lee@uhccux.uhcc.hawaii.edu (Greg Lee) (02/23/89)
From article <Feb.22.15.20.26.1989.931@elbereth.rutgers.edu>, by harnad@elbereth.rutgers.edu (Stevan Harnad):
" ...
" (5) The Modularity Assumption:
" Searle implicitly adopts a strong, untested "modularity" assumption to
" the effect that certain functional parts of human cognitive performance
" capacity (such as language) can be be successfully modeled
" independently of the rest (such as perceptuomotor or "robotic"
" capacity). This assumption may be false for models approaching the
" power and generality needed to pass the Turing Test.
This seems to me correct, except I'm not sure we could say that the
modularity assumption for language is untested. The construction of
(putatively) complete grammars has been attempted, and since none have
come close to correctly describing a natural language, the evidence
that's in suggests the assumption is false.
On the other hand, the proposal or conjecture found elsewhere in
Stevan's discussions that finding a way to ground the symbols will lead
us somehow to a better theoretical understanding is unlikely to be
correct. I think. In saying why, I'd prefer the terms 'syntactic' for
the symbol manipulation approach and 'semantic' for grounding symbols
(but without intending to imply that theories customarily called
'semantic' are properly so called).
A reasonable way to rate the prospects of an analytic approach is to ask
(and answer) the question: what has it helped us find out? Looking at
the score for the last few years, and sticking to fundamental
discoveries, I make it syntax: 3, semantics: 0. The discoveries are:
(1) Movement constraints (Haj Ross) -- constituents cannot occur
"too far" from where they belong,
(2) Cross-over (Paul Postal) -- nominals cannot come on the wrong
"side" of coreferents,
(3) One per sent (Charles Fillmore) -- when nominals are classified
by role (agent, patient, ...) one finds at most one of each
role represented per clause.
(Disclaimer: probably few linguists would agree with my scoring.)
My conclusion is that semantics as currently conceived has not
gotten us anywhere, and probably never will.
Greg, lee@uhccux.uhcc.hawaii.edu