|
How do speakers of a language
infer the grammatical structure
of a newly heard sentence? In
exemplar-based theories of
language, this inference process
is said to consist of comparing
the novel sentence to sentences
that the speaker has already
heard, and making analogical
generalizations from the
observed partial similarities.
There have been attempts at
fully specifying and
computationally implementing
such a process, but it remains
an open question whether this is
a feasible approach, given the
fact that even relatively short
word sequences tend to occur
very infrequently in corpora. In
this talk, I propose a novel
exemplar-based grammatical
inference mechanism and present
some experiments testing its
feasibility.
|
