Personally I've come around to a pragmatist position on this, after some
time as a radical Free The Text purist. It's just not possible to
encode information without also encoding something about what we're
supposed to do with it, any more than it's possible to draw a "real"
line segment. But we can certainly do some very useful things by
trying.
An interesting paper on a similar topic is at
http://www.sil.org/sgml/ohco1.html, "Refining Our Notion of What Text
Really Is: The Problem of Overlapping Hierarchies."
--