Re: Lee's feature proposal

On 1 May 2009, at 09:31, Bijan Parsia wrote:

> On 1 May 2009, at 09:10, Steve Harris wrote:
>
>> On 1 May 2009, at 08:59, Ivan Herman wrote:
>>> It is not clear to me (lack of my technical knowledge!) whether  
>>> Bijan's
>>> SPARQL/OWL proposal covers both semantics of OWL or not. OWL DL  
>>> is, in
>>> many respect, a loose sub thing to OWL Full, so it might, but we  
>>> have to
>>> be very explicit (at charter time, too!). So it would be good to  
>>> put my
>>> mind at ease:-) How would we handle the others like RDFS?
>>
>> I'm not really hot on the logical underpinnings, but I don't  
>> remember running into any substantial problems when applying  
>> SPARQL over RDFS. There are some questions around how you handle  
>> certain queries that theoretically have infinite solutions, but  
>> there are pragmatic workarounds for those.
> [snip]
>
> So there are sort of two cases (off the top of my head): Bnodes  
> being entailed and the infinite number of axiomatic triples. BNodes  
> entailment is already worked out in SPARQL as it stands
[snip]

The relevant section.

http://www.w3.org/TR/rdf-sparql-query/#sparqlBGPExtend

Which imposes a finiteness constraint. There is also:

http://www.w3.org/TR/rdf-sparql-query/#BGPsparqlBNodes

Which shows how finiteness is ensured in basic SPARQL in the presence  
of blank nodes.

Without getting into technical details too much in this thread, an  
analogous criterion for blank nodes can be adopted for RDFS and even  
OWL (which, I believe, is necessary in order to maximize the  
uniformity of the user experience when shifting semantics). SPARQL/DL  
copes with infinite possible class expressions being entailed  as the  
value of class variables by syntacticly restricting the set of  
possible solutions to atomic names (which can be liberalized to the  
subconcept closure of the original graph). Essentially, this puts a  
constraint on the set of terms in the solution mapping:
	http://www.w3.org/TR/rdf-sparql-query/#sparqlSolutions

Ensuring finiteness is the challenge for these regimes and, even  
then, only in the three cases of bnodes in the answer, infinite  
axiomatic triples, and syntactically higher order variables. All  
other cases are perfectly standard and easy peasy.

Cheers,
Bijan.

Received on Friday, 1 May 2009 09:09:57 UTC