- From: Jonathan Borden <jonathan@openhealth.org>
- Date: Mon, 14 Jul 2003 13:44:29 -0400
- To: "Williams, Stuart" <skw@hp.com>, "Roy T. Fielding" <fielding@apache.org>
- Cc: <www-tag@w3.org>
Roy T. Fielding wrote: > > The resource referred to by the URI does not vary. What varies is the > target that is ultimately referred to by the "sentence" surrounding the > URI referral. While I generally agree with your stance on this issue (as far as I can tell), I am slightly concerned with this introduction of the terms "target" of a "sentence". If not a "resource", what is such a "target"? In RDF terms, such targets of sentences are RDF resources, sometimes so called "anonymous resources" or "b-nodes" (blank nodes) in RDF terms. If this is what you mean then we have a readily available formalism to work out these issues. >... For example, if I say > > I want one of these cars: <http://www.vw.com/touareg/>. > > Then I have used that URI to identify a category of vehicles by > reference to an HTTP resource identified by an http URI. The URI > is acting as an identifier for that VW brand of car, and it seems > unlikely that www.vw.com will reuse that identifier for something > else, even though it is clear that <http://www.vw.com/touareg/> > on its own is a website for the vehicle brand and not the brand itself. In trying to work through this issue bear with me, if you will, as I try to translate your descriptions of these "sentences" into OWL/RDF http://www.w3.org/TR/owl-ref statements. OWL does provide a logic for making inferences about resources (owl:Thing's) based upon such "sentences". In N3: :I :WantCar <http://www.vw.com/touareg/> . Now let's assume that the property :WantCar is constrained to target things that are cars (sounds resonable :-) We can do this with an RDFS domain/range restriction. :WantCar rdf:type rdf:Property . :WantCar rdfs:domain :Person . :WantCar rdfs:range :Car . This means that when the :WantCar property is used in a statement (sentence), the subject is constrained to be a :Person and the object is constrained to be a :Car given this we can conclude that: :I rdf:type :Person . <http://www.vw.com/touareg/> rdf:type :Car . > > In other words, context matters even when the URI itself is > context-independent, and use within a given context is what defines > the meaning of a reference. That is why there is no conflict at all > between the references <a href="http://example.com"> and > <foo xmlns="http://example.com">; the context surrounding > the reference defines meaning by its use, not by the URI scheme. > Whether there is or is not a conflict depends. For example suppose (edited for better RDF clarity) and assuming that both "href" and "xmlns" can be considered properties (xmlns is not normally so considered, but allow this example). <a id="A1" href="http://example.com"> <foo id="FOO1" xmlns="http://example.com"> => :A1 :href <http://example.com> . :FOO1 :xmlns <http://example.com> . and :A1 rdf:type :a . :FOO1 rdf:type :foo . :href rdfs:range :WebSite . :xmlns rdfs:range :XMLNamespace . now I'd be able to conclude that http://example.com rdf:type :WebSite . http://example.com rdf:type :XMLNamespace . which _might be_ a contradiction if I also say that web sites and namespaces are _disjoint_ i.e. :WebSite owl:disjointWith :XMLNamespace . Bottom line: although this N3 might be painful for those of us who are not RDF/OWL inclined, I believe it is important to try and define these things in a precise fashion, otherwise there will be a danger of introducing terms such as "sentence", "context" and "target" to the already muddled URI discussions. I do think that the RDF/OWL treatment of such issues is appropriate, and hope this will be considered in deciding these issues. Jonathan
Received on Monday, 14 July 2003 13:44:42 UTC