COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	smolen[f88,jmc]		Notes on Proper Treatment of Connectionism (PTC)
C00005 ENDMK
C⊗;
smolen[f88,jmc]		Notes on Proper Treatment of Connectionism (PTC)

1. My own commentary raised certain issues, but I forgot to discuss them.

2. Smolensky's answer doesn't address words with special pronunciations
or words that are homonyms with English but because of context
are given a foreign pronunciations.  The Prince and Pinker example
of ``high-sticked'' is even more telling.

3. Connectionism seems to deal with only one kind of brittleness ---
that which is mitigated by partial conclusions.  It doesn't deal
with lack of elaboration tolerance.

The word ``brittleness'' which doesn't have a single meaning fuzzes
up discussion.

4. Smolensky discusses the exact, complete and precise theories
that are to be sought at the subsymbolic level.  Maybe there
won't be complete theories, in the same sense that the
specification of a flipflop is incomplete.

5. Stich's notion that symbols might not exist needs refinement
to be a definite notion.  Imagine an entirely ordinary computer
running LISP, except that we replace the memory by a holographic
memory that is nevertheless addressable in the ordinary way.
There is then no localized structure in the memory corresponding
to any instruction or ordinary LISP datum.  However, the use of
such memories wouldn't correspond to any revolution in cognitive
science.  Mere lack of localization doesn't amount to much.