RE: RDF/XML shorthand for RDF reification

Hi Ian!

Ian Horrocks wrote:

>I'm not an expert in OWL Full matters, but I guess that Peter is
>concerned that the root node of the reified axiom will now become a
>named node, that there could/would be different names in different
>ontologies, and that the entailment would thus be broken.
>
>Ian

Hm, I feel a bit uncomfortable, because I cannot easily tell that I
understand the whole issue here (got lost in the discussion). So I have to
ask: Is it that the RDF mapping would produce fresh "random" URI names as
the root node of an axiom annotation?

Sure, if there are two RDF graphs representing the same axiom annotation,
but having different URIs as their root nodes, then the one annotation graph
does not entail the other one. But what I was about in my original private
mail to Alan was that this situation isn't avoidable in OWL Full, anyway. It
definitely will be possible for OWL 2 Full authors to write an ontology such
as

  michaels:axiom rdf:type owl:Axiom ;
    rdf:subject A ; rdf:predicate rdfs:subClassOf ; rdf:object B ;
    rdfs:label "Foo"^^xsd:string .

  ians:axiom rdf:type owl:Axiom ;
    rdf:subject A ; rdf:predicate rdfs:subClassOf ; rdf:object B ;
    rdfs:label "Foo"^^xsd:string .

where the two axiom annotations only differ in their root node:
'michaels:axiom' vs. 'ians:axiom'. I cannot specify a semantic condition in
OWL Full which only matches on bNodes, but forbids (or ignores) the use of
URIs.
 
So my point was that nothing new will happen from the point of view of OWL
Full semantics, if the result of the RDF mapping of annotated axioms was an
RDF reification graph with a (random?) URI instead of a bNode. My current
proposal of OWL 2 Full already covers this case out of the box (because I
cannot prevent it), so there would be no additional impact on OWL 2 Full's
semantics. 

The only difference would be that the OWL Full semantics of the result of
such a modified RDF mapping would be slightly different compared to the
semantics of the result of the current RDF mapping. But this would be no
surprise: Whenever the RDF mapping for some OWL feature changes, then the
OWL Full semantics of the output graph may change, too. In many cases, this
semantic difference is negligible, though. IMHO, it would be negligible in
our case, since the relevant point of my current proposal of axiom
annotation semantics is simply that the respective axiom triple (or graph)
is entailed from the reification graph (because it isn't yet stated by the
reification itself). This is, however, true in either case.

Cheers,
Michael

>On 1 Jul 2008, at 13:19, Michael Schneider wrote:
>
>> Hi Peter!
>>
>> Peter F. Patel-Schneider wrote:
>>
>>> I'm not sure that I believe this.
>>>
>>> I think that one consequence of using the shorthand would be that an
>>> annotated axiom might not entail itself in OWL Full.
>>>
>>> For example, how would one arrange it so that
>>>
>>> SubClass(Label("Foo") A B)
>>>
>>> entails
>>>
>>> SubClass(Label("Foo") A B)
>>>
>>> in the OWL Full arena?
>>
>> I am not certain that I correctly understand the question. In
>> general (by
>> Simple Semantics) every RDF graph entails itself. So as long as a
>> DL axiom
>> has a mapping to RDF, the respective RDF graph entails itself in
>> OWL Full.
>>
>> In the case of your example axiom above, the RDF mapping would be
>> (but using
>> a URI instead of a bNode for the axiom):
>>
>>   ex:axiom rdf:type owl:Axiom
>>   ex:axiom rdf:subject A
>>   ex:axiom rdf:property rdfs:subClassOf
>>   ex:axiom rdf:object B
>>   ex:axiom rdfs:label "Foo"^^xsd:string
>>
>> The RDF/XML serialization would be pretty compact in this case, but
>> this is
>> has no impact on OWL Full, which only deals with the (abstract) RDF
>> graph.
>>
>>> peter
>>
>> Cheers,
>> Michael
>>
>>> From: Alan Ruttenberg <alanruttenberg@gmail.com>
>>> Subject: Re: RDF/XML shorthand for RDF reification
>>> Date: Tue, 1 Jul 2008 07:11:54 -0400
>>>
>>>> On Jun 26, 2008, at 7:16 AM, Peter F. Patel-Schneider wrote:
>>>>
>>>>> 2/ Changing to named nodes would change the OWL Full semantics.  A
>>>>>    careful check would have to be made to see whether any
>>> interesting or
>>>>>    useful inferences could change.
>>>>
>>>>
>>>> According to Michael this is one not a problem. I'm still thinking
>>> about issue 1/.
>>>>
>>>> He says:
>>>>
>>>>> Short answer: There are no additional consequences for OWL Full.
>>>>>
>>>>> Longer answer: Just have a look at the OWL Full Wiki:
>>>>>
>>>>>   <http://www.w3.org/2007/OWL/wiki/
>>>>> FullSemanticsAxiomAnnotations#Semantics>
>>>>>
>>>>> The "Main semantic condition" there is applicable to both cases,
>>> with a bNode
>>>>> or with an URI at the LHS. That's not because I wanted to make this
>>> semantic
>>>>> condition more general as in OWL DL - I promise that I always try
>>> hard to be
>>>>> as close to the DL semantics as possible. The point is that in an
>>> RDF
>>>>> compatible semantics, it is not possible to restrict axioms to such
>>> RDF graphs
>>>>> which only have a bNode as their LHS. This would be a syntactic
>>> restriction,
>>>>> which is not possible in OWL Full - which is a major distinction
>>> between OWL
>>>>> Full and OWL DL.
>>>>>
>>>>> So this whole current discussion about named axioms is really only
>>> an OWL DL
>>>>> topic. It's the question whether to extend the reverse RDF mapping
>>> to cases,
>>>>> where the LHS of an axiom annotation may be an URI, or not. OWL
>>>>> Full
>>> is
>>>>> completely indifferent about this question. In particular, in my
>>> current
>>>>> proposal, you will always receive the axiom triple as a result,
>>> whether the
>>>>> LHS node is a bNode or a URI.
>>