Re: [OWLWG-COMMENT] Punning and the "properties for classes" use case

From: "Michael Schneider" <schneid@fzi.de>
Subject: RE: [OWLWG-COMMENT] Punning and the "properties for classes" use case 
Date: Mon, 5 Nov 2007 21:42:37 +0100

> Hi, Peter!
> 
> Peter F. Patel-Schneider wrote:

[...]

> >[...]  From the RDF mapping document:
> >
> >	OWL 1.1 ontologies serialized in RDF/XML are interpreted by
> >	translating them into the functional-style syntax and applying
> >	the OWL 1.1 semantics.
> >
> >So the OWL 1.1 meaning of the triples is defined in the OWL 1.1
> >Semantics document, just as the semantics of an OWL DL ontology is
> >defined by the direct model-theoretic semantics for OWL DL no matter
> >what form it is written in.
> 
> Ok, this is pretty clear to understand, but it surprises and concerns me. As
> far as I have understood the facts, in OWL-DL there hasn't been a conflict
> between the direct model theoretic semantics, and the DL version of the RDF
> compatible semantics. But now it looks different to me. Let's start from 
> [an OWL 1.1] ontology encoded in RDF [...].  Now the mapping document says
> that in order to learn the semantics of this ontology, I have to map the RDF
> graph into functional style syntax, and then look up the OWL-1.1 semantics
> from the semantics document. 
> 
> But how is an RDF compatible semantics possible in this case? An RDF
> compatible semantics won't allow interpretations for which a class denoted
> by "<u>" is semantically unrelated to an individual denoted also by "<u>",
> right? 

Right.

> But this is allowed by OWL-1.1 "direct" semantics, if I correctly
> understand it. 

Correct.

> Or to make it more explicit: Is it correct or false that in
> OWL-1.1 with punning, I am allowed to interprete some given URI ref <u> with
> both
> 
>   (a) an /arbitrary/ individual from the universe
>   (b) an /arbitrary/ class from the universe
> 
> ?

Well, not exactly, but close.   When u appears as an individual, you get
an element of the domain of discourse (universe).  When u appears as a
class, you get a set of elements of the domain of discourse.  You can't
pick which one you want - it is determined by the context.

> In OWL-DL, this problem did not exist, because the RDF graphs of OWL-DL
> ontologies never used URI refs both as individuals and classes. 

Right.

> And for
> OWL-Full, where such a duplicate usage was actually allowed, there was only
> RDF compatible semantics, no direct semantics, so no conflict could arise.

Right.

> So if punning allows interpreting the same URI refs by semantically
> unrelated individuals and classes (i.e. arbitrarily chosen combinations of
> individuals and classes): Dosn't this mean that an RDF compatible semantics
> will not be possible for OWL-1.1?

Well, it is probably possible to come up with an RDF-compatible
semantics that is completely equivalent to the OWL 1.1 semantics by
using name splitting, which is the other option for translation that you
mentioned.   Without using name splitting it is probably not possible.  

Punning and enhanced annotations are places that expose problems with
the RDF view of the world.

> Cheers,
> Michael

peter

Received on Tuesday, 6 November 2007 10:10:30 UTC