Generalized quantifiers & distributed KIF
Message-id: <>
Date: Wed, 8 May 91 14:25:21 EDT
Subject: Generalized quantifiers & distributed KIF
I would be happy with Len Schubert's recommendation for generalized
quantifiers.  In defining a translation from conceptual graphs to
predicate calculus, I first translate to the syntax,

   (Q <var>:<type>) <body>,

where <type> is either a constant that is defined in the ontology or
an arbitrary lambda-expression that specifies a temporary type or sort
that has not been given a name.  For the sentence, "Most dogs that
live in New York are pampered pets," this would give you,

   (Most x:(lambda y)(dog(y) & live-in(y, New York)))(pampered(x) & pet(x)).

This isn't exactly the form I would generate from the conceptual graph,
but it is close enough that I could translate back.  But I would despair
of trying to recover this form from a low-level KIF.

Since conceptual graphs are fairly close to natural language structures,
the issues involved in translating to them are essentially the same ones
that would come up if you were trying to build a help or explanation
facility for KIF that generated answers in English.

Re distributed KIF:  Carl Hewitt's note about a KIF that was suitable
for distributed interchanges raises another interesting point.  Natural
languages are especially suited to distributed environments with heavily
context-dependent communications.  Perhaps NL features like indexicals
("you", "I", "this", etc.) may be needed in a distributed KIF.

This introduces a very large can of worms, since each interpreter would
have to maintain the equivalent of a mental model for each speaker it
talks to.  That would introduce multiple contexts where each one would
have to keep track of who knows what and why.  This might be more than
one would want to implement in the first edition of KIF, but it might
be necessary to allow for such things as future extensions.