There were a number of sessions I found particularly interesting either because they presented novel ways to utilize and process XML or because they gave an insightful glance at how others view the XML family of technologies. 

Imperative Programming with Rectangles, Triangles, and Circles - Erik Meijer
This was a presentation about a research language called Xen that experiments with various ways to reduce the Relational<->Objects<->XML (ROX) impedance mismatch by adding concepts and operators from the relational and XML (specifically W3C XML Schema) world into an object oriented programming language. The main thesis of the paper was that heavily used APIs and programming idioms eventually tend to be likely candidates for including into the language. An example was given with the foreach operator in the C# language which transformed the following regularly used idiom

IEnumerator e = ((IEnumerable)ts).GetEnumerator();
  try {
     while(e.MoveNext()) {  T t = (T)e.Current; t.DoStuff(); }
  } finally {
     IDisposable d = e as System.IDisposable;
     if(d != null) d.Dispose();


foreach(T t in ts){ 

The majority of the presentation was about XML integration. Erik spent some time talking about the XML to object impedance mismatch and how cumbersome programming with XML could be.  Either you wrote a bunch of code for walking trees manually or you queried nodes with XPath but then you are embedding one language into another and don't get type safety, etc (if there is an error in my XPath query I can't tell until runtime). He pointed out that various XML<->object mapping technologies fall short because they either don't map a rich enough set of W3C XML Schema constructs to relevant object structures but even if they did one now looses the power of being able to do rich XPath queries or XSLT/XQuery transformations. The XML integration in Xen basically came in 3 flavors; the ability to initialize classes from XML strings, support for W3C XML Schema constructs like union types and  sequences into the language and the ability to do XPath-like queries over the contents fields and properties of a class.

There were also a few other things like adding the constraint "not null" into the language (which would be a handy modifier for parameter names in any language given how often one must check parameters for null in method bodies) and the ability to apply the same method to all the members of a collection which seemed like valuable additions to a programming language independent of XML integration.

Thinking about it I am unsure of the practicality of some features such as being able to initialize objects from an XML literal in the code especially since Xen only supported XML documents with schemas although in some cases I could imagine such an approach being more palatable than using XQuery or XSLT 2.0 for constructing or querying strongly typed XML documents. Also I was suspicious of the usefulness of being able to do wildcard queries (i.e. give me all the fields in class Foo) although this could potentially be used to get the string value of an XML element with mixed content.

The language also had integrated SQL like querying with a "select" operator but I didn't pay much attention to this since I was only really interested in XML.

The meat of this presentation is available online in the paper entitled Programming with Circles, Triangles and Rectangles. The presentation was well received although sparsely attended (about two or three dozen people) and the most noteworthy feedback was that from James Clark who was so impressed he kept saying "I'm speechless" in between asking questions about the language. Sam Ruby was also impressed by the fact that not only was there a presentation but the demo which involved compiling and running various samples showed that this you could implement such a language in the CLR and even integrate it into Visual Studio.

Namespace Routing Language (NRL) - James Clark
This was a presentation for a language for validating a single XML document with multiple schemas simultaenously. This was specifically aimed at validating documents that contained XML from multiple vocabularies (e.g. XML content embedded in a SOAP envelope, RDF embedded in HTML, etc).

The core processing model of NRL is that it divides an XML document into sections each containing elements from a single namespace then each section can be validated using the schema for its namespace. There is no requirement that the same schema language is used so one could validate one part of the document using RELAX NG and use W3C XML Schema for another. There also was the ability to to specify named modes like XSLT which allowed you to match against element names against a particular schema instead of just keying off the namespace name. This functionality could be used to validate interleaved documents (such as XHTML within an XSLT stylesheet) but I suspect that this will be easier said than done in practice.

All in all this was a very interesting talk and introduced some ideas I'd never have considered on my own.  

There is a spec for the Namespace Routing Language available online.