Re: Entailments and normativity

Jeremy-






At 11:15 +0100 1/21/03, Jeremy Carroll wrote:
>I have made, a possibly successful, case to my HP colleagues that OWL is about
>documents:
>- syntactic forms
>- semantic consistency

interesting, I don't agree with that, but if your HP colleagues are 
comfortable I am okay with saying that OWL (among the many other 
things it is good for) is about documents

>
>The WG seemed minded to restrict our concerns to precisely those when it came
>to specifying software. In particular there was no desire to specify any more
>complicated reasoning component, such as one that could actually run our
>entailment tests.

   I find myself  puzzled.  I thought what the group agreed at the WG 
is that we were only willing to define compliance with respect to OWL 
documents, not that we thought that other tools (including reasoners) 
were somehow less important -- we spent the great part of the past 
year and a half making sure OWL would apply to reasoners, I suspect 
we don't want to throw that away.  I also point out the markup tools, 
crawlers, scrapers, etc. that already work with OWL -- let's see what 
we actually have before we make decisions.
  In short, OWL has already been being put to use in many ways (and if 
we count DAML+OIL in VERY many ways including applications people are 
making money off of and/or using daily in govt applications).  I 
personally don't feel comfortable defining compliance of any software 
- that's different from saying I don't think we have a good 
specification of reasonable reasoning capabilities -- in fact, just 
the opposite - our entire language design was influenced by this 
(otherwise we could just publish Lite and Full - why have Owl DL if 
we don't care about reasoning???)


>So we have:
>- a semantics that permits discussion of the meaning of a single document
>(specifically its consistency)
>- a semantics that permits discussion of the relationship between two
>documents (entailment)
>- three syntactic levels
>
>What I have heard we want to have as CR exit criteria is adequate exploration
>of the first and third of these; and not much on the second.
>
>This could possible be better reflected by making the entailment part of the
>semantics informative rather than normative.
>
>Possible text near the beginning of the AS&S:
>[[
>OWL specifies an RDF vocabulary and three syntactic levels: OWL Lite, OWL DL,
>OWL Full, each of which specifies a set of RDF documents. OWL normatively
>specifies through the use of model theoretic semantics the consistency of any
>such document. The model theoretic semantics also permits the discussion of
>entailment relationship between pairs of documents. These relationships are
>informative. In particular, implementors of OWL Consistency Checkers are free
>to use alternative semantics as long as these are logically equivalent to an
>OWL Consistency Checker built using the semantics specified here. No other
>constraints on the semantics of OWL implementations are normatively
>specified.
>]]

I would oppose this (strangely enough) - we spent so long working out 
a normative semantics that would include entailments, and have 
developed a very clean system with respect to this -- let's not throw 
out the baby with the bathwater -- and again, we have Euler, Dan's 
cwm stuff, Racer (an Owl interface being written as we speak), Fact 
(which I assume Ian will put some sort of Owl front end on), and the 
Network Inference reasoner - so I think we can safely claim to have 
some

>
>(with appropriate 'normative' and 'informative' markers elsewhere)
>
>The motivation here is that much of the difficulty HP has with the current
>specification of OWL is to do with what is expected of OWL implementations,
>and an apparent mismatch between that and the sketch of CR exit criteria
>created at the f2f. Specifically we do not intend to migrate Jena to the OWL
>Semantics any time soon, nor do we see competing systems planning such a
>migration either; nor do we see a WG intending to make the migration of such
>a system to OWL (and its semantics) a CR exit criterion.

I heard several people at the f2f say they were willing  or expecting 
to do this, and I know of at least two ongoing efforts.  I would be 
happy to have the existance of OWL DL reasoners be a criterion for 
exit (and I never heard anyone say otherwise at the f2f).  However, 
we do need some notion of what we test against - unless you want our 
implementation report to include theoretical proofs of algorithm 
completeness (instead of implementations) - I've been going through 
the DL Handbook, and I'd be happy to offer the theory -- there's 
plenty of it :->



>To give a specific example, the comprehension principle is part of OWL
>semantics, i.e. that various classes and restrictions necessarily exist.
>These can be turned into consistency tests at the OWL Full level but not at
>OWL DL or OWL Lite. e.g.
>
><owl:Restricition>
>   <owl:onProperty>
>      <owl:ObjectProperty rdf:ID="p"/>
>   </owl:onProperty>
>   <owl:hasValue>
>       <owl:Thing rdf:ID="v"/>
>    </owl:hasValue>
>    <rdf:type>
>       <owl:Class>
>          <owl:complementOf rdf:resource="&owl;Class"/>
>       </owl:Class>
>    </rdf:type>
></owl:Restriction>
>
>This says that a particularly restriction is not an OWL Class, and this is
>inconsistent.
>This:
>(a) seems plausibly straight forward to implement (I wouldn't be surprised if
>Euler already could prove this)
>(b) is anyway in OWL Full where we are being explicit about the lower
>implementability goals.
>
>The alternative OWL DL entailment:
>
><owl:ObjectProperty rdf:ID="p"/>
><owl:Thing rdf:ID="v"/>
>
>DL-entails
>
><owl:Restricition>
>   <owl:onProperty>
>      <owl:ObjectProperty rdf:ID="p"/>
>   </owl:onProperty>
>   <owl:hasValue>
>       <owl:Thing rdf:ID="v"/>
>    </owl:hasValue>
></owl:Restriction>
>
>seems to present more difficulty, particularly in a forward chaining
>environment, and the WG does not seem minded to want to see even one complete
>OWL DL reasoner that could do this before exiting CR.

umm, what do you mean "the WG does not seem minded" -- if you propose 
this as a test, we can discuss it as we do any test that someone 
wants to propose.  If the group feels this is a fair test, put it in 
the test suite and we'll see if Euler and the others can do it.

>(I note that a unionOf example would be a lot less compelling, partly because
>both the entailment and the inconsistent file are at the same OWL DL level)
>
>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 Tuesday, 21 January 2003 08:30:34 UTC