Re: A possible way of going forward with OWL-R unification (ISSUE-131)

Michael Schneider wrote:
> Ivan Herman wrote:
> 
>>> I need to speculate, please comment: If I have an RDF graph, which is
>> intended
>>> to be an OWL-R ontology in RDF graph form, then:
>>>
>>>   (A) If I want to do RDF-based reasoning, I just use the triple rules
>> as I
>>> would do it in current OWL-R/Full.
>>>
>>>   (B) If I want to do DL-style reasoning, then
>>>
>>>       (1) I try to map the graph to functional style syntax by means
>> of the
>>> reverse RDF mapping.
>>>
>>>       (2) If the RDF graph could successfully be mapped,
>>>           then the resulting functional syntax is checked against the
>> OWL-R
>>> constraints
>>>           as provided by the Profiles document (still to be defined,
>> of
>>> course).
>>>
>>>       (3) If the functional syntax turns out to be a valid OWL-R
>> ontology,
>>>           then the model-theoretic semantics as specified in the
>> Semantics
>>> document
>>>           are applied. These are then the OWL-R/DL semantics of this
>> ontology.
>>> Is it this?
>>>
>> I think the point of OWL-R is that in, say, 95% of the cases (A) and (B)
>> (well, by re-serializing the result of (B) into RDF, that is) the
>> resulting set of RDF triples will be identical by some mathematical
>> magic:-)
> 
> I don't understand this statement. What do you mean by "the resulting set of 
> RDF triples"? My question started from an /existing/ RDF graph, I didn't talk 
> about /producing/ RDF.
>

You are right. I actually thought of this yesterday evening (you know, 
the shower is the best place to think properly:-). From an RDF point of 
view I think the best possible answer is: if I make a SPARQL query on an 
endpoint that reasons with (A) and contains the initial graph, and I do 
the same in (B), then in the cases where I do not use those conflicting 
terms (less sure about the 95%, actually, it may be less, but let us not 
try to get numbers) I would get the same results.

> Let me say this: As long as the triple rules don't change, then things seem to 
> move in the direction which I prefer. It should then even be possible without 
> a problem to add the RDFS axiomatic triples and perhaps additional ones for 
> OWL R. Why? Because we can then state the following: "If used in "RDF mode", 
> the additional axiomatic triples belong to the rule set. If used in "DL mode", 
> then they are not taken into account." I suppose that this won't have big 
> ramifications for the semantic comparison of the two modes, because such a 
> comparison will probably be done in a "Theorem 2" way: Only such entailments 
> are actually compared, where the LHS and the RHS are valid OWL-R-DL ontologies 
> in RDF graph form. This should cancel out most stuff resulting from the 
> additional axiomatic triples. I am not perfectly certain about this claim, but 
> this was the way how OWL-DL and OWL-Full (having all the RDFS axiomatic 
> triples, of course) have been compared in the past, so there is at least some 
> hope that it might work. :)
> 

I think you are right. The issue is still, I guess: what _are_ the 
axiomatic triples? The RDFS ones are clear, but are there axiomatic 
triples to add on the OWL terms?

Ivan

> Cheers,
> Michael
> 

-- 

Ivan Herman, W3C Semantic Web Activity Lead
Home: http://www.w3.org/People/Ivan/
PGP Key: http://www.ivan-herman.net/pgpkey.html
FOAF: http://www.ivan-herman.net/foaf.rdf