- From: Ian Horrocks <horrocks@cs.man.ac.uk>
- Date: Tue, 20 Aug 2002 18:41:36 +0100
- To: "R.V.Guha" <guha@guha.com>
- Cc: Richard Fikes <fikes@KSL.Stanford.EDU>, www-rdf-logic@w3.org
On August 19, R.V.Guha writes: > > Ian, > > I hope I am not pulling this too much out of context ... but you seem > to be implying that somehow an implementation may or may or match up to > the semantics? If so, what is the point of the semantics in a practical > sense? Even when an implementation is based on a "correct" algorithm, the implementation itself may introduce errors. It is also possible that some implementors will deliberately or accidentally sacrifice completeness, and perhaps even soundness, for whatever reason - there isn't much we can do about that. These cases only seem to highlight the need for a clear specification of what it means to be correct. > > On a more general note, I've been wondering about the complexity issues > involved on the SW. Even with the comparatively more tractable > computation afforded by DLs (as compared to FOL), we are still dealing > with algorithms that are far from sub-linear. And with millions of > axioms distributed around the web, I doubt there is any chance of > providing complete inference for the SW ... I'd like to hear your > opinion on this. It isn't clear that computation w.r.t. the whole web will be feasible via whatever mechanism. Fortunately, many applications will only need to compute w.r.t. a very small subset of the total content of the web. Ian > > guha > > Ian Horrocks wrote: > > >> This situation might be acceptable, even inevitable, for an > >> implementation, but is highly undesirable for a semantic > >> specification. > > >> Regards, Ian >
Received on Tuesday, 20 August 2002 13:37:55 UTC