Modeling issues and guidelines?

Find below the cc of a message I've sent today to the Protégé mailing list.
Not to get answers on this forum, of course (although I would welcome any
knowledgeable answer privately), but sending it, I was wondering if it is
intended by WebOnt group to develop and somehow annex to OWL specification
some informative document(s) about that kind of modeling issues, and/or
guidelines or references to best practices, modeling in OWL-FAQ, whatever

I know this is not explicitly in the charter nor deliverables. It might
even be contradictory with the requirement "The products of the WebONT
group should not presuppose any particular approach to either ontology
design or ontology use."

OTOH the specification have made a considerable, and I would say, unusual,
effort of pedagogy. I figure this is about to trigger ontology development
and maintenance to spread outside the original core community of AI and
Knowlegde Engineers, and become mainstream technology. From what I see
around, this trend is already there. Based on experience in helping
customers in Topic Maps modeling - which implies mastering of a handful of
not so hard to grasp concepts - I'm concerned with the perspective of a
rapidly growing demand of modeling support from OWL potential users, and
some good tutorial literature about it would be indeed very important to
face this demand and help adopters to avoid syntactic hacking and really
use OWL in a way that "makes sense".

I would of course gladly volunteer to contribute to such an effort by any
means that this group would find relevant.


Bernard Vatant
Senior Consultant
Knowledge Engineering
Mondeca -

-----Message d'origine-----
De : Bernard Vatant []
Envoyé : mardi 22 juillet 2003 12:24
À : protege-discussion@SMI.Stanford.EDU
Cc : Jean Delahousse
Objet : Modeling issue : Should subclassing always go along with
property specification?

This is a quite fundamental modeling issue that I have not yet seen
explicitly addressed either in various litterature I have gathered on
ontologies, no more than in Protégé guidelines.

In short : Does it make sense to create a subclass without specifying some
new property for it, or some specific restriction on properties of the

The rationale of this question is that e.g. in biological taxonomy, the
creation of a subclass is always, AFAIK, linked to specification of some
characteristic(s), enabling with the less ambiguity possible, to
distinguish an instance of the subclass from a generic instance of the
superclass, and/on from other subclasses. This best practice I have learnt,
as an occasional amateur bird-watcher.

But, as a matter of fact, outside biological taxonomy, I've not came across
so many ontologies where this rule is really enforced, and every subclass
explicitly defined by specific set of properties or restrictions. "Lazy"
ontology building is often in fact re-engineering of a Thesaurus,
extracting implicit classes and class-subclass relationships hidden under
generic-specific relationships, resulting in a rich class-subclass
hierarchy without any properties underlying the subclassification - simply
because they are not asserted in the Thesaurus source.

Is it correct to consider that practice as underspecified modeling, and
that such ontologies can only be used by human editors making somehow sense
of the meaning of subclasses names and context - assuming they are more or
less "domain experts". But in no way such ontologies can be used e.g. for
automatic classification, since no system can make use of non-explicit

A contradictory argument, in favor of that kind of underspecified modeling
is that enforcing such a constraint for large ontologies building and
maintenance is a serious overhead, both from modeling cost and system
scalability viewpoints.

Any clues on ideas, practices on that issue, or references to literature
where it has been addressed, are welcome.

Thanks for your attention.


Bernard Vatant
Senior Consultant
Knowledge Engineering
Mondeca -

Received on Tuesday, 22 July 2003 07:14:50 UTC