- From: Ed Barkmeyer <edbark@nist.gov>
- Date: Thu, 20 Apr 2006 11:28:00 -0400
- To: Francois Bry <bry@ifi.lmu.de>
- CC: public-rif-wg@w3.org
François Bry wrote: > There is a clean and simple way out: RIF been made of 3 complementary > yet distinct languages: a language of deduction rules, a language of > normative rules, and a language of reactive rules. We...ell. I'm not sure that is either clean or simple. First, I still don't understand the semantic model for "reactive rules", but that's a minor point. Second, Michael Kifer has argued that we can't get a useful semantics for a language that supports "normative" and "deductive" rules in combination. So, bowing to his expertise, I can agree these must be separate languages. But... Many applications for "normative rules" also want some kind of "deductive" (or perhaps "reactive"?) capability whose application in some sense "precedes" the application of a "normative ruleset". So, if these are "separate languages", they have to be part of a "family", as Gerd says, and that family has to occasionally get together at dinner. Third, (IMO) "at dinner" that family has to include the adopted children of FOL ethnicity: RDF and OWL. Fourth, I'm not sure how quickly we will get agreement on the semantic model for a single "language of deduction rules". We all know that there are multiple well-known semantic models for "rules-based inferencing". Will we agree to use one semantics, with its limitations, as the standard? or agree to several semantics models, each of which corresponds to an "intended semantics" attribute of a ruleset? And in the latter case, isn't the "normative" semantics just yet-another-semantic-model? Finally, if all we can agree on is syntax that can support several possible semantic models (some of which will yield identical results for some rulesets), we will violate our Web responsibilities: We can't know all of the users/recipients of the ruleset. So we can't privately agree on the intended semantics for the exchange. A web-published ruleset with an unspecified semantics is meaningless. So I think what François suggests is only "simple" in theory: We have to agree on more than one semantic model, document all the ones we agree on, and give them names, which are possible values for some attribute of the ruleset. [And OBTW, if W3C had done that with OWL, we would have DL+ 22 combinations of letters for add-on features, guaranteeing that nearly no ontology would transfer successfully between engines -- exactly the opposite of the result that Bijan observed.] -Ed -- Edward J. Barkmeyer Email: edbark@nist.gov National Institute of Standards & Technology Manufacturing Systems Integration Division 100 Bureau Drive, Stop 8263 Tel: +1 301-975-3528 Gaithersburg, MD 20899-8263 FAX: +1 301-975-4482 "The opinions expressed above do not reflect consensus of NIST, and have not been reviewed by any Government authority."
Received on Thursday, 20 April 2006 16:08:57 UTC