文档介绍：This excerpt from
Modularity in Knowledge Representation and Natural-Language
Understanding.
Jay L. Garfield, editor.
© 1991 The MIT Press.
is provided in screen-viewable form for personal use only by members
of MIT .
Unauthorized use or dissemination of this information is expressly
forbidden.
If you have any questions about this material, please contact
******@..
4
Context Effects in Lexical Processing:
A ConnectionistApproach to Modularity
Michael K. Tanenhaus , Gary S. Dell, and Greg Carlson
Our aim in this chapter is to bring together two controversial ideas in
cognition : modularity (the claim the processing is divided among autono -
mous subsystems) and connectionism( an approach to cognitive modeling
that emphasizesthe parallel spread of activation among neuronlike pro-
cessing units ).!
To say the least, modularity and connectionism are strange bedfellows .
They are both broad claims about cognitive structure, but beyond that most
researcherssee the two approachesas either orthogonal or antithetical.
The view that modularity and connectionism are orthogonal stems from
the observationsthat connectionism, with its emphasison quasi-neural ele-
ments, is a claim about microstructure whereas modularity is a claim about
macrostructure. The view that the two approachesare in opposition is due
to the fact that connectionist modelers often have mitment to interac -
tive (that is, nonmodular) treatmentsof processing. Spreadingactivation , in
which diverse sources of information are translated into mon coin
of activation , is a natural mechanism to handle interaction (McClelland and
Rumelhart 1981). Claims that the structure of cognition is highly modular
are therefore claims that the connectionists ' most potent mechanisms are
only minimally useful.
Here we challengeboth the idea that connectionismand modularity are
irrelevant to each other and the view that they are patible. We will
argue that co