to figure that out.

There's another possibility, which I don't think should be ruled out. As far as we understand, the overwhelming weight of the parameters, almost all of them, are on the morphological/phonological side – that is, they are in the mapping from the syntax/semantics to the sensorimotor system. I mentioned before that I think there's very good evidence that this is a secondary process, and this came along later in evolution. First people sort of learned how to think, and then later when there were enough of them they somehow tried to figure out a way to externalize it. Externalization is a very non-optimal sort of process. You have two systems that have nothing to do with each other. One of the systems has evolved for basically the semantic interface (the conceptual thought system), and maybe it turned out to be almost perfect and provide a perfect matching to that system. Maybe it's even tautologous in that this system reads off it. The other system is the sensorimotor system, which has just been sitting around there for a long time. And somehow you've got to get them together and there are a lot of ways of doing this. Here's another nice theorem waiting to be proven. The ways in which this is done are optimal. That is, if you take the very messy systems of phonology and morphology, maybe they turn out to be the best possible ways of handling this very difficult problem. Conceivably, it's a long-term project.

Another possibility arises from the fact that the anthropological evidence doesn't tell us much about this group of people who underwent this amazing change. Presumably it was a small group, and small hunter-gatherer tribes tend to separate. Often pretty quickly they split up into very small groups. This means that they may not have much contact with one another – remember that it's all happening within a very small window of evolutionary time, maybe 50– 100,000 years or so. There could have been enough differentiation so they started externalizing independently, and if they externalized independently they might have just solved the problem independently. And then later on they get together again and it looks as if there're a lot of languages and here are the parameters. If that's true then there wouldn't be too much rhyme or reason to the choice of parameters;
¹⁰
it would be partly historical accident – “here's the way we tried to do it and here's the way those guys tried to do it,” they entangle them all and it looks like a system of parameters. And so I think there's every option open from a perfect solution to a minimax problem to a worst possible solution, which is one damn thing after another. Anywhere in there could be some kind of answer to this question. I think it's an interesting question. And then there's Janet Fodor's possibility as she explains in her paper
(see
Chapter 17
): maybe some of them are hidden and we never see them because we just can't get to them; you just pass them by in the lattice.