Re: attempts to reconciliate quote-semantics and "context"-semantics (Was: Re: RDF dataset semantics again)

Ivan,


Le 23/08/2012 15:11, Ivan Herman a écrit :
> Hi Antoine,
>
> (Sorry, this mail is a little bit long. tl;dr: I actually like the
> direction of where this is going:-)

I'm glad and relieved to read that :)


> I'll try to summarize the different pieces of the puzzle in your
> proposal in one place, and in somewhat laymen's term. I do that to
> check my own understanding but also to make it easier for the group
> to follow; this may be helpful so that we could make an informed
> decision. I hope you don't mind me doing that. (I will also have
> technical questions on some details... see below,)
>
> 1st piece.
>
> We already have a series of 'Entailment Regimes' scattered around the
> various documents: Simple, RDFS, OWL-RL, RIF... These are already
> reused in the SPARQL Entailment regime document (and all have a URI
> to identify them, b.t.w.). We would add one more, namely the
> no-semantics one [1]. As far as I can see the major difference
> between this 'No-Semantics' and the 'Simple' is that 'No-Semantics'
> ignores all the intricacies of blank nodes and is just a mapping for
> RDF graphs in terms of a model.

Yes.


>
> 2nd piece
>
> What Richard has defined a while ago in [2] for datasets is,

I think you used the wrong URL for [2]. Richard did this:

http://www.w3.org/2011/rdf-wg/wiki/Yet_Another_Dataset_Proposal


> actually, a kind of a parametrized interpretation of datasets. If I
> take a (G,<n1,G1>,<n2,G2>,...,<nk,Gk>), the parameters are the
> choices of the entailment regimes, that may be different for each Gi,
> as well as for the default graph.

Richard did not go that far. The entailment regime is a global parameter 
that define one regime for all graphs in the dataset.

In its current state, it does not address the problem of making 
entailments inside graphs. So:

<g1> { <s> <p> <o> }

does not DS-RDF-entail:

<g1> { <p> a rdf:Property }

This is due to the fact that the definition of "state relationship" 
imposes that a resource can be associated to only one graph. But maybe 
there is something to take out of his proposal, with some modifications.


> This means that, in theory, the
> application/data provider can choose the combination of entailment
> regimes per graphs: G1 would be considered for OWL-RL, whereas Gk is
> subject to RIF, for example.
>
> As a base line, ie, if no other information is known, then the
> No-Semantics is the entailment regime used for all the G1,G2,...,Gk
> graphs. (Does this also apply to the default graph?)

If it is not applied to the default graph, then what semantics apply to it?

>
> 3rd piece
>
> The 'extension' point which, until now, relied on typing in our
> earlier brainstormings, is performed through a separate set of data
> description statements (as bit VoiD-like, if I may say so). To copy
> from your page:
>
> @prefix  ds:<datasetInferenceVocabulary>  . [ a
> ds:DefaultGraphInferenceRegime ; ds:onGraphIRI
> "http://www.example.com/"^^xsd:anyURI ;
> ds:withRegime<http://www.w3.org/ns/entailment/RDFS>  . ] [ a
> ds:GraphInferenceRegime ; ds:onGraphIRI
> "http://www.example.com/"^^xsd:anyURI ;
> ds:withRegime<http://www.w3.org/ns/entailment/Simple>  . ]
>
> would specify the parameters for the interpretation per [2].
>
> As you say, these statements may be part of the Default graph of a
> dataset, but may also be external, and an inference engine would then
> use those.
>
> First of all, is this a fair summary of what you propose?

A very accurate summary, good job.


> _Personally_, I like where this is going, although I am not sure it
> would cover all the use cases we had in the past. For example, one of
> the use case that did come up is what we used to call 'merge
> semantics', meaning that all inferences and constraints checks are
> made on the merge of all the graphs (including the default graph).
> This is not covered by this structure but, maybe, this is not
> important.

Yes, it does not cover the merge/union semantics. But they are simple 
semantic extensions, for which we can provide (or a future WG) 
vocabulary to specify.


> (I am also a little bit worried about the complexity for users. We
> shall see the feedbacks.)

The formal semantics will certainly look quite complex, but it can be 
explained in a rather simple way. In a nutshell, it says that:

"All RDF graphs in an RDF dataset don't mean the same thing. To be 
explicit about what they mean, we provide a vocabulary that specify the 
semantics of each graph. We call the semantics assigned to a 
<name,graph> pair its entailment regime, because it determines what 
entailments are valid for that pair. For example ..."

Then we have to warn the people who would like to read the formal 
details should be well armed.


