Re: [BlankNodeRefs] Questions on blank node refs

On 18 Mar 2009, at 22:52, Axel Polleres wrote:

> Hi Bijan, all,
[snip]
> In more detail: In RIF itself, there is no such thing as blank nodes.
> RDF triples are mapped to slots in RIF. When mapping RDF(S)  
> entailment to RIF, blanknodes of the entailing graph are skolemized  
> whereas the entailed graph is mapped to a RIF condition formula such  
> that blank nodes are mapped to variables.
> As for more details, cf. [1]
>
> * As for OWL:
> Can you elaborate a bit more why are blank nodes a problem in OWL?
> Per se, some RDF triples with blank nodes can be mapped to a OWL1  
> statements already, i.e.
>
> s p _:b.
>
> would become:
>
> SubclassOf(oneOf(s)  restriction( p minCardinality(1) )

The problem is that no user expects or wants that behavior and no  
implementation supports that behavior. The semantics is simply at odds  
with the market.

I mean, prima facie, would *you* expect a BGP s p ?o to return a BNode  
when querying both the KBs above? (Not to mention that the later KB  
would return a lot of other stuff as well.)

> but that is more problematic with bnodes in subject positions, right?

More pressingly is handling cyclic sets of BNodes which would cause  
the undecidability of entailment (and be a real PITA).

This means that very natural RDF graphs cannot be handled by most OWL  
reasoners even when interpreted as ABoxes.

Recall such problems can extend down into RDFS, since, I believe,  
conjunctive query over RDFS with blank nodes as existentials is NP- 
complete in data complexity.

(For comparison, data complexity of cc over SHIQ is also NP-complete.)

> Is that the kind of problems you are referring to or the problem to  
> define the semantics of a SPARQL query on top of OWL entailment, on  
> datasets with blank nodes such as co-reference in solution sets,  
> etc., cf. your part of our ESWC Tutorial [2]?

Those too. It's just not clear what to bind variables to in the semi- 
distinguished case (i.e., when you bind to bnodes).  If I have a  
SomeValuesFrom should it generate a BNode? How many? What if I have a  
minCardinality 5..should it generate at least 5?

It all would be very simple if BNodes were local, slightly shifting,  
local names.

> If you could summarize/review the issues briefly, it might be nice  
> for the group.

This email seems to never have gotten sent :)

Cheers,
Bijan.

Received on Friday, 10 April 2009 11:15:31 UTC