SUO: SUO Best Practices
JOHN SOWA WROTE:
>
> Developing the tools and methodologies for extending, refining,
> and sharing ontologies are just as important, and perhaps even
> more important than any particular choice of categories we may
> develop or propose for the SUO.
>
John Sowa's remark of a week ago (as well as Robert Kent's highly
elaborated endorsement of it, and the methodological unease that Pat
Hayes, Chris Partridge and others have expressed in explaining their
emphatic NO votes) perhaps makes it germane to raise again the issue of
"best practices" that was deferred by the acting chair at an earlier
stage:
Is there some existing or best practice that leads us to believe that an
integrator's approach to ontology building (as in Teknowledge's current
"merged" KIF artifact) will be productive? What purposes, if any, have
been successfully served by such an approach at the upper level? From
existing practice, what kind of institutional arrangements, competence and
resources are needed to support such an artifact?
The previous experience of the ad hoc ontologies standards group led very
early to distinguishing ontology merger from ontology alignment. "Merger"
was used to mean an authoritative combining of artifacts in terms of a
stipulated semantics. Since merger of this kind requires an ongoing
Authority, which was beyond the resources of an ad hoc group, it was
explicitly rejected in favor of "alignment"-- grounded in a semi-automated
methodology that is made fully transparent to the user, who can then
resolve the inevitable "bow ties" between ontologies constructed on
differing principles to serve differing (user-defined) purposes. (In the
metadata community people speak of "crosswalks" or "field mapping" among
metadata sets developed on differing principles.) Nevertheless, it was
thought feasible to attempt an *example* of an upper level provided it
were labeled as a reference object and constructed in a transparent way
that enabled users to adapt it, extend it, or disable parts of it as they
saw fit-- and also to preserve the integrity of any sub-ontologies or
parallel ontologies with which it might be aligned.
The intellectual investment in an alignment approach is of course
conceptually higher at the outset (it is an "academic" task for which the
current interpretation of IEEE procedures, disclosure norms and philosophy
of voluntarism fueled by 'material interest' would have required some
creative adaptation) but it offers a greater prospect of self-maintenance,
whereas the merger approach, if it is to be more than a glossary, appears
to require a stipulative 'semantics' maintained by an ongoing authority or
institution (currently Teknowledge) to respond to the inevitable anomalies
and criticisms and then to amend, extend and revise the ontology to
accommodate them.
We should ask ourselves once again whether there is some 'best practice'
upon which the current base document builds.
Are there some existing practices that support the Teknowledge approach at
the upper level? For what purposes are they adequate-- search, NLP
disambiguation, inference down to domain ontologies, decision support,
pedagogy (these being purposes mentioned in the SUO discussions thus far).
What does the previous experience with this approach suggest about the
likely strengths and limits of a standard based on it? What are the
institutional and financial prerequisites of developing and maintaining
such a standard? Are these prerequisites likely to be satisfied by some
committed subgroup of the current discussants?
Until someone can root the activity in existing or best practice, and show
a movement towards settled norms, in what sense is it a standards
activity?
Lee