Re: Minimal dataset semantics

On 23 Aug 2012, at 16:17, Sandro Hawke wrote:
>> The quoting semantics makes it a contradiction if dataset A and dataset B contain the same graph IRI with different associated graphs. We cannot do semantic extensions that produce useful additional entailments from a contradiction.
> Not true.   There are two versions of the quoting semantics -- partial-graph semantics and complete-graph semantics.

The "partial-graph quoting semantics" is really just a broken form of "truth semantics". It is broken because it arbitrarily emulates one effect of entailment (the fact that dropping any triple from a true graph yields a true graph), while not emulating all the other effects of entailment (e.g., the fact that substituting any subject or object with a blank node in a true graph yields another true graph, or that adding redundant inferred triples doesn't affect truth, or that substituting one literal with an equal-valued literal doesn't affect truth, or that replacing one blank node with another fresh blank node doesn't affect truth).

>   The discussion here in recent days has focused on the complete-graph quoting semantics, but in previous telecons we had near consensus (everyone but Eric) on using the partial-graph semantics.

No, we had near consensus that "complete-graph quoting semantics" is not desirable. I think what you interpret as support for "partial-graph quoting semantics" was mostly support for "some form of truth semantics".

> Partial-graph quoting semantics might also be called "quad" semantics.   You can decompose the dataset into quads that each stands on its own; merging datasets is just the set-union of the quads.  Each quad tells you that a particular triple is in a particular named graph.    There is no way to form a contradiction among such quads.

I don't understand what that means. I don't think that "contradiction among quads" is an interesting or relevant notion. The question is what datasets contradict or entail each other or are equivalent.

Should these two datasets contradict each other?

  :g { :a :a :a }

  :g { :b :b :b }

My answer would be no, because the RDF graphs { :a :a :a } and { :b :b :b } don't contradict each other. (Neither do they entail each other or are equivalent. They are simply different.)

If your answer is "yes" (which I assume it is), then, given that each dataset consists of one quad, we do have a contradiction between quads after all?

>> No other proposed semantics does have that problem. All of the other proposed semantics can be easily extended with an additional clause that requires equal graph names to be associated with equal graphs.
>> Therefore, the quoting semantics is *not minimal*. Quite the opposite. It is not a "weak" semantics at all, because it makes it very easy to derive contradictions, and contradictions are *very strong* semantic effects.
>> I also agree with Antoine that formalizing the notion of "no semantics" is pointless.
>> My conclusion is that our viable options are:
>> a) to say nothing regarding the semantics of datasets, or
> Can we do that and still do your point (1) and (2) above?

No. Saying nothing about the semantics means that we don't formally say what the graph IRIs denote, so that fails #1.

> If so, I like that idea.    I'm just not sure if that's possible (or how to do it).

Well, this would be what Antoine called "formalizing the notion of 'no semantics'". It's probably possible, but I don't see the point. We would be writing some no-op formal machinery.


>> b) to define a minimal version of a "truth-based"/"entailment-based" semantics (where [[ :i1 { G } ]] entails [[ :i1 { G' } ]] if graph G entails graph G').
> (I'm going to stay out of this option for now; need to think about it more.)
>    -- Sandro

Received on Thursday, 23 August 2012 15:47:03 UTC