- From: pat hayes <phayes@ai.uwf.edu>
- Date: Wed, 18 Sep 2002 19:51:49 -0500
- To: www-webont-wg@w3.org
Heres a non-technical summary of the basic idea of how to resolve issue 5.3 along the lines currently being worked out. I am sending this now in case the issue comes up at the telecon. 1. The OWL 'universe' consists of some subclasses of the RDFS universe; the OWL vocabulary is restricted (by domain and range assertions, mostly) to this sub-universe (consisting of owl:Thing, owl:Class, owl:Property). However, exactly how the OWL universe relates to the RDFS universe is not yet specified. 2. The 'weak' meanings of the OWL vocabulary can be given without making any assumptions about the relationship between the OWL universe and the RDFS universe. These meanings correspond roughly to the DAML-style semantics that Peter refers to in issue 5.10, ie they do not incorporate any assumptions about the existence of classes that are not explicitly named. So they do not support many of the 'intuitive' entailments that Peter has pointed out. (Though see 7. below.) 3. These weaker meanings can be applied to the entire RDFS domain if one wishes, ie it is perfectly coherent to assert things like rdfs:Class rdfs:subClassOf owl:Class . in order to claim that the OWL universe and the RDFS universe are the same. This allows the OWL vocabulary to be used (with its weak meaning) in situations where classes may be instances, for example. 3a. The 'Russell paradox' RDF graph is a simple contradiction in this semantics. 4. If one makes additional assumptions about the OWL universe, so that its various parts are disjoint in the ways described in the OWL MT, and that this restricted universe is closed under the various class constructors (such as union/intersection and restrictions) then one gets the 'strong' meanings of full OWL, but on a more restricted universe. This supports a lot more entailments and also allows efficient DL reasoners to work properly, providing a computational guarantee of inferential power, but only on a suitably restricted ('layered') use of the OWL vocabulary. Still, many users may be happy to stay inside the restrictions in order to gain the extra utility. 4a. If you were to add these assumptions to the unrestricted RDFS universe, the Russell paradox graph would be a genuine paradox, since these semantic assumptions would require it to be true. 5. There is a syntactic criterion which can be applied to an OWL/RDF graph to detect when it 'fits' inside the restricted full-OWL category (and therefore it is kosher to use the stronger inference procedures on it) but the one I have is rather complicated and may be too expensive to use in practice. Probably a simpler criterion could be invented; I think Peter has one, but I can't follow the details. Of course, one way to say this would be to refer to the translation from the OWL abstract syntax, but it would be nice to have a quick check that could be applied directly to an RDF graph. 6. It is possible to transfer more of the 'strong' meaning to the unrestricted RDFS-wide use of the OWL vocabulary - for example, it is certainly harmless to add closure under unions and intersections - but determining the exact limits of this is a research problem. Basically, we have to invent a new formal set theory and prove it consistent. (My own view is that this isn't worth doing: if we need to get this technical in order to provide a web language for general use, we are barking up the wrong tree.) 7. I am pretty confident that (modulo minor bugs) this semantics agrees with the picture one would get by starting from the abstract syntax, applying Peter's MT to it, and then translating it into RDF. That is, it produces the same entailments on any OWL/RDF that is a translation of any well-formed abstract OWL. I confess however that I have not written out a detailed proof of this. 8. When all this is done by translation into Lbase (or equivalent), the 'weak' meanings correspond to straightforward FOL transcriptions of the semantic meanings, and the 'strong' meanings involve adding a number of extra comprehension axioms, most of them of the form (forall <class, property> (exists <class> .....)) 9. There are some unresolved issues about the need to 'strengthen' parts of the RDFS MT. We are working on those. They are not show-stoppers but we need to get the details right. Write-up coming soon (Friday). Pat PS. Peter's help has gone from the acknowledgments stage to the co-author stage at this point, but he isn't responsible for *this* message and may want to disown parts of it. -- --------------------------------------------------------------------- IHMC (850)434 8903 home 40 South Alcaniz St. (850)202 4416 office Pensacola, FL 32501 (850)202 4440 fax phayes@ai.uwf.edu http://www.coginst.uwf.edu/~phayes
Received on Wednesday, 18 September 2002 20:51:51 UTC