with altogether. (More detailed discussion of this point appears in Gleitman and Wanner 1982) and Shatz 1982.)
To besure, Batesand Elman endbysaying“Evenifwe assume that a brain...containsno innateknowledge at all,we
havetomakecrucialassumptionsaboutthestructureofthelearningdevice,itsrateand styleof learning,and thekinds
of input that it‘prefers’to receive.”Still, it is my impression that many advocates of such arguments either are not
awareofor explicitlywishtodenythecomplexityof linguisticstructure; and a less complexstructurenaturallyrequires
a less elaborate learning theory. As in Chapter 1, I insistthat we cannot adequatelyassess theories of language learning
without understanding the character of what is learned: Fig. 1.1 represents a bare minimum that all linguists (not just
unreconstructed Chomskyans) agree upon.Visionseems intuitivelysimpletoo, yet no onein cognitivescience believes
any more that the visual system has a simple structure or that the brain just“learns to see”without any specialized
genetic support.
It is useful to put the proble mof learning more starkly in ter ms of what I like to call the Paradox of Language
Acquisition: The community of linguists, collaborating over many decades, has so far failed to come up with an
adequate description of a speaker's f-knowledge of his or her native language. Yet every normal child manages to
acquire thisf-knowledge by theage of tenor so, without reading any linguistics textbooks or going to any conferences.
How is it that in some sense every single normal child is smarter than the whole community of linguists?
The answer proposed by the Universal Grammar hypothesis is that the child comes to the task with some f-
preconceptions of what language is going to be like, and structures linguistic input according to the dictates (or
opportunities!) provided by those expectations. By contrast, linguists, using explicit reasoning—and far more data
fro mtheworldthanthechild—havea muchlarger designspacein whichtheymust localize thecharacter ofgrammar.
Hence their task is harder than the child's: they constantly come face to face with therealpoverty of the stimulus. No
child has to decide the sorts of issues we have been sketching here: whether there is a separate prosodic tier in
phonology; whether or not there are derivationalrules; whether such-and-such a phenomenon belongs to derivational
rules, lexical formation rules, or interface constraints; and what kind of a rule is responsible for Noun Incorporation.
And surely no child has to choose among major architectural alternatives such as GB, LFG, HPSG, Cognitive
Grammar, OT, and many other yet-to-be-devised alternatives. The Universal Grammar hypothesis supposes that at
some level(wemight say intuitivelyor instinctively),thechildf-knows theright choices. It may take children a whileto
sort the phenomena out, but, as stressed in Chapter 2, they all come up with essentially the same solution.