notations within linguistic theory don't allow us to extract just the non-redundant parts of idioms and irregular
morphology as distinct lexical entries. Hence we cannot codify their“cost”in terms of number of symbols on the
page, the traditional methodological criterion for generality. I have therefore tentatively suggested that, as far as
traditional linguistic notation is concerned, these sorts of relation are tacit or implicit; they are a consequence of how
memory for related items is organized in the brain. I am given some encouragement in this conclusion from the
behavior of some connectionist models, but much remains to be worked out, both within linguistics and within
neuroscience. Yet again, though, the same problems accrue to the taxonomicorganization of concepts, so this is not a
proble mpeculiar to idio ms and irregular verbs.
What we have learned about the brain fro mstudyinglanguage, then, is how to fra me the hard questions more sharply.
I don't think that's so bad. My guess is that these questions are likely not going to be answered in terms of language
alone. If they are answered through experiments on, say, a pigeon's homing behavior plus some creative combination
ofbrainrecording andcomputationalmodeling,that'sfinetoo. Butlanguagegivesus a benchmark ofuneliminableand
well-understood problems that make clear what is at stake.
ThegoalofPartIIwastomovelinguistictheoryitselfcloser tothepossibilityofexplanationinterms ofthebrain. Not
by reducing it or denying any of its complexity—quite the opposite: by embracing its complexity and reorganizing its
architectureaccordingly. Thecrucialinsightcomes initiallyfrom phonology. Phonologicalstructuresincethelate1970s
has been viewed as composed of a number of independent generative subcomponents, each of which is responsible
for its own characteristic for mof structure ortier. The tiers are linked by interface rules that establish not one-to-one
correspondences among the structures, but rather partial homologies in which mismatches of structure are to be
expected.
Here we have extended this approach to the whole of language. Each of the major components of linguistic
structure—phonology, syntax, and semantic/conceptual structure—is the product of an independent generative
system, and each is further subdivided into independent tiers. The notion of a generative system is of course nothing
new here—even within the individual components. What is new, I think, is the explicit recognition that this is the way
the grammar is put together globally.
The key to formulating the parallel generative architecture more clearly is a full-blown theory of the interface
components: how an interface component is structured in general, exactly what interface components are necessary,
and what role interfaces play in the determination of grammatical structure. In particular, all the usual problems of
learnabilityand balanceofpower among componentspertaintointerfacesas wellas tothegenerativecomponents. IfI
have