- From: Peter F. Patel-Schneider <pfps@research.bell-labs.com>
- Date: Thu, 20 Dec 2001 10:15:53 -0500
- To: jos.deroo.jd@belgium.agfa.com
- Cc: las@olin.edu, phayes@ai.uwf.edu, jjc@hplb.hpl.hp.com, Frank.van.Harmelen@cs.vu.nl, horrocks@cs.man.ac.uk, mdean@bbn.com, lynn.stein@olin.edu, www-webont-wg@w3.org, www-archive@w3.org, hendler@cs.umd.edu, connolly@w3.org
From: jos.deroo.jd@belgium.agfa.com Subject: Re: UPDATE: initial message concerning syntax Date: Sat, 15 Dec 2001 14:43:29 +0100 > Hi Peter, > > [...] > > I think SWOL is really nice. > In the RDF/RDFS MT we have, among other things, a bunch > of entailment rules. I think SWOL should just extend that > idea via additional SWOL entailment rules. > So using SWOL vocabulary brings extra SWOL entailment rules. > Let's start with a simple example: > > :Person a swol:Class; swol:disjointUnionOf ( :Man :Woman ) . > > where the range of swol:disjointUnionOf a list of classes > (and it is always a list, even when it is a singleton) > Now we can entail stuff via following entailment rule > for all :C, :D, :L, :x > > { :C swol:disjointUnionOf :L . :L :member :D . :x a :D } > log:implies { :x a :C } . > > (maybe we should also add { :L a swol:List } to the LHS) > and for :member one has also entailment rules > for all :x, :a, :b > > ( :x / :b ) :member :x . > { :b :member :x } log:implies { ( :a / :b) :member :x } . > > where ( :x / :y ) is syntactic sugar for > [ a swol:List; swol:first :x; swol:rest :y ] > > I think if we can do that for all SWOL vocabulary > there should be no trouble for SWOL entailment > and actually for ANY entailment. Unfortunately, as I have already indicated in several places, this approach does not work, or, at best, only works with a lot of difficulty and fiddling. To make it work correctly you have to include a full theory of lists and other syntactic constructions in your theory. Once this is done, semantic paradoxes, or, if you prefer, the ability to derive a contradiction from the empty knowledge base, are very hard to avoid. Even if the whole formalism does not fail, there are quite a number of related issues that affect interpretations and inference. > It is also interesting to use SWOL vocabulary for > RDFS entailment rules e.g. > > rdfs:subPropertyOf a swol:TransitiveProperty . > rdfs:subClassOf a swol:TransitiveProperty . > > instead of the RDF/RDFS MT rules rdfs5 and rdfs8. > (just as we already use log:implies) so it's > not hierarchically layered, but wild web like :-) It is precisely this absence of layering that opens up the possiblity of semantic problems. > -- > Jos De Roo, AGFA http://www.agfa.com/w3c/jdroo/ peter
Received on Thursday, 20 December 2001 10:16:49 UTC