Re: "FAST" OWL and "Greater" OWL

From: "Jonathan Borden" <jonathan@openhealth.org>
Subject: Re: "FAST" OWL and "Greater" OWL
Date: Tue, 24 Sep 2002 23:18:48 -0400

> Peter F. Patel-Schneider wrote:
> > >
> > > The risk is that if Greater OWL is proven untenable, then we are left with
> > > Fast OWL. Sounds fine to me.
> >
> > Actually the burden of proof has to be the other way.  Large OWL has to be
> > proven tenable to be the basis for a standard.
> 
> I shudder to say this, but there is no _requirement_ that Large OWL be
> proven tenable to be the basis for a "standard" -- you assume that
> "standardization" work is more precise than it is. I know of a number of
> cases where we have pushed standards (real standards such as ASTM, ISO etc)
> out the door somewhat prematurely with full knowledge that we can correct
> the mistakes at a later date. My allusion to the IETF process what to
> highlight this fact, for example the RFC process where an RFC spends quite a
> long time as an RFC, and typically undergoes many many revisions, before it
> has any prayer of being a real "Internet Standard".

I do know that several standards have had errors, so, perhaps, ``proven''
is too strong.  

I learned, from a lawyer of all people, the ``burst of laughter'' test.  If
you need to convice a jury that your company's way of doing business is
reasonable, and the jury bursts out laughing when it hears them, you have
lost your case.  I believe that this would be the reaction if we put out
Large OWL without proofs that it actually was a viable logic.

> In any case for the W3C REC process, there is _no requirement_ that the
> model theory for Large OWL be proven tenable before our documents move to
> full REC. More important are working implementations, test cases etc.

Without a working model theory, the very idea of working implementations,
test cases, etc., etc., for Large OWL doesn't make sense.

> That's not to say that I advocate pushing something half baked out the door.
> Rather I'd be quite comfortable saying (something like): Here is "Fast OWL"
> which is well characterized and has good, fast implementations. Here is
> "Large OWL" which we have high hopes for on the Semantic Web and which is
> more consistent with the "RDFS way" of doing things. "Large OWL" remains a
> research project for the following reasons...

> I'm totally comfortable with such an approach.

I suppose that this might be possible, but it would have to be very
carefully stated, and I would view it with deep skepticism. 

> > > What is the real problem with that except that
> > > it requires a bit more work to develop _both_ model theories? (on the other
> > > hand if Pat is willing to take on this work then I for one am in no position
> > > to complain...)
> >
> > I have already put considerable work in on Large OWL, at least compared to
> > the time I have available between now and the next deadline, so it is not a
> > matter of ``a bit more work''.
> 
> Fair enough. When I say "a bit more work", I say that with a wink. It is a
> real issue about _how much_ work it will be and who has the spare cycles to
> devote etc. This real cost has to be weighed against the potential benefits,
> research contribution etc. e.g. is it going to be something interesting
> enough for you (or whomever) to put your name on the document, and invest
> the time to work at least some of the issues out.

The cost does have to be weighed against the benefits, but some things
count for very little in the Working Group's cost-benefit analysis, and
research contribution should have a zero direct weighting.  (It may, of
course, have an indirect weighting, as a way of inducing a Working Group
member to do some work, but having a standard have some research
contribution, has, at best, zero benefit.)

> I guess my question comes down to this: Assuming we have "Fast OWL" and it
> works, what is the downside of "Large OWL" even if it is proven not to work
> in the future. I'd say that implementors, as long as they are made aware
> that "Large OWL" may not pan out, would be taking a risk that their hard
> work will be a waste of time. On the other hand there are scores of
> implementors willing to put in the time on RDF/RDFS without even having a
> model theory of any sort, so these folks already seem willing to take these
> risks. Again, assuming that we have "Fast OWL", why do I care if "Large OWL"
> doesn't work? (speaking loosely).

Yes, assuming that we already have Fast OWL, there is much less reason to
care about whether Large OWL is being investigated.  However, we do not yet
have Fast OWL, or even Weak OWL.  

My view here is quite strong.  Weak OWL (and Fast OWL as well, I think)
have been presented to the Working Group and received very strong support.
The Working Group chairs should be doing everything in their power to get
Working Group members to work on Weak OWL and Fast OWL so that they can be
completed in time for consideration at the upcoming Face-to-Face.  Either
that or they should pick some other acceptable semantics for OWL, either a
stand-alone semantics or an RDF-compatible semantics, and work to have that
ready for consideration at the upcoming Face-to-Face.

> Jonathan

Peter F. Patel-Schneider
Bell Labs Research

Received on Wednesday, 25 September 2002 06:59:54 UTC