W3C home > Mailing lists > Public > public-rule-workshop-discuss@w3.org > August 2005

Re: A Rule Interchange Format VS a Rule Language for Interoperability and other issues

From: Christian de Sainte Marie <csma@ilog.fr>
Date: Thu, 25 Aug 2005 10:49:48 +0200
Message-ID: <430D862C.9040705@ilog.fr>
To: Dieter Fensel <dieter.fensel@deri.org>
CC: public-rule-workshop-discuss@w3.org

Dieter Fensel wrote:
> 
> Dear Christian,
> 
> It is always refreshing to get valid requirements from industry on what is
> needed, but not always similar insightfully to listen to how it can be 
> achieved.

Thanx Dieter. It is always so cool to hear the nutty professor explain 
what the really beautiful solution is! If only it was practical...

:-) (just in case somebody wasn't sure)

> Christian, with all respect you seriously contradict yourself on two 
> essential points:
> 
> 1) You want to have an exchange language that cover 100% of all rule 
> languages
> (even you admit later that you expect only 80% of the things actually 
> exchanged
> between different rule languages).

Sorry, I was not clear enough: I meant, 80% on average: 100% of my 
requirements, 60% of yours. So, no contradiction on my side.

> Therefore, your requirements could only be fulfilled in the way you propose
> either by second-order logic or by using FOL only as an exchange syntax 
> without
> attached semantics BUT NOT BY FOL WITH MULTI-MODEL SEMANTICS as it is 
> defined
> in any text book.

Good point on deductive closure: is there a way to get around that? You 
mention minimal model semantics somewhere: can you tell me more (sorry, 
I left active research a long time ago -about the time you got your PhD; 
and I was never very good with formal logic anyway).

I am not sure I understand what you mean with multi-model semantics: is 
the fact that a formula can have multiple interpretations a problem? 
Why? For all practical purposes, rules come with a model, anyway: you 
have to define the meaning of your (application-specific) 
predicates/functions, in most cases type your variables etc. Sections 
2.6, 2.7 and 2.8 in Sandro's draft are exactly about how the RFI deals 
with that: actually, I think that this is probably the difficult part 
(but only the import mechanism is in the scope: making sure the other 
side gets the semantics is a more general semantic web problem).

So, ok, typed FOL may not be FOL (and FOL+SNAF is not FOL anymore, 
anyway): what if we said that the "the starting point will be a notation 
for FOL with (e.g.) Tarski-like semantics; it will then be extended to 
cover the pragmatic requirements (SNAF, types, actions in the 
consequent, etc), while making sure to keep it sound", or something like 
that. Does it make sense? That is what was meant, I think, since the 
draft also mentions all of those extensions.

Or is it something else (a small example like the one on deductive 
closure would help greatly)?

  In conclusion; I only see three ways to go (truly
> spoken only
> one actual pathway to go):
> 
>         - Using second-order logic (I wait for the genius that is going 
> to propose this)

Yes, I mean: no, I do not think anybody wants to go that way.

>         - Using only an exchange syntax which brings us close to the 
> usage of XML

Yes, yes! Is it easy to do? RuleML did most of it alredy? Great! Let's 
do that (ahum... would it be possible to have it with a tiny bit of 
semantics, though?)

>         - Asking ourselves whether there is a way out of this dilemma. 
> And in fact,
>         there is!
> 
> Why not going the other way around? Identifying the 80% that vendors and 
> users thinks
> they need (and are therefore subject for broader exchange), put them in 
> a simple
> language and define a unique model semantics for it as it is used by 
> most systems
> anyway (in case they use a model theoretical semantics at all)?

1. Efforts to define a good enough universal rule language have been 
on-going for most of the last 40 years. There is no reason to believe 
that they will succeed anytime soon, and why anymore within the WG life 
time than elsewise? Ok, it could be next week: but what if it is not?

2. Even if we could be sure that the language could be specified within, 
let say, 2 years, there is a huge legacy of rule bases and rule-based 
systems that would need be ported to the new language, all the tools, 
commercial or not commercial would have to be redeveloped on the new 
basis; all this while the installed base is growing at a very fast pace 
(so, even more to port)! That probably means no broad interoperability 
before, let us be optimistic, 10 years.

With the less than perfect interchange language solution, I believe that 
we can reach a honest level of interoperability (enough to relieve a big 
part of the users pain) within 2 or 3 years. And, as I said already, 
that RIF solution does not preclude the RLI one, either: actually, on 
the contrary, a widely deployed RIF would help in the adoption of the 
RLI (once one is defined).

Christian
Received on Thursday, 25 August 2005 11:00:39 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:16:23 GMT