Of Minds and Language (40 page)

This is the general picture that many people assume at present. Now, as there are many more parameters than we originally thought, it turns out that the different parametric choices will enter into various complex kinds of interactions,
generating many possible configurations of properties, so that the superficial diversity to be expected is great. Nevertheless, the deductive interactions between principles and parameters still are quite tight, so that there are many logical possibilities that are excluded even in a system which has a richer parametric specification of the kind I am describing.

I would like to conclude with a brief discussion of the reanalysis that Guglielmo Cinque (2005) proposed of one of the universals that Joseph Green-berg (1963) had identified in his very important work in the sixties. Greenberg had observed that if you look at a variety of languages, you notice that certain elements that enter into the structure of the nominal expressions can vary in order, although there are limits to order variation. If we limit our attention to cases in which the noun is either at the beginning or at the end of the string of modifiers, we basically find three types. One type is realized by English and by the Germanic languages in general, where the order is demonstrative, numeral, adjective, noun (Dem Num Adj N) giving something like:

(8)    these three nice books

One also finds quite a few other languages in which the order is the mirror image: N Adj Num Dem. Thai has that property, so a noun phrase in Thai has the order

(9)    books nice three these

– an exact mirror image to English. Then, by restricting our attention to cases in which the noun is either final or initial, a third case that is found, instantiated by the African language Kikuyu, is N Dem Num Adj, like English except for the fact that N is at the beginning of the string:

(10)    books these three nice

Apparently, we never find the fourth logical possibility given this pattern, that is to say, a language which would be like Thai, with a mirror-image order of adjective, numeral, and demonstrative, but with the noun in final position (*Adj Num Dem N):

(11)    *Nice three these books

Now Guglielmo Cinque (2005) has shown that this systematic gap can be derived from very reasonable computational principles. Just in a very simplified manner, what we can say is that we can take the Germanic order as the basic order. So (8) – demonstrative, numeral, adjective, noun – is the initial, first-merge order. Other orders can then be derived by Move, but movement is always driven by movement of the noun, so that the noun may move alone, and then you get a structure like
(10), with the same order of elements as in English except that the noun has moved stepwise to initial position. Or you have another possible instance of movement, which some linguists have called Snowballing Movement. The noun moves step-wise, but at each step it pied-pipes the whole structure it has moved to, a procedure which ends up producing the mirror-image effect. In this case, you start with something like the English order, you move the noun to the left of the adjective, and now you take the newly-created constituent, noun plus adjective, to the specifier of the numeral, and so on. If you repeat this movement a number of times, you obtain the exact mirror image of the Germanic order. But there are no other possibilities. Particularly, one cannot get the order in (11) because the noun is in final position in this case, which indicates that the noun has not moved, but noun movement is the engine of the whole process, so in the absence of noun movement the order cannot be subverted. In this case there is simply no way to get the reversal of the order with respect to the basic order. Cinque shows that the gap observed by Greenberg is not an exception, it follows from reasonable principles of linguistic computation. Following this model, it may be possible to give principled explanations to much important empirical work within the typological tradition. In conclusion: there are more parameters than previously assumed, because parameters are properties of functional heads, and the inventory of functional heads is rich, particularly if the cartographic view is correct. Still, deductive interactions between principles and parameters are tight, and therefore the attested patterns of variation are only a fraction of the logical possibilities.

HIGGINBOTHAM: In relation to Luigi's point (after Cinque), you can easily derive the fact that you can say
these three nice books
but not
books nice three these
just from compositionality – you know, just from a hierarchy. It's not clear to me that we need anything else.

C
HOMSKY
: Part of the sequence just comes, independently of precedence and c-command, from the composition (presumably D and NP). So the D is going to remain outside anyhow, and then what is left is just the relation between
three
and
nice
. And here there seems like a fairly clear semantic property. I mean,
nice books
are a kind of books, but
three books
aren't a kind of books. There is an old paper by Tom Bever from years ago on adjectives,
²¹
where he tried to argue, with some plausibility, I think, that there is a kind of squishiness in adjectives and some of them are more noun-like. For instance,
red
can be a color, whereas
nice
can't be a something, and he argued that the more noun-like ones tend to be closer to the noun. So these kinds of considerations could be the answer to the
three nice
order, in which case you'd get the ordering.

R
IZZI
: Okay, so suppose you can derive the hierarchy from the needs of semantic compositionality and some related factors, as Jim and Noam suggest. This gives the Germanic order
These three nice books
. What about the other permissible orders? And the impossible one? Take the mirror image order
Books nice three these
: this could also be a direct reflection of compositionality on external merge, but here the syntactic assumptions you make become crucial. Suppose we adopt Kayne's antisymmetry, which rules out a structure like
[[[[books] nice ] three ] these ]
: then, within Kayne's system there must be a computational procedure (snowballing movement) deriving this order from the basic order. Consider now the order
Books these three nice
: here, basically under anybody's assumptions, you need movement of N (or NP) to derive this particular order. And then you must make sure that the movement computation, which is needed anyhow, does not overgenerate, and can't give rise to the unattested order *Adj Num Dem N, a fact that Cinque plausibly tries to derive from the assumption that only N can move in this configuration (possibly pied-piping some other material), so if N doesn't move, there is no way to alter the basic order Dem Num Adj N. So, Cinque's point is that under reasonable assumptions on the fundamental hierarchy of projections in nominal expressions and on possible movement processes, one can derive the typological facts. This approach looks very plausible to me.

