Of Minds and Language (11 page)

Another difference – it is kind of an intuitive argument, but a powerful one – is that Ken Hale (whose intuition was better than any other human being I've ever known) thought that external arguments didn't belong inside the VP. That
doesn't sound like a very convincing argument for people who don't know Ken Hale, but he had kind of like a God-given linguistic intuition. Anyway, there is enough information around aside from that to suggest that it is something we don't understand about where external arguments fit in. And if that case is out, then every case of what we call headedness just follows from a minimal search operation, which would mean that what we have to say is, “This is correct.” I agree with you about the decomposing, but we should decompose the unbounded Merge operation into the fact that essentially everything is (head, XP). That looks special to language, but then that has plausible sources, like in theta-theory and in the cartographic approach, which adds the rest of the stuff.

B
OECKX
: For me the Ken Hale argument is of course, given where I come from, a powerful one. So I am happy you agree with me that decomposing Merge, regardless of how we do it, is an important next step. Some of the things you said actually illustrate a few of the things that Marc Hauser and I have been running into, namely translating, for example, theta-theory, or notions like external arguments, or even head vs. XP – this is actually the hard part for the next step, i.e. testing the FLN/FLB distinction. Because how do we do, for example, theta-theory independently of the very specific linguistic structures that linguists know for sure, but people like Marc do not, or at least do not know how to test yet? That is the hard part. Similarly for notions like external arguments, or even XP – how do we go about testing that? But if you agree about the next step, about decomposing Merge (no matter how we do it), that is one point that I wanted to make.

P
IATTELLI
-P
ALMARINI
: I have a question for Noam. You say the status of the head emerges somehow. So for example, if I have “red wine,” how do I put together “red” and “wine”? It seems that “wine” is the head. What is the phenomenon there?

C
HOMSKY
: Well, first of all, it is not really true that we put together “red” and “wine.” We put together an XP, which is an adjective phrase, and it could be “very red” or, you know, “formerly red,” or “redder than this,” or whatever. It just happens that the case that you gave is a reduced XP, but in fact it is an XP. So we are putting together the XP (“formerly red,” or “redder than that”) with a head, a noun, so that is a head-XP relation. And in fact just about everything you look at is a head-XP relation. We sometimes mislead ourselves, because we select as the XP something which is in fact a head, but that is just a special case. For example, that is why “many” cannot be a determiner. You can't have a determiner “many” because it could be “very many” or “more than you thought” or something like that, so it really is an XP. You look through the
range of structures, and they are almost entirely head-XP. The only exceptions that I know of are internal Merge, which has reasons to be different, and then has the interesting property that Caterina Donati noticed,
¹⁰
that if it is a head, it behaves differently – the thing that is extracted, and the external argument. That is the sticking point, both for NPs and for the sorts of clauses, and in both cases it has exceptional properties, which makes one think that there is something else going on there.

P
IATTELLI
-P
ALMARINI
: So what about EPP, the Extended Projection Principle?

CHOMSKY
: The Extended Projection Principle remains, in my opinion, simply mysterious. Actually, since Tom Bever is not here,
¹¹
I can maybe speak for him. He was going to give a paper with a proposal about that, and it is an interesting proposal. I don't understand exactly how to make it work, but it is a different take on the matter. The EPP is the one that says that every sentence has to have a surface subject, so for example in English you cannot just say *
Is a man in the room
; you have to say
There is a man in the room
. You have to put in a fake subject to make it look like a subject, and as a matter of fact that is a source of EPP. It is English. Now I think there is a kind of historical accident here. The first language studied in any depth was English, and English happens to be one of the very rare languages that has an overt expletive. It just is not common. Almost no language has them, and in the few languages that do appear to have them, like Icelandic, it is a demonstrative and only appears in special cases. Most of the time you don't put it in at all. And then there is an argument about whether it is really a specifier of T or whether it is somewhere in something like Luigi Rizzi's left periphery, but the point is that it is very rare. Well, when people started looking at null-subject languages, they kind of modeled it on English, and they assumed that since there is no subject (you don't hear it, if it is a null subject), there must be a null expletive because then you get EPP. But suppose there isn't a null expletive. There is really no strong evidence for it that I know of. It just satisfies EPP. So maybe EPP is just wrong, just some idiosyncrasy of English, which we could look into.

Well that suggests a different way of looking at null-subject languages, but then comes Tom Bever's proposal. I don't feel right about giving it, because I'm probably not doing it the way he would have done it, had he been present, but what he is arguing is that there are for every language what he calls
“canonical sentence forms,” of a kind that are sort of standard, the things that you are most likely to hear, especially a child,
¹²
like
John saw Bill
, or something, and these canonical sentence forms are simply different for different languages. For VSO languages, they are different in that you don't hear any subjects. There may be one in Irish sometimes, but it is not the canonical sentence form. For null-subject languages the same. You don't typically hear Subject Verb Object, because they have a different canonical sentence form. Then what he argues is that there is a kind of general learning procedure of some sort that utilizes the canonical sentence form and sort of forces the other forms to look like the canonical sentence form. So in English you would stick in this pointless expletive to make it look like the canonical sentence form. When you look at the proposal in detail, it is hard to work out, because there are plenty of sentences in English …

P
IATTELLI
-P
ALMARINI
: He thinks that EPP is linked to a general cognitive strategy.

