semantic unification

davis@ai.mit.edu (Randall Davis)
Date: Thu, 12 Aug 1993 09:08:54 -0700
Message-id: <9308121610.AA11604@hatcher.ai.mit.edu>
Comment: List name: SRKB-LIST (do not use email address as name)
Originator: srkb-list@isi.edu
Errors-To: aberg@ISI.EDU
Reply-To: <davis@ai.mit.edu (Randall Davis)>
Sender: srkb-list@ISI.EDU
Version: 5.5 -- Copyright (c) 1991/92, Anastasios Kotsikonas
From: davis@ai.mit.edu (Randall Davis)
To: Multiple recipients of list <srkb-list@ISI.EDU>
Subject: semantic unification

   Date: Thu, 17 Jun 1993 08:46:45 -0700
   Posted-Date: Thu, 17 Jun 1993 08:46:45 -0700
   Originator: srkb-list@isi.edu
   Reply-To: <petrie@mcc.com>
   Sender: srkb-list@isi.edu
   From: Charles Petrie <petrie@mcc.com>

   >> Your (and Mr. Petrie's) remarks about "semantic unification" 
   >> seem very interesting.  Unfortunately, they are also a bit 
   >> difficult to follow, since I did not participate in the 
   >> earlier discussion, where, appearently, the term "semantic 
   >> unification" was defined in the first place.  
   >> 
   >> If you have hints, pointers to the literature, or something else, 
   >> I'm very interested, since what you talk about appears to be of interest 
   >> to me --- as Fano indicates.  

   But suppose two systems, A and B, have been translated into KIF, (or
   conceptual graphs) resulting in Ak and Bk respectively . And suppose a
   few connections have been made. Supose we know that constant "a" in Ak
   is exactly "b" in Bk.  Then, trivially, if "a --> c" in Ak and "b -->
   d" in Bk, an automated theorem prover scanning the two systems could
   at least ask the users if "c" and "d" were not related. That is,
   perhaps more connections could be discovered once a few had been
   found.

   Alternatively, such a question could await problem solving in which
   there was a need to relate these two constants. But it's an
   interesting notion for me to take advantage of the common intermediate
   representation and have a theorem prover search for possible
   additional connections off-line.

   Charles

Response delayed until we had a firm citation for the latest article.
See:

Trice A, Davis R, Heuristics for reconciling the knowledge of multiple
experts, {\em Information Systems Research}, Vol 4, #3, pp 1--27, September
1993.

Trice, A, and Davis, R, "Consensus Knowledge Acquisition," {\em Proc 6th Banff
Knowledge Acquisition for Knowledge-Based Systems Workshop}, pp.33.1-33.20.
(Available through SRDG Publications, Dept. of Computer Science, University of
Calgary, Calgary, Alberta, Canada, T2N 1N4)

Trice A, Davis, R \quotes{Consensus Knowledge Acquisition} MIT AI Lab Memo
1183, December 1989.


All three of these articles deal with an implemented system created two years
ago that does just the sort of thing speculated about in the two para aboves.
In particular, it finds common reps and searches for possible additional
connections offline, using a collection of heuristics to determine when two
concepts are likely to mean the same/different thing.