Typed KIF

• Mail folder: Interlingua Mail
• Next message: David McAllester: "KIF "
• Previous message: Ramesh Patil: "Re: KIF "
• Reply: Tom Gruber: "Typed KIF"

Message-id: <9201071706.AA27740@quark.isi.edu>
To: interlingua@isi.edu
Subject: Typed KIF
Date: Tue, 07 Jan 92 09:06:34 PST
From: Robert MacGregor <macgreg@isi.edu>

I would suggest that a facility for translating an
arbitrary KIF database into a typed logic version of
KIF database be developed.  Ramesh has indicated that
a part of a KIF to LOOM translator involved
scanning the KIF to build up a set of definitions,
since LOOM requires that predicates be defined before
they appear in statements of fact.  LOOM does not have
a typed logic, but I would imagine that KR systems
that do would have even more stringent translation
requirements.

Therefore, I am proposing that a typed KIF database
would place the definitions ahead of other sentences,
and it would indicate the types of all quantified
variables.  Thus, the builder of a KIF-to-Typed Brand X
translator could save some effort by having to write
only a Typed KIF-to-Typed Brand X translator.

There are other possible transformations that would
make KIF easier to translate out of.  For example,
if the typed KIF also guaranteed that standard math
operators (+, *, etc.) appeared with only two input
arguments (rather than a variable number), that 
would contribute another small reduction in the
effort needed to write a translator.

In general, the KIF current philosophy seems to be to make
it easy to translate INTO KIF, at the expense of making
it harder to translate OUT of KIF.  Inventing a stricter,
and thus friendlier version of KIF, plus developing
a (single) translator from full KIF into the restricted
KIF, would reduce the effort required for a KR system
to build a KIF interchange facility.

- Bob