Re: OWL Lite semantics

>(This responds in part to the points raised by Ian, Pat and Peter).
>
>I think I would rather pose the proposal as a proposal to bless a 
>particularly style of incomplete reasoning.

I would suggest that as a (very, very) general principle, that 
incomplete reasoning should not need to be blessed. We should expect 
that many reasoners on the Web will be incomplete, for a variety of 
reasons. Which is fine, and shouldn't need to have prayers said for 
their intercession.

>Seen from an OWL Full perspective ...
>OWL DL reasoners must be complete with respect to OWL DL semantics 
>and sound with respect to OWL Full semantics.
>Any non-OWL DL entailments found can be distinguished since either 
>the premises or conclusions are syntactically excluded from OWL DL.
>However, at one level an OWL DL reasoner is an incomplete OWL Full reasoner.
>
>Similarly, under my proposal an OWL Lite reasoner must be complete 
>with respect to the proposed OWL Lite semantics, and sound with 
>respect to the OWL Full semantics.

Is that supposed to be a prediction, or a requirement?

>This seems wholly analogous, and will introduce no new practical 
>problems, since the reality is that not every OWL Lite system will 
>provide complete reasoning with respect to the OWL DL semantics over 
>the OWL Lite subset. By giving developers a lower bar to aim at, we 
>can reasonably expect them to hit it, thus aiding interoperability 
>by giving clear (and achievable) subsets to aim at.

I don't think completeness has got anything to do with 
interoperability. Two incomplete systems can interoperate. In fact, 
they may be able to help one another by each one handling some 
inference that the other cannot. Barriers to interoperability are 
caused by having different semantics for the same syntax, a situation 
I thought we had just spent the best part of a year's hard work 
managing to avoid.

>I will respond to the requirement to justify the achievability claim 
>later. (tomorrow or wednesday).
>
>In my opinion, it is unrealisitic to expect all OWL Lite developers 
>to achieve complete reasoning with repect to the current OWL DL 
>semantics over the current subset.

I agree, so they don't. Fine. If we had done RDF datatyping properly, 
all extant RDF reasoners would have been incomplete. Which would also 
have been fine.

>We need to do one of:

We don't need to do anything. Anyone can use an incomplete reasoner 
any time they want. If they also want to try inventing a tweak to the 
standard MT which makes their reasoner seem complete (WRT to the 
tweaked MT), then also fine, if it  makes them feel better. That's 
what LP theoreticians have been up to for the last decade or more. It 
doesn't make the software run any faster, but it gives them something 
to do. But that's up to them to do, not up to us. And we couldn't do 
it anyway since there will probably be about a hundred different 
incomplete reasoners out there eventually, all incomplete in slightly 
different ways but all able to send messages to one another, because 
all sharing a common semantics for the interchange language.

Pat



-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam

Received on Monday, 9 December 2002 18:19:51 UTC