W3C home > Mailing lists > Public > public-rif-wg@w3.org > January 2006

Re: [SWC] Re: RIF and QL

From: Enrico Franconi <franconi@inf.unibz.it>
Date: Thu, 26 Jan 2006 02:23:14 +0100
Message-Id: <58D73E57-2A86-4D6E-B6C2-C520F21ADD25@inf.unibz.it>
Cc: W3C RIF WG <public-rif-wg@w3.org>
To: edbark@nist.gov

On 26 Jan 2006, at 01:22, Ed Barkmeyer wrote:
>> I am sure (or hope) that what are you talking about is only *one*   
>> option out of the *many* that RIF has to characterise and define  
>> in  order to interoperate with ontology languages (such as RDF, in  
>> this  example): let's call this option the 'trivial'semantics.
>
> YES!  This 'trivial' semantics is what I thought François was  
> recommending. What I describe is only one option out of many even  
> for doing this.  And it is not about "interoperating with RDF", as  
> Chris Welty said in the telecon, it is only about interoperating  
> with an "external" SPARQL service.

... even though there is not much difference among the two, since I  
(roughly) proved in my email how inter-operating with 'trivial  
semantics' with RDF (or OWL) is a *special* case of interoperating  
with an "external" SPARQL service as defined by you and Francois.

> The issue of "interoperating meaningfully with ontology languages"  
> is a MUCH more complex question.

Of course :-)
But remember that what you call interoperation with an "external"  
SPARQL service is just a special case of it, with a special semantics.

>> However, there may be several kinds of these 'connections': most  
>> of  them are based on a model-theoretic characterisation rather  
>> than on  entailment (see [1] - and I can really think of at least  
>> three  additional important classes: FOL semantics (à la SWRL), LP- 
>> weak-safe  semantics (à la Rosati), and autoepistemic semantics).  
>> So, we are  really talking about at least 4 different semantic  
>> options to  characterise the interoperability between a knowledge  
>> base and rules  by means of 'query expressions' (as defined above)  
>> appearing in the  body of some rules.
>
> I agree that it is possible and interesting to investigate this  
> option.  What you are talking about is "rule language expressions"  
> that interrogate the "integrated knowledge base" available to a  
> rule engine with some integrated or "attached" inferencing  
> capabilities that have other model theoretic bases.

yes, but let's better say: "rule language expressions" that interact  
with the "integrated knowledge base" available to a rule engine with  
some integrated inferencing capabilities that have other global model  
theoretic bases.

> This requires understanding of how to integrate the model theories  
> and/or how to "restrict" and "contextualize" rule language  
> expressions for the "federated" or "unified framework" of model  
> theories of the attached inferencing engines.

yes, and a (almost) complete survey on this is in [1].

>> So, while I believe that there should be room for the 'trivial'   
>> semantics above, RIF should try to characterise also the several   
>> alternative approaches as known in the literature.
>
> No argument here.  Yes, the WG should be required to investigate this.
> I only say that the "query language" that deals with "integrated  
> model theories" is a *new* query language, and the unified model  
> theory is needed to interpret any such query.

yes.

> Even if it *looks like* SPARQL, the semantics of the query is not  
> RDF semantics per se; it is rather the RIF-meets-RDF semantics.

Yes, as a general statement. However, given the minimal model  
property of RDF, the two things will coincide if also the RIF has the  
minimal model property (such as f-f horn). The slightly annoying  
things are the bnodes.

> The issue that motivated the contribution (from both François and  
> me) is whether we can "extend" a query language like SPARQL into  
> RIF with any notion of compatibility.

If we restrict 'sparql' to happen only in the body, it seems to me  
doable (for the reasons above). However, the unavoidable non trivial  
matter, as I argued in my message on RDF/SPARQL compatibility <http:// 
lists.w3.org/Archives/Public/public-rif-wg/2006Jan/0062>, is when you  
have an RDF triple in the head as well, since you may have unsafe rules.

> What I understood François to say is that you only get "SPARQL  
> compatibility" by encapsulating SPARQL queries -- the 'trivial'  
> semantics -- and I agree with that point.  Whether/when it is  
> possible to get RDF-compatible results from a RIF query that uses  
> the integrated RIF-meets-RDF model theory, and how the syntax of  
> that query relates to the SPARQL syntax, are (IMO) open issues.

Not really, they are well understood semantically: the DL community  
has deep results on the integration of rules with KBs since almost 15  
years, and we have now more recent results even deeper and  
interesting; similarly you can say of the FLogic community.
This is to say that all the semantic options are well known, and we  
can easily reuse those results to build the wide spectrum of  
possibilities that RIF should have in defining the various RIF-meets- 
RDF/OWL model theory.

Specifically to RDF, I am convinced that the problem may be really  
much easier than we think. We just need to stop for a while and start  
writing down exactly (i.e., formally) all the different things we  
want, and then the formalisation of the semantics, and the  
computational behaviour will hopefully come easily; all from the  
known literature.

> The point is that RDF and SPARQL do not define "compatibility" in  
> this case; the unified RIF-meets-RDF semantics is the definition of  
> "compatibility".

Sure.

> And in that area, I am inclined to agree with François that we will  
> not solve that problem to anyone's satisfaction in a year.  (I see  
> from your paper that you have reason to disagree with that  
> position.)  But that is all the more reason why we should not begin  
> by trying to "extend" a language like SPARQL. We can't do that  
> without first agreeing on the "unified framework", and the  
> resulting relationship to SPARQL will be "a superset of a subset  
> with a modified interpretation".

I somehow disagree with this conclusion. I am much more optimistic.

[My effort in the SPARQL WG (DAWG) proved that: we have now (it is  
not public yet) a semantics for SPARQL that is fully compatible with  
all conceived logic-based extensions, but of course it is defined  
only for the trivial task of getting syntactically subgraphs out of  
RDF graphs]

Look, I am not saying we should make any choice that will make  
somebody unhappy: we just have to *study* all the choices in  
parallel, and then to define the semantics of the RIF part 1 (pure  
horn) compatible with all/most of them. You can't say in advance that  
this will fail, unless you produce a counterexample :-)
And if we don't fail, everybody will be happy, which is my goal as a  
scientist.

cheers
--e.


[1] Enrico Franconi and Sergio Tessaris (2004). Rules and Queries  
with Ontologies: a Unified Logical Framework. Workshop on Principles   
and Practice of Semantic Web Reasoning (PPSWR'04).
<http://www.inf.unibz.it/%7Efranconi/papers/ppswr-04.pdf>
Received on Thursday, 26 January 2006 01:23:27 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 2 June 2009 18:33:26 GMT