Merging Rulesets

Michael Kifer writes:
>             Merging rules is a whole different ball of hair. As far as I
> can tell, logical reasoning (monotonic or nonmonotonic) is not what you
> would use here. This is an intellectual activity, which can be automated
> with the help of heuristics (which are not logical inference rules),
> machine learning, etc.
> 
> I bet, however, that **if** you decide to encode those rule-merging
> heuristics in a rule-based language (and not in Java, for instance) then
> you would prefer a Prolog-like language with S/NAF rather than FOL.

The Semantic Web applications I'm familiar with (including everything
based on cwm, some of my prolog work, and various things I hear about)
all merge rulesets freely by just puting them together (concatenation,
set-union, conjunction, etc).  It's trivial with RDF, OWL, N3, and FOL
(when you use URIs for names).   It sounds like it might be very hard
or impossible with rules written in a non-monotonic language.

Is this the root of all the tension around NAF/LP/etc?

     -- sandro