Of Minds and Language (14 page)

C
HOMSKY
: Randy's comment sort of suggests Liz Spelke's experiment,
¹
i.e. using language for intermodal transfer (visuo-spatial, for instance).

G
ALLISTEL
: You're right, it does seem to, but in fact I'm not sympathetic to that. I don't agree with Liz on the interpretation of those experiments, but what I said does seem to point in that direction.

G
ELMAN
: I'd like to modify what Randy said, to say that what seems to be unique to humans is a representational capacity. Language is one that can be used for a wide range of activities, but notational capacities are also representations. Drawings can be representations, plans, and so forth – there are many options. And I have yet to see data that animals can go invariably from one representational format to another.

P
ARTICIPANT
: It's only a simple question. Do the systems of communication of bees and birds display feedback? For example, if they make a mistake and then realize that they've made a mistake, do they communicate it?

G
ALLISTEL
: Ahhhh [scratches head; laughter]. That's tough! Sort of implying that as a result, where the bees that are following the dance consult their map, sort of implying that they conclude that the dancer didn't know what the dancer was talking about, right? [Chuckles to himself.] Because if the information conveyed by the dance is sufficiently inconsistent with the information on their map, they appear to discount the information in the dance. I'm not sure whether that isn't correcting themselves, of course. I'm not sure this is relevant, but there are recent experiments by Laurie Santos,
²
one of Marc's many good students, who has gone on to do work that Marc has also done on observing the mind sort of thing, where you have to represent whether the other animal knows what you know, in order to choose. This has been a big issue for a long, long while. But I thought her recent experiments, which I cannot reproduce (I'm sure Marc can, as they were partly or mostly undertaken with Marc)
were very persuasive on that score. Part of Marc's genius has been to exploit naturalistic circumstances, and they exploited naturalistic circumstances in a way to make a much more compelling case that the animal knew that the other animal didn't know X.

P
ARTICIPANT
: I was wondering if you have feedback when you have something similar to negation. It is usually claimed that negation is unique to human language …

G
ALLISTEL
: Ohhhh, like where the catcher in a baseball game shakes off the signal? I can't quickly think of a clear example that one could regard as equivalent to negation. But negation is certainly a kissing-cousin of inversion, and animals invert all the time. I mean, they invert vectors, right? Not only do they calculate the home vector themselves when they are out there and they have found food, but when they get back, what they are dancing is not the vector they calculated coming home, but the inverse vector, the vector for going the other way. About negation, I always remember that tee-shirt that says, “What part of No don't you understand?” [Laughter]. It seems to me about as elementary as you can get.

P
IATTELLI
-P
ALMARINI
: Concerning foraging, I have seen work by my colleague Anna Dornhaus, concerning some of the optimal criteria that honeybees meet in foraging,
³
which is rather astounding, because they have constructed a graph of how many bees are proactive (they go out and look for food) versus the reactive foragers that wait for the dance. So they have calculated the percentages of proactive versus reactive, and the graph you get depends on how long the food is available. And you have a triple point like in second-order phase transitions in physics and chemistry. It's extraordinary. They have a number of predictions that sound very weird, but then they observe them in nature or in the laboratory. So it seems that, when we approach foraging in a quantitative way, among other things, it is one of those fields in which the species seem to be doing the best thing that they could possibly do. Have you any comments on that, because it is a question of great current interest in linguistics. It wouldn't be the only case in which you have biological systems that are doing the best that can be done.

G
ALLISTEL
: Yes, this question of optimality is apt to provoke very long arguments in biological circles. I can give you sort of a general view, and then my own particular view. If you look on the sensory side, you see spectacular optimality. That is, sensory transduction mechanisms are, most of them, very near the limits of what is physically possible. So the threshold for audition, for
example, is just above the threshold set by physics – there's a slight vibration on the eardrum due to the fact that on a small surface there is stochastic variation in how many molecules of air hit that surface, and that produces a very faint vibration in the eardrum that is an ineliminable noise in the system. And the amount of additional vibration that you need from another source is just above that limit. The most essential thing is to calculate how much the eardrum is moving at that threshold. It is moving less than the diameter of an atom! So that's a lot better than you would have thought at the beginning.

Similarly with the eye. One of the proofs before it was directly demonstrated that the absorption of a single photon by a single rhodopsin molecule in a single rod generated a signal that could make its way all the way through the nervous system came from a famous experiment by Hecht, Shlaer, and Pirenne in which they showed that there was a clearly detectable effect.
⁴
This was subsequently studied by Horace Barlow and Barbara Sakitt,
⁵
and they showed that for every quantum or photon of light absorbed, there was a quite sizeable increase in the probability that a human would say that he had detected the flash. There are ten million rhodopsin molecules in the outer segment of a single rod, and there are a million rods in the retina. So it is a little bit like one of these huge soccer matches and someone burps and the referee says, “Who burped?” There are a hundred million spectators and somehow the burp is centrally detectable. That's pretty impressive.

There is wide agreement about this – the facts are extremely well established. When you come to computational considerations, that is where the arguments begin, but of course that reflects the fact that we, unlike the sensory things, don't know what's going on. Most neuroscientists think that the computations are just one spike after the next, right? But this seems to me nonsensical. Any engineer will tell you that the contradictions that follow the transduction of the signal are more important than the transduction in the first place. That is, if you've got a good signal but lousy signal processing, then you've wasted your time producing a good signal. So it seems to me that the pressure to optimize the computations is at least as great as the pressure to optimize the signal transduction, and we know that the signal transduction is very near the limits of what is physically possible. So I tend to think that the computations, or processing of the signal, are also at the limits of what is computationally possible. But since we know practically nothing about how the nervous system computes, it's hard to say.

