W3C home > Mailing lists > Public > www-rdf-interest@w3.org > December 2003

RE: WordNet and Concepts - versioning conceptual schemes

From: Sampo Syreeni <decoy@iki.fi>
Date: Wed, 10 Dec 2003 03:52:40 +0200 (EET)
To: "Miles, AJ (Alistair) " <A.J.Miles@rl.ac.uk>
Cc: "'Karl Dubost'" <karl@w3.org>, leo@gnowsis.com, www-rdf-interest@w3.org
Message-ID: <Pine.SOL.4.58.0312081529080.22126@kruuna.Helsinki.FI>

On 2003-12-08, Miles, AJ (Alistair) uttered:

>I'm imagining that a way to handle evolution and change of concept
>schemes such as wordnet is to release discrete versions, and then define
>a complete backwards mapping from each new version to the last.

Agreed. The mechanism might be a bit heavy, but it also admits incremental
upgrades, where not all of the terms are mapped at once. Furthermore it
allows us to describe simple mappings between sufficiently similar -- but
completely separate -- vocabularies. If you have full first order
predicate logic and a closed world assumption at your disposal, you can
also map between your ontology and OWL's versioning mechanism.

I also think you are correct in not equating your exactMatch with OWL's
(or DAML's) equivalence relations -- yours is a subtype. The reason is
that equality relations have symmetric semantics, while the evolution of
vocabularies usually implies preference for one term over the other. The
symmetricity holds for pure reasoners, but not for aggregators which
prefer one of the terms.

exactMatch seems to be transitive (actually it's an equivalence relation
as well), so you should declare it as such. The lesser matches aren't, so
don't go overboard with it.

Both DAML and OWL make a distinction between classes and properties, and
at least OWL clearly separates classes from instances, as far as
equivalence goes. This means that if you want to bring your ontology into
correspondence with OWL, you'd probably want to subtype your relations for
classes and properties, separately. After this you can sensibly declare
the ranges, domains and OWL equivalences of your predicates. Such
declarations are the Thing to Do.

I'd also suggest that you beat your properties into a hierarchy or two.
For instance, all of the somesuchMatch properties are mappingRelation's,
true, but an exactMatch is also clearly a broadMatch, and a narrowMatch
would appear to be an exactMatch. There are plenty of such connections in
your schema. General connections like inter-type relations are precisely
why we bother with ontology languages -- when you bring in an agent which
can infer things, the connections can prove useful beyond belief.

Finally, the class intersection/union/negation capability of your ontology
doesn't seem to easily fit in with OWL. This might necessitate some
further type magic, and I doubt you could leverage OWL's facilities. I'm
also a bit unclear as to the details of how AND, OR and NOT would
precisely translate into triples.

Otherwise your ontology seems quite useful, as it is.
-- 
Sampo Syreeni, aka decoy - mailto:decoy@iki.fi, tel:+358-50-5756111
student/math+cs/helsinki university, http://www.iki.fi/~decoy/front
openpgp: 050985C2/025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
Received on Tuesday, 9 December 2003 20:52:46 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:52:03 GMT