Re: xlxp-dev: RDF and XLink

Paul Prescod wrote:

> According to the definition of link as "an assertion of a relationship",
> links are "machine-understandable" assertions. It follows then, that XLink
> could either be superceded by or based upon RDF -- whether or not links
> are metadata (which depends on definitions and context).

But this misses the point of why you need an XLink (or a HyTime or an
HTML) *in addition to* an RDF: they are operating at different levels of
generality and addressing different detailed applications.  

It's like saying that you can get from Austin to Dallas by air so you
have no need for cars.  Sometimes you use cars, sometimes you use
planes. They both reflect the same basic data model (vehicle transports
people) but meet different requirement sets. 

What's missing is a definition of the fundamental model that both
mechanisms reflect so that you can formally say that both match so that
you know whether or not question "should I fly or should I drive" is
even meaningful.

I've observed a lot of people in a lot of venues arguing over what they
think RDF is. That suggests two things to me:

1. RDF is general to the point of uselessness in the absence of a
specific application. I've heard statements like "it's just tupples".
Hot tip: So are all data structures. Doesn't help. There must be more to

2. The RDF specification is not clear itself about what RDF is: is it a
syntax? Is it a modeling formalism? Is it an amorphous pile of goo? All
three? None of the above? Very difficult to tell.

XLink and HyTime, as examples, both suffer from problem 1 as well: they
are so general that you can do almost anything with them, which makes it
difficult to decide what you actually should do with them.  However,
both XLink and HyTime at least clearly define their scope of
application, even if it is artificially (and, in the abstract,
unnecessarily) narrow.

It's also the case that at a certain level of abstraction all problems
in information management come down to relationship representation and
management, so any mechanism, whether it's RDF, HyTime, XLink, or
relational databases will be attractive once you're at the point where
more mundane problems are solved.  The real issue can't be "is RDF
better than/worse than {relationship representation x}?", it can only be
"what fundamental truths about relationships and their management does
{relationship representation x} give us insight into?"  Because at the
end of the day, syntax doesn't matter--only data models matter and in
this domain, the data models are pretty much the same no matter how you
slice it.

It's also the case that we've tended arrive at the same abstraction by a
variety of paths, so it's only to be expected that there would be
different ways of representing the same things. I don't consider it a
problem except to the degree that people are going out of their way to
define unneeded or gratuitous ways to do things that are already done in
a satisfactory way.


<address HyTime="bibloc">
W. Eliot Kimber, Senior Consulting SGML Engineer
ISOGEN International Corp.
2200 N. Lamar St, #230, Dallas, TX 75202

Received on Tuesday, 13 April 1999 21:30:41 UTC