Re: Sketch: reasoning conformance levels

From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Subject: RE: Sketch: reasoning conformance levels
Date: Fri, 8 Nov 2002 15:14:38 +0100

> > I like the clearly defined levels of conformance.
> > "Almost" conforming software is one of the reasons
> > DoD still buys mostly all Microsoft products.
> 
> I think there is some buy-in to trying to get two clear statements of
> reasoning conformance:
>  OWL DL  - this seems relatively easy, basically complete with respect to
> the OWL DL semantics.
> 
>  OWL Lite - this seems more problematic.
>    - can we specify a non-onerous relationship owl-lite-consequence
> (subrelationship of owl entailement) 

Sure this is possible.  Go right ahead, but don't expect to get a
relationship that will be non-contentious.

> so that all owl-lite reasoners:
> MUST
>  If A owl-lite-consequence B prove that A owl-entails B.
> MUST NOT
>   Prove A owl-lite-entails B if A does not owl entail B.

> A very simply way would be to mark some of the test cases as
> owl-lite-conformance tests. 

This would not be a reasonable way of proceeding.  Test cases only provide
points in an infinite space.

> We probably also want a statement of principle.

Yes, but what statement?  There is a pile of trouble here.

> But there could be an OR there.
> e.g.
>  A owl-lite-entails B if
>    either there is a test case marked as an owl-lite entailment
>    or some words to be defined.
> 
> We would then try to mark all the test cases which were covered by the
> words, and maybe some others that we thought appropriate.



> Jeremy
> 
> ===
> That MUST NOT clause looks very general:
> 
> Soundness:
> Any conforming OWL reasoner (at any level) MUST NOT prove B from A if the
> OWL entailment does not hold in OWL Full.
> 
> Anyone unhappy with that?

You have to be careful here, because of the relationship between OWL/DL and
OWL/Full.  

> Jeremy

peter

Received on Friday, 8 November 2002 11:34:46 UTC