SUO: Re: Do we need an upper ontology?
pat hayes wrote:
> The last time I met Doug (about 6 months ago) I talked to him about
> the SUO, briefly. I asked him whether it wouldnt make sense to simply
> remove the 'upper ontology' from CYC altogether, and have an
> inheritance hierarchy which simply had a large number of top nodes
> (which would be, roughly, the tips of the upper ontology
> classification.) He agreed, and said that he had in fact seriously
> suggested making that change some years earlier, but that his
> customer base all protested that they preferred to have a single top
> node. He shrugged, with the implication that if thats what they want,
> he was willing to let them keep it. So I think that your and Doug's
> viewpoints are not quite identical; he thinks the upper levels of the
> ontology are irrelevant and distracting.
>
Unfortunately, Doug Lenat is also of the opinion that there should
be only **one** middle ontology, namely CYC. OF course, if we have only
one middle ontology, then any upper ontology consistent with it
will be as good as any other, and no upper is in fact needed. But
if we have **more than one** middle or domain ontology, the concern
is that there will be incompatible concepts defined in the disparate
ontologies, and the only way to avoid that is to construct an upper
ontology from the top levels down; if lower ontologies are defined in
terms of the upper levels, some interoperability should result.
This seems a priori like a methodology far more likely to result
in interoperability than to hope that by some fortuitous coincidence
two separately developed ontologies can be reduced to common terms.
I thought it was precisely the inability to make a common logical
system from separately developed ontologies that was the motivation
for work on a SUO. It is certainly my motivation.
[Pat Hayes]
> I might add that this methodology is supported to some extent by
> empirical work on concept mining, where a key strategy in identifying
> useful conceptual dimensions is to find the ways that concepts can be
> teased apart, ie how they differ as well as the ways they are
> similar. The key question to ask is, given three concepts A, B and C:
> in what way would B be closest to A and furthest from C? The point
> being that this kind of analysis requires a level of detail which
> only becomes apparent when one has a reasonable amount of particular
> detailed knowledge to apply it to.
This is certainly a useful method to discover precisely what
different
people intend when they formalize their conceptualizations, but I was
under the impression that:
(1) there is no easy way to make one formal ontology to communicate with
another
developed separately; the kind of analysis that Pat Hayes suggests is
very
time-consuming and at the end, the two different ontology builders need
to come to some agreement as to which common set of concepts can be used
to allow interoperability; but
(2) doing it this way is no faster, and probably a lot slower than
building a common upper ontology of defining concepts and then
having users build their specialized ontologies in those terms.
(3) There has already been a lot of "concept mining" from middle-level
ontologies, but in the absence of a coordinated project to share
these insights and integrate them into a logically consistent whole,
they remain largely unusable in systems other than those in which
they were developed.
I agree that this is an important issue for the SUO, but I thought
that
the whole project of the SUO was predicated on at least the first two
assertions
above. If Pat Hayes is aware of successful examples of merger of
middle-level ontologies, it could be very helpful to study and learn
from
such cases.
Pat Cassidy
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================