Why A Data Model Does Not An Ontology Make
On 2007-08-02, Robert Meersman gave a lecture at SUNY in Buffalo, at the Center of Excellence for BioInformatics and Life Sciences, led by directors Barry Smith, Lew Goldberg and Werner Ceusters. He addressed an active audience of philosophers, computer and AI scientists, biologists, and bio-engineers. The talk was scheduled for 1 hour but developed into a lively discussion lasting 2 1/4 hours. The presentation slides are available online.
Almost all examples of so-called ontologies in today's research literature on the Semantic Web are in fact just their author's (extended) data model for a particular, a priori known, application that author has in mind. Ontologies as computer-based repositories of a domain's semantics however should not be bound to the context of a single "application". Indeed they must express a form of agreement, in an "application-independent" language, on the concepts, relationships, events, rules, and processes present in that domain. Agreements in turn imply (virtual) communities of users and/or developers collaborating towards such a shared, formal, understanding.
In the DOGMA Framework (Developing Ontology-Grounded Methodology and Applications) of VUB STARLab we study the theoretical foundations of ontologies and of collaborative ontology engineering, and are building an experimental tool suite to illustrate the principles that we claim are involved. This seminar reported on some results and applications as well as difficulties.