- From: Jim Hendler <hendler@cs.rpi.edu>
- Date: Thu, 8 Nov 2007 10:42:05 -0500
- To: Boris Motik <boris.motik@comlab.ox.ac.uk>
- Cc: OWL Working Group WG <public-owl-wg@w3.org>
Apologies - Boris pointed out to me that he did have an example in an earlier message - it was at the end of a long message, so I have cut it out and pasted it below - Thanks Boris! From Boris: Let me give you a concrete example. Assume that O1 contains the following ABox assertion: (13) hasParent(Bob,Mary) As long as O2 contains named individuals (Bob, Mary, and so on), you will get exactly the same answers. Now let O2 be an ontology containing the following ABox assertion: (14) hasParent(Bob,_:1) Here the difference becomes important. Under the standard "true" semantics, O2 follows from O1. This is because, in first-order logic, hasParent(Bob,Mary) entails \exists x : hasParent(Bob,x). Under the approximative semantics, O2 *does not* follow from O1. This is because (14) is actually equivalent to the following ABox O2': (14) hasParent(Bob,some-invented-constant) Now in first-order logic, it is not the case that hasParent(Bob,Mary) entails hasParent(Bob,some-invented-constant), so O1 does not entail O2. The last example might suggest that the approximative semantics might be undesirable. I would just like to point out, however, that the anonymous individuals in the ABox make ABoxes a kind of a query language. It might be cleaner not to mix the two roles of an ABox. An ABox might be thought of as just the data; for querying, we might use languages such as SPARQL, which would then be designed appropriately such that we get everything we want (such as the correct answer to (14)) without the drawbacks. For example, if we modify SPARQL to disallow inequalities in the query, we can be decidable for SHIQ. Regards, Boris
Received on Thursday, 8 November 2007 15:42:26 UTC