rapaport@sunybcs.UUCP (04/07/87)
STATE UNIVERSITY OF NEW YORK AT BUFFALO
GRADUATE GROUP IN COGNITIVE SCIENCE
THOMAS G. BEVER
Department of Psychology
University of Rochester
THREE PARADIGMS FOR THE STUDY OF COGNITION AND LANGUAGE
Three current metaphors for the study of language behavior are the modu-
lar, the nodular, and the linguistic. I argue for a particular version
of the linguistic model.
Evidence for the modular model depends on the fact that language is com-
posed of distinct levels of representation, the units of computation of
which are incommensurable. The output of the computations can be
matched as wholes. Thus, it is the quantal nature of the representa-
tions that guarantees modular-like behavior, not the architectural
``impenetrability'' of the modular processes. Several experiments sup-
port the argument against architectural modularity. The nodular model
is most strongly instantiated in current connectionist treatments, which
have the virtues of rich computational power and the failings of associ-
ationism. The crucial problem is that they cannot explain the structure
underlying performance. These models are a kind of structural ``clay''
that conform to the structure of language, but that cannot explain why
the structure exists and why it is the way it is. For that, we have to
turn to linguistic investigations.
The linguistic metaphor takes the problem for performance models to be
that of relating linguistic structure and behavior. There have been two
kinds of postulated relations between grammars and performance: direct
(the grammar is the performance model) and indirect (the grammar defines
structures that the perfromance model must compute). Direct models are
contaminated by the need to explain how abstract structures are related
to real concepts and physical signals. That relation requires that
every model in fact be indirect, involving some kind of assumption about
how grammar and behavior are linked. Although current theories tend to
ignore this requirement, it is implicit in each theory, and probably
wrong, too. Examples will be drawn from the assignment of antecedents
to referents, access by different kinds of anaphors to different mental
representations of their antecedent, and the priming of antecedents by
empty categories.
Monday, April 27, 1987
3:30 P.M.
Park 280, Amherst Campus
Informal discussion at 8:00 P.M. at David Zubin's house, 157 Highland
Ave., Buffalo. Call Bill Rapaport (Dept. of Computer Science, 636-3193
or 3181) or Gail Bruder (Dept. of Psychology, 636-3676) for further
information.
COMING ATTRACTION:
JOHN HAUGELAND, UNIV. OF PITTSBURGH, ``UNDERSTANDING AND PERSONALITY,''
APRIL 23, 4 P.M., KNOX 4