- From: Bijan Parsia <bparsia@isr.umd.edu>
- Date: Sat, 11 Feb 2006 13:28:18 -0500
- To: Igor Mozetic <igor.mozetic@ijs.si>
- Cc: Gerd Wagner <wagnerg@tu-cottbus.de>, Francois Bry <bry@ifi.lmu.de>, Dieter Fensel <dieter.fensel@deri.org>, public-rif-wg@w3.org, bonatti@na.infn.it, edbark@nist.gov
On Feb 10, 2006, at 3:34 AM, Igor Mozetic wrote: > Dieter Fensel wrote: >> At 14:21 09.02.2006 +0100, Piero A. Bonatti wrote: >>> +1 !!!!!!!!!! >>> >>> On Thursday 09 February 2006 08:55, Francois Bry wrote: [snip] >>> > c. RDF bnodes are a serious challenge for efficient reasoning >>> which, to >>> > the best of my knowledge, is far from being solved. >>> > >> Indeed, this is something very stupid about RDF. Why is this a >> recommendation >> to repeat similar mistakes for RIF? > > Shouldn't be OWL added to the list as well? Which species of OWL? > In general, it seems that IETF process of accepting Internet > standards/RFCs (the requirement for prior implementation and testing) > works well. The W3C has a similar requirement. The CR (candidate recommendation) phase is where you are required to gather implementation experience of the complete design. Hence, while designing, you look for current implementation experience in anticipation of the CR. This was, in fact, done for OWL, esp. the DL fragment. My organization (the MIND Lab) started our OWL DL ++ reasoner Pellet specifically to show that traditional tableau reasoners were not *that* difficult to implement (i.e., there had been some concern in the WebOnt group that only people deep in the DL community could do it; not so). > Maybe at least for RIF we should try to stay closer to > operational recommendations. Of course, we don't know what the implementation requirements will be, exactly. But we ought to be keeping them in mind. (For example, I could see a phase one approach that did a superset rather than a subset approach, wherein while the superset was semantically fairly straightforward to specify, the relations between the superset and various subsets similarly easy, there was no practical or even known complete proof procedure and little hope of one. Does that mean it is unimplementable? Well maybe. It still could be the case that *analysis* tools short of full inference (e.g., for checking compatibility of fragments; even incomplete ones could be very useful) would be implementable and I could see a CR criteria set that argued that that was what was needed. Indeed, this might be exactly the way in which RIF isn't itself a "rule language" but an interchange format. Like a "rule language" it has a formal syntax and semantics, but unlike a rule language there's no necessity for there to be any practical entailment determining procedure. (Just thinking aloud)) Cheers, Bijan.
Received on Saturday, 11 February 2006 18:28:37 UTC