On ontologies and modelling

As already noted, writing a good, reusable ontology is difficult. Clear evidence is all around in (what I consider to be) flaws in existing schema/ontologies, where the authors have made restrictive or invalid modelling assumptions or choices.

At the moment I'm working on relationships, for use in an address book and other applications that refer to people.

Even trivial things like siblinghood are a challenge. For instance, GEDCOM in RDF supposes that the “sibling” relationship is symmetric, which is fine. But is it transitive? After all, if I'm your brother, and you're someone else's sister, aren't I their brother?

The key is to look for failure cases. In this instance, I can share one parent with you, and you can share another with that other person. We are half-siblings, but that relationship isn't transitive. Only pure genetic siblings (i.e. those sharing two parents) have a transitive relationship. This kind of fine-grained consideration appears everywhere, from these genetic aspects to whether spouses must foaf:know(s) each other*. I have been discussing issues such as these with Ian Davis, regarding the relationship vocabulary: e.g. the symmetry of the rel:friendOf property.

Two further aspects are precision of modelling and language complexity. The rel schema uses owl:sameAs on properties, which renders the schema OWL Full — no good for DL reasoners. Also, relationships such as rel:siblingOf and rel:parentOf are quite general, expressing a broad range of roles. These are adequate for social software (expressing social roles — “he was like a son to me!”), but no good for a technical family tree, or any medical use. When designing for reuse and broad applicability, one must really watch one's entailments!

Some day soon I hope to have finished a bunch of ontologies. Yeah, right!

* True in some contexts (e.g. the British legal system, where both bride and groom must be present to sign the register) but possibly not in others.

Posted at 2004-07-20 16:04:13 by Richard • Link to On ontologies and …