- From: Jon Hanna <jon@hackcraft.net>
- Date: Mon, 20 Sep 2004 17:39:33 +0100
- To: <www-rdf-interest@w3.org>
> In particular the use/mention distinction has really helped locate a > problem that has kept bugging me when thinking about rdf. The analogy is false IMO. Mentioning a URI is saying something like "The URI 'http://www.example.net/blah' has 27 characters, uses the common convention of naming a webserver 'www', has no query-string and doesn't use any of character escapes defined in RFC 2396". On the other hand both using a URI to identify a resource and to retrieve a representation are using URIs. > This can be summarized by considering the following triples which > though formally contradictory, can in fact be seen to be compatible: > > (A) http://www.paris.org/Monuments/Eiffel > http://purl.org/dc/elements/1.1/creator > http://www.gustaveeiffel.com/ > > (B) http://www.paris.org/Monuments/Eiffel > http://purl.org/dc/elements/1.1/creator > http://www.endex.com/gf/ > > If we only understand the above names to be used then the two > sentences > are contradictory. > > (A) states that the Eiffel Tower was created by Gustav > Eiffel, whereas > (B) states that > it was created by Gary Feuerstein. (A) is true in this > interpretation, > whereas (B) is false. > > But we can guess that what was probably intended by (B) was not that > <http://www.paris.org/Monuments/Eiffel> refer to the Eiffel > Tower, but > that it refer to the Web Page which refers to the Eiffel > Tower, whereas > in the first sentence we have the intuition that it refers to the > actual building located in Paris. We can guess what was meant by a lot of incorrect statements, they are still incorrect though. Actually it could be that A is correct and B is incorrect - it is perfectly reasonable for URIs to identify the kind of "conceptual document" that the HTTP-URIs-are-limited camp say all URIs identify (sometimes a cigar is still a cigar). Neither possibility for what http://www.paris.org/Monuments/Eiffel identifies is problematic though. There is a clear relationship between the document and the monument in either case: All we need is some sort of framework for describing resources which defines relationships between two or more resources. Preferably those relationships would be identified with URIs to keep it all webby and RESTful. Maybe something using a labelled acyclic graph might be the way to go about it... > Harry's solution is to assume that rdf resources by default point to > the representation of the resource and not the thing itself, and then You can't even assume that a resource has a representation, or that there is only one. Assuming that this is what is being referred to is hence falls long before the first fence. (Also, I dispute the use of the term "rdf resources". There are resources. RDF is a way to describe them.) > to add a new wpn:// scheme to help create names that refer > directly to > an object. It follows from the argument that URIs don't identify resources that we need some sort of uniform resource identifier. I don't buy the problem, so I don't buy the solution. It occurs to me right now that from a philosophical > perspective this cannot quite be the correct solution, since the > arguments have always been that names can be used in both ways. I've hardly ever heard this argued. What I have heard argued is: 1. URIs identify any sort of thing. When used with a given system (the web) they result in representations being returned. When used otherwise the system in question does whatever is appropriate for it to do with respect to that thing. 2. URIs identify "conceptual documents". When used with the web they result in a particular document being returned, which in regards to such matters as content-type, language and other negotiable features is an instance of the more Platonic "conceptual document" (which is not in a particular language or content-type). Just because you get a representation when you do a particular action with it no more makes that representation the thing the URI identify than using my name in an enquiry makes "Jon Hanna" intrinsically bound to "Quite tall, shaven head, wears black a lot". It certainly doesn't make it bound to that and nothing else. > Would it perhaps not be easier to extend RDF so that one can > point to a > resource in either way, for example by allowing the following: > > <Entry rdf:about="http://www.paris.org/Monuments/Eiffel"> > or > <Entry rdf:refers="http://www.paris.org/Monuments/Eiffel"> That isn't extending RDF, it's extending RDF/XML. I think the way to extend RDF to deal with this is to define a predicate that defines the relationship between the resource identified by the URI and the zero or more representations that may be returned when you dereference it's URIs. (Amusingly one could usefully mention URIs for real here since a resource can have more than one URI, but the representation returned may depend on which one is used - this is a feature of the character strings the webserver received, not of the URI qua resource identifier.) > I have been confused a few times about this exact problem, > and am very > thankful to Harry Halpin, and Henry S.Thompson for the clear > explanation of this problem. I think the waters are muddier now.
Received on Monday, 20 September 2004 16:39:54 UTC