COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "let.pub[let,jmc]" source
C00004 ENDMK
C⊗;
.require "let.pub[let,jmc]" source;
∂CSL Professor Patrick Winston↓Artificial Intelligence Laboratory
↓545 Technology Square↓Cambridge, MA 02139∞

Dear Pat:

	Thanks for your %2Learning by augmenting rules and accumulating censors%1.
It indeed involves a form of non-monotonic reasoning.  It resembles
a formalism I have thought about but didn't really work out until
I read your paper.

	Mine works as follows:

1. Sentences are labelled, and the labels have parameters corresponding
to the free variables in the sentences.  An example is

Lbird(x):	bird(x) ⊃ can-fly(x)

2. The conclusion of a sentence can be an inhibition.  An example
of a suitable sentence is

Lostrich(x):	ostrich(x) ⊃ inhibit(Lbird(x)).

3. If a sentence is inhibited for particular arguments, then we
don't use it to draw conclusions.  Thus if we have  bird(Joe)  and
ostrich(Joe), we won't conclude  can-fly(Joe).  My inhibitions play
the role of your censors and "unless clauses".

4. Cross inhibitions between sentences can lead to difficulties in
interpretation, and I would be inclined to ban them unless