About the model: Question 5
5. The FRBR-RDF people seem to have followed OWL and defined two properties for every property, each an inverse of the other. Thus, they have defined the class responsibleentity, and the two properties creator and creatorof. The creator property is defined as “an entity in some way responsible for the creation of a work,” the domain is work and the range is responsibleentity. The creatorof property is defined as “a work that was in some way created by an entity,” the domain is responsibleentity and the range is work. Is that correct RDF logic, and would every statement of authorship have to be duplicated, e.g. 1. this person is the author of this work and 2. this work has this author? (This may actually be related to point 1 above.)
“Good questions. I don’t know. We’re entering the world of free modeling as opposed to the default taxonomical model, and as such people are free to make up what they want. I’m positive you can’t say there’s anything such as “correct” RDF logic. I’m sure that model makes sense to whoever did it, but if it feels unnatural to others reading it, I’d say that’s an indication of someone getting it wrong. Mind you, getting models such as this *right* is close to impossible (but not absolutely impossible :).As to the bi-directionality of the statements, I’m not sure how the property is defined (optional, mandatory?) but it shouldn’t be necessary to duplicate the expressions, but then, it comes down to the use of the ontology. There’s a slight difference between ; “Herbert” has_written book_1 book_2and ; “Book 1” written_by “Herbert” “Book 2” written_by “Herbert”It comes down to practicalities of where you model *from*. Sometimes, if you model from the book level, repeating author is a must (which is modus operandi in the cataloging world, as you know) but you have to make a judgment call on what other parts of the author file you should express, given your knowledge on what other things exist in the vicinity of the thing you’re expressing. This is of course quite impossible, and leads to a lot of replication which inference engines need to sort out (and indeed bring a lot of trouble to the semWeb world). The trick with any of this is to have control over the URLs that are used as persistence identifiers.I think we’re seeing just how messy RDF can get without proper RDF tools. :)”–September 11, 2007 email from Alexander Johannesen.
DECISION: To keep the model to a reasonable size, inverse classes were not defined, for the most part; it is hoped that they can be viewed as being implicit in the model as is. Exceptions: whole-part relationships; words for music/music for words; broader/narrower term relationships between subjects.