RE: pun in ontolingua KB
Tim Lethbridge <tcl@csi.uottawa.ca>
From: Tim Lethbridge <tcl@csi.uottawa.ca>
Message-id: <199406212043.QAA18513@mail.csi.UOttawa.CA>
Subject: RE: pun in ontolingua KB
To: gruber@HPP.Stanford.EDU
Date: Tue, 21 Jun 94 16:40:42 EDT
Cc: kuipers@cs.utexas.edu, Doug@SURYA.CYC-WEST.MCC.COM,
ontolingua@HPP.Stanford.EDU, srkb@cs.umbc.edu
In-reply-to: <XLView.772225353.7590.gruber@hpp-ssc-1>; from "Tom Gruber" at Jun 21, 94 11:38 am
X-Mailer: ELM [version 2.3 PL11]
Sender: srkb-owner@cs.umbc.edu
Precedence: bulk
> At 8:26 AM 6/21/94 -0500, Benjamin J. Kuipers wrote:
> >Doug [Lenat] is right. This is a (the?) critical issue in knowledge-sharing.
I agree wholehaeartedly. Dealing with this issue was one of the major themes
of my recently-submitted PhD thesis.
The approach I developed in CODE4 can be summarised as follows:
* Have separate concepts (frames, units etc.) for 1) things (length, car, etc.) and
2) for the *terms* that are used in natural language when manipulating these
concepts. A term can map to several user concepts, and several user concepts
can be represented by several terms.
* When users are specifying knowledge they can either: 1) make *direct* links between
concepts, or 2) specify a term and have the system point out
when the term is ambiguous.
I prefer this scheme to relying on 'context' for several reasons:
1. Forcing a unique namespace even within a context is too restrictive. For
example, the knowledge enterer may very well want to have two concepts
represented by the term 'length' within a very narrow context.
2. Contexts overlap. It can be hard to remember which context is active
and it can be complicated to constantly specify context in
an interdisciplinary knowledge base.
- Tim
--
Timothy C. Lethbridge tcl@csi.uottawa.ca
H(voice/fax): 237-6642 am970@freenet.carleton.ca
W: (613) 564-8155 FAX: (613) 564-9486
Department of Computer Science, University of Ottawa
150 Louis Pasteur, Ottawa, Canada K1N 6N5