Then the question arises which is raised by your remarks: where does the initial hierarchy come from? Here I think it is entirely plausible that the hierarchy is grounded in semantics, that the requirements of compositional semantics impose certain orders and are inconsistent with others. The cartographic endeavor tries to determine what the functional hierarchies are for different kinds of expressions across languages, what varies and what remains constant. As far as I can tell, this is fully compatible with the attempt to trace the observed hierarchies to the interpretive considerations raised by Chomsky and Higginbotham. In fact, in my opinion, the cartographic projects and results invite such efforts to provide “further explanations” in terms of interface requirement.

Rochel Gelman

I usually start my presentations on this topic by asking the members of the audience to participate in an experiment. I show them slides with a pair of items and ask them to rate their similarity using a scale of 1 to 10, where 1 is,
Couldn't be less similar
, and 10 is,
Very, very similar
. Their task is simply to call out a number that reflects how similar they perceive the pair of stimuli in the slide to be. A sample stimulus pair is presented in
Fig. 15.1
.

As expected, they normally rate the pairs as very similar, presumably because they look very much alike on the surface. Then I inform them that the items in the slide were taken in two different places. One of the pair was taken at a zoo, and one was taken on the shelf of a store that specializes in fine ceramic copies. Now, with this as background information and a mindset that distinguishes these environments, I ask them to rate the pair of items again. This time the adult audience also does as expected: they now rate the exact same pair of stimuli as very dissimilar, switching from the top end of the similarity scale to the bottom end of it.

Let us turn now to what 3- and 4-year-olds do when they are shown the zoo and store pictures. When a child comes into the room, he finds the experimenter on her knees, surrounded by forty-two pictures, taken of twenty-one pairs of real and fabricated animals. She tells the child that she just dropped her pictures and asks if they will help to put the zoo pictures in the zoo book, and the store pictures in the store book. The child is then given the items, one at a time. Both
age groups do this extremely well. They do not fall for the overall surface similarity as might be expected given any Piagetian, stage, or association theory about preschool competence. According to such theories, preschoolers are perception-bound. If so, our young subjects should treat pairs that are perceptually very similar on the surface as the same. Therefore their placements should be at chance. But they are not. In fact, in one such study (Gelman and Brenneman 2004), 67% and 100% of the 3- and 4-year-olds, respectively, turned in performance that met a criterion of p < .026. For the children to succeed on this task, they had to be able to look for details in the photographs of the live and fabricated version of the same kind that provided clues regarding their different ontological categories. But to do this, they had to have available a framework providing hints as to what constitutes relevant information for animate as opposed to inanimate objects.

Fig. 15.1. Photographs of dogs that are similar on the surface, although one is of an animate and the other of a fabricated dog. An example of displays used in Gelman and Brenneman (2004).

Results like the above have led me to the view that there is a core domain which involves a high-level causal–conceptual distinction, one that makes principled distinctions between the nature of relevant energy sources for the movements and transformations of animate and inanimate separably moveable objects. For inanimate objects to move or be transformed, there has to be a transfer of external energy. Although animate objects obey the laws of physics, their particular motion paths and transformations are due to the generation of energy from within. I have dubbed these the Innards-Agent and External-Agent principles (Gelman et al.1995). The idea is that the children benefited from an implicit, abstract causal framework, which informs the kind of perceptual information they take to be relevant and therefore salient for descriptions of similarity and actions. Thus, the framework provides input about what kind
of data are relevant to each sub-domain, in this case, cues for biological/living or inert things. The cues include ones that are relevant to the potential actions on the one hand, and potential functions, on the other hand. That is, the possible forms and details of each kind of object are part of implicit skeletal “blueprint” characterizations of the two ontological kinds.

Further evidence for this view was obtained in Massey and Gelman (1988). Children aged 3 and 4 were asked whether a series of objects could move themselves up and down a hill or whether they needed help. The objects all were novel. They included vertebrates and invertebrates, wheeled objects, statues that represented and shared parts of mythical human or animal creatures, and complex inanimate objects that resembled stick-like human figures. No graduate students could tell us what they were. Neither could the 3- and 4-year-olds, who successfully told us which objects could move by themselves both up and down a hill. What these young children said was most informative, as illustrated in the following sections from our transcripts.

Experimenter:
Could this (a statue)
[go up the hill by itself]?

Child:
No
.

Experimenter:
Why not?

Child:
It doesn't have feet
.

Experimenter:
But look, it does have feet!

Child:
Not really
.

In her own way, this child was telling us that the statue was not made up of the right kind of stuff. Another child told us that a statue was just a furniture statue, again an example from an inert category.