XML information modeling best practices

I'm reading Jeffrey Veen's _The Art & Science of Web Design_ (New
Riders)  this weekend.  I strongly recommend it for its perspective on
Web design, but it's also fun to read a lot of it from an XMLish
information-modeling perspective.  It's getting beaten up on Amazon by
Jakob Nielsen and David Siegel fans as well as the usual suspects, but
there's more than a little brilliance here.

I was thinking that it would be cool to do a book that was more or less
_The Art & Science of XML_, though it's not exactly the kind of approach
my employer (O'Reilly) usually publishes.  Then I got to wondering
whether any book on XML and information modeling would get the same sort
of blasting response from people in different schools of thought.

For all the discussion that takes place on xml-dev, we spend very little
time discussing the structures that people are actually creating with
XML.  To some extent, that's a natural outgrowth of XML 1.0's emphasis
on syntax, and certainly worthy.  On the other hand, I have to admit
that I frequently marvel at the XML I come across 'in the wild'. 

I had a very hard time keeping myself from apologizing a few weeks ago
as I presented some raw RPC-style SOAP markup. I can barely stand even
to look at the XML used for XML-RPC (despite having written a book on it
and discussed it in various books).  The XML-based file formats I've
looked at lately - the format for Tinderbox, for instance - seem pretty
plainly to be raw object serializations without much (any?) thought
given to the markup.

None of these examples prevents developers from doing what they need, at
least in so far as their use in the program context for which they were
intended.  In the case of Tinderbox, XSLT can extract what I want
without any particularly strange trickery, and the program's well worth
exploring whatever I think of their XML.

I know that information modeling is hard - no question about that.  Is
expressing information models in XML really _this_ hard, though?  Or is
it just that people don't value taking the effort?

The relational database people, whatever their other sins, did an
excellent job expressing things like normalization approaches. 
Object-oriented programming has school after school battling out these
issues and best to accomplish particular information representations and
processing.  XML seems to have left all this to the wind.

These are kind of Saturday-afternoonish thoughts, but I'm really
wondering whether a lot of the confusion around XML (and some of the
hideous results we're seeing) comes from a lack of discussion - and
tools built around the results of those discussions.

-- 
Simon St.Laurent
Ring around the content, a pocket full of brackets
Errors, errors, all fall down!
http://simonstl.com