Re: RDF and OWL compatibility

Francis McCabe <fgm@fla.fujitsu.com> wrote:
> 
> Please forgive my lack of arcane logic ..
>   but it seems to me that the use of gensym here:
> 
> > kb:works-with(Z,X) :- rdf:type(X,db:employee), gensym(Z).

You are right. The above has no logical meaning. Especially since David was
appealing to such implementational concepts as "returns some freshly minted
bNode on each call". (We shouldn't be reinventing Prolog -- there is already
an ISO standard for it.)


> is not sound.
> Presumably, what is really meant is:
> 
> kb:works-with(Sy,X) :- rdf:type(X,db:employee)  where Sy is some  
> gensym().

No, this is not going to model Enrico's use case correctly. This was the
whole point of Enrico's message to which David was replying.

See my response
http://lists.w3.org/Archives/Public/public-rif-wg/2006Jan/0037.html
for a correct representation. ("Correct" should be understood loosely -- in
terms of getting satisfactory results for the given use case, not that
Enrico's first-order representation has the same semantics as my
representation that relies on default negation.)


> I.e., the skolemization should be done once and for all, not on each  
> use of the rule.
> 
> However, at the moment, I can't see why you would want to skolemize  
> at all .. what is wrong with the rule with the universal quantifier  
> in it?

Because he wanted to mimic an existential variable, which Enrico has in his
use case.

	--michael  


> Frank McCabe
> 
> 
> 
> On Jan 9, 2006, at 9:50 AM, Dave Reynolds wrote:
> 
> >
> > Enrico Franconi wrote:
> >
> >> On 5 Jan 2006, at 15:51, Dave Reynolds wrote:
> >>> On the question of bNodes in the head, I hear the argument that  
> >>> it  is not sufficient to just treat these as new Skolem constants  
> >>> but  my intuitive understanding of the issue is too weak. It  
> >>> would be  really helpful if someone could construct a test case  
> >>> which  demonstrates the difference in results that arise between  
> >>> correct  treatment of bNodes in the head versus treatment as  
> >>> Skolem  constants. In the concrete cases I've seen where bNodes  
> >>> are used in  the head of rules they seem to be intended as a form  
> >>> of anonymous  gensym - so the Skolem constant semantics may be  
> >>> the more  practically useful interpretation.
> >> A a naive gensym would fail the use case  <http://www.w3.org/2005/  
> >> rules/wg/wiki/Managing_incomplete_information>, where two  
> >> examples  (in section "9.4. (Rules involving generation of  
> >> unknown)") show how  you can make things wrong with a naive use of  
> >> skolem constants to  implement the existential variables in the head.
> >
> > Sorry to be slow on the uptake but I didn't follow why the examples  
> > break with naive gensym.
> >
> > The example in 9.4 on that page shows the rules:
> >
> > [[[
> >       rdf:type(X,db:employee) :- kb:works-with(X,Y).
> >       kb:works-with(Z,X) :- rdf:type(X,db:employee).
> > ]]]
> >
> > A naive implementation of bNodes-by-gensym would rewrite the unsafe  
> > second rule to something like:
> >
> >       kb:works-with(Z,X) :- rdf:type(X,db:employee), gensym(Z).
> >
> > where "gensym/1" returns some freshly minted bNode on each call;
> >
> > or, better,
> >
> >       kb:works-with(Z,X) :- rdf:type(X,db:employee), gensym(X, Z).
> >
> > where "gensym/2" binds Z to some bNode which is a (skolem-)function  
> > of X.
> >
> > As far as I can see either of these would return the same answer as  
> > the classical semantics for both of your test cases 9.4.1 and 9.4.2.
> >
> > Dave
> >
> >
> 
> 
>

Received on Tuesday, 10 January 2006 21:03:42 UTC