- From: Jim Hendler <hendler@cs.umd.edu>
- Date: Fri, 9 May 2003 18:22:25 -0400
- To: Ian Horrocks <horrocks@cs.man.ac.uk>, Jeremy Carroll <jjc@hpl.hp.com>, Brian McBride <bwm@hplb.hpl.hp.com>
- Cc: www-webont-wg@w3.org
THE FOLLOWING IS NOT A PROPOSAL, IT IS A SOCRATIC ARGUMENT TO HELP ME UNDERSTAND THE ISSUE (and Jeremy's position) Jeremy- I don't advocate this, but let's play hypothetical. Suppose the WG decided to drop Lite and DL completely. We would say "OWL is a vocabulary for which we think there are good uses" and release the current vocabulary and all the current tests labeled Full. Your objection woudl seem to go away, because it is an objection to the particular design of DL v. specific issues. OK, we then take everything currently called an OWL Lite or OWL DL tool, and we argue that each and every one of them implements a proper subset of Full, some provide some guarantees the whole of full doesn't (i.e. some guarantee decidability, some only work in a smaller subset, but have effective algorithms for all non-datatype reasoning, etc.). It is clear that we would then have a lot of implementations, and for most features we'd have more than two interoperable tools. Even better, we could trivially write a program to XSLT any document into DAML (some lose a little expressivity, but that's okay since there's no objection to Full being powerful enough to go beyond many tools) and could claim every DAML tool as an implemented OWL tool *. So, given the argument to date, I would argue that the WG in that situation could move to PR with little or no CR, since we would have lots of implementations and at least two things handling every subset and no expectation of decidability. OK, so why is it then a bad thing that would require 6 months of CR to include two named subsets of OWL that happen to be decidable in one case, and effective in another? - JH * - I am actually considering doing this for real -- for a large subclass of OWL (most of OWL DL) this would give us 40+ implemented OWL tools including RACER and other reasoners - the "demapper" looks like it would take about 20 lines of Perl code, At 18:03 +0100 5/8/03, Ian Horrocks wrote: >On May 5, Jeremy Carroll writes: >> >> >> A short message. >> >> I am not expecting reasoners to work fast, just to reason >>successfully within >> some time period that is less than forever. >> >> This is not about quantitative capabilities of reasoners, but qualititave >> capabilities. Can they reason about finite classes and their cardinalities - >> or are they really only capable of advanced reasoning about >>infinite classes. >> >> OWL DL and Full permits descriptions that describe both infinite and finite >> classes. >> The DL community have much more experience in reasoning about >>descriptions of >> infinite classes than finite ones. > >Jeremy, > >It is hard to understand why this is, in itself, a good reason for >having a lengthy CR period, or what you expect would be achieved >during such a period (unless it were to be years rather than >months). We know that Full is not even decidable. We know from >DAML+OIL that no effective algorithm for all of DL is available. We >have known for a long time that even our "Lite" language requires, in >the worst case, time exponential in the size of the ontology in order >to perform common reasoning tasks. All of this was accepted by the WG >at the time we were designing the language(s). > >As far as your expectation for reasoners to reason "successfully" is >concerned, success may be rather hard to quantify (or even qualify): >the medics here deemed their reasoner to be successful so long as it >was able to classify their ontology overnight - this would obviously be >considered an abject failure in some applications. It is likely that >Lite reasoners will be, in general, more "successful" than DL >reasoners, which will be more "successful" than full reasoners. Your >point about reasoning with finite classes is well made, but a complete >investigation/solution of this problem may take an indefinitely long >time. It might be worth pointing out that DAML+OIL is also able to >express finite classes, and this does not seem to have been a barrier to >its widespread adoption. > >Regards, > >Ian > > > >> >> Jeremy >> -- Professor James Hendler hendler@cs.umd.edu Director, Semantic Web and Agent Technologies 301-405-2696 Maryland Information and Network Dynamics Lab. 301-405-6707 (Fax) Univ of Maryland, College Park, MD 20742 240-731-3822 (Cell) http://www.cs.umd.edu/users/hendler
Received on Friday, 9 May 2003 18:26:40 UTC