>
> Also, some technical issues/questions:
>
> 1. (I should have looked at that yesterday, sorry.) I found the
> terminology used in [2] a bit confusing and I have the feeling that
> there is an overload of the letter 'E'... At present, it says
>
> [[[ Moreover, dataset interpretations are defined with respect to an
> entailment regime E, as defined in SPARQL 1.1 Entailment Regimes. Let
> KE be the set of all E-interpretations. The interpretation of an RDF
> Dataset (G, (<n1>,Gn1), ..., (<nk>,Gnk)) over vocabulary V is a pair
> (I,Con) where I is an E-interpretation of G (the default graph) and
> Con is a mapping from V to KE.
>
> A dataset-interpretation (I,Con) of a vocabulary V wrt entailment
> regime E satisfies an RDF Dataset (G, (<n1>,Gn1), ..., (<nk>,Gnk))
> iff I E-satisfies G, and for all i in [1..k], Con(ni) exists and
> E-satisfies Gni. ]]]
>
> I *could* read this sentence as asking that *all* constituent graphs,
> including the default graph, must use the same interpretation. Isn't
> it:

Yes, that was the idea when I wrote it. The idea of having distinct 
regimes for distinct <name,graph> pairs only made sense to me when I 
drafted the new proposal yesterday.

This is why I marked the following in the new proposal:

"TODO: modify the semantics such that different entailment regimes can 
be used for different named graphs."


> [[[ Moreover, dataset interpretations are defined with respect to an
> entailment regime E, as defined in SPARQL 1.1 Entailment Regimes. Let
> KE be the set of all interpretations as defined in SPARQL 1.1 plus
> the No-Semantics. The interpretation of an RDF Dataset (G,
> (<n1>,Gn1), ..., (<nk>,Gnk)) over vocabulary V is a pair (I,Con)
> where I is an E-interpretation of G (the default graph), and Con is a
> mapping from V to KE.
>
> A dataset-interpretation (I,Con) of a vocabulary V wrt entailment
> regime E-satisfies an RDF Dataset (G, (<n1>,Gn1), ..., (<nk>,Gnk))
> iff I E-satisfies G, and for all i in [1..k], Con(ni) exists and
> Con(ni)-satisfies Gni. ]]]
>
> (Note that last 'Con(ni)-satisfies Gni'.)

Con(ni) assigns a term of the vocabulary to an E-interpretation. It does 
not assign a term to an entailment regime. This is something I have to 
think about, but I have an idea. It's not trivial, as far as I can see.


> Terminologically and in line with the outline you had, I actually
> o.k. to call that an 'E-interpretation of the dataset'; after all,
> each entailment regime is parametrized, wether it applies to the
> default graph or not.
>
> 2. If I choose to use, say, OWL as an entailment regime for Gi, how
> do I specify *which* OWL Ontology should be used for that purpose? I
> know this is also a question one may about current graphs, too, but
> the situation becomes a bit more complex if we have many different
> graphs. Would we require some sort of a follow-your-nose on the terms
> used in Gi and allow the system to find the relevant ontologies? Or
> would we suppose that, at least conceptually, Gi has all the
> ontologies as part of the graph already, and we decide that this is
> not our problem?

Entailment regimes are special semantic conditions that are hard coded 
in consensual standards. We don't define an entailment regime 
per-ontology. If you want the knowledge of the ontology to influence the 
inference on a named graph, then you can put the ontology inside the 
graph. But I understand that it would be more convinient to be able to 
refer to an external document (whether it is OWL or plain RDF). 
Something like an "import". Truely useful but I doubt we should take 
care of this in this WG. Let us leave some work to do for the future 
generations :)


> For RIF, SPARQL/RIF introduced the rif:usedWithProfile. Maybe
> something similar, but also for OWL, should be added to those
> parameters that you defined.

Maybe. Let us see first how people react to this proposal.


> 3. My other issue is a question. The No-semantics means that there
> are no semantic conditions on the graphs. However, looking at first
> table of semantic conditions in section 1.4 of the RDF Semantics
> document, although that section says 'denotation of ground graphs',
> the first four condition is not dependent at all on blank nodes, it
> just defines a bit what a graph structure is.

Seems possible too. I guess it does not matter that much, as the 
no-semantics has only one trivial entailment: all graphs entail 
themselves and only themselves.

