- From: Dan Brickley <danbri@danbri.org>
- Date: Tue, 30 Sep 2008 15:12:58 +0200
- To: public-owl-dev@w3.org
- Cc: Alistair Miles <alistair.miles@zoo.ox.ac.uk>, Thomas Baker <tbaker@tbaker.de>
Hello, Ontology experts. I'm looking for advice on current OWL-friendly best practices for modelling mass-produced items and their (possibly varying) characteristics. The motivation here is discussions during the Dublin Core conference last week, and from something in the Library and bibliographic metadata world called FRBR - the 'Functional Requirements for Bibliographic Records', http://en.wikipedia.org/wiki/FRBR A FRBR-driven perspective is slowly getting traction in the library world, and is important also in discussions about the future directions for Dublin Core. FRBR attempts to give a better account of what kinds of things Libraries describe in their bibliographic records, such that we can better deal with the digital age (DVDs, "multimedia", translations, versions and editions etc.). FRBR conceptualises things in terms of 'Works', "Expressions", "Manifestations" and eventually, countable "Items". See the Wikipedia reference above to track down full details. There are efforts (notably by Ian Davis of Talis, see http://www.frbr.org/2005/10/03/ian-davis-frbr-in-rdf) to express these notions directly as RDF/OWL classes. The Library cataloguing standard AACR2 is currently being revised (by the Resource Description and Access group, see http://www.rdaonline.org/) and taking on board FRBR ideas. I have long had a suspicion that some of the distinctions that FRBR make sare more general, and deal with issues that can have common class-based modelling idioms. Basically the problem of the library is that it needs to keep track of countable, locatable, damage-able items, as well as model their shared characteristics. So a book's title, topic, authorship etc, expected number of pages, etc are attached at a different level of abstraction to information about its location, physical state, owner, actual number of pages, and so on. One of the problems we've seen with FRBR deployment is that, because it has only these 4 "buckets" to put things in, some lack of agrement about whether something is a "Work" or an "Expression". The FRBROO project attempts to address this by combining it into the larger CIDOC CRM ontology, see http://en.wikipedia.org/wiki/FRBRoo . This is useful I think, but dauntingly large, and doesn't exploit the richness we have in OWL for describing the membership rules for classes. So my thought here is that it is also worth considering an alternate model, in which "Work", "Expression", "Manifestation" and "Item" are thought of as functional requirements on our ontologising, rather than as directly modelled classes. Hence my question here. How do ontologists lately tend to model things like an aircraft part, or other mass-produced item, when we have a situation in which (a) the design of these copied itself needs modeling (b) their instances may be flawed, damaged or lack adherance in various ways towards their stereotypical ideal. I've been thinking that we could partially model this by some annotation on the class which pointed to a description of an indicative instance. For example, a T-Shirt design might typically be associated with T-Shirt instances that have a certain weight, colour, and so on. I don't want to make strong OWL claims that each actual shirt has just this weight, colour, ... any more than a library wants to imply that the actual number of pages in a shelved book is necessarily what we'd expect from the ideal. Hope I'm making some sense here! I guess the issue I'm skirting around is how to handle default reasoning in RDFS/OWL, and whether there are deployed patterns that work for describing typical manufactured instances which might be re-usable in the bibliographic world. Thanks for any thoughts / pointers. cheers, Dan ps. I made a couple of sketches in diagram / slide form, which might help indicate what I'm on about... http://www.flickr.com/photos/danbri/2891150205/ (static class view) http://www.flickr.com/photos/danbri/2892286406/in/photostream/ (timeline view)
Received on Tuesday, 30 September 2008 13:13:39 UTC