RDF is, as the model and syntax draft is perfectly clear, really
about labelled directed graphs. At its core, therefore, is the notion
of a labelled, directed edge. (You could define this either in terms
of a single ternary relation
link(from,label,to) or in
terms of a collection of nonsymmetric, irreflexive, intransitive
relations, one for each label type.) I will ignore second-order
for now, except to observe that although the ternary approach allows
quantification over labels without going second order, as
labels are just constants, it would still have to go second order to
or any other intentional label type.
Again, in RDF the nodes of the graph are resources, that is, loci in hyperspace, or strings.
Now the relevant observation is clear: In the XML-link draft, links include, although they are not limited to, labelled directed connections between resources.
In the interests of avoiding the unnecessary proliferation of W3C recommendations with overlapping domains, I think it is appropriate to then ask why RDF is not first recognised and then presented as an XML-link application, rather than an independent standard.
Consider the following example, taken from the RDF Model and Syntax draft recommendation (August 1998 version):
The first description above can be accommodated to XML-Link with minimal restructing, as follows:
Slightly more restructuring makes things a bit simpler in the DTD, but maybe that's a red herring:
The obvious advantage of this is it means you can have multiple properties with no additional work in the DTD.
One obvious problem here is that although complex things (i.e. resource-valued properties) are easy, simple things (i.e. string-valued properties) are ridiculously hard. The best you could do as things stand would be:
which is less than wonderful.
The obvious solution to this is to allow the inline attribute on locator link elements, with the obvious meaning: no href allowed, link end is the locator element itself.
Now we can do the whole example:
Everything else (except for second-order stuff, which I'll return to in a subsequent note) is syntactic sugar, a fit subject for negotiation between the XML Schema, XLink and RDF Schema groups.