Web applications are not the same as web pages. A web page is a document, which ought to have simple clean markup. A web application on the other hand is a tool that can edit said documents. It's more "meta."
To expect the tool to have the same sort of semantic purity as the documents it interacts with is overly idealistic. It doesn't matter as much what we can systematically infer about an application, it is more important that the content be accessible in some sort of standard format.
Similarly, Scribd's HTML5 viewer, is just that, a viewer. Not a full fledged PDF to HTML converter. It's primary purpose is to allow a user to experience a document in its entirety in the browser. In a world of trade-offs, and priorities, the markup of the viewer itself is near the bottom.