Re: Re 2: Agenda for 19 Sep 2012


still trying to understand, bear with me.

RDF entailment is described in terms of I-s, so I try to translate what you describe in terms of those; this would allow me to compare with the original proposal. What you seem to say that a model for a Dataset (G, <n1,G1>,...,<nk,Gk>) is a (I0,I1,...,Ik) so that I0 satisfies G, and Ii satisfies Gi. Ie, the models for each element of the dataset are mutually 'independent' from one another.

If this is true, then I vaguely remember that this approach was discussed at some point in the past, and Pat was pretty much against it. I do not remember his exact arguments, but essentially what it leads to is that if the same 'u' appears in G1 and G2, this would amount to assigning a very different meaning to the same 'u', depending on which graph it is contained in. And that was, for Pat, a big no-no:-) (Pat, is my recollection correct?)

Also, there is a vague inconsistency then in terms of the current RDF Semantics. *I know* this is not exactly the same but, nevertheless, here is what the current text says in section 2

Following conventional terminology, I satisfies E if I(E)=true, and a set S of RDF graphs (simply) entails a graph E if every interpretation which satisfies every member of S also satisfies E.

Ie, the current semantics document talks about a set of graphs, and the structure clearly refers to an I satisfying *every* member of the set. Again, *I know* this is not the same as the datasets but, nevertheless, for a casual user, it would make it a bit inconsistent if we had a different approach for a collection of graphs in a dataset...


On Sep 18, 2012, at 13:58 , Peter F. Patel-Schneider wrote:

> There are other differences.
> As far as the entailments go, the minimal semantics proposal has the property that if the default graph is inconsistent, then the named graphs do not matter for entailment.
> peter
> On 09/18/2012 06:58 AM, Ivan Herman wrote:
>> Peter,
>> On Sep 17, 2012, at 20:02 , Peter F. Patel-Schneider wrote:
>>> My participation this week is at risk.
>>> I cast my vote AGAINST:
>>> PROPOSED: Editors should integrate the "Minimal Semantics" design and included test cases into Working Drafts and the Test Suite, with the understanding that this is the approach we will use unless serious problems emerge. Any specific issues people have with this design should be RAISE'd soon.
>>> As indicated multiple times in the past, I am in favour of a weaker semantics for RDF datasets either no semantics at all or maybe something like.
>>> 1/ No change to the semantics for RDF graphs.
>>> 2/ Entailment between RDF datasets  D1=<G1,N1> and D2=<G2,N2> is defined as
>>>    a) G1 entails G2
>>>    b) for each <n,g> in N2 there is some <n,g'> in N2 such that g' entails g,
>>>                                  or if there is no such <n,g'> then the empty RDF graph entails g
>>> I could also live with a semantics that eliminated this last disjunct.
>>> My reasons for preferring a weaker semantics are that I don't want to have certain uses for RDF datasets ruled out by a particular semantics at this time, when no one know just how RDF datasets will be used.   One kind of use that I am thinking of is just using RDF datasets to record multiple graphs, which no commitment to any interchange of information between the named graphs or between the default graph and the named graphs or even any commitment to having the names being formally related to the graphs.
>> Again for my understanding. Unless I miss something fundamental, everything you say is possible with the proposed semantics up to the statement "any commitment to having the names being formally related to the graphs". Indeed, the current IGEXT mechanism does impose a (albeit fairly weak) relationship between the name and the graph.
>> Is there any other major difference between what you propose and what is on the wiki page?
>> Ivan

Ivan Herman, W3C Semantic Web Activity Lead
mobile: +31-641044153

Received on Tuesday, 18 September 2012 12:33:34 UTC