Re: rdfs:subClassOf and metaclasses

At 09:28 03/07/2002 -0700, Piotr Kaminski wrote:

>From: "Brian McBride" <bwm@hplb.hpl.hp.com>
> > The first step in this process is to establish exactly what the problem
> > is.  The only "problem" I see below is that you don't think the current
> > proposal "makes sense".  We need something a little stronger than that,
> > e.g. an internal contradiction.
>
>
>I'm still thinking through the ramifications and hope to come up with
>something stronger soon.  I think the only way to reach an internal
>contradiction is through the domain and range constraints, but I'm not sure
>they're strong enough for the job.

Its going to be hard; RDFS has no concept of negation.

>   I'm convinced I could find something in DAML+OIL, but that's a much 
> weaker argument At this point some help from people who've been doing 
> this kind of stuff much longer than me would be very welcome.
>
> > i.e. I conclude that the class of resources defined by Brian is a subClass
> > of the resources defined by Piotr, i.e. every resource defined by Brian is
> > also a resource defined by Piotr. That is not a valid inference.
>
>Precisely.  You're getting an unwanted inference, so there's an error in the
>model you started out with.

There is something a little circular in your argument.  I unwind it thus:

   1. the test case I gave is one that RDFS can (and should be able to) 
represent
   2. adding your rule results in an invalid inference
   3. ergo your rule is broken

To break this argument, we need evidence that your rule brings greater 
benefit than the set of models which it rules out.

Brian

Received on Wednesday, 3 July 2002 13:55:16 UTC