COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub[let,jmc]" source
C00006 ENDMK
C⊗;
.require "let.pub[let,jmc]" source;
∂CSL Professor John Haugeland↓Philosophy Department
↓Pittsburgh University↓Pittsburgh, PA∞

Dear John:

	Thanks for your "An Overview of the Frame Problem".  I have
the following comments.

	1. I doubt that many of the language translation people were
as dogmatic as Weaver.  By the way it seems to me that Weaver was
more a publicizer than a creator of statistical information theory.

	2. You have gotten into technical ground in your discussion
of resolving the ambiguity of "pen".  The point is that there is
much scope for invention in finding the relevant facts in large
databases, and I
think some of the methods the computation linguists and AI people have devised
will work quite well in the "pen" case.
Perhaps some of them have already told you that.

	For example, pursuing both meanings at once will lead
quickly to a scenario in the play pen case and to inconclusive
results in the other.  The use of hash tables of associations
can quickly get the relevant facts even among millions of
irrelevant facts.  If you are not familiar with hash coding,
you will find it surprising.

	The general point about translation often requiring
understanding has become accepted except by the most recalcitrant
Chomskyans.  As to what understanding is, there is no accepted
dogma.

	The frame problem has been tackled in several ways in AI,
none of which do you precisely describe.  It seems to me you
invent your own approach.  My own approach is via circumscription,
and it is now generally agreed that some form of non-monotonic
reasoning should be invoked.  I enclose a draft of a paper that
will be given at the October conference on non-monotonic reasoning.
I believe that the level of logical technicality of that paper
is essential for even a philosophical discussion of the AI
approach to the frame problem.

	I hope you will continue formulating problems for AI
to try to solve.  Try to find the simplest examples that seem
difficult to you.  The lack of sharp challenges, clearly expressed,
is one reason people keep making programs with the same weaknesses.

.reg