Re: named graphs and rdfg:subGraphOf

On Oct 7, 2009, at 3:46 PM, Jakub Kotowski wrote:

> Hello,
>
> I am trying to understand the definition of the rdfg:subGraphOf  
> property
> from [1,2]. It says that:
>
> <f,g> is in IEXT(I(rdfg:subGraphOf))  iff
> rdfgraph(f) is a subset of rdfgraph(g)
>
>
> What confuses me is probably the "syntactic part" of the definition:
> rdfgraph(f) is the ("syntactical") set of triples of the named graph  
> f.

Yes, graphs are indeed syntactic entities.

> I am wondering whether it means that if rdfgraph(g) does not contain  
> all
> the triples from rdfgraph(f) I can infer them - so that after the
> inference proces I'll get a new, enriched rdfgraph(g) for which the
> condition already holds (rdfgraph(f) is a subset of rdfgraph(g))...?

It was not intended to have this kind of 'procedural' interpretation,  
but if you had a system which tried to generate graphs from their  
descriptions, then this might make sense. It would not be inference,  
however, but something more like a graph construction process.

> Instead of this description I would almost rather like to ask  
> whether it
> means that the triple f rdfg:subGraphOf g entails:
> g {
> 	rdfgraph(f)
> 	plus whatever was in g before

Before what, exactly? You seem to have a process or procedure in mind,  
but I do not follow what it is. The rdfg vocabulary is intended for  
describing relationships between graphs which already exist.

> }
> But that doesn't seem to be correct because entailment is not defined
> over a set of named graphs.
>
> The alternative would be that a knowledge base containing:
> 	the triple		f rdfg:subGraphOf g
> 	the named graph		f
> 	the named graph		g ...(the original, not enriched one)
>
> is inconsistent because rdfgraph(f) is not a subset of rdfgraph(g) but
> the triple f rdfg:subGraphOf g is asserted.

Well, indeed, this combination as described is inconsistent. Now the  
question is, what to do about this inconsistency when it has been  
detected, if anything. One possibility is simply to report it as an  
inconsistency. Another is to try to 'repair' it, by changing the world  
so as to make it consistent. This could be done in various ways: the  
subgraph assertion could be rejected, for example, or the graph named  
g could be expanded to include the triples of the graph f. The formal  
semantics does not specify which of these, if any, is correct or  
appropriate: it simply specifies the inconsistency. What can be done  
depends in practice largely on who has ownership of the various graphs  
involved.

> Well, this alternative maybe
> does not even make sense because a set of (accepted) named graphs is
> defined to have the usual RDF semantics of the merge of the respective
> graphs.
>
> On the one hand the first interpretation would seem more plausible
> because it would make the rdfg:subGraphOf property usable as a way of
> nesting named graphs, on the other hand, the provenance motivation for
> named graphs seems to be in favour of the alternative interpretation
> which sees the property as rather descriptive.

Certainly, the property is intended to be purely descriptive, as far  
as the normative specification is concerned.

> After all, if a graph
> changes because of inference (inferred triples are added) then the new
> graph should probably be associated with provenance information which
> documents that it was inferred and possibly how.

Certainly it will be a different graph, one with a different name.  
Named graphs are not intended to be names of *dynamic* graphs. The  
usual rules of 'cool URI' usage apply here just as they would to any  
other document or information resource.

>
> At least the following is true, right?
>
> :f rdfg:subGraphOf :g does not hold if :f and :g are specified as  
> follows:
>
> :f {
> 	:a rdfs:subclassOf :c .
> }
>
> :g {
> 	:a rdfs:subclassOf :b .
> 	:b rdfs:subclassOf :c .
> }
>

That is correct. In this case f is a subgraph of the RDFS closure of  
g, but not of g itself.

Hope this helps.

Pat Hayes

> Am I just making things too complicated and overlooking something?
> By the way, esentially this question has already been asked but
> unfortunately left unanswered:
> http://lists.w3.org/Archives/Public/semantic-web/2006May/0108.html
>
> Best regards,
> Jakub Kotowski
>
>
> [1] Named graphs, provenance and trust Export
> Jeremy J. Carroll, Christian Bizer, Pat Hayes, Patrick Stickler
> In WWW '05: Proceedings of the 14th international conference on World
> Wide Web (2005), pp. 613-622.
>
> [2] Named graphs
> J. Carroll, C. Bizer, P. Hayes, P. Stickler
> Web Semantics: Science, Services and Agents on the World Wide Web In
> World Wide Web Conference 2005------Semantic Web Track, Vol. 3, No. 4.
> (December 2005), pp. 247-267.
>
>
>

------------------------------------------------------------
IHMC                                     (850)434 8903 or (650)494 3973
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes

Received on Thursday, 8 October 2009 14:59:19 UTC