Re: Shapes, Individuals, and Classes - OSLC Motivations

I'm not an expert on SPARQL but it appears to me that base SPARQL doesn't use 
any notion of entailment.  Instead it operates directly on the RDF graph 
itself.  The details of this are given in Section 18.3.1 of the SPARQL spec 
http://www.w3.org/TR/sparql11-query/#BGPsparql

peter

PS:  The SPARQL documents talk about simple entailment, but it seems to me 
that basic SPARQL matching does not depend on simple entailment at all. SPARQL 
appears to be based on RDF graphs, subgraph matching, and instance maps, none 
of which depend on entailment.


On 11/13/2014 08:08 AM, Eric Prud'hommeaux wrote:
> * Peter F. Patel-Schneider <pfpschneider@gmail.com> [2014-11-13 06:12-0800]
>> Well "RDFS entailment" is certainly a term that can be used, but I
>> don't think that "graph entailment" is defined in the RDF spec.
>
> OK, let's see if we at least agree on what the term means and then
> find an acceptable term. Backing down from a fairly expressive regime,
> QL allows one to determine that two IRIs are the same individual. This
> data would match a constraint that x:mother have a cardinality of 1.
>
> data:
>    foaf:mbox a owl:InverseFunctionalProperty .
>
>    <Foo> x:mother <A>, <B> .
>    <A> foaf:mbox <C> .
>    <B> foaf:mbox <C> .
> schema (Resource Shapes):
>    :FooShape a rs:ResourceShape ;
>      rs:property [
>        rs:name "mother" ;
>        rs:propertyDefinition x:mother ;
>        rs:valueType oslc:AnyResource ;
>        rs:occurs rs:Exactly-one
>      ] .
> schema (ShExC):
>    :FooShape { x:mother NonLiteral }
>
>
> RDF entailment allows one to lean bnodes:
>
> data:
>    <Foo> x:mother _:a, _:b .
>    _:a foaf:givenName "Madonna" ; foaf:familyName "Ciccone" .
>    _:b foaf:givenName "Madonna" .
> schema (Resource Shapes):
>    :FooShape a rs:ResourceShape ;
>      rs:property [
>        rs:name "mother" ;
>        rs:propertyDefinition x:mother ;
>        rs:valueType oslc:AnyResource ;
>        rs:occurs rs:Exactly-one
>      ] .
> schema (ShExC):
>    :FooShape { x:mother NonLiteral }
>
>
> In the absense of an entailment regime, SPARQL's behavior treats all
> bnodes and IRIs as distinct. What term would you propose for that?
>
>
> For contrast, here's an example of what might be RDFS entailment:
>
> data:
>    my:mbox rdfs:subPropertyOf foaf:mbox .
>    foaf:Person rdfs:subClassOf foaf:Agent .
>    <Bar> foaf:knows [ a foaf:Person ; my:mbox <X> ] .
> schema (Resource Shapes):
>    :BarShape a rs:ResourceShape ;
>      rs:property [
>        rs:name "knows" ;
>        rs:propertyDefinition foaf:knows ;
>        rs:valueShape [
>          a rs:ResourceShape ;
>          rs:property [
>            rs:name "type" ;
>            rs:propertyDefinition rdf:type ;
>            rs:allowedValue foaf:Agent ;
>            rs:occurs rs:Exactly-one
>          ] ;
>          rs:property [
>            rs:name "mbox" ;
>            rs:propertyDefinition foaf:mbox ;
>            foaf:mbox oslc:AnyResource ;
>            rs:occurs rs:Exactly-one
>          ]
>        ] ;
>        rs:occurs rs:Exactly-one
>      ] .
> schema (ShExC):
>    :BarShape {
>      foaf:knows {
>        a (foaf:Agent),
>        foaf:mbox NonLiteral
>      }
>    }
>
> The instance data uses a subproperty (my:mbox) and a subclass
> (foaf:Person) but presumably an "RDFS entailment" would mean that
> instance would match the schema requiring foaf:mbox and foaf:Agent
> respectively.
>
>
>> peter
>>
>>
>> On 11/12/2014 11:09 PM, Eric Prud'hommeaux wrote:
>>> * Peter F. Patel-Schneider <pfpschneider@gmail.com> [2014-11-12 15:13-0800]
>>>> On 11/12/2014 02:31 PM, Holger Knublauch wrote:
>>>>>
>>>>> On 11/13/14, 8:24 AM, Peter F. Patel-Schneider wrote:
>>>>>> RDFS is part of the RDF spec, so we do not have to go beyond the RDF spec to
>>>>>> get the benefits of RDFS.
>>>>>
>>>>> Which benefits do you mean? Inferencing?
>>>>>
>>>>> (In general I admit I often find your statements enigmatic. It's often unclear
>>>>> to me what solution you are proposing).
>>>>
>>>> Well, then I'm having the effect I want.
>>>>
>>>> At this point I'm not advocating or even proposing a solution.  I
>>>> am, however, trying to tease out just what others are advocating or
>>>> proposing and pointing out statements that are not correct.
>>>>
>>>> The statement here was that RDFS is not part of the RDF spec.
>>>> That's not true.  To get the benefits of RDFS (any maybe I should
>>>> have said, if any) you don't need to go beyond the RDF spec.
>>>
>>> We need (relatively non-controversial) labels to say that we do or do
>>> not expect a shape definition to include RDFS inferencing (not
>>> addressing fragments thereof at this point). If we are careful not to
>>> say "RDF spec" can we use "graph entailment" and "RDFS entailment"?
>>>
>>>
>>>> I do also believe that RDFS has benefits, namely its notions of a
>>>> class and property hierarchy and domains and ranges on properties.
>>>> (I'm not saying that RDFS is the best ontology language, or even a
>>>> very good one, but RDFS does provide some interesting capabilities.)
>>>>
>>>>
>>>>> Holger
>>>>
>>>> peter
>>>>
>>>>
>>>
>

Received on Thursday, 13 November 2014 16:35:09 UTC