phrase structure, but it is violable. For instance, English gerundive noun phrases such asJohn's walking the dog last night
(as inJohn's walking the dog last night was a mistake) have no head noun; rather they contain a verb phrase, headed by a
verb (Jackendoff 1977: ch. 9).
This is roughly the Head Constraint on argument structure discussed in section 5.9.4. It says that the head X of a
syntactic phrase XP expresses a semantic function F, and its syntactic argument(s) YP (and optional ZP) express the
semantic argument(s) W (and optionally V) of F; moreover, the whole syntactic phrase expresses the combination of
the semantic function and its arguments.
Many other candidates suggest themselves as structural fragments that might act as UG“attractors”for structural
patterns of languages. Among the more elaborate ones are the syntactic“tricks”such as long-distance dependencies
(therelationbetween a wh-phrase or topicalized phrase at the front of a clause and its trace elsewhere in the sentence)
and the other major types of violation of the Head Constraint enumerated in section 5.9.4.
More generally, the content of UG boils down in large part to (a) architectural universals—what kinds of structure
there are in the grammar and how they interface, and (b) what particular fragments of structure (or“tricks”) are
prespecified to bias the child's acquisition of generalizations. These piecesdidhave to evolve in the history of the
species; it is hoped that their evolutionis more plausible than that of a standard syntactocentric“grammar box.”(We
return to the question of evolution in Chapter 8.) In addition, the present approach permits us to make sense of the
“markedness”properties of Universal Grammar, in terms of independent characteristics found in concept formation
in other areas of cognition.
I must confess that at this point I hear the linguists screaming: There's not enough machinery! How can this be
constrained? He's sold out to the connectionists! And I hear the network modelers screaming: There's too much
machinery! How can all this be built in? Will those linguists ever learn? I can only respond that the position here is far
from a weak-kneed compromise: it is an attempt to make the most sense I can out of the genuine insights of both
sides. Both classes of objections indeed must be met; I take this as another challenge for future research.
Above all, what I would like the reader to take away fro mthis discussion is the i mportance of psychological
considerations in thetheory of grammar. By takingvery seriously thequestionof what is stored and what is computed
online, we have managed to justify a major reorganization of the theory of grammar. If this is letting the theory of
performance intrude on the theory of