I wonder if some actual examples might help  clarify this. What kind of 
things are you thinking of Courtney?

In transcribing speech for the BNC we had lots of arguments about the 
need to distinguish normalization from correction. For example:

1) you're transcribing a speech in which someone says "couldn't of" 
(people do) and you'd like to correct/normalise this to "couldn't have"
2) in the transcript someone else made, you think they mis-heard 
"couldn't have" as "couldn't of" (or vice versa)

The first is a case for <reg> and <orig>. The second is a case for 
<sic>  and <corr>

However, in deciding how to render the markup, I assume you'd be guided 
by your intended/perceived audience. Some kinds of audience absolutely 
must have access to the full weirdness and uncertainty of the original. 
Others want to be offered a more sanitized version of reality. Yet 
others like to have access to both -- so you need to dