Another aspect of the theory that comes out quite unfamiliar is the theory of learnability. Here I must be still more
speculative. But the conviction has grown on me that the dominantview of Universal Grammar as a a highly complex
specification of all possible grammars—whether in terms of parameters or ranked constraints—is untenable. It does
not allow enough roo mfor the range of idiosyncrasy in language, as evidenced for instance by the variety of
constructions in English. The present framework tentatively offers the possibility of seeing Universal Grammar as a
much more limited (though hardly trivial) set of“attractor”structures that, through inheritance hierarchies, guide the
course of the child's generalizations over the evidence. Whether this proposal can be worked out depends not only on
facts fro mlanguage and language acquisition, but also—heavily—on the resolution of the questions above about the
relation of language to thebrain. At the moment, anyway, this seems to be a promisingdirection for bringing linguistic
learnability within reach of the brain.
The second half of Part II and all of Part III were spent exploring ways in which this reconfiguring of generative
grammar offers a new degree of psychological and biological relevance. A standard criticism of mainstream generative
grammar is thattheprinciplesof linguisticf-knowledge thatthecompetencetheory ascribes tothelanguage user relate
to models of language processing only in the most indirectfashion. Linguistictheory has often replied to this criticism
by hiding behind the competence/performance distinction. But if the goal is to relate language to the mind, such a
reply is less than satisfactory in the long run.
The parallel architecture, which is motivated largely on grounds internal to the competence theory, proves to offer a
more positive response to this criticism. It translates quite directly into a model of processing, one in which the
constraints on processing follow directly from the rules of grammar. The interface components play a particularly
important role in the processing model, for they are what enable a processor to relate sound and meaning
“opportunistically” or“incrementally.”In particular, the effects of modularity emerge as a consequence of which
components of the grammar and of the rest of the mind are related by interfaces, and how rich and direct such
relations are. Moreover, treating lexical items as part of the interfaces immediately brings the competence theory to
bear directly on heavily studied issues of lexical access in comprehension and production.
A less immediate but still telling criticism of generative grammar concerns evolutionary issues. Even if there is no
direct evidence for the evolution of the language faculty, it is reasonable to ask for a plausible scenario by which it
could have evolved in all its complexity. Again, mainstream generative grammar does not lend itself to incremental
evolutionary development, largelybecause theparts of language that seem likelytohaveevolvedfirst—phonology and
semantics—are