QuickTopic's Quick Doc Review system does this, both for HTML and MsWord documents (which it renders to HTML). It posts a document with ANAMEs inserted, and lets each paragraph spawn a QuickTopic thread.
Bruck Eckel has done something similar, making a BackTalk feature in Zope. His Thinking In Python book draft is delivered in static HTML, but with an "Add a Comment" link at the end of each paragraph, which calls the Zope feature on his site, passing a paragraph ID.
I wonder whether it makes sense to think of this in terms of Annotation Systems?
A big semantic question is whether paragraphs are the right level of granularity. As Robert Horn has pointed out, sometimes a paragraph contains multiple concepts/ideas. Maybe we need to teach people to do more Structured Writing (OutLining as final format)?
Making Wiki paragraph-addressable is a challenge, because nodes are intended not to be static. One could try and do it anyway, any maybe tie into a complicated way of walking back through diffs. Or one could argue that it isn't necessary because you can edit wiki nodes directly and insert comments right where they are applicable. But this has it's own huge tracking problems...
The OHS crowd inserts Purple Numbers into EMail archives. But the problem is that these only appear in the archives, not in the distributed messages. So they are rarely referred to in the process of the discussion. I think one needs greater integration between the listserve host and the web-archive generator, so that a message:
hits the listserve server
then the rendered email gets distributed to the list members
(there's plenty of issues here regarding: whether to author in Structured Text, HTML or PlainText; whether everyone has to be able to read HTML vs maybe inserting Purple Numbers into a still-PlainText version to distribute; when replying to a message with Purple Numbers, whether/how to strip the Purple Numbers out of the original message.
Edited: | Tweet this!