Re: RIF: A thought about requirements --> PRR

From: "Vincent, Paul D" <PaulVincent@fairisaac.com>
Subject: RE: RIF: A thought about requirements --> PRR
Date: Wed, 31 May 2006 11:43:08 -0500

> Thanks Peter: this is the first time I've seen some written argument
> against PR support in RIF. Comments below...
> 
> PS: I'm assuming this is a serious response, and not a joke (difficult
> to identify sarcasm from my perspective).
> 
> Paul Vincent
> for Fair Isaac Blaze Advisor  -- Business Rule Management System
> @ OMG and W3C standards for rules
> 
> > -----Original Message-----
> > [...]
> > 
> > > Qu: What are your arguments *AGAINST* PRR support for the widespread
> > > adoption [goal]?
> > 
> > From viewpoint that I espouse:
> > 
> > 1/ PRR support opposes alignment with the Semantic Web, because it
> > requires
> >    concepts that are alien to the Semantic Web, and alignment with the
> >    Semantic Web supports widescale adoption.
> 
> [PV>] This is an interesting viewpoint that we should study in more
> detail: if RIF is part of the semantic web and PR opposes the semantic
> web, then clearly I am wasting my time in RIF! 

This doesn't follow.  There is no sense that there *has* to be complete
success in all the goals, CSFs, and requirements.  Instead, our job is to
find a result that does a decent job of deciding where to stand when
resolving or alleviating conflicts.

> However:
> - what PR concepts are alien to the semantic web? The fact that
> production rules are usually defined against data terms, and such terms
> are decomposed into RDF in the SW? 

Production rules are based on operational concepts, which are alien to the
declarative basis of the Semantic Web.

> - who claims "widespread adoption" for the SW? I don't see any facts to
> support this.

What?  Are there not applications in the SW?  Do they not count?  

I would argue the other way.  The RIF WG is in the Semantic Web activity so
*only* SW applications should count for the widescale adoption goal.
Covering other applications is useful, but not necessary.

> > 2/ PRR support opposes low cost of implementation, because it requires
> >    a complex, non-coherent formalism.
> 
> [PV>] I'm not sure I understand this, as there are opensource PR tools
> available (ie low implementation cost), and I would be interested in why
> (ie supporting evidence) for the complexity / lack of coherence for the
> "formalism" (you mean semantics?).

Well, if existence of tools means low implementation cost, then there is
low implementation cost overall because tools exist for all the proposed
aspects of the RIF.  

I maintain that the production rule paradigm (operational specification,
conflict resolution, attached procedures, etc.) is a very complex formalism
and that there is indeed a lack of coherence in it.  I would make the same
claim for just about any other formalism that is currently used a complete
programming paradigm.

> > 3/ PRR support opposes no surprises, because PRRs are inherently
> >    surprising.
> 
> [PV>] I assume "surprising" has some technical meaning here I am not
> familiar with. Exciting maybe... :-) 

The conflict resolution mechanisms in production rules make their behaviour
quite surprising (at least to me).

> > 4/ PRR support opposes the support of logical rules, because PRRs are
> >    non-logical.
> 
> [PV>] I think you mean to say:
> PRR support opposes the idea of RIF as an exclusively logic language,
> because PRs are non-logical.

I would go further.  I would say that PRR support opposes the idea of RIF
as a language with a logical component, because PRs are non-logical.  This
is *not* to say that the RIF could not contain both logical and non-logical
components, just that there is an opposition here.

> Which is true.
> However, why does RIF need to be exclusively about logical rules? 

It doesn't, although I would argue that it would be best if the RIF was
exclusively about non-operational rules.  This is not to say that the RIF
could not support certain aspects of production rules, however.

> Or do
> we need to consider RIF(logic), RIF(behavioural) etc ie subtypes of RIF
> for different fundamental rule types?

I do not believe that this is a desirable way to proceed.

> > Most of the above arguments depend on the PRR support being
> > interpreted as support of the inherently operational aspects of PRRs.
> > If instead, PRR support can be done from a simple, standard logical
> > language, then these arguments have much less weight.
> 
> [PV>] It would be interesting to know if PRs can be represented in a
> form of logic - chiefly this is the action / RHS / behavioural part. Can
> logic be used to describe the behaviour(s) seen in IT systems?

Well, in some sense yes, as logic is in some sense universal.  However, I
don't view that as a desirable way to proceed.  But what says that all
aspects of production rules *have* to be captured in the RIF?  

> > > Qu: What alternative to PRs will fulfil the widespread adoption CSF?
> > 
> > The diagram has already several alternatives.
> 
> [PV>] Per
> http://www.w3.org/2005/rules/wg/wiki/Goals%2C_Critical_Success_Factors_a
> nd_Requirements (thanks Evan for correcting my terminology!) there are
> indeed other requirements that fulfil this goal. But note the diagram is
> not weighted, and the terms are not defined. For example, if widespread
> adoption is defined as being amongst the EXISTING semantic web
> community, you are probably correct. If widespread adoption means
> amongst EXISTING rule users, then probably adding weights to the links
> would indicate importance for the PR support - although of course this
> would be difficult to quantify.
> 
> > [...]
> 
> [PV>] As a counter argument, how many of the use cases assume PRs and
> can RIF be viable without such use cases being supported? 

Yes, this is indeed an interesting argument.  Who would like to "count
noses"?  I would suggest two counts, one for "mentions" of non-logical
rules and one for need for non-logical rules.

peter

Received on Saturday, 3 June 2006 06:49:30 UTC