- From: Richard Cyganiak <richard@cyganiak.de>
- Date: Wed, 5 Oct 2011 12:44:47 +0100
- To: Ian Davis <Ian.Davis@talis.com>
- Cc: Pierre-Antoine Champin <pierre-antoine.champin@liris.cnrs.fr>, "public-rdf-wg@w3.org" <public-rdf-wg@w3.org>
Ian,
On 4 Oct 2011, at 22:47, Ian Davis wrote:
>> Inference is defined over RDF graphs, not over collections of unasserted RDF graphs.
>
> While I agree with this statement, isn't it the case that most graph stores do inference over the collection of graphs in a dataset?
Most graph stores don't do inference at all.
> Given these two graphs in a single graph store and a schema that says ex:ancestor is transitive then I'm sure many inference enabled systems will answer true if asked whether :c is an ancestor of :a
>
> :G1 { :a ex:ancestor :b }
>
> :G2 { :b ex:ancestor :c }
>
> I'm pretty sure BigOWLIM does this, possibly others too.
Like some other stores, BigOWLIM probably just by default places the union of all named graphs into the default graph. If it then does per-graph inference, then you get the effect you describe above when querying the default graph.
Anyways, the relevant spec for inference in triple stores will be SPARQL 1.1 Entailment Regimes, and it only defines entailment over individual graphs.
It's worth pointing out that BigOWLIM's default setup is unsuitable for any use case that uses named graphs for versioning, trust, or authentication.
Best,
Richard
Received on Wednesday, 5 October 2011 11:45:27 UTC