Re: yet another strawman, was Re: generic XML to RDF triple mapping

On Fri, 8 Sep 2000, Jonathan Borden wrote:

> James,
> >
> > I believe that it should be possible to map arbitrary XML into RDF
> triples.
> funny you should mention this :-)
> I generally agree, and would put forward that the 'mapping language' is an
> RDF Schema itself. On the other hand with or without an RDF schema, it is
> possible to 'direct' an RDF processor given 'minimally invasive'
> modification of an XML document.
> The simplified RDF syntax by Sergey Melnik itself modified from TimBL's
> strawman serves as a basis for doing so.
> The main problems are that anonymous instances are generated unless XML
> elements are tagged with rdf:instance="..uri..." attributes. I would like to
> propose a mechanism that does not rely upon explicit naming of anonymous
> resources, rather uses an XPointer to a particular node within the XML tree
> as a URI fragment to create a URI reference to the resource e.g.
> t)
> My strawman is here:

Thanks, I've added this to

I'm still digesting Brian's note on anonymous nodes, so may yet change my
mind, but I have to say I've learned to live with anonymous nodes in
RDF. Having hacked around with various implementation strategies, I'm now
happy with anonymous nodes so long as we have strategies for identifying
those resources through description.

For example, a Person node might lack a URI in some chunk of RDF data, but
come with a description of their foo:personalMailbox or
bar:personalHomePage, either of which is sufficient to uniquely pick out
some individual.

Regarding the spec-theology of whether the 'URI is in the model', I find
it hard to argue with ignorance. My RDF robot often comes across data
files that mention a resource without mentioning its URI. While as a
resource it is the sort of thing that should have a URI name, you can't
blame the datasource or the indexing robot for not having that information
to hand. So while I believe the RDF model says that nodes have URIs, that
doesn't automagically get us into a situation whereby each RDF
processor/database always *knows* the URI for every node it has some
representation of. Sometimes this information costs money, for example.

 On the syntax front, it looks like some syntax proposals will make
it easier than others to associate URIs with nodes when exchanging
data. For example during the SOAP/RDF discussion at WWW9, it seemed the
main syntax difference between RDF and SOAP was that, in a SOAP
serialization, URIs (if present) would be explicitly represented as
attributes of some node, ie. they weren't specialcased: URI names were
just another property of objects.

I do like the idea of picking out the xpointer information though, but
would probably be inclined to represent these as properties of the node
rather than as its URI. This of course strays into the territory the list[1] is currently revisiting, ie. the old chestnut of
whether one resource might be named by two different URI names etc etc.



Received on Friday, 8 September 2000 15:22:59 UTC