Re: Guide tests

On Tue, 2003-10-07 at 07:51, Jim Hendler wrote:
> At 10:03 AM +0100 10/7/03, Jeremy Carroll wrote:
> >Jim:
> >
> >>  I think we should wait to see if these get passed - I'm not yet 
> >>convinced that these are not passable
> >
> >
> >Martin has indicated to me that he would be happy with a single 
> >system that passes these tests. Other HP developers have heard 
> >informally from other groups who seem to believe that the current DL 
> >profile is implementable, which is part of the reason for our 
> >surprise.
> >
> >I take the action to propose reclassifying these to be withdrawn, 
> >for now at least.
> >
> 
> at least on hold
> 
> >
> >>  p.s. Even if these do prove to be too hard to prove consistent at 
> >>the current time, I don't see how that would change the response to 
> >>Merry -- proving things consistent is not the only reason to have 
> >>OWL...
> >>
> >
> >
> >There are choices such as:
> >- do what Martin suggests and redefine DL downwards to match 
> >implementations. While this would require a second last call, we 
> >could then go straight into PR without another CR. (Note: I do not 
> >know whether HP would want this or not - I have heard arguments both 
> >ways).
> >- add sufficient health warnings to the guide (where 'sufficient' 
> >would be a topic of debate) and reply again indicating that these 
> >are the main defence against Martin's concerns about the 
> >implementability of DL
> >
> >I would like to see test results from the heavyweights: Racer, Cerebra etc.
> 
> 
> I still await an argument that proving large and complex ontologies 
> consistent is the only reason d'etre for OWL

I don't think Martin's position depends on such claims.
Meanwhile, proving large and complex ontologies consistent
is among the very few tasks the WG chose to highlight
with a software conformance clause. He evidently finds
insufficient implementation experience to justify
advancing the spec to Proposed Rec as is.
I think his point is well made, though perhaps not
compelling. I haven't made up my mind whether I find
it convincing or not.



>  - I've never believed 
> that and continue not to.  Here's a thought - the NCI ontology was 
> carefully crafted to be in OWL Lite, and passes various syntax 
> checkers at the Lite level.  It also doesn't use any particularly 
> complex OWL - however, I suspect proving it consistent would be hard 
> for Lite implementations (because it contains about 17000 classes - 
> so the problem is simply bulk, not complexity) - yet several groups I 
> know about are using it routinely -- other examples include OpenCYC 
> and our wine ontology -- so I would resist any non-editorial changes 
> caused by a mistaken notion that consistent checking is more 
> important than other kinds of inference on the Semantic Web...

"mistaken notion"? I don't think it's a mistake to conclude,
from the fact that our specs have a conformance clause for
consistency checkers and not other sorts of software, that
consistency checking is more important.

If we don't want people to come to that conclusion, we
should change our spec.

(Recall that I don't support having the "consistency checker"
conformance clause; I abstained. I'd be happy to see it go.
http://www.w3.org/2001/sw/WebOnt/ftf5.html#Test
)

-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/

Received on Tuesday, 7 October 2003 09:58:53 UTC