W3C home > Mailing lists > Public > www-webont-wg@w3.org > October 2003

Re: Guide tests

From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
Date: Wed, 08 Oct 2003 09:52:05 +0100
Message-ID: <3F83D035.9050605@hplb.hpl.hp.com>
To: Ian Horrocks <horrocks@cs.man.ac.uk>
Cc: Dan Connolly <connolly@w3.org>, Jim Hendler <hendler@cs.umd.edu>, www-webont-wg@w3.org



Jeremy:

>>I am not sure what Martin's current position is - merely that without a 
>>single implementation passing these two tests that the response given 
>>during LC is invalidated.
>>


Ian: 
> Surely you can't be serious! It has always been crystal clear
> (including in our responses to LC comments) that we don't currently
> have a decision procedure for OWL DL (we removed the definition of
> "complete OWL Lite consistency checker" from test for this reason).
> The inability of existing implementations to reliably prove the
> consistency of OWL DL ontologies is an obvious consequence of this
> fact. Moreover, the spec allows conforming consistency checkers to
> return "Unknown" when they are unable to determine if a document is
> consistent or inconsistent.
> 


Of course you are right that there was much in the response that remains 
pertinent, and requesting reopening discussion is not intended as a 
free-for-all.

However text such as

[[
The WG has been made aware of implementations of OWL DL that include
both inverseOf and oneOf and which seem to be performing well in
practice.  The working group will definitely consider their status
and usability before deciding on our schedule with respect to
Candidate Recommendation and Proposed Recommendation.
]]

suggests that straightforward examples including both features should be 
tractable.

The decisions made about what was in DL were known to be pushing the 
boundaries, and the changes we made during LC reflect that we were not 
expecting implementations to be able to pass maybe some of the more 
difficult DL tests (e.g. 3-sat encoded with oneof).

The guide examples, in contrast, are not attempting to be clever or to be a 
challenge to systems, merely exemplory use of DL.

As far as I can tell, most of my colleagues see the current failure to 
prove these tests as justifying our earlier scepticism about the usefulness 
of the DL boundary, but accept that given where we are in the process, the 
likely outcome will be adequate text in the guide exlaining how to remain 
within a smaller more tractable subset.

Jeremy
Received on Wednesday, 8 October 2003 05:00:49 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:58:02 GMT