RE: Non-Text Literals

Geoff,

> From: Geoff Chappell [mailto:geoff@sover.net] 
>
> The problem is that referring to a web resource solely by its 
> URL (as a URI) in RDF doesn't fix its meaning. 

That's what we have model theoretic interpretations for. Some of this
got thrashed out on the TAG list not so long ago. When it comes to URLs,
the naming authority gets to fix meaning (assuming 'meaning' means what
is denoted by a URI). Arguably, all URIs that support authoritative
naming (such as http) have their meanings fixed by the naming authority.



> Applications 
> may choose to believe otherwise and assume the missing 
> knowledge that can't be specified in RDF - i.e that we really 
> mean a particular URI to be treated as a URL so it's ok for 
> an application to go and retrieve something from it.

That's not missing knowledge, that's RDF not having a model for
dereferencing URIs. I used to think this was a serious problem; now I
think it can be dealt with in a layered design. 

 
> I've been wondering recently if datatyping could play a role 
> here. An rdf datatype is essentially a function, the details 
> of which are unknown to rdf, that is able to fix the 
> interpretation of a name. The web through its mechanisms 
> plays a similar role of resolving names to resources. So why 
> not have a web datatype in rdf? For example:

It sounds like you're talking about pluggable datatypes? I think the wg
have/are considering this, you might want to go through rdf-core's
archives. Though I imagine any such semantics can also be provided
through the property's domain and range without a mechanism for
datatyping the literals. The thing is, surely Literals are of the type
Literal, not of type
UnknownTypeTunnelledThroughLiteralsLetsRunAProcessToFigureOutWhatAndDoAD
ynamicCast.
 
> _:a rdf:lex http://www.example.org/someimage.gif
> _:a rdf:dtype ex:HttpResource
> ex:HttpResource rdf:type rdf:DataType
> 
> (hopefully that approximates one of the datatype proposals 
> closely enough so that my point is clear)

Nice; I'm not sure it's necessary to provide datatypes to do this, or
even if it's a good idea. It might be, I'm just not sure.


Bill de hÓra 
--
Propylon
www.propylon.com

Received on Saturday, 31 August 2002 11:05:31 UTC