Yes, these are late and I have no reasons, only excuses. But, for your
perusal, my minutes of the Ad-Hoc Committee on Ontology Standard's
Meeting of September, 1996 are submitted, below.
If you have any changes, please email me directly. I shall summarize
these changes and post them -- by January 15, 1997, say?
------------- draft minutes ------->
The ANSI Ad-Hoc Committee On Ontology Standards, 9/16-17/96
Held at Stanford's Knowledge Systems Laboratory
Host: Bob Englemore
Minutes: Steve Sawyer (With assistance from Tony Sarris and Monique Sugimoto)
Attending (in order of signature)
Name................Affiliation.........E-mail address........
Woojin Paik, Textwise woojin@textwise.com
Lee Auspitz Textwise lee@textwise.com
Andres Kornai IBM kornai@almaden@ibm.com
Bob Spillers IBM spillers@vnet.ibm.com
John Sowa SUNY Binghampton sowa@west.poly.edu
Martin v.d. Berg IBM berg@turing.stanford.edu
Bob Englemore Stanford rse@ksl.stanford.edu
Frank Olken LBNL olken@lbl.gov
Mark Sastry Compuware mastry@sastry.com
Nicolo Guarino LADSEB-CNR guarino@ladseb.pd.cnr.it
Adam Farquhar KSL afarquhar@ksl.stanford.edu
Monique Sugimoto EDR monique@cogent.net
Bart Emanuel IBM-Internet Div. bart@dev.infomkt.ibm.com
Bill Mark National Semi. bmark@ampere.nsc.com
Fan Hsu IBM Fanhsu@vnet.ibm.com
Fritz Lehman Cycorp fritz@cyc.com
Tony Sarris Ontek Corp. tony@ontek.com
Eduard Hovy USC/ISI hovy@isi.edu
Nancy Lawler U.S. DOD e6nl001@coe.coppin.umd.edu
Bruce Bargmeyer U.S. EPA bargmeyer.bruce@epamail.epa.gov
Steve Sawyer Syracuse University ssawyer@cat.syr.edu
Introduction of attendees (see above)
Review of Agenda
Attendance of most major' ontology standards
George Miller/Wordnet people invited but declined
Many large specialized ontology representatives not here
engineering and medical fields
Distribution of the White Paper'
Ontology Alignment of Pangloss and CYC Upper Structures
(Presentation by Ed Hovy)
(Ed's slides are available and cover this topic in more detail)
Pangloss includes the WordNet top-level, but it has been much realigned.
That means that the basic concepts are there, but they were poorly
structured in WordNet, so Pangloss rearranged them.
The 'merged ontology' has about 1000 concepts, because when in doubt a
composite approach was used and CYC's larger number of concepts were hung
off a concept common to CYC and Pangloss. But it should be pruned back to
about 300 or so.
Only deals with ISA (class specialization) hierarchy.
Next Steps:
Add middle-structures (?)
Add facilities to allow domain models from all over.
Integrate other ontologies (e.g.,):
1. EDR (more than the current 100 concepts)
2. Ontolingua
3. CG 'starter set'
4. Mikrokosmos
5. Naming
Discussion on the relationship of STEP generic resource models to the
draft 'reference ontology'?
Could the RO be used to position the STEP models?
Current model is 'thing-based'; not as good for processes, events,
relationships -- because of its linguistic perspective (which is biased
toward objects).
Other Things to do:
1. Check links and add more links when possible.
2. Prune Pangloss upper structure and remove redundancies
3. Add EDR concepts.
4. Merge more CYC concepts into upper region.
5. Place on WWW -- in two stage
a. First for meeting members to use & review, then
b. For the public.'
