My thoughts on OWL etc.

["John Colgrave" <colgrave@hursley.ibm.com>]

Having read the minutes of the FTF discussion on Taxonomy Management etc. I
think that some of my views were not accurately presented, and some of them
have clarified, so I thought I would try and write down what my current
thoughts are.  I also have a few comments on the discussions minuted.

My concern with Luc's approach is that it is using OWL to define a schema,
which is not what OWL is intended for.  OWL is intended for defining
ontologies, and an equivalent of a UDDI value set can be viewed as a subset
of an ontology.  I think the choice to be made is between using (something
like) OWL as it is intended to be used, similar to my excerpt of a UNSPSC
ontology/taxonomy, and using XML Schema to create a UDDI-specific schema for
value sets.

I think this choice will have an impact on how many of the other
requirements it is necessary to address in UDDI, so we need to make it
before progressing much further with the other requirements. 

I think it would be better to profile something like OWL than to invent
something specific for UDDI.  I said in my response to Luc that I thought
that OWL was the best candidate at the moment, and now that it has become a
full W3C Recommendation it is an even better candidate.

I am not convinced of the need to have standard APIs for saving and managing
ontologies in UDDI, and navigation is one area where a standard API may
evolve in the context of OWL so it would be better for that to be used than
an equivalent API defined just for UDDI.

Contrary to the FTF minutes, this view is not "largely based on the premise
that semantic search capabilities should be exploited by inferencing engines
external to a UDDI server", it is based on the assumption that it is better
to use a standard such as OWL (in the way in which it was intended to be
used) rather than inventing a specific language just for UDDI, so that we
can benefit from ontologies/taxonomies created by other groups, with perhaps
no knowledge of or interest in UDDI, and also so that we can benefit from
tools and APIs that are being developed for OWL.  I am not thinking about
inferencing at the moment.

I was a little surprised by the discussion of equivalency as I did not
realize that it was a requirement, or something that was currently
supported.  In OWL you can express equivalence between classes.  This is
intended for equivalence across ontologies but I guess you could use it
within an ontology if that made sense.

I don't understand why both "Define a standard format for representing a
value set" and "Define an XML-based schema for representing value sets" are
listed as being in the minimal set of features.

I am concerned that restricting value set relationship support to a single
root node relationship tree will prevent UDDI from using standard ontologies
defined without such a restriction in mind.  For example, if someone were to
produce a full OWL version of UNSPSC, we would probably not be able to use
that in UDDI as it would probably not have a single root node.

As another example, the National Cancer Institute Thesaurus is available as
an OWL ontology that is 32MB and which contains just over 500,000 RDF
triples.  I think that there are multiple top-level classes in this
ontology, there is no single root class.  I would like to be able to use
that ontology directly, rather than have to transcribe it to a UDDI-specific
format, or even add an unnecessary root node.

John Colgrave
IBM