ANSI KIF, CGs, & Ontologiessowa@west.poly.edu (John F. Sowa)
Date: Sun, 5 Nov 1995 13:36:13 +0500
From: firstname.lastname@example.org (John F. Sowa)
To: email@example.com, firstname.lastname@example.org, email@example.com, firstname.lastname@example.org
Subject: ANSI KIF, CGs, & Ontologies
Cc: email@example.com, firstname.lastname@example.org, spillers@VNET.IBM.COM
The fall meeting of the ANSI X3T2 committee on Information Interchange and
Interpretation was held at Stanford from October 24 to 27. Mike Genesereth
and John Sowa reported on their progress on writing and editing the draft
proposed standards for the Knowledge Interchange Format (KIF) and conceptual
graphs (CGs). The committee also worked on editing and commenting on the
draft for the ISO Conceptual Schema Modeling Facilities (CSMF). But an
interesting new development was a proposal by Bob Spillers from IBM, who
would like to organize a consortium of commercial and academic groups to
develop a framework for registering and coordinating ontologies.
Everyone agrees that it is probably impossible to get a single standard
ontology that everyone could agree upon. But one of the major goals of the
knowledge sharing effort is to establish a framework that can facilitate
the exchange and sharing of ontologies by groups that can come to some sort
of agreement. But that goal raises many more questions than anyone has
so far been able to answer satsifactorily:
1. How can we support commercial groups that need standards for ontologies
without impeding the innovations and free exploration needed for research?
2. Commercial groups have been developing very detailed ontologies for
specialized domains such as medicine, law, engineering, library
classifications, etc. Funding is available from various sources to work
on the details, but there is little or no funding for working on the upper
levels, which are necessary to form the bridges and fill in the gaps of
the more detailed ontologies. Is it possible to coordinate the independent
efforts in order to develop a gneral framework that would be maintained
by a neutral organization such as ANSI or ISO?
3. For natural language processing, the WordNet hierarchy of about 60.000
words, which was developed by George Miller's group at Princeton, has
become the most widely used ontology because the price is right: free.
Yet free for research and academic uses does not always imply free for
commerical use. It would be important to have such ontologies maintained
by a recognized standards organization, such as ANSI, for which the
legal issues have been resolved.
4. The issue of natural language raises a question about the differences
between the lexicons and ontologies currently used in NLP projects vs.
the ontologies used for knowledge-based systems. The NL ontologies,
such as WordNet, tend to be very large, but with very few constraints
other than the subtype or subsumption relation between concepts. The
ontologies used for detailed reasoning in knowledge-based systems tend
to be much smaller, but with a much greater amount of detailed knowledge
(i.e. axioms, constraints, definitions, defaults, etc.). An exception is
Cyc, which is both broad and detailed, but even the Cyc developers found
a need to break up their originally monolithic knowledge base into a
collection of smaller _microtheories_, which could be developed and used
in a "mix or match" fashion. How can the different needs and approaches
for NLP and KR systems be reconciled? Is the effort to reconcile them
worth pursuing? What would the costs and benefits be?
5. The explosion of information available on the Internet has raised
another need: indexing and classifying techniques for finding information.
A broad ontology, such as WordNet, might be very useful for indexing
information, which might have a lot of detail that would be necessary
for reasoning, but not as necessary for indexing. The Cyc developers
are currently planning to develop links and associations between WordNet
and their 100,000 concept types. Such associations could be used to
interpret questions and direct them to a more detailed microtheory, which
might then be used for answering them. Can such an approach be applied
to other projects, such as Ontolingua?
6. Besides the technical issues, there are a great many practical ones.
How do you get different groups to cooperate? Funding is usually easier
to get for detailed projects rather than general ones. Will the groups
that have the needs and expertise be able to get together and work
together? How and under what auspices?
Bob Spillers presented his suggestions on Wednesday, Oct. 25, and the group
had a lively discussion about how to proceed. Everyone was interested and
would like to see such a collaboration succeed, but no one had clear answers
to the above questions. At the closing plenary session on Friday, the
X3T2 committee approved the following resolution:
Robert Spillers from IBM presented suggestions for top-level ontologies
that would be needed for a variety of purposes related to conceptual
modeling, knowledge representation, knowledge sharing, and information
retrieval. After some discussion, the committee decided that the
suggestions presented by Mr. Spillers were closely related to current
X3T2 projects on conceptual schemas and CSMF. Therefore, the X3T2
committee authorizes Mr. Spillers to convene an ad-hoc working group
on ontologies to meet prior to the next X3T2 plenary in March, 1996,
and to report on the results and recommendations of the ad-hoc group
at the next X3T2 plenary.
Bob Spillers volunteered to host the March meeting at the IBM Santa Teresa
Laboratory (just south of San Jose). The ontology group would meet on
Monday and Tuesday (March 4 & 5), followed by the regular X3T2 meeting
on March 6 to 8.
Meanwhile, Spillers plans to contact various groups that might be interested
in participating. Richard Fikes, who is in charge of the Ontolingua efforts,
attended the discussion on Oct. 25. Spillers has already contacted Doug Lenat
at the Cyc project; Lenat is very interested and would like to make the Cyc
hierarchy of concept types available (but without the detailed knowledge that
is associated with each type). Other groups will be contacted between now
and March. Anyone who may be interested in participating in the effort
should send a note to Bob Spillers at IBM (email@example.com) or call
him at 408-463-4021.