Re: Draft for a "minimal dataset semantics"

On 06/09/12 14:52, Ivan Herman wrote:
> Interesting, I did not realize that.
>
> So, if I understand things right, SPARQL has already made some
> decision so that (translated into the issues in this proposal):
>
> - ISSUE 1: is actually yes per SPARQL; even more, it allows for a
> different entailment regime to be applied for each graph (something
> Antoine has considered and we discussed at some point and we thought
> it would be too complicated to get a consensus
>
> - ISSUE 3: is not 100% the same in SPARQL, but SPARQL seems to have a
> possibility to describe what a specific service does with a graph,
> ie, which entailment it uses. The issue here is to use something
> similar for datasets (not for processing endpoint). But that is
> certainly something similar.
>
> What is interesting here is the SPARQL approach to ISSUE #1. The
> question is whether this WG should follow the SPARQL approach. But
> the fact that SPARQL already does it looks like a very strong
> argument to do it in general, too... (which may invalidate my
> cautious approach I had so far)

(from memory ...)

Because entailment is a graph relationship, and query is based on BGP 
matching of entailment graphs, it seemed to be natural that each graph 
has an entailment regime, and so a dataset is a collection of graphs 
(with entailment regime).

Only if an entailment is based triples from different places or triggers 
a conclusion in a different graph does that break down (I think).

The plus side is that if you have a graph-based reasoner, you don't have 
to do anything to put graphs in datasets - just catch all the BGP matching.

 From a personal POV: one graph having one set of rules and another a 
different set of rules is the important case.  It's a simple step from 
there to one entailment regime per graph.

In practice, the same entailment regime everywhere in the dataset is 
common - the service description does have one property that provides a 
default.

I don't recall much discussion of it being a dataset feature and 
honestly, "it just happened" as far as I remember as being the natural way.

 Andy

Received on Thursday, 6 September 2012 14:07:59 UTC