Re: SEM: circular primitive vs. defined was Re: was: comprehensive entailments without dark triples

On Mon, 2002-04-22 at 16:57, Jonathan Borden wrote:
> Dan,
> 
> (changing the subject because the thread is getting long, and I am
> introducing a new question later on)...
> 
> > At a glance, I'm pretty sure I could live with a theory
> > of classes that was too weak to deal with that thing, at least
> > for a few years. But I'll have to mull it over.
> 
> That worries me. The cost seems to be "dark triples" which are simple to
> deal with in the RDF MT, and there are relatively simple syntactic
> constructs which can encode them.

I don't understand 'dark triples' well enough to estimate
their cost at anything lower than 'unbounded'.
In particular, the costs outlined in

  Problems with dark triples approach
  From: Jeremy Carroll (jjc@hplb.hpl.hp.com)
  Date: Wed, Apr 17 2002
  http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0132.html

are unacceptable to me.

> I also note that RDF M&S has language
> which indicates that 'reification' is to be used for something like 'dark
> triples', but of course reification was intended to be used for a variety of
> purposes, it just seems that providing something that was in RDF M&S in
> perhaps a slightly different form, isn't all that out of charter.
> 
> My fear is a 'theory of classes' that is too weak to do real world tasks,
> particularly those that _already use description logic_. Medical
> applications should be a slam dunk for OWL, it would be a real shame if the
> language were crippled to the point of losing that.

Well, some medical applications should be a slam dunk for OWL, yes.

Whether all of them need to be a slam dunk for OWL 1.0 -- even
all of the ones supported by DL reasoners -- I'm
not so sure.

I wouldn't expect anybody to take my word over yours about
what the medical community needs, however. I'm only talking
about the applications I intend to build/deploy: automating
W3C process/operations, calendaring/scheduling,
that sort of thing.


> But beyond that, suppose we cripple the language, even then we aren't sure
> that the solipsistic semantics will layer correctly on RDF i.e. won't
> introduce yet some other paradox.

Huh? I'm quite confident that the solipsistic semantics introduces
no paradoxes.

> Of course we could spend another 6 months
> trying to work that out... It seems a much more efficient use of the time
> and energy of the combined RDFCore+WebOnt team to fix this problem and move
> on. Dark triples is the only proposal on the table for which there appears
> any sort of consensus that it would actually work.

Please point me to this dark triples proposal -- the one that
explains how they work. I'm pretty sure there isn't one,
and I've done quite a bit of looking (and phoning people,
and so on.)


[...]

> > > <Restriction rdf:ID="PaternalDominantInheritance">
> > >     <onProperty rdf:resource="#father">
> > >     <toClass rdf:resource="#PaternalDominantInheritance">
> > > </Restriction>
> >
> > I don't think that's what you meant.
> >
> > That says that PaternalDominantInheritance is *exactly*
> > those things whose fathers have type PaternalDominantInheritance;
> > that there are no other conditions (like actually
> > having a disease) for being in this class.
> >
> > I think you meant that PaternalDominantInheritance
> > is a *subClass* of the (father hasClass PaternalDominantInheritance)
> > restriction, right?
> >
> > OK, that's still circular, but it's a "primitive" class;
> > i.e. you're going to be able to communicate to the
> > machine *necessary* conditions for being in this class,
> > but not *sufficient* condidions, right?
> >
> 
> I don't quite get "primitive" vs. "defined" to be honest. I understand the
> necessary and sufficient,

That's all there is to it.
(that is: assuming *I* understand it!)

> but for example:
> 
> <Restriction rdf:ID="BlueThing">
>         <onProperty rdf:resource="#color"/>
>         <toValue>blue</toValue>
> </Restriction>
> 
> That is primitive, right?

nope. It's defined.

> I don't get why it would be wrong to label
> _anything_ which has the property color="blue" as a :BlueThing.

it wouldn't.

> I mean
> anything can have any number of super classes, right?

yes.

> Is primitive vs.
> defined simply an 'implementation' mechanism of reducing the number of
> classes a thing belongs to? i.e. are "defined" classes, classes that you are
> telling the system that you want it to make inferences on etc.? Otherwise
> what is the harm of letting things be members of the instance sets of
> primitive classes?

There's no harm; I just didn't think you meant to say that there
were no other conditions for being in the PaternalDominantInheritance
class... I presume one has to actually have some disease
to be in that class.

-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/

Received on Monday, 22 April 2002 18:43:03 UTC