Re: OWL semantics ( with focus on an axiomatization)

A bit about the history of the DAML axiomatization might be useful.
This is aimed at supporting that an axiomatization such as [1]
has value and I think that we will want to do one.  Also, Richard Fikes and I have
already discussed doing an update to [1] for OWL once OWL stabilized.

after daml-ont was released it was clear that there were disagreements (among
committee members and the public) as to what terms meant.  Richard Fikes and I
agreed that we should write a more precise specification of the meaning of the
terms.
I suggested a denotational semantics.  Richard suggested that specification in KIF
might be useful to a significant segment of people.  We decided to go with
Richard's hypothesis and provide a KIF-based specification as the first
specification.  (as you all know of course there was also a model-theoretic
specification done later by Peter et al).

We did find a number of people who stated that the axiomatization Richard and I
generated was quite useful since it was operational for their systems.  Also, we
received feedback that our axiomatization was in a form that people found easy to
read and understand.  (I am not claiming that that is true for everyone, I am just
relaying feedback from some readers of the document who stated that this was true
for them..  I state this just to support the notion that some people will prefer an
axiomatization form.)
Internally, we used our specification to encode our own special purpose daml
inference rules for our FOL reasoner (JTP at Stanford).  We had other people tell
us that they took the set of rules as their basis for a coding (and sometimes
checking) effort.

As a related piece of evidence, my thesis [2] includes in it a set of inference
rules that were used to encode CLASSIC (an earlier description logic system).  That
set of inference rules (in natural semantics style) was used for the explanation
module [3] for CLASSIC and that set of inference rules is the set that formed the
basis of the implementation.
Most CLASSIC papers include the denotational semantics in them (and do NOT include
the set of inference rules) and I completely agree that for many purposes, the
short semantics specification such as the model-theoretic one for daml+oil has
great value.

I do not dispute  that it is work to get a mostly correct, mostly complete
axiomatization done.  Getting it to the point that you believe it is complete is
more work.  Getting to the point that you believe it is totally correct is more
work.  Proving it correct is something that people rarely do.
Richard Waldinger took our axiomatization and worked with specware to do some
validation work.
We have a draft paper on that work (it does not claim correctness for the entire
axiom set though.).

My main reasons for writing this email are to:
a - provide additional support that an axiomatization is useful
b - state that I am willing to provide work on generating one
c - point out that many implementors will need to have a set of axioms in order to
do their encoding.
(we can provide one or we can leave it up to them to do the work that I believe we
all think is difficult to generate one).
d - point out in particular that even without a proof that our entire
axiomatization is correct (and in fact without a hope that we would ever have a
such a proof), we received feedback that stated that the axiomatization was useful


This email is not trying to dispute that:
- a model-theoretic semantics is useful   (and more strongly, i think we definitely
want one)
- proving correctness of an axiomatization is hard (and more strongly is something
that I do not think we want to make the committee take on).

Deborah

[1] http://www.w3.org/TR/daml+oil-axioms
[2] http://www.research.att.com/~dlm/papers/thesis-abstract.html
[3] http://www.research.att.com/~dlm/papers/explain-abstract.html

"Peter F. Patel-Schneider" wrote:

> From: "Jos De_Roo" <jos.deroo.jd@belgium.agfa.com>
> Subject: Re: OWL semantics
> Date: Tue, 6 Aug 2002 15:02:45 +0200
>
> > [...]
> >
> > > However, what does that have to do with proving that a large number of
> > > axioms (say, about 140) are correct with respect to a model theory, or even
> > > that they capture the meaning of a complex language (such as OWL)?  To
> > > prove the former requires considerable work - work that can build on
> > > similar efforts in the past, but still new work.  To show the latter also
> > > requires considerable work - work of a very different kind, but still
> > > work.  Will this work have to be done before the WG is finished?  Who will
> > > do it?
> >
> > I don't see an alternative to avoid the work
> > I would even recommend that we have many, many more
> > such attempts as http://www.w3.org/2002/03owlt/ontAx.n3
> >
> > -- ,
> > Jos De Roo, AGFA http://www.agfa.com/w3c/jdroo/
>
> If we don't produce an axiomatization, then we don't need to have this work
> done.
>
> Simple.
>
> peter

--
 Deborah L. McGuinness
 Knowledge Systems Laboratory
 Gates Computer Science Building, 2A Room 241
 Stanford University, Stanford, CA 94305-9020
 email: dlm@ksl.stanford.edu
 URL: http://ksl.stanford.edu/people/dlm/index.html
 (voice) 650 723 9770    (stanford fax) 650 725 5850   (computer fax)  801 705 0941

Received on Tuesday, 6 August 2002 16:05:24 UTC