RE: [RIF] Reaction to the proposal by Boley, Kifer et al

Guys

There are a few general concerns we have about the points raised by
Peter in that RDF/OWL should be first-class citizens in RIF. Firstly,
the production/reaction rules should play a good part of the standard.
As I understand it, it has been an industrial requirement all along,
certainly from our company. PR/RR have their own semantics that is not
directly compatible with, say, RDF deduction.

Secondly, tying RIF closely to OWL would effectively assume that
well-designed OWL ontologies become a prerequisite to do any meaningful
form of inference. The reality of large software projects is that it may
be prohibitively expensive to construct and maintain a consistent and
complete set of reference ontologies even for a subset of the main line
of business. In research, it may work well, while in the real world,
only very stable parts of the business could be covered while other more
volatile parts may stay outside of the OWL glasswork. What I am saying
is that rules may have to play a pragmatic role than other parts of the
Semantic Web technologies hence tying them closely with these parts may
hinder the success of the rules technologies per se. Bad ontologies can
unfortunately also be easily constructed and what about reasoning in
their presence, I'm worried about how the system will behave in this
case.

I was wondering if you think these two observations are making sense.

Alex Kozlenkov

Advanced Technologies Group
Betfair Ltd.

-----Original Message-----
From: public-rif-wg-request@w3.org [mailto:public-rif-wg-request@w3.org]
On Behalf Of Peter F. Patel-Schneider
Sent: 03 May 2006 12:27
To: kifer@cs.sunysb.edu
Cc: public-rif-wg@w3.org
Subject: Re: [RIF] Reaction to the proposal by Boley, Kifer et al 


From: Michael Kifer <kifer@cs.sunysb.edu>
Subject: Re: [RIF] Reaction to the proposal by Boley, Kifer et al 
Date: Wed, 03 May 2006 01:35:08 -0400

> 
> "Peter F. Patel-Schneider" <pfps@inf.unibz.it> wrote:
> > 
> > I, on the other hand, have some serious concerns with the proposal.

> > 
> > I worry about using a substitution in the basis of the semantics of
> > conditions, in particular because the rest of the semantics of
conditionals
> > (including the semantics of existentials) are unspecified in the
proposal.
> > Will this actually end up close to a known logic?
> 
> This is based on a known logic: infinite Herbrand interpretations.

Reference, please.

> It is easy to extend this semantics for general, non-Herbrand domains.
> But this will be useful only for the FOL dialect. The others will have
to
> fall back on Herbrand interpretations.

I worry about situations like the following.  

Suppose that there (only) are two constants, T and F, and (only) one
one-place function, N.  Does it then follow from the following two facts

	N(T) = F
	N(F) = T

that

	Ax Ey x = N(y)


[...]

> > I worry about the connection between the proposal and RDF and OWL.
I do
> > not view it as appropriate to relegate existing Semantic Web
languages to
> > an add-on query interface.
> 
> Why?

Because we are supposed to be working within the confines of the
Semantic Web.

> A query interface is a very general thing. 

Is it really?  It instead seems to me to be much more like a ghetto.  In
particular, the proposal uses a ternary predicate for RDF triples,
divorcing RDF facts like ex:a rdf:type ex:b from a representation
as ex:b(ex:a).

> It can range from seamless
> integration (e.g., in FOL-based dialects) 

How?  Can a query interface actually achieve complete integration?  For
example, can a query interface reasonably support a rule language that
is
oblivious to the source of information, i.e., what sort of rule could be
used to produce the transitive closure of the binary relation R, where
some
facts about R are in RDF and others are in the form of conditionless
rules?

> to Rosati and Eiter et al. style integration, 

Perhaps.  These sorts of integrations do make a firm divide between
"rule"
predicates and other predicates, so it may be more natural to "query"
the
other predicates.  I am not totally convinced, even in this case.  I
note
that the papers by Rosati et al. do not have a special query interface
for
the other predicates.

> to external calls to ontologies. 

Well, I suppose that if the ontology cannot be changed then there is no
problem with a query interface.

> Different degrees of
> integration will be appropriate for different dialects. (For instance,
what
> would be the appropriate integration for PR rules?)

Good question.  What answers are reasonable in this proposal? 

> 	--michael  

Peter F. Patel-Schneider




________________________________________________________________________
In order to protect our email recipients, Betfair use SkyScan from 
MessageLabs to scan all Incoming and Outgoing mail for viruses.

________________________________________________________________________

Received on Wednesday, 3 May 2006 11:56:27 UTC