Re: Issue timbl-03 "collection clutter" proposal to close

At 07:59 24/04/2003 -0400, Dan Brickley wrote:

>* Peter F. Patel-Schneider <pfps@research.bell-labs.com> [2003-04-24 
>07:47-0400]
> > From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
> > Subject: Re: Issue timbl-03 "collection clutter" proposal to close
> > Date: Thu, 24 Apr 2003 12:21:34 +0100
> >
> > > > The OWL specs reference rdf:List, but they don't care how one 
> learns about
> > > > these Lists, ie. whether an RDF/XML parser tells you directly versus
> > > > whether they are inferred from the semantics of rdf:first and rdf:rest.
> > > >
> > > > Could a member of the WebOnt WG confirm this?
> > >
> > > Sorry I should have read the thread before replying.
> > >
> > > Due to the rather odd way that OWL DL works, these triples are needed.
> >
> > I would instead say
> >
> >       Because of requests from a WebOnt member (hi Jeremy) the syntactic
> >       definition of OWL DL when written in RDF triples requires that list
> >       elements be subjects of triples with predicate rdf:type and object
> >       rdf:List.
> >
> > > i.e. without these triples the current OWL document do not work, and 
> more
> > > difficult things would be needed to be done - the correspondence 
> proof is
> > > an important example.
> >
> > I do not believe that the correspondence proof would need much change.
> >
> > > Of course everyone is right to say that if they were not there they 
> could
> > > be inferred but ...
> > >   they are needed in the OWL DL syntax - before the semantic 
> reasoning part,
> > > yes they could be inferred there, but that would be additional work, 
> and a
> > > change.
> > >
> > > I am interested in Peter's view; if we made this change Peter would 
> have to
> > > do some work - if he were largely positive, I would change my 
> position from
> > > favouring a reject to favouring an accept.
> > > (cc-ing Peter on this message)
> > > [The proposal is that RDF/XML parsers should not emit triples
> > > _:x rdf:type rdf:List .
> > > since they are redundant).
> >
> > I oppose this change.  OWL already depends on them being there.
>
>OK, thanks for the the additional context everyone. I withdraw my concerns
>and am now happy to go along with Dave's advice on the issue.

I have to say I'm uncomfortable:  if I understand Peter correctly, OWL 
could be reworked to not require the [... rdf:type rdf:List] statements.

In my own work, I've made some quite extensive use of lists.  In the long 
run, requiring to generate the extra triples will, I think, add 
significantly to the number of triples for a typical model.  Would we be 
better serving the future users of RDF if we jointly worked out ways to 
avoid the un-needed generation of triples by RDF parsers?

If this really is a major problem to change, then I'll back down.  But so 
far I haven't heard enough to convince me that it's not worth a little more 
effort now to significantly contain the size of RDF graphs for future 
applications.

...

A question:  suppose that one could infer through (say) rdfs closure rules:
   <a> rdf:first <b>
then one can also infer (though rdfs rules):
   <a> rdf:type rdf:List

The current requirement by OWL that rdf:List be included explicitly seems 
to suggest that an *inferred* rdf:first property is in some respects opaque 
to an OWL DL reasoner.  Is this intended?

#g


-------------------
Graham Klyne
<GK@NineByNine.org>
PGP: 0FAA 69FF C083 000B A2E9  A131 01B9 1C7A DBCA CB5E

Received on Friday, 25 April 2003 11:30:39 UTC