- From: Christian de Sainte Marie <csma@ilog.fr>
- Date: Wed, 24 Aug 2005 20:18:19 +0200
- To: Michael Kifer <kifer@cs.sunysb.edu>
- CC: Jim Hendler <hendler@cs.umd.edu>, Sandro Hawke <sandro@w3.org>, public-rule-workshop-discuss@w3.org
Michael Kifer wrote: > > RuleML people have realized that there can't be a single super-language > into which everything can be translated with the same semantics. So, their > approach is that the semantics rests with the rules languages and RuleML > defines their XML serializations. > > The dream of being able to take any rule-based language, map it into a > "superlanguage", push through the wire, and then map it to a different > rules language at the other end of the wire (with an equivalent semantics) > is a pipe dream - unachievable. Fair enough. But who would want to do that, anyway (mapping any rule from any one language to any other one, be it through a single pivot language or elsewise)? People write and/or use rules for a purpose and they will want to share/retrieve rules for a purpose, too. So, for all practical purposes, we are talking about a "superlanguage" such that rules written in any language can be map into it, and then mapped back into any other language, with an equivalent semantics, provided that there exist a semantics-preserving mapping for that rule between the initial and end languages. Or a "superlanguage" that would work like that for a reasonable subset of the rules that can be written in a reasonable subset of the rule languages (where reasonable must be large enough for the superlanguage to add sufficient value to be worth its development). My understanding is that this is what the charter describes. > At least, not through FOL. So, RuleML takes > a more pragmatic approach Yes, RuleML is indeed a possibility, afaIu. My concern with RuleML is that it seems to take a prescriptive (top-down) approach to which languages can map into each other, whereas my natural tendency would be to leave it to the users (I mean, e.g. rules or application developers) to decide. The bottom-up approach requires that the standard specifies graceful failure behaviours, but it seems to me that it is more robust against unforeseen usages (in that it allows them, which is what we want it to be). And profiles can help, anyway (profiles differ from the RuleML typology in that a user community can define -and publish- any profile it finds useful). Again, maybe I am dead wrong, but, then, I really need somebody to explain why... Christian
Received on Thursday, 25 August 2005 11:00:57 UTC