W3C home > Mailing lists > Public > w3c-rdfcore-wg@w3.org > April 2003

Re: Proposed response for timbl-02 (reification semantics)

From: Dan Brickley <danbri@w3.org>
Date: Thu, 10 Apr 2003 07:15:39 -0400
To: Patrick.Stickler@nokia.com
Cc: bwm@hplb.hpl.hp.com, gk@ninebynine.org, w3c-rdfcore-wg@w3.org
Message-ID: <20030410111539.GD24100@tux.w3.org>

* Patrick.Stickler@nokia.com <Patrick.Stickler@nokia.com> [2003-04-10 13:28+0300]
> 
> 
> > -----Original Message-----
> > From: ext Dan Brickley [mailto:danbri@w3.org]
> > Sent: 10 April, 2003 10:15
> > To: Stickler Patrick (NMP/Tampere)
> > Cc: bwm@hplb.hpl.hp.com; gk@ninebynine.org; w3c-rdfcore-wg@w3.org
> > Subject: Re: Proposed response for timbl-02 (reification semantics)
> > 
> > 
> > * Patrick.Stickler@nokia.com <Patrick.Stickler@nokia.com> 
> > [2003-04-10 09:51+0300]
> > > > From: ext Dan Brickley [mailto:danbri@w3.org]
> > > > write a test case (would be a new kind of test case for 
> > us) that uses
> > > > *OWL* semantics for inferring identity (eg. via 
> > > > InverseFunctionalProperty).
> > > > 
> > > > The goal would be for inferences justified by the OWL 
> > rules to cause
> > > > 'annoying' onclusions involving the resources named by 
> > > > rdf:predicate, :subject 
> > > > or :object properties. We could even ask TimBL if he could 
> > > > re-couch his 
> > > > concerns about the semantics using such an approach...
> > > > 
> > > > I understand this would need a fwd reference to OWL, but 
> > > > since it would be 
> > > > a 'health warning' note for implementors rather than a formal 
> > > > rdf core test
> > > > case I don't see that as a problem.
> > > 
> > > Do you mean basically telling folks not to write rules that infer
> > > things from reifications directly, but only from asserted triples?
> > > 
> > > If so, then I agree.
> > 
> > I didn't mean that, although that should also be discouraged. 
> > 
> > I meant something like the following:
> > 
> > (this to do with the De Re / De Dicto issue with rdf:predicate etc
> > directly referencing the thing that is denoted by the 
> > predicate etc of the
> > reified statement, ie. not safely quoting it and hence 
> > allowing substitution
> > of coreferents)
> > 
> > here's a sketch towards testcase, sorry haven't polished this up but
> > the intent should hopefully be clear. It takes your sample and 
> > shows the kinds of interferences that OWL-happy systems might make in 
> > the face of knowing some things have multiple URIs and OWLs ability to
> > express equality, directly (sameFooAs) or indirectly 
> > (InverseFunctionalProperty).
> > 
> > [[
> >   _:s rdf:type rdf:Statement ;
> >    _:s rdf:subject <s1> ;
> >    _:s rdf:predicate <p1> ;
> >    _:s rdf:object <o1> ;
> >    _:s #source <http://some.server/some_schema.rdf>
> > 
> > <s1> owl:sameIndividualAs <s2>
> > <p1> owl:sameIndividualAs <p2> 
> > <o1> owl:sameIndividualAs <o2>
> >    
> > ...which with OWL semantics I believe gets us to:
> >    
> >    _:s rdf:type rdf:Statement ;
> >    _:s rdf:subject <s2> ;
> >    _:s rdf:subject <s1> ;
> >    _:s rdf:predicate <p2> ;
> >    _:s rdf:predicate <p1> ;
> >    _:s rdf:object <o2> ;
> >    _:s rdf:object <o1> ;
> >    _:s #source <http://some.server/some_schema.rdf>
> > 
> > 
> > which includes the subset of triples,
> > 
> >    _:s rdf:type rdf:Statement ;
> >    _:s rdf:subject <s2> ;
> >    _:s rdf:predicate <p2> ;
> >    _:s rdf:object <o2> ;
> >    _:s #source <http://some.server/some_schema.rdf>
> >    
> > which strongly suggests that the triple
> > 
> >    <s2> <p2> <o2> .
> > 
> > ...can be found in some_schema.rdf, rather than 'can be 
> > deduced from...'.
> > ]]
> > 
> > since the author of some_schema.rdf may know nothing about 
> > those 3 uris,
> > anyone defining predicates such as 'source' above which relate 
> > a rdf:Statement and its parts 
> > to documents etc should bed very careful with their
> > definitions, to avoid mistakes. 
> 
> Well, I think that precluding inference based on reifications also
> covers the above. I.e., you shouldn't have any reification statements
> as part of an IF portion of an inference rule. And even if you had
> those OWL equivalence assertions in the graph, if you avoid inferences
> involving reifications, you won't end up with the dubious source
> assertion
> 
> >    _:s rdf:type rdf:Statement ;
> >    _:s rdf:subject <s2> ;
> >    _:s rdf:predicate <p2> ;
> >    _:s rdf:object <o2> ;
> >    _:s #source <http://some.server/some_schema.rdf>
> 
> since to get that inferred statement, you'd have to have a reification
> as a premise of your inference rule.
> 
> So it seems to me that so long as the warning is to avoid allowing
> any of the reification vocabulary in the IF portion of rules, things
> will be just fine.
> 
> Eh?

Eh indeed. Do I understand you right? You sound like you are suggesting
that your use case for reification will work OK, so long as RDF database,
query and inference systems (eg. OWL imlementations) have special case 
knowledge that rdf:predicate/subject/object triples cannot be used to 
conclude anything.

This would mean a major change to the OWL specs (which have just gone to
last call). Currently they treat RDF statements that use pred/sub/obj just 
as they treat any other statements. If this conflicts with the reification
use case you present, they ought to be told asap. Otherwise implementations 
of the OWL specs will do damage to RDF data written in the style of reificaiton
that you endorse. There's nothing we can do to stop OWL systems drawing 
such inferences, given sufficient input data; or nothing much short of 
reintroducing the notion of dark triples and claiming that all triples to
do with reification are 'dark'. Which is not somewhere I'm in a hurry to go.

So Patricks, re-reading your comments above to be sure... Is it fair to say:

[
PatrickS's use of RDF reification as currently specified in M+S and rdfcore docs
 will only meet its application goals if inferences can somehow be 
restricted to dissalow the use of 
any of the reification vocabulary in the IF portion of any inference rules.
]

If that's the case, I suggest either (i) we've lost a use case that 
justified rdf reification's current design, or (ii) OWL and RDF semantics
need to be changed to include the restriction on inferences based on 
statements that use RDF:statement/predicate/subject/object constructs.

Dan 
Received on Thursday, 10 April 2003 07:15:45 EDT

This archive was generated by hypermail pre-2.1.9 : Wednesday, 3 September 2003 09:56:54 EDT