- From: Piero A. Bonatti <bonatti@na.infn.it>
- Date: Wed, 8 Feb 2006 17:38:57 +0100
- To: Dieter Fensel <dieter.fensel@deri.org>, "Gerd Wagner" <wagnerg@tu-cottbus.de>, <edbark@nist.gov>, <public-rif-wg@w3.org>
On Wednesday 08 February 2006 17:13, Dieter Fensel wrote: > >But OWL/SWRL have already introduced disjunctive > >conclusions (which btw are not a problem for the > >model-theoretic semantics, even not when combined > >with NAF; they are only a problem for the inference > >engines), so this is not PhD research! > > 1) They are ONLY a problem for the inference engines!!!!!!!!! > So for the ONLY thing that really counts. there are very interesting inference engines - notably, DLV - based on disjunctive logic programming (i.e. they support disjunctions in the head). they are nicely and efficiently integrated with widespread databases and have already attracted the attention of some companies and there will be more and more engines supporting more and more useful features, as time goes by. so why should a rule INTERCHANGE format exclude the features supported by such engines? in general, WHATEVER FEATURE WE EXCLUDE A PRIORI IS A POSSIBLE CAUSE OF EARLY OBSOLENCE OF THE RIF STANDARD I don't see why interoperability should be guaranteed among arbitrary pairs of reasoning systems. one can't plug a PCMCIA device into a USB port. similarly, the RIF should allow - disjunctive engines to interoperate with each other - "minimalistic" rule engines to interoperate with each other - disjunctive and "minimalistic" rule engines to interoperate, *when the shared rulebase is simple enough* but of course an engine is not obliged to understand or make sense of every possible rule encoded in the RIF what I would really expect from RIF is support for fast and effective identification of the features actually used by a rulebase, so that an engine can quickly check whether it can interpret those rules. here all sorts of properties may come in (range restrictions, hierarchic or recursive rulebase structure, datalog or function-supporting fragments, etc.). the metadata attached to a rule encoding would more or less directly specify the requirements on the inference engine this approach yields a far more scalable standard piero
Received on Wednesday, 8 February 2006 16:39:11 UTC