KIF review"Matthew L. Ginsberg" <firstname.lastname@example.org>
Date: Thu, 24 Mar 94 08:04:43 PST
From: "Matthew L. Ginsberg" <email@example.com>
Subject: KIF review
I passed the KIF document off to someone whom I respect, asking him to
have a look at it. His comments (much more favorable than mine would
have been!) are appended.
This'll be short (mainly because I can't type).
The language looks reasonable to me as a KR language. There are a few funny
things and misleading parts, but they are small and probably not too hard to
fix. I liked how easy it was to do defun equivalents, e.g. in the number
section. The set theory used is not very standard, but I understood what they
were after. The non-mon aspects are very simple-minded compared to current
research, while the stuff of definitions seemed needlessly complex and rococo.
But overall: not too bad at all.
I guess you mainly want to know what I think of it not as a KR language but as
*the* KR language. Like you (and many others) I see problems left, right, and
center in settling on KIF 3.0 as the language for KR work. The main reason,
as far as I'm concerned, is that a lot of this research is involved with
looking at different sorts of languages and examining their properties. KIF
3.0 has a built-in form of non-mon entailment, for example, whereas it remains
an active area of research as to what form is best for what purposes.
Similarly, it doesn't handle modalities at all except as predicates over
quoted expressions (which doesn't get you very far). And it has absolutely no
provision for representing probabilistic uncertainty of the sort being
developed by Bacchus, Halpern, and others.
But this is not to say that I know of a better language or that I have some
specific features that I think should be added. Any language I can think of
would run into similar trouble. That's not so surprising: if your research is
concerned with pushing the envelope in KR languages, you shouldn't expect that
a language that attempts to categorize the least controversial parts of KR
would capture everything you needed. But as far as it goes (with a few minor
problems), KIF 3.0 looks fine; I can even imagine putting it to useful work in
certain application areas.