I think their choice of distinguishing stream level semantics from tree level semantics is probably correct if you want to talk about general documents. I mean

I think their choice of distinguishing stream level semantics from tree level semantics is probably correct if you want to talk about general documents. I mean an HTML document is much better modelled as a tree of nodes than a long string.