system = logic + control
pfps@allegra.tempo.nj.att.com (Peter F. Patel-Schneider)
Message-id: <9008131527.AA25649@vaxa.isi.edu>
Date: Mon, 13 Aug 90 11:27:22 EDT
From: pfps@allegra.tempo.nj.att.com (Peter F. Patel-Schneider)
To: dam@AI.MIT.EDU
Cc: interlingua@vaxa.isi.edu
In-Reply-To: "David A. McAllester"'s message of Mon, 13 Aug 90 10:32 EDT <19900813143229.0.DAM@WITHERING.LCS.MIT.EDU>
Subject: system = logic + control
I think that the requirement that translation into the interlingua be SOUND
is not completely appropriate. In many systems, inference is not complete
with respect to the standard semantics (whatever that is). For example,
subsumption in CLASSIC is not complete with respect the the standard
semantics for terminological logics. Requiring that the translation into
the interlingua be sound means that the incompleteness of CLASSIC's
subsumption must be retained (and, perhaps, increased). This would require
the creation of a quote-based axiomatization of subsumption that is
incomplete, and, again, it would be very hard to determine if this
axiomatization is anywhere close to subsumption in LOOM, for example.
I think that a better requirement for translatation would be that the
translation be approximate, i.e., that the meanings of constructs in the
representation language be close to the meaning of their translations in
the interlingua, and that questions of inference be relegated to secondary
status, at best. As quotation does not preserve "meaning" well, if at all,
this strongly argues for the sort of interlingua I proposed.
I realize that this requirement puts a very large load on "meaning", but,
as I am really trying to represent declarative knowledge, I think this is
appropriate.
peter
PS: It is possible to move things from inference to semantics, and vice
versa. Consider the attempts to create a semantics for PROLOG with
negation by failure.