Re: KIF Ramesh Patil <email@example.com>
To: firstname.lastname@example.org (David McAllester)
Subject: Re: KIF
In-Reply-To: Your message of Tue, 07 Jan 92 14:23:44 -0500.
Date: Tue, 07 Jan 92 11:35:34 PST
From: Ramesh Patil <email@example.com>
This brings us back to the expressibility/tractability issue. Can I
interpret your statement as saying that full higher order is avoided
because of tractability issues? What possition does (should)
knowledge sharing standards take on the expressibility/tractability
issue? I would think that the interlingua should be maximally expressive
so that any data base can at least be expressed in the interlingua.
For example, the interlingua should allow one to exactly express the
Reimann hypothesis (suppose someone has a data base of mathematical knowledge).
Allowing the expression of arbitrary mathematics gets a little tricky,
but a simple higher order logic would handle all of "ordinary" mathematics
(including the Reimann hypothesis).
I agree with you that the interchange language should be fully
expressive. My comments about transitive relation relating to
KL-ONE family relate to the objectives of KRSS group which is trying
to come up with common language for KL-ONE like KR systems. The
hope there is that by providing a common language at-least the
systems in KL-ONE family will be able to share. The objective of
that group is to specify a conservative core language that is
tractable and a less conservative outer core language which
specifies features found in various members of the family etc.
There, the issues of expressivity/tractability are important. At
least, the group as a whole now agrees that achieving completeness
of inference over the entire KR system is unobtainable for any
interesting KR language.
In dealing with KIF the only tractability issue is one of
translation. Can two KR systems communicate with each other using
KIF language EFFECTIVELY. What the systems do (or are capable of
doing) with the content of the knowledge being transferred is up to
them and in my openion not the focus of the KIF effort.
In summary, I agree with the your argument for complete expressivity
in interchange language.