W3C home > Mailing lists > Public > public-owl-wg@w3.org > January 2008

Re: Proposal to close ISSUE-69

From: Jeremy Carroll <jjc@hpl.hp.com>
Date: Wed, 23 Jan 2008 14:22:06 +0000
Message-ID: <47974D8E.4070606@hpl.hp.com>
To: Bijan Parsia <bparsia@cs.man.ac.uk>
CC: "Web Ontology Language ((OWL)) Working Group WG" <public-owl-wg@w3.org>


There are at least two incompatibilities of concern:

- user expectation
- requirement that implementations have equivocal URIs

On user expectation - the current situation is that users are taught 
that each URI means one thing. In the DL world, it has to be one of the 
different major classes. In the Full world, it can have multiple 
expressions - but it is still a single unified concept, and, for 
example, test cases in the classes-as-instances paradigm demonstrate 
this univocality.

In the post OWL 1.1 world, users taught in the DL paradigm will learn 
scoping rules - the scope of a name depends on its context of use - 
which simply do not hold in the RDF compatible view. In fact, many 
proponents of the RDF compatible view, and much of the software 
available, is built on architectural principles that insist that the DL 
scoping rules are simply wrong (personally I don't much like that 
architectural principle, but still).
Having learnt these rules it will be harder for the users to migrate 
from DL systems, to RDF compatible systems, and so a key part of 
interoperability - the lack of lock-in - is lost.

On implementation, currently within a triple based system, such as Jena, 
it is possible to work fairly successfully with DL ontologies. This 
occurs both in obvious interoperation such as the Jena & Pellet 
combination, and also by using Jena rules - which, while weaker than a 
DL reasoner - can have specific additional rules added to cope with the 
needs of specific ontologies etc.

If correct reasoning over a DL ontology will require that say, the 
individual and class corresponding to http://example.org/car behave as 
separate objects, then the Jena implementation, which as a single object 
corresponding to each URI will be inadequate and require extensive 
change. This change becomes all the wrose since the full semantics still 
treats both the class and the individual as a single thing.

Jeremy









Bijan Parsia wrote:
> 
> ISSUE-69 claims that punning is incompatible with OWL Full because 
> punning is "role bases" i.e., a URI "plays"  different roles (sometimes 
> a class, sometimes an instances). In contrast, in OWL Full "this [ed. 
> doesn't know what 'this' is] is identity".
> 
> While it's true that punning is weaker than Hilog semantics (in the 
> sense of missing entailments) there is clear sense in which the 
> semantics are compatible...punning is strictly weaker than the Hilog 
> semantics. (Analogously, possibly finite owl:Thing is weaker than 
> necessarily infinite owl:Thing.)
> 
> Jeremy claims that if we keep Hilog semantics for OWL Full and punning 
> otherwise, we "weaken" the relationship between OWL Full and OWL DL. But 
> this is arguably not true. For any legal OWL DL ontology, to the extent 
> that its entailments coincide with the entailments under OWL Full 
> semantics, the coincide in the presence of punning. That's because 
> without punning, no ontology where a term is used in more than one role 
> is an OWL DL ontology.
> 
> So everything that was true in OWL 1.0 is still true in OWL 1.1 with 
> punning.
> 
> However, we can say some new stuff, that is, that there some ontologies 
> (a subset of those with non-separated vocabularies) which now are legal 
> OWL 1.1 ontologies (under punning semantics). Now we can only make 
> somewhat weak claims about the coinciding of the entailments under the 
> OWL 1.1 DL and Full semantics *of these new ontologies* (to wit, that 
> the entialments under punning are a subset or equal to the entailments 
> under Hilog). But note that the don't *always* diverge, or anything like 
> that. So the situation is exactly the same as before, except the set of 
> legal OWL 1.1 dl ontologies is a larger subset of all the RDF graphs.
> 
> We may want to discuss whether we thing hilog semantics is desirable 
> over punning, but *this issue* claims that there is an incompatibility. 
> I claim that there is no incompatibility. Thus we should close it. As 
> there is no proposal in this issue, there's nothing to accept or reject.
> 
> Cheers,
> Bijan.
> 
Received on Wednesday, 23 January 2008 14:22:46 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Wednesday, 23 January 2008 14:22:48 GMT