The way connectionism is used to segregate linguistic from non-linguistic modes of cognitive processing bothers me. Rockwell says: (p. 12 l 5 from bottom) "A connectionist translation of 'cultivate the mind of no-minedness' would be 'don't let your linguistic mental processes get in the way of your vector-transformation processes'." But surely linguistic processes are yet another brand of vector-transformation processes. Even Fodor would grant me that. His only complaint against (some) connectionists is that they think that the vector processing for language and higer thought can be done without being describable at the cognitive level as a symbolic processor. But everybody thinks that neurons do the heavy work, whether the work be computing sensations or doing exercises in symbolic logic. It is just that those who believe in LOT believe that the brain IMPLEMENTS a classical processor (by means of a species of vector transformation of course). So I do not think connectionist concepts can used directly to craft a distinction between knowledge that and knowledge how, or between the lingustic and the non-linguistic aspects of cognitive processing. The divide must be crafted, by distinguishing a brand of vector processing called symbolic processing that in turn resembles what happens in digital computers: i.e. we have discrete sematically evaluable symbols which play causal roles in the processing. We have a virtual symbolic processing machine implemented in neural wetware. Now if this is right, the distinction you want to make is between those vector processes that implement a symbolic processor and those that do not.
That is one way to look at matters. But many connectionsts think the moral of connectionsm is that we now see how to break down the distinction between clearly symbolic and non-symbolic processing. The idea is that no cognitive processing meets Fodor's criteria, though some may approach it to some degree. Here lingustic processing and perceptual processing are of a piece. From this point of view, use of connectionism to enforce a bogus divide between knowledge that and knowledge how undermines the very lesson connectionsm teaches.
For similar reasons I object to using connectionsm to distinguish the "non-inferential" from the inferential (which is presumed to be done by the help of symbolic represetnations). Again, one possiblity raised by connectionist research is that human inference is not mediated by language-like represetnations at all. On page 15 (top) Rockwell considers the possiblity that a sensation of a pink ice cube could allow us to navigate the world skillfully to put it in my drink. Another thing the same sensation might do is allow us to respond "yes" when queried "Is that icecube blue or orange?". Is it so clear that the latter way of navigating the world must be mediated by some radically "non-connectionst??" process? Why not consider the possibility that this too is the product of vector processing without the employment of explicit symbolic representation, (even though one is inclined to describe the response "Yes" as the result of a reasoning process).
Back to CQ HomepageJames W. Garson
Department of Philosophy
University of Houston
Houston, Texas 77006-3785
e-mail: JGarson@uh.edu