Re: exchanging OWL through RIF

My response is below.

Ed Barkmeyer <edbark@nist.gov> wrote:
> 
> Michael Kifer wrote:
> 
> > I am having a second thought about the requirement that OWL should be
> > exchangeable through RIF by encoding it in FOL (which I am guilty of voting
> > for also and now attribute it to sleep deprivation :-).
> > 
> > I think this requirement is completely misguided.
> 
> I think the above statement of the requirement can be misunderstood, but I 
> don't think the intent is at all misguided.
> 
> It is not a matter of "exchanging an OWL ontology"; the requirement is to 
> deliver the semantic content of an OWL ontology as a ruleset, so that a rules 
> engine can incorporate that content into its rulebase.
> 
> A pseudo use-case: A given site "Uhu" using OWL ontologies and an OWL engine 
> may find it necessary to communicate with a site "Rex" that has only 
> "rulesets" and "rule engines" for some task in which Uhu needs the support of 
> Rex.  In this case, it is important that Uhu be able to convert the relevant 
> OWL ontology to a "rules" (RIF) form, so that it can be used by Rex in 
> performing its supporting task.  And the requirement for RIF is that its "FOL 
> subset" be able to capture the semantics of the OWL ontology.
> 
> The alternative view of this scenario is that Uhu simply sends the OWL 
> ontology, and it is incumbent on Rex to convert the OWL ontology to its 
> internal "rules" form.  There is nothing wrong with this view, except that it 
> has no role for RIF -- it makes the OWL->rules conversion a software project 
> for the Rex engine, and another project for the ILOG engine, and another for 
> the Jena engine, etc., creating lots of work for the engine providers and many 
> third parties who are familiar with the proprietary rules forms.  By 
> comparison, any tool that can convert OWL to RIF without loss (standard form 
> to standard form) gives Uhu what is need to work with Rex, and also Rudi and 
> Regina, no matter what rules engines they have.

This kind of scenario will *never* fly.
For one rule set to incorporate another rule set there must be an agreement
on the common terms, predicates, etc. A "standard" translation of OWL would
not conform to that and, therefore, it will require hand-crafted mediators.
These mediators would need to connect the predicates used in Uhu with those
in Rex.

A much more realistic scenario is that each side provides well-defined
APIs by which clients can query the ruleset/ontology providers.
This is what we called "interoperability" in design goals.


> > For interoperability, we will need to be able to send queries to OWL
> > engines. Representation of those queries will need to be hashed out later.
> 
> This is, of course, exactly the inverse use case.  Here the Rex site needs the 
> assistance of the Uhu site in making some inference.  But Rex does not need 
> RIF for this at all, only something like SPARQL.

Exactly. And some people think that interoperability between rule/ontology
languages is more important than interchange.

> But suppose that Rex needs 
> to send this ruleset to Regina, so that Regina can use its local KB to assist 
> Rex in making some inferences.  Then when Rex sends the RIF ruleset to Regina, 
> the SPARQL queries to Uhu that appear in some of the antecedents must have a 
> RIF representation.  (And I think this is in some sense the degenerate case. 
> It is entirely possible that Regina is a 'hybrid' site, combining both DL and 
> Rules reasoning capabilities, with the consequence that Regina wants to 
> "understand" the SPARQL query, not just blindly send it to Uhu.)

"Understand" the sparql query? Are you talking about meta-reasoning about
rule programs? You are dragging the whole thing into the direction of
impossible or, at least, of something that is well beyond the current
technologies.

The exact form of RIF to OWL/RDF queries is yet to be determined, but I
doubt that they will be some kinds of translations from SPARQL or OWL to RIF.
More likely they will resemble current interfaces from rule languages to
databases.

> It seems to me that "Web-based Rules exchange" demands that we support
> BOTH of these two use cases, not just the latter.

We should support things that are

   1. Possible
   2. Within reach of current technologies
   3. Not pie-in-the-sky scenarios, like REX/UHU, which would require us to
      burden RIF with components designed for scenarios that are better
      served by other means (interoprability)


	--michael  

> 
> -Ed
> 
> -- 
> Edward J. Barkmeyer                        Email: edbark@nist.gov
> National Institute of Standards & Technology
> Manufacturing Systems Integration Division
> 100 Bureau Drive, Stop 8263                Tel: +1 301-975-3528
> Gaithersburg, MD 20899-8263                FAX: +1 301-975-4482
> 
> "The opinions expressed above do not reflect consensus of NIST,
>   and have not been reviewed by any Government authority."
> 
>

Received on Thursday, 2 March 2006 22:24:46 UTC