Re: Inferencing

Yes these are two topics:

1) Constraint checks may want to see additional matches ("in the WHERE 
clause"). For example, a SPARQL query may contain ?x rdf:type ex:Person 
and this should also include all ex:MalePersons. This problem is already 
solved by OWL's triple mapping and corresponding implementations - most 
RDF databases have a mode in which these inferences are made on the fly 
and then become visible to the query.

2) Constraint definitions attached to a superclass (assuming 
Shape=class) need to also apply to instances of a subclass. This is 
covered by the Charter and is essential, but it's something that the 
surrounding engine can handle when it triggers the queries.

I knew this would start a lengthy discussion, but it is necessary to 
resolve at some stage anyway ;)

Holger


On 8/9/14, 11:15 AM, Peter F. Patel-Schneider wrote:
> I don't see the interaction here.  Hierarchies of shape definitions 
> appear to allow more things to match when doing constraint-based 
> recognition.  Here instead there is the dual notion of more things 
> being required to match.
>
> peter
>
>
> On 08/08/2014 05:43 PM, Holger Knublauch wrote:
>> Yes I would also be disappointed, but I believe this is already 
>> covered by the
>> charter section 3 (Scope):
>>
>> "Hierarchies of shape definitions
>> Permit any of a set of shapes to stand for a specified shape, e.g. to 
>> say that
>> either a User shape or an Employee shape can be used in place of a 
>> Commentor
>> shape."
>>
>> Holger
>>
>>
>> On 8/9/14, 10:39 AM, Peter F. Patel-Schneider wrote:
>>> I would be very disappointed if the RDF graph
>>>
>>> foo rdf:type bar .
>>> bar rdfs:subClassOf bbb .
>>>
>>> satisfied the constraint
>>>
>>> bbb <= atleast 2 prop
>>>
>>> I thus think that inferencing has a lot to do with constraint checking.
>>>
>>>
>>> peter
>>>
>>>
>>> On 08/08/2014 05:33 PM, Holger Knublauch wrote:
>>>>
>>>> On 8/8/14, 10:24 AM, Eric Prud'hommeaux wrote:
>>>>>
>>>>> > 3. OPTIONAL A specification of how shape verification interacts 
>>>>> with
>>>>> > inference.
>>>>>
>>>>> I think this one feel off radar. Did you see any support for this?
>>>>>
>>>>
>>>> In general, constraint checking should not *require* inferencing. 
>>>> However, I
>>>> believe we should make sure that the topic of inferencing does not get
>>>> prohibited by the charter. If the WG decides there is a chance to 
>>>> improve the
>>>> semantic web stack, then it should be allowed to do so. For example 
>>>> I do like
>>>> the idea in one of the ShEx papers to use structural information to 
>>>> produce
>>>> new output (e.g. XML trees or other RDF triples). Another example is
>>>> spin:rule, which is in our experience tremendously useful for defining
>>>> mappings between ontologies, and to calculate the ex:area of a 
>>>> ex:Rectangle
>>>> from ex:width and ex:height. Once we have a mechanism to attach 
>>>> SPARQL and
>>>> templates to classes for constraint checking, we could use exactly 
>>>> the same
>>>> mechanism to define such production rules - it becomes a rather 
>>>> trivial
>>>> addition that would keep the solution consistent. All this could go 
>>>> into a
>>>> separate, non-normative deliverable, but we should not exclude it.
>>>>
>>>> Holger
>>>>
>>>>
>>

Received on Saturday, 9 August 2014 01:57:49 UTC