Building on Presentation, there are six interest areas for the
members/attendees to pursue:
1. Theory (products: theories, criteria, validations)
- 'crystal' (facet-based)
- relations to KIF, Ontek
- validation of reference ontology
2. Reference Ontology Construction (products: intermediate ontologies,
merger tools)
- current merged ontology
- upper WordNet
- Mikrokosmos
- EDR
3. Distribution (products: web pages/browsers, APIs, etc.)
- API
- web form
- tools
- advertising
4. Standards (products: standards, guidelines/technical reports)
- other related committees
- standardization procedures and timeframes
5. Funding (products: proposals, consortia)
- DARPA
6. Thematic roles and lexicon of terms.
- CYC has 100
- UMLS has many
- CG (via CSLI input) has a couple dozen
DISCUSSION/COMMENTS
Nancy Lawler discussed her need for help with queries across multi-data
stores, some of which are structured and some of which are text. In both
cases, there are indices or tags which may not be consistent. This would
provide a structure for navigating across those data stores.
Lee Auspitz outline how this will not be helpful to searches as they are
currently conducted. Most of those searches reflect terminological
differences at the specialized term level. If a document is on the web
and it isn't structured or its structure isn't positioned into the bottom
of the RO, then this may not help.
This was countered with: but having a structure of commonly used (based
on linguistics at least) concepts might help relate things even when
searching across specialized collections of terms.
John Sowa presented his lattice concept with examples using the Gospel
and notes how this provides conceptual clarity. A discussion of tree
versus lattice representation follows. Conclusion seems to be that a
lattice has many appealing virtues, while a tree is simple. And, a tree
can be seen as a collapsed lattice. Or, at least a lattice can be
super-imposed on a tree.
Nicolo Guarino presented his concerns with how generalizations in
ontology design can lead to dangerous and unintended consequences. He
showed this in examples of place and physical object. This opened a
discussion of how to represent certain concepts and Pierces ideas of
firstness, secondness and thirdness. This was brought back to the
efforts of creating a reference or standard model.
Discussion (beginning in the PM of 9/16 and beginning on the AM of 9/17)
focuses on if/how the reference ontology (RO) is to serve:
1. as an example and describe how to interact with it or
2. as _the_ standard and demand conformance.
Consensus built around the latter. That is, conformance to the reference
ontology is more likely, more useful, and more do-able than is creation
of some fixed standard.
Consensus also seemed to say that adherence to the RO guidelines should
be assessed by some organization. Should this be a standards body? Or
will vendors employing this have the leeway to pick and choose what they
agree/don't agree with?
Other guidelines for developing, aligning and interchanging Ros were also
felt must include (perhaps as a technical report/ draft?):
1. General Discussion of Ontology
2. Elements of a Reference Ontology (see above)
3. Structuring Relations
4. APIs (functional capabilities of tools)
This latter point led to a sub-group of people (Adam, Andres, Bart,
Martin, Fan) to meet and discuss the API for a guiding RO. This would
provide for:
1. Syntax
2. Batch
3. Traversal tools
4. Interface
A useful reference ontology would be a first example for applying the
above. This would be produced by the consortium and maintained by the
consortium (or by Stanford on its behalf).The Center for National
Research Initiatives might manage it. The RO should be put on web and
use spring symposium to advertise. The RO in KIF with html plus KIF to
sgml converter to be done at KSL.
Bruce Bargmeyer, of EPA, addressed the group with two issues:
1. That an RO would be valued by many domains (example he
used is of EPA efforts to make sense of their world)
2. That standard creation has many steps, and that draft
standards could be very useful long before formal
approval.
Bob Spillers, of IBM, returned to the white paper and spoke of creating a
consortium of institutions who would work together (with an oversight
board comprised of delegates from the member institutions, funders, and
customers/stakeholders). This consortium could be used to direct funds
and steer research in a concerted effort. Discussion seemed to be that
this was a novel, untried, but potentially valuable idea worth continued
discussion.
Funding and community building issues:
1. DARPA will have large program announcement in October, 1996.
2. March 24-26, 1997 AIII Spring Symposium at Stanford.
Next meeting might be week before at USC ISI.