> I believe that can be
> safely added to the No-semantics as a semantic condition. I *think* I
> understand why the other two are not added as conditions: we do not
> want to get into the blank node store at this point.
>
> Whether we expand the no-semantics with those or not is not *really*
> important, though; I just wanted to understand why those were left
> out.
>
> Thanks
>
> Ivan
>
>
> [1]
> http://www.w3.org/2011/rdf-wg/wiki/TF-Graphs/Dataset-semantics-2.0
> [2] http://www.w3.org/2011/rdf-wg/wiki/TF-Graphs/Dataset-semantics
>
>
> On Aug 22, 2012, at 16:57 , Antoine Zimmermann wrote:
>
>> So, I made a new wiki page in a tentative to reconciliate the
>> different semantics.
>>
>> http://www.w3.org/2011/rdf-wg/wiki/TF-Graphs/Dataset-semantics-2.0
>>
>> Basically, I define a super-weak semantics of RDF graph, which can
>> be used as an underlying entailment regime for the dataset
>> semantics of [1].
>>
>> Then, I put an example of a possible vocabulary to allow more
>> expressiveness. The vocabulary is mirroring some of the terms of
>> SPARQL 1.1 service descriptions [2].
>>
>> This truly makes the "base" semantics very very weak but allows one
>> to extend it to any variant on top of it.
>>
>>
>> [1] http://www.w3.org/2011/rdf-wg/wiki/TF-Graphs/Dataset-semantics
>> [2]
>> http://www.w3.org/TR/2009/WD-sparql11-service-description-20091022/
>>
>>
>>
>>
AZ
>>
>> Le 22/08/2012 16:33, Ivan Herman a écrit :
>>>
>>> On Aug 22, 2012, at 15:54 , Pat Hayes wrote:
>>>
>>>>
>>>> On Aug 22, 2012, at 2:04 AM, Ivan Herman wrote:
>>>>
>>>>>
>>>>> On Aug 21, 2012, at 21:48 , Pat Hayes wrote: [snip]
>>>>>>>
>>>>>>>
>>>>>>> Antoine, I have the impression that we are actually in
>>>>>>> agreement. The document we have put forward has two
>>>>>>> essential points:
>>>>>>>
>>>>>>> - we would have a default semantics in the form of the
>>>>>>> quoting semantics
>>>>>>
>>>>>>
>>>>>> Whoa. I do not know what y'all mean by a "default
>>>>>> semantics". Is this a default that can be overridden? If
>>>>>> so, I know of NO semantic theory  anywhere in logic or
>>>>>> linguistics that can provide this. If y'all want this, you
>>>>>> are on your own.
>>>>>>
>>>>>> If not, what exactly is it supposed to mean?
>>>>>>
>>>>>> Pat
>>>>>
>>>>>
>>>>> What I meant is: this is the semantics that is standardized
>>>>> to be valid in the absence of any other indication. I did not
>>>>> say anything else.
>>>>
>>>> And Antoine agrees. OK, then a better term would be "weak" or
>>>> "minimal" semantics. "default" sounds like nomonotonicity
>>>> (being overridable) to me.
>>>
>>> Agreed, sorry for my sloppiness. 'Minimal' sounds indeed good to
>>> me.
>>>
>>> Ivan
>>>
>>>
>>>>
>>>> Pat
>>>>
>>>>>
>>>>> Ivan
>>>>>
>>>>>
>>>>> ---- Ivan Herman, W3C Semantic Web Activity Lead Home:
>>>>> http://www.w3.org/People/Ivan/ mobile: +31-641044153 FOAF:
>>>>> http://www.ivan-herman.net/foaf.rdf
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>> ------------------------------------------------------------
>>>> 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
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>> ---- Ivan Herman, W3C Semantic Web Activity Lead Home:
>>> http://www.w3.org/People/Ivan/ mobile: +31-641044153 FOAF:
>>> http://www.ivan-herman.net/foaf.rdf
>>>
>>>
>>>
>>>
>>>
>>>
>>
>> -- Antoine Zimmermann ISCOD / LSTI - Institut Henri Fayol École
>> Nationale Supérieure des Mines de Saint-Étienne 158 cours Fauriel
>> 42023 Saint-Étienne Cedex 2 France Tél:+33(0)4 77 42 66 03
>> Fax:+33(0)4 77 42 66 66 http://zimmer.aprilfoolsreview.com/
>>
>
>
> ---- Ivan Herman, W3C Semantic Web Activity Lead Home:
> http://www.w3.org/People/Ivan/ mobile: +31-641044153 FOAF:
> http://www.ivan-herman.net/foaf.rdf
>
>
>
>
>
>
>

-- 
Antoine Zimmermann
ISCOD / LSTI - Institut Henri Fayol
École Nationale Supérieure des Mines de Saint-Étienne
158 cours Fauriel
42023 Saint-Étienne Cedex 2
France
Tél:+33(0)4 77 42 66 03
Fax:+33(0)4 77 42 66 66
http://zimmer.aprilfoolsreview.com/

Received on Thursday, 23 August 2012 14:20:59 UTC