Re: Draft for a "minimal dataset semantics"

Le 13/09/2012 20:13, Peter F. Patel-Schneider a écrit :
>
> On 09/13/2012 12:50 PM, Antoine Zimmermann wrote:
>> Le 13/09/2012 03:08, Peter F. Patel-Schneider a écrit :
>>
> [...]
>
>>
>> Here is another test case:
>>
>> { :n owl:sameAs :m }
>> :n { :s :p :o }
>> :m { :x owl:differentFrom :x }
>>
>> OWL-dataset-entails:
>>
>> :n { :t owl:sameAs rdf:type }
>>
>>
>
> Well, depending on which version of OWL. OWLfull, yes.
>
>>> I don't think that this means that 2.16 is not tricky. There is some
>>> interaction. However, this means that named graphs are not independent
>>> from each other. In fact, there is a much easier situation showing that
>>> named graphs are not independent, namely 2.13 T11.2 Similarly, it means
>>> that named graphs are not independent from the default graph.
>>>
>>> It is also the case that an inconsistent default graph makes the named
>>> graphs irrelevant.
>>
>> It makes the dataset inconsistent, which is fortunate.
>
> I don't consider this to be particularly fortunate. Why should an
> inconsistent default graph make the named graphs irrelevant? Why should
> the default graph situation bleed into the named graphs at all?

This semantics consider that datasets are sentences and quotes of 
sentences (not quote in the sense of Sandro's "quoting semantics, but in 
the usual conversation sense). For instance, consider the following 
statements:

Joe is a person. Joe is not a person. Joe told me "I am person".

This is inconsistent. However:

Joe is a person. Joe told "I am a person and I am not a person".

This is consistent. What Joe says is inconsistent, but it's irrelevant 
for the consistency of my statements.


>>> This last is, I think, a particularly strong point against providing
>>> this sort of semantics at all.
>>
>> By "this last", what do you mean? This last test case (T14.1) or this
>> last sentence that you wrote above?
>>
>> I think you mean the former (if it's the latter, I don't see why). Do
>> you think that, if the graphs--named and default--were independent, it
>> would be acceptable? That's the alternative proposal where IGEXT maps
>> IRIs to graphs, instead of resources to graphs.
>
> I prefer having the named graphs independent of each other, and
> independent of the default graph. I might be able to live with the
> situation where an inconsistent default graph makes the named graphs
> irrelevant, but why should I have to? We could just go back to the
> proposals from much earlier where these sorts of issues do not arise.

Which proposals?


>>
>> But even with IGEXT mapping resources, I find the test case to be even
>> more satisfying if it's consistent. It makes it easy to implement a
>> reasoner in that case. Take a reasoner for entailment regime E.
>> Check if any two of the graph IRIs denote the same thing in the
>> default graph. For each group of owl:same graph IRIs, reason on the
>> merge of the graphs in the group, using the E-reasoner.
>
> This is not sufficient. You can do the forcing n into m trick to require
> lots of case splitting.

Hmm, let me try to find an example of what you are saying.

{ :p  rdfs:range  xsd:boolean .
   :x  :p  :n1, :n2, :n3 .
   :n2  owl:differentFrom  :n3 . }
:n1  {  :a  :b  :c  }
:n2  {  :a  owl:sameAs  :d  }
:n3  {  :a  owl:sameAs  :d  }

OWL-Full-dataset-entails:

:n1  {  :d  :b  :c  }

Right. Difficult. But not more difficult than OWL Full. Maybe for 
OWL-DL-dataset-entailment, there could be syntactic restriction. It will 
not shock DL people to have a syntactic restriction such as "you cannot 
use a graph IRI in the default graph".


> If named graphs are just named by IRIs then this is not needed.

I'm personnally pro this alternative but not to the point that I'll 
block the other proposal.


>
> peter
>
> PS: Here is a reiteration of the old proposals.
>
> There is no independent notion of interpretations of RDF datasets. If
> you want to do something like entailment between RDF datasets you can
> either look at one component of the dataset, so you ask whether the
> default graphs sit in an entailment relationship or ask whether the
> graphs with a particular name sit in an entailment relationship, or you
> can look at the entire dataset, so you ask whether the default graphs
> sit in an entailment relationship and all the similarly-named graphs sit
> in an entailment relationship. In each case you probably want to
> consider a missing named graph to be the empty graph. This ends up being
> more flexible and considerably simpler than the minimal semantics
> currently being proposed, as well as requiring no new reasoning machinery.
>


To me, it is important to have entailment between datasets.

{ <n,G> }  entails  { <n,G'> }

is not the same as "extract G from <n,G>, G entails G'".

I want that the inferred triples are labeled by the graph IRI, for 
various purposes including version management, provenance management, 
temporal validity, etc.

-- 
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 Friday, 14 September 2012 13:27:08 UTC