RE: Cyclic Classes/Properties [was: Re: DAML Correction: Same Is Not A Sub Of Sub]

> From: Sean B. Palmer [mailto:sean@mysterylights.com]
> > Apologies if this has been raised a million times before.
> 
> Well, not a million, but very nearly (O.K., three references).
> 
> http://lists.w3.org/Archives/Public/www-rdf-comments/2000AprJun/0045
> http://lists.w3.org/Archives/Public/www-rdf-logic/2001Feb/0106
> http://www.w3.org/2000/03/rdf-tracking/#rdfs-no-cycles-in-subClassOf

Those all raised the issue of classes rather than properties.  However, if
you have a hierarchy of properties, exactly the same approach works with
properties.  If you look at Ian Horrocks' FaCT reasoner [1], for example, it
deals with role equivalence by adding each role to the other's list of
ancestors.

> Sure is an interesting topic though. The RDF Schema specification only
> restricts this by prose, which makes it very difficult for people to
> reason with... how much more expressive would RDFS need to be in order
> to constrain this properly?

For my $0.02... roughly as expressive as DAML+OIL.

> An additional problem is that if you let class hierarchies by cyclic,
> then you might start getting inconsistencies quite easily

The last time this came up (late Feb/early March), a lot of heat was
generated.  The DAML+OIL approach is that a cycle in a class hierarchy
defines a set of equivalent classes --- so if A is a subclass of B, B is a
subclass of C, and C is a subclass of A, then A, B, and C are equivalent.
The last two links to which you refer appear to make this point rather well,
showing where RDFS is not ideal.

> Couldn't DAML or something
> introduce a daml:CyclicClass / daml:CyclicProperty?
[...]
> Unfortunately, all instances of rdf:Property and
> daml:CyclicProperty would have to be daml:Disjoint...

Also known as 'if I don't look at it, maybe it'll go away' :-).

If one of the goals of RDF-Logic is to allow as much as possible to be
expressed using RDF and RDFS, I think a more appropriate approach is to
request (hopefully) small changes to the RDF(S) specs.  Yes, it would be
possible to define a whole new structure on top of triples that completely
ignores RDFS; but I don't think that helps.

I know I'm re-opening a discussion that DanC closed remarkably succinctly in
http://lists.w3.org/Archives/Public/www-rdf-logic/2001Apr/0217.html, so I'll
try to connect to some nuts and bolts:

What would break in RDFS if the wording "A property can never be declared to
be a subproperty of itself, nor of any of its own subproperties" were
removed?

		- Peter

Received on Sunday, 6 May 2001 07:28:06 UTC