Re: Importing RDF documents from RIF - part 2

On 2010-02-23 11:38, Axel Polleres wrote:
> I realise I never followed up on that...
> 
> On 19 Jan 2010, at 16:09, Jos de Bruijn wrote:
> 
>>
>> On 2010-01-19 16:47, Axel Polleres wrote:
>>> FWIW, the semantics of owl:imports is defined by OWL as well, plus  they have defined an
>>> RDF serialisation, and that makes perfect sense to me, if I look at it from the
>>> So, I can't really follow the argument why this would be in the scope of RDF.
>>
>> I'm not saying it is. I'm just saying it's an extension of RDF, and thus
>> seems out of scope for RIF.
>> OWL 2 DL defines an RDF serialization, which does not have much to do
>> with the RDF semantics. OWL 2 Full is an extension of RDF.
>>
>>>
>>>>  think it would be better to refer to the RIF ruleset
>>>> from the SPARQL query, and have the RIF ruleset import the RDF dataset.
>>>
>>> there is no way to do this directly, I mean without changing the definitions of
>>> BGP extensions, that is at the heart of the definition of entailment regimes for SPARQL.
>>
>> I don't really understand the difference from the point of view of
>> SPARQL between RIF importing RDF and RDF importing RIF.
>> Why does "RDF importing RIF" not require changing the definition of BGP
>> extensions, while "RIF importing RDF", considering that the proposed
>> semantics are the same?
> 
> BGP matching is entirely defined in terms of an RDF dataset, if you can't refer 
> from that dataset to the RIF ruleset, that's a problem.

I still don't see how this affects the definition of BGP matching. You
just have an input RDF dataset. I am assuming this is a virtual dataset
that is obtained from "applying" the RDF/RDFS semantics and a bunch of
RIF rules to an RDF graph. Now, the definition of BGP matching does not
need to know how this dataset is obtained, now does it?

> 
>>>
>>> I agree with the concern that it might be too late for  RIF to squeeze that in the current specs,
>>> it'd be great though, if we'd get a WG note out at least. If only its to endorse that we'd use
>>> the URI rif:usedWithProfile?
>>>
>>> Since OWL has defined an RDF serialisation, handling that from SPARQL is
>>
>> I guess this sentenced was cut off?
>>
> 
> yes... the missing word was "easy", meaning to say handling OWL from SPARQL is easy, since the OWL ontology is 
> encoded in the RDF graph. 
> 
>>>
>>>> By the way, I would be very interested in the semantics you have in mind
>>>> for the SPARQL queries on RIF ruleset.
>>>
>>> For very limited rulesets (strongly safe) and simple RDF its fairly straightforward,
>>> since the closure is finite, finiteness of answers to conjunctive queries (i.e. BGP matching)
>>> should be straightforward. I am unsure whether we'd get to any more complex combinations,
>>> i.e. mixing OWL/RDFS, etc. plus rulesets, this might also depend on how the resp
>>> OWL, RDFS, etc. entailment regimes evolve in SPARQL.. feedback welcome at [1], though you may
>>> want to wait for the next pub round which is pending (next week hopefully).
>>
>> I would be interested in your definition of "closure".
> 
> As I said, the closure is finite for strongly safe core (minimal Herbrand model, definiable 
> by the common least fixpoint operator).

I'm still interested in the definition.

> 
>> It is not in the
>> mentioned spec. I am particularly concerned about how blank nodes are
>> treated. When not being careful, the closure is infinite, even for very
>> simple graphs and empty rulesets.
> 
> in what sense are blanknodes a problem? (when thinking only of strongly safe RIF Core?)
> What speaks against just treating them as skolem constants? That's, BTW also what we do in the 
> OWL entailment regimes in SPARQL already.

I'm not saying they are a problem. It's just that one needs to be
careful with them, as well as with the infinite axiomatic triples.


Jos

> 
> Axel
> 
>>
>>
>> Cheers, Jos
>>
>>>
>>>
>>> Axel
>>>  
>>> 1. http://www.w3.org/TR/sparql11-entailment/
>>>
>>> On 19 Jan 2010, at 15:07, Jos de Bruijn wrote:
>>>
>>>> This sounds like a reasonable extension of the RDF semantics. Not
>>>> convinced, though, that this is in the scope of the RIF-RDF+OWL spec.
>>>> For this reason, and for procedural reasons, I suggest that if such an
>>>> extension of RDF is of wider interest, a separate spec be written for
>>>> this extension.
>>>>
>>>> Furthermore, I am not convinced that extending RDF is the best solution
>>>> to your problem of querying RIF-RDF combinations using SPARQL: RIF is
>>>> "on top of" RDF. I think it would be better to refer to the RIF ruleset
>>>> from the SPARQL query, and have the RIF ruleset import the RDF dataset.
>>>>
>>>> By the way, I would be very interested in the semantics you have in mind
>>>> for the SPARQL queries on RIF ruleset.
>>>>
>>>>
>>>> Cheers, Jos
>>>>
>>>> On 2010-01-19 15:55, Axel Polleres wrote:
>>>>> Here the easy-to-digest version:
>>>>>
>>>>> In order to enable the import of RIF documents from RDF, I suggest the addition of a new section
>>>>> 6 (or alternatively a new subsection to section 5) to
>>>>> http://www.w3.org/2005/rules/wiki/SWC#Semantics_of_RIF-RDF_Combinations
>>>>> which essentially says the following:
>>>>>
>>>>> ====================================================================================
>>>>>
>>>>> = 6 Importing RIF rulesets in RDF =
>>>>>
>>>>> The models of a RIF-X-combination (R,S) where any of the graphs in S contains a ground triple
>>>>>
>>>>>    R1 rif:usedWithProfile P .
>>>>>
>>>>> such R1 is an IRI referring to a RIF document and P is a URI referring to an imports profile as defined in
>>>>> Section http://www.w3.org/2005/rules/wiki/SWC#Profiles_of_Imports
>>>>> are restricted to those models which are also models of (G, R') where R' is identical to the RIF document R,
>>>>> with the addition that R' has additional imports clauses
>>>>>
>>>>>   Imports( R1 )
>>>>>   Imports( G P )
>>>>>
>>>>> Together with the conditions in section 5.2 this ensures that RIF-X-combinations where R is empty, i.e. which
>>>>> are only defined by a set of RDF graphs, can also import RIF rulesets.
>>>>>
>>>>> ====================================================================================
>>>>>
>>>>> I think this should work apart from that there is one small issue which I consider possibly "suboptimal"
>>>>> as opposed to "owl:imports".
>>>>>
>>>>> The triple
>>>>>    R1 rif:usedWithProfile P .
>>>>> is in such case still be considered part of the "data" whereas owl:imports triples are in fact not considered in the OWL direct semantics, right?
>>>>>
>>>>> maybe those better filled in about "owl:imports" can help, I could actually also try to solicit help/alternative suggestions from people that
>>>>> have a more complete understanding on that matter.
>>>>>
>>>>> best,
>>>>> Axel
>>>>
>>>> --
>>>> Jos de Bruijn
>>>>   Web:   http://www.debruijn.net/
>>>>   Phone: +39 0471 016224
>>>>   Fax:   +39 0471 016009
>>>>
>>>
>>
>> --
>> Jos de Bruijn
>>   Web:   http://www.debruijn.net/
>>   Phone: +39 0471 016224
>>   Fax:   +39 0471 016009
>>
> 

-- 
Jos de Bruijn
  Web:   http://www.debruijn.net/
  Phone: +39 0471 016224
  Fax:   +39 0471 016009

Received on Wednesday, 24 February 2010 10:07:43 UTC