Re: A vision for the RIF

Francois Bry <bry@ifi.lmu.de> wrote:
> 
> Dear Gary,
> > My big concern is finding a single declarative semantics for both
> > deduction rules and reactive rules.  
> 
> Dop not try. Deduction rules are declarative or stateless, while
> reactive rules are imperative or state dependent. A common semantics for
> deduction and reactive rules would as ill-defined a goal as a common
> semantics for Java and Haskel. A common semantics for deduction and
> reactive rules would be like marrying fire and water.

Transaction Logic gives a clean model theory, which extends predicate
calculus with reactive rules (among other things). There are many users of
FLORA-2 who are apparently not unhappy playing with water and fire at the
same time.

> > Maybe a better way to state this is to have a semantics for RIF that
> > would let me evaluate a given rule set using either forward chaining
> > or backward chaining (with the same results obviously).  
> I advocate for a declarative semantics for Horn deduction rules defined
> in terms of fixpoint and, equivalently, as minimal model.

I am fine with that, but several people here want FOL too.

> In addition, I
> advocate for requiring Interpreters of deduction rules to fullfil a few
> (rewasnable) termination requirements for  (syntacxtuically defined)
> classes of (finite) Horn deduction rule sets

I think this would be too intrusive. But we should think of a mechanism by
which the sender could convey to the receiver that there is a way to eval
the given rule set using some algorithm. One general way (if a known set of
queries is intended to be asked against the ruleset)
could be to tell the receiver that there is a finite number of
answers N. Then a reasonably general inference engine (not Prolog's) would
know to stop churning out answers after N of them are found.

> 
> > Harold Boley proposed at the last F2F that the rule head would be an
> > implicit assert when run on a forward chaining engine.  That sounded
> > like a good idea to explore.
> Might be good for diploma theses or PhD. Leave it gthere. Do not try to
> marry fire and water. We should not misuse the WG for doing in a few
> weeks research that has not been done by a much largewr community over a
> few decades.

Can you point *specifically* to where is a PhD thesis (or even an MS
thesis) in the roadmap that Harold was talking about?


	--michael  


> > Of course, there would be lots of restrictions on the rules, but I
> > don't see any way to have RIF Core be anything but the semantic
> > intersection of deductive, normative, and reactive rules.  
> There is a clean and simple way out: RIF been made of 3 complementary
> yet distiunct languages: a lanfguage of deduction rules, a language of
> normative rules, and a languiage of reactive rules.
> 
> Regards,
> 
> Francois
> 
>

Received on Thursday, 20 April 2006 18:20:50 UTC