Re: Dark triples motivation

Dan Connolly wrote:


> On Mon, 2002-04-15 at 12:33, Peter F. Patel-Schneider wrote:
> > The only problem that I see with your examples is that they concentrate
on
> > daml lists (daml:collection) and containers.  I see the problem much
more
> > as having to do with defined classes and restrictions, and lists and
> > containers only showing up because they are needed in some places in
> > defined classes and restrictions.
>
> I'm still at a loss to see how 'dark triples' solves anything.
> So I'd love to see an even simpler example of how it can help...
> one that doesn't use lists/collections would be great.

We spent a fair amount of time going over aspects of this at the F2F. Here
is my take: I barely understand the debate but I have great respect for both
Peter and Pat, and when the two _disagree_ I have no idea who is right and
who is wrong, on the other hand, when the two _do agree_ I tend to listen
very carefully. As you may have noticed there has been a long debate on the
list regarding the semantics of OWL, layering, paradoxes etc, mostly Peter
and Pat (this is of course simplified, but the rest of the folks please bear
with me) arguing about stuff that I am not qualified to have an independent
opinion on. At the F2F, after the semantics break out, Peter and Pat
announce that they have a mutually agreed upon solution to the problem,
namely triples that are not asserted, the so-called "dark triples". Now I
may not understand the details of description logic, but as a simple country
neurosurgeon, I figure that if this is good enough for them, it is fine by
me.

In any case, this is _my_ reason for agreeing with the request. The entire
group did discuss the issue at length, and a consensus did emerge.

>
> I think I explained this in a telcon, but I don't think
> it got recorded very well, so I'll reiterate:
>
> The best way for group X to make a request to group Y
> is for X to state its requirements *and* propose a solution,
> as an existence proof that the requirements can be met.

The requirement is for mechanism to define triples that are "dark" or
unasserted, as a way to represent part of the OWL syntax.

The group did discuss several solutions: one, which I proposed
http://lists.w3.org/Archives/Public/www-rdf-comments/2002JanMar/0036.html to
use an embedded <rdf:RDF> element as a container around the stuff intended
to be unasserted. Ora Lassila suggested use of a special container element,
e.g. <rdf:dark> or some other appropriate name. We spent some time
discussing the pros and cons of either approach, and ultimately decided that
this was something that RDFCore might debate in further detail.

Of importance, I understand that "RDF reification" is no longer an option to
do something like this given the new RDF MT.

What was decided was that Jim and Guus were going to discuss this at
semweb-cg and if appropriate a small group from WebOnt would discuss the
details with RDFCore.

> Group Y then gets to either
> -- accept the requirements and the proposed solution
> -- accept the requirements but provide another
> solution that meets the requirements
> (and if this solution does meet the requirements,
> group X is pretty much compelled to accept it)
> -- push back on the requirements.
>
> I'm starting to get a sense of what WebOnt's requirements are
> here (a sane model theory with the ability to express
> stuff like complement, while using RDF/xml syntax) but
> I can't see a proposed solution anywhere.

ok, so there is is.

>
> Well, actually, I've seen a proposal to be able to
> put rdf:RDF wherever a typedNode can go, ala:
>
> <rdf:Description rdf:about="#bob">
>   <com:says
>     <rdf:RDF>
> <rdf:Description rdf:about"#sky">
>   <my:color rdf:resource="http://example/vocab#blue"/>
> </rdf:Description>
>     </rdf:RDF>
>   </com:says>
> </rdf:Description>
>
> I think that's a nifty idea; very much like N3's {} mechanism...
> but I can't see how it's relevant to the layering/paradox
> issues.
>

The idea is that triples _inside_ the embedded <rdf:RDF> or <rdf:dark> etc.,
would not be asserted i.e. the set of triples is divided into the set of
asserted triples and the set of unasserted triples.

Pat and/or Peter might comment for the archive why this solves the layering
problem.

Jonathan

Received on Monday, 15 April 2002 15:36:37 UTC