C
HOMSKY
: It is a general cognitive strategy, coming from generalizing from canonical sentence forms. It is pretty tricky to get it to work out, because, say, English has many sentences without subjects, like every yes/no question, for example. But still, there is something there that I think is attractive.

G
LEITMAN
: Yes, I think it is very attractive too, but there is this little problem, that if you look at what an [English] input corpus looks like, it is 10 percent Subject Verb Object, but I'm only counting 10 percent of the things you would say in
sentences
. A whole lot of it is just noun phrases. So let's just take the cases that are sentences. If you look at a corpus from a mother to kids aged 0 to 3, only 10 percent of the sentences are SVO. Imperatives and questions, that's what it is. “Shut up,” and “Would you shut up” – that's what most of it is.

C
HOMSKY
: I'll answer in the way Tom would answer, I think.
¹³
He has talked about it and I don't know the numbers, but I think what he would say at this point is that the child knows that some things are not declarative sentences, and they are constructing their canonical sentence form for declarative sentences. That is the attractive part of the argument; then come the nuts and bolts that make it work.

G
LEITMAN
: Yes, the nuts and bolts are not the reasons I study it, but I think it is a very attractive hypothesis and also I think it is probably true.

P
IATTELLI
-P
ALMARINI
: Something like that seems to come out with Broca's aphasics – some such strategy where they use a canonical order and they seem to pay attention to the canonical order. When it is inverted they are lost.

G
ELMAN
: Yes, in languages where the subject is not first, there are people who have predicted that verbs would be preferred, and it turns out not to be the case.

C. R. Gallistel

By way of prelude, I make a rapid – and necessarily superficial – tour of familiar philosophical terrain, because the material on animal cognition that I then review has substantial bearing on long-standing philosophical issues of relevance to contemporary cognitive science.

In this epistemology, the newborn mind knows nothing. But it has the capacity to experience elemental sensations and to form associations between those sensations that recur together. Thus, all representation derives from experience: “There is nothing in the mind that was not first in the senses” (Locke 1690). The mind's capacity to associate sensations makes it possible for experience to mold a plastic mind to reflect the structure of the experienced world. Thus, concepts derive their form from the form of experience. The farther removed from sensory experience a concept is, the more derived it is.

In this epistemology, our concepts of space, time, and number are maximally derivative. They are so far removed from sensory experience that they do not seem to have sensory constituents at all. Nor is it clear how their highly abstract, essentially mathematical form can be derived from experience. Neither the nature of the relevant experience, nor the inductive machinery necessary to derive them from that experience are in any way apparent. And yet these abstractions seem to play a foundational role in our representation of our experience.

Kant famously responded to this puzzle by arguing that the empiricists were wrong in attempting to derive our concepts of space, time, and number from our experience of the world. On the contrary, Kant argued, these organizing concepts are a precondition for having any experience whatsoever. We always represent our experiences, even the most elementary, as ordered in time and localized in space. The concepts of time and space are not derivable from our experience; rather, they are the foundation of that experience.

Descartes famously argued that the machinery of the brain explains unmindful behavior. But, he argued, some behavior – behavior informed by thought – is mindful. He further argued that the operations of thought cannot be the result of mechanical (physically realizable) processes. He was among the originators of a line of thought about mind in human and non-human animals that continues to be influential, not only in popular culture but in scholarly and scientific debate. In its strongest form, the idea is that only humans have minds. In its weaker form, it is that humans have much more mind than non-human animals. A corollary, often taken for granted, is that the farther removed from humans an animal is on the evolutionary bush, the less mind it has. The most popular form of this idea in contemporary thought is that animals, like machines, lack representational capacity. Therefore, abstractions like space, time, number, and intentionality do not inform the behavior of non-human animals.

The popularity of the view that non-human animals know nothing of time, space, number, and intentionality owes much to the lingering effects of the behaviorism that dominated scientific psychology until relatively recently, and that still dominates behavioral neuroscience, particularly those parts of it devoted to the investigation of learning and memory. The more extreme behaviorists did not think that representational capacity should be imputed even to humans. Radical behaviorism fell out of favor with the rise of cognitive psychology. The emergence of computers, and with them, the understanding of the physics and mathematics of computation and representation played an important role in the emergence of contemporary cognitive psychology. The fact that things as abstract as maps and goals could demonstrably be placed into the indubitably physical innards of a computer was a fatal blow to the once widespread belief that to embrace a representational theory of mind was to give up the hope of a material theory of mind. The realization that a representational theory of mind was fully compatible with a material theory of mind was a
critical development in scientific thinking about psychology, because, by the early twentieth century, a theory of mind that made mind in principle immaterial was no longer acceptable in scientific circles.

By the early twentieth century, the progress of scientific thought made Descartes's concept of an immaterial mind that affected the course of events in a material nervous system unacceptable to the great majority of scientists committed to developing a scientific psychology. The widespread belief in a uniquely human mind did not, however, die with the belief in a materially effective immaterial mind. Rather, the belief in a uniquely human form of mental activity came to rest largely on the widely conceded fact that only humans have language. If one believes that language is
the
(or, perhaps, a) medium of thought, then it is reasonable to believe that language makes possible the foundational abstractions. One form of this view is that it is language itself that makes possible these abstractions. Alternatively, one may believe that whatever the unique evolutionary development is that makes language possible in humans, that same development makes it possible to organize one's experience in terms of the foundational abstractions.