- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Mon, 5 May 2003 23:18:29 +0300
- To: www-webont-wg@w3.org
> Would you argue there isn't enough evidence for > Prolog to move to PR if we were the Prolog WG? There are some standard execution models for Prolog. So if my chess program terminates in a month on swi-pl I would be surprised if it takes 300 years on yap. Admittedly there are some variant execution models but those that are likely to impact whether your program works at all are usually prominently contrasted with the standard implementation (Warren Abstract Machine). === So, agrreing that P !+ NP, we see that even OWL Lite is at some level problematic ... and I haven't expressed anxiety about that. What's the difference? I think that the type of constructs one needs to use in Lite to get to encode computationally hard problems are really quite unnatural, i.e. I think they are of theoretical interest but not ones that real data are likely to exhibit. On the other hand, the feature interactions in OWL DL that cause the problems seem, in my judgement, to be ones that happen quite naturally. Given the ability to make closed world comments about the cardinality of properties then it seems natural to say a property is N-to-M. Given three of these connecting three finite or infinite classes then you get the test cases I have just posted. This does not feel to me like an academic game, but a serious engineering point - if we don't expect the users to use such facilities we probably should not define them. My triangle pictures - at least for me, it is natural to describe them to highlight certain points - their triangularity and their vennness. These descriptions are topolgocial and can be encoded in OWL DL, yet reasoning with them involves working out who is who, with the absence of unique names. (The 3-SAT examples are artificial problems steming from the absence of unique names). More another day ... Jeremy
Received on Monday, 5 May 2003 17:18:21 UTC