- From: Peter F. Patel-Schneider <pfps@research.bell-labs.com>
- Date: Tue, 05 Mar 2002 08:53:48 -0500
- To: jos.deroo.jd@belgium.agfa.com
- Cc: www-webont-wg@w3.org
From: "Jos De_Roo" <jos.deroo.jd@belgium.agfa.com> Subject: Re: more on a same-syntax extension from RDF(S) to OWL Date: Tue, 5 Mar 2002 14:43:40 +0100 > > > > > I agree that http://www.agfa.com/w3c/euler/owl-rules.n3 does not > sanction > > > > any oneOf consequences. Therefore, you will not get > > > > John a person . > > > > to imply > > > > John a [ owl:oneOf ( John ) ]. > > > > from http://www.agfa.com/w3c/euler/owl-rules.n3. > > > > > > > > All this says, however, is that there are desirable inferences that > > > > http://www.agfa.com/w3c/euler/owl-rules.n3 does not sanction, i.e., > > > > http://www.agfa.com/w3c/euler/owl-rules.n3 is incomplete. > > > > > > good point, I've added > > > { ?L owl:item ?x } log:implies { ?x a [ owl:oneOf ?L ] } . > > > so now > > > the-empty-graph > > > log:entails > > > :John a [ owl:oneOf ( :Frans :John :Mary ) ] . > > > but that still doesn't give us an empty hypothesis to entail > > > _:1 owl:oneOf ( _:2 ) . > > > I will think further... > > > > Actually it does, as this is (assuming I've got the _ and : in the right > > order) a simple existential version of part of > > > > :John a [ owl:oneOf ( :John ) ] > > But here it is written in the implied consequent graph > so if one would have written the rule as > { ?L owl:item ?x } log:implies { ?x a _:1 . _1 owl:oneOf ?L } . > and normalize (eliminate existentials) that to following rules > { ?L owl:item ?x } log:implies { ?x a [ :skf ?L, ?x ] } . > { ?L owl:item ?x } log:implies { [ :skf ?L, ?x ] owl:oneOf ?L } . > (using Skolem functional terms) > then the _:c owl:oneOf ( _:r ) entailment follows, > but we haven't written the rule like that... > For ``facts'' such as :John a [ owl:oneOf ( :John ) ] . > there is no problem to write them as > :John a :skc . :skc [ owl:oneOf ( :John ) . > (using Skolem constants) and we happen to also do that. > Maybe we have confused the whole thing with [ ] notation > but it is quite clear that one should not re-re-re-skolemize [ ]'s > > -- > Jos Well what I think that all this points out is the difficulty of writing things in this way. In particular, you have to be *very* careful how you label what you are doing. Note that you said that it doesn't *entail*, which is a semantic relationship. I pointed out that it does *entail*, which it does. Your rules won't infer it, as they are incomplete. I think that it would be much better if everyone using N3 refrained from using log:entails as the predicate in rule. It would be even better if they refrained from using log:implies, log:Truth, and other such properties without pointing to a readable description of what they meant by such properties. Peter F. Patel-Schneider Bell Labs Research
Received on Tuesday, 5 March 2002 08:55:45 UTC