Re: Layering OWL on RDF: the case for unasserted triples

Brian McBride wrote:

> At 13:17 30/05/2002 -0400, Jonathan Borden wrote:
> >I have been actioned by WebOnt WG to send this to RDFCore WG as a
statement
> >of a problem we have been having, with two possible solutions which
involve
> >unasserted triples.
>
> Thank you Jonathan.  I'm sure that preparing this has been a demanding
task.
>
> For clarity, please could you tell us:
>
>    o whether webont is considering any other solutions to this problem,
i.e
>      has webont resolved that some form of dark triples is the only
acceptable
>      solution

The OWL model theory itself remains an open issue and different members of
the WG have different ideas about how this might be accomplished.

Besides the "Comprehensive Entailments" proposal from Jeremy, he had
previously made the "Solipsistic" proposal but withdrew that after
discussion at the Amsterdam F2F.

Dan Connolly has just posted a new "Solipsistic" proposal

http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0235.html and then

http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0264.html

note that in the F2F we discussed whether (in OWL):

<#John> rdf:type <#Student>
<#John> rdf:type <#Employee>

ought entail

<#John> rdf:type _:Ci
_:Ci rdf:type owl:Class
_:Ci owl:intersectionOf _:L
_:L rdf:type owl:List
_:L owl:first <#Student>
_:L owl:rest _:L2
_:L2 rdf:type owl:List
_:L2 owl:first <#Employee>
_:L2 owl:rest owl:nil

which seems to be a common sense entailment (according to OWL) yet would not
be licensed under a solipsistic system.

At the F2F we agreed that such entailments would require that the Query
system assert the existence of such classes, and that:

1) we need commonsense entailments
2) whether or not solipsism would work remains unclear at present
3) it would be hard to predict the effect on future specifications
(particularly a query system).

Hence proceeding with solipsism would involve some at present unclear amount
of risk. At this point Jeremy withdrew the proposal however Dan (who was not
at the A'dam F2F and hence did not participate in the discussion at that
time) has just reintroduced it, and adequate discussion has not yet
occured -- except for Peter Patel-Schneider's restatement of his oft stated
positions:

e.g. http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0239.html and
http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0246.html

Of course Pat's question:
http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0290.html and
Peter's response
http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0292.html and we
again find ourselves back at "unasserted triples" which actually is a
_compromise_.


>
>    o what is webont asking of rdfcore?

Besides a solution to the world's problems? :-)

the _minimum_ that would be required would be: nothing, OWL could introduce
the _owl:Unasserted_ class into its namespace, however I would rather
RDFCore introduce the _rdf:Unasserted_  class into its namespace which would
(IMHO) produce a cleaner layering of the RDF and OWL mt semantics.

From the WebOnt perspective, this would be preferable to a nested <rdf:RDF>
or <rdf:context> syntactic signalling of unasserted triples (e.g. a nested
context) as the syntax of OWL ontologies would be cleaner.

_Alternatively_ one might require each RDF instance document to specify
'dark' or unasserted namespaces in an attribute on the root <rdf:RDF>
element e.g.

<rdf:RDF rdf:unassertedNS=http://www.w3.org/2002/owl-unasserted#
                xmlns:owlUn="http://www.w3.org/2002/owl-unasserted#">

    <owl:Class>
        <owl:intersectionOf>
            <owlUn:List>
                    <owlUn:first rdf:resource="#student"/>
                    <owlUn:rest>
                            <owlUn:List>
                                    <owlUn:first rdf:resource="#employee"/>
                                    <owlUn:rest rdf:resource="owlUn:nil"/>
                            </ownUn:List>
                    </owlUn:rest>
            </owlUn:List>
        </owl:intersectionOf>
  </owl:Class>

</rdf:RDF>

in any case, the minimum that would be required would be the existence of
the rdf:Unasserted class (or the owl:Unasserted class and hence nothing from
RDFCore, if that is acceptable to the SemWeb CG)

Jonathan

Received on Thursday, 30 May 2002 16:02:20 UTC