Re: proposals for the underlying principles of OWL

Having looked through Pat's latest:

   http://www.coginst.uwf.edu/users/phayes/RDFS2OWL-D.html

I believe that it goes a very long way to reconciling, Peter's


   http://www-db.research.bell-labs.com/user/pfps/owl/semantics.html

and the RDF(S) model theory.

While this is certainly worth a big cheer, I am not ready to celebrate by
closing off as many issues as Peter would like. Nor is it clear that Pat's
work, which shows sufficient compatibility between the RDF(S) approach and
the Description Logic approach, necessarily results in the group chosing the
DL approach (Peter's MT) as normative, and Pat's as non-normative support.

I think it may be helpful to formally record some decision that progress is
being made, I will think a bit more and try and suggest a substantial weaker
proposal that I could support.

This may partly be about speed of closure. In terms of process, if the WG
feels that Peter's proposals for how to address OWL semantics are the way to
go, I would prefer a decision to publish WDs showing that semantics followed
by community feedback, and then the formal decisions that close the
associated issues.

I will try and indicate here some of the issues that I think merit further
discussion.

> Issue 4.6 equivalentTo
  I think there are other resolutions to this than simply deleting
equivalentTo.
I would prefer dropping all the sameXXXAs and keeping equivalentTo. This
could remain strongly typed so that
  xx rdf:type owl:Object .
  yy rdf:type owl:Class .
  xx owl:equivalentTo yy .
is a contradiction.
  I also am increasingly believing that there are positive advantages in
having the syntactic expressions in the domain of discourse (which is the
case in Pat's approach). Thus issues about intensional and extensional
sameClassAs are not, at least for me, resolved by Pat's work. What Pat as
shown is that the extensional sameClassAs is at least an option which is
compatible with RDF(S) model theory.

  Peter's MT cannot extend beyond the abstract syntax, with it's
limitations.
  Pat's MT is automatically extended beyond that, and explicitly cut back to
match Peter's world. Maybe there are areas outside Peter's proposed abstract
syntax that would be useful

> Issue 5.1 Uniform treatment of literal/data values
>	There is a strict separation between OWL object and data values.
>	Removing the separation has computational consequences.

We have had some discussion of this, but I didn't feel we reach closure
(i.e. consensus that the computational consequences were sufficiently
compelling to maintain this separation). I find Dan's same state example a
compelling use case and would be happy to have another way of making an OWL
reasoner not perform as the cost.

> Issue 5.10 DAML+OIL semantics is too weak

You and Pat have shown one approach to resolving it.
At times both I and Dan have argued that the alternative of simply
dismissing this issue, and asserting that the DAML+OIL semantics is actually
adequate is also a possible way forward.
I personally find Pat's herbrand closure clever, but perhaps too clever.
Certainly worth an appendix, but the community might be better served by the
weaker semantics ...

> Issue 5.19 Classes as instances
I suspect that the advocates of this issue will not be pleased to simply
have it dismissed.

Jeremy

Received on Thursday, 29 August 2002 01:20:47 UTC