Timed Text roles

The current Timed Text (DXFP) draft specifies a ttm:role attribute with 
a set of standard values:

http://www.w3.org/TR/ttaf1-dfxp/#metadata-attribute-role

But it fails to explain what these standard values actually mean, which 
makes it very difficult to know precisely how to use them. For example, 
should the Timed Text equivalent of a narratorial voiceover be "caption" 
or "narration"? Should a subtitle in the same language as spoken 
material be "caption" or "transcription" or "dialog"? What are the 
"quality" and "reproduction" roles for? What's the difference between 
"action" and "kinesic"? Would it be possible for the TT WG to clarify 
the standardized values with definitions and examples?

public-tt has only ever had two postings, neither of which was relevant 
to Timed Text and one of which was simply spam, so I'm not sure that it 
has any active readership. So with apologies I'm copying in w3c-wai-ig 
as the most general accessibility group for discussion. I hope it's not 
off-topic.

--
Benjamin Hawkes-Lewis