lOOKS
459
advances from
Syntactic Structures.
All three are in part highly technical,
but the hypothesis they articulate has a general importance that can
scarely be exaggerated.
Although generative grammar, like any rich theory, suffers from
resume treatment, a brief sketch can give some idea of the stakes.
Conceive a grammar, then, as a theory of a particular language,
analogous to the "theory" that a speaker carries in his head. For a
start it must specify ("generate") the sentences of that language, and do
so with a finite number of rules, since the speaker's brain is finite.
Yet every speaker can speak, recognize, and understand infinitely many
sentences; the grammar too must possess this creative capacity. Here
the mode of syntactic analysis proposed by descriptive linguists falls
short, because, as Chomsky showed, one cannot specify all the sentences
of a language through a manageable inventory of elements. One can,
of course, account for all the sentences in a
corpus,
but doing so con–
stitutes a fairly unexciting achievement, since the challenge in writing
grammars is precisely
to
anticipate
new
sentences.
Furthermore, a grammar must provide each ·sentence with a
structural analysis that corresponds to the way in which speakers in–
tuitively analyze it. For instance, the grammar will have to indicate
that the grammatical subject differs from the logical subject in "The
books were written by a Swede," that the two parallel sentences
John is easy to please
John is eager to please
have very different underlying structures, that "I had three books
stolen" is structuraIIy ambiguous, and so on. It is necessary to abandon
the requirement that structural categories be tied concretely to the data,
and to posit a set of abstract rules (called "transformations") that relate
surface structures to deep structures. The deep structures alone express
significant relationships, and
it
is upon them that the semantic rules
operate to yield meanings. But the phonological rules, which indicate the
sound of each sentence, work on surface structures. The picture, then,
is of a three-part grammar that mediates, through syntax, between the
sound and the sense of infinitely many sentences, in a way that explains
how finite human beings do the same astonishing thing.
It is crucial to notice that generative grammar takes as part of its
data the "linguistic intuition" of the speaker-what he
knows
(though
he cannot describe his knowledge explicitly) about traffic rules within
any sentence. Correspondingly, Chomsky insists that the focus of linguistic
investigation is not the speaker's performance, which will incorporate
all the irregularities that attend on haste and human fallibility, but his