Marc D. Hauser

I want to begin by saying that much of what I will discuss builds tremendously on the shoulders of giants and couldn't have been done if it hadn't been for the thinking and experimental work of people like Noam Chomsky, Randy Gallistel, and Rochel Gelman, who significantly inform what I will be telling you about. Today I want to develop an idea of a new research path into the evolution of language, which I'll call “evolingo,” parasitizing the discipline known as “evo-devo,” and I will tell you a little about what I think the label means. Then I want to give you a case example, some very new, largely unpublished data on quantifiers. Finally, what I will try to argue is that there is really a new way of thinking about the evolution of language that is very different from the earliest stages of working on this problem.

Definitionally, what I want to do is anchor thinking about this in terms of viewing language as
a mind-internal computational system designed for thought and often externalized in communication
. That is, language evolved for internal thought and planning and only later was co-opted for communication. This sets up a dissociation between what we do with the internal computation as opposed to what the internal computation actually evolved
for
. In a pair of papers that we published a couple of years ago (Hauser et al. 2002; Fitch et al. 2005) we defined the faculty of language in the broad sense (FLB) as including
all the mental processes that are both necessary and sufficient to support language
. The reason why we want to set up in this way is because there are numerous things internal to the mind that will be involved in language processing, but that need not be specific to language. For example, memory is involved in language processing, but it is not specific to language. So it is important to distinguish those features that are involved in the process of
language computation from those that are specific to it. That is why we developed the idea of the faculty of language in the narrow sense (FLN), a faculty with two key components: (1) those mental processes that are unique to language, and (2) those that are unique to humans. Therefore, it sets out a comparative phylogenetic agenda in that we are looking both for what aspects are unique to humans, but also what aspects are unique to language as a faculty.

Evolingo, then, is a new, mostly methodological, way of thinking about the evolution of language, whose nature can be described in terms of the three core components described by Noam Chomsky in his opening remarks here and in his recent work (Chomsky 2005b) – that is, the system of computational rules, semantics or the conceptual intentional system, and the sensorimotor or phonological system, and their interfaces. What the evolingo approach then puts forward is that we are looking for the study of mind-internal linguistic computations, focusing on those capacities that are shared, meaning both in terms of homologies (traits that have evolved through direct, common descent) as well as homoplasies (traits that have evolved largely from convergence or independent evolution, but arise due to responses to common problems), looking at those aspects that are unique to humans and unique to language as a domain of knowledge.

The real change with the prior history of work on the evolution of language is that it focused almost entirely on non-communicative competencies, using methods that tap both spontaneous capacities as well as those that involve training. I want to make just one quick point here, because I think some of the work that I have done in the past has confused this. Much of the work in animal learning that has gone on in the past has involved a particular kind of training methodology that, by its design, enables exquisite control over the animal's behavior. In contrast, much of the work that we have done in the past ten or so years has departed, not intellectually, but I think methodologically, from prior approaches by looking at what animals do spontaneously, in the
absence
of training, as with an experiment that Tecumseh Fitch and I did. We did not train the animals through a process of reward or punishment to show what kinds of patterns they can extract. We merely exposed them, passively, in much the same way that studies of human infants proceed.
¹
We are trying to use very comparable methods to those used with human infants so that if we find similar kinds of behaviors, we can be more confident about not only the computation, but how it was acquired and implemented. I'll pick up on these points later in the talk.

So the two very important empirical questions that I will address in a moment are: (1) to what extent are the conceptual representations that appear to uniquely enter into linguistic computation built from nonlinguistic resources; and (2) to what extent have linguistic conceptual representations transformed in evolution and ontogeny some of our ontological commitments? The reason why I think this is important, and the reason why I think the evolingo change in approach has been important, is that almost all the work at a phylogenetic level that has addressed questions of interest to linguists about the nature of language, language structure, and computation, has looked almost exclusively at the communication of animals, either their natural communication or what we can train them to do with sign languages or symbols. What it has generally failed to do, except in the last few years, is to ask about the computational capacities that may be seen in completely different domains and never externalized. This is why, in my first paper with Noam and Tecumseh Fitch (Hauser et al. 2002), we made the analogy that some of the computations that one sees in language may well appear in something like spatial navigation – the integration of spatial information that Randy elegantly described in his talk (see
Chapter 4
) about the notion of landmarks and bearings. Those kinds of computations may have some similarity to the kinds of computations we see in language.

A couple of examples of how I think the structure of the questions has changed in the field, away from questions like “Can animals vocalize and refer to things in the world?” or “Do animals have any syntactic structures?” to other kinds of questions. I think in terms of conceptual evolution there are two issues, one having to do with the nature of animal concepts. And here I will just take the lead from Randy's elegant work, and argue that in general, the way that people in the field of animal cognition have thought about them is exactly the way that Randy describes,
²
namely as isomorphisms or relationships between two distinct systems of representation. Critically, and as Randy describes (I'm not going to go through this, although interestingly we picked out the same terms), they seem to be abstract, not necessarily anchored in the perceptual or sensory experiences for things like number, space, time, and mental states. Importantly, there seems to be virtually no connection in animals, perhaps with the exception of honeybees (which is why I asked that question),
³
between the sensorimotor output of signaling and the richness of the conceptual systems they have. Notice there is nothing remotely like a word in animal communication. I take it to be the case that what is debated in the field, and I think what
should be of relevance to people working in language, are the following issues: the details of the format and content of the representations in animals; how the language faculty transforms the conceptual space; and lastly, whether there are language-specific conceptual resources. And it is really the latter question that I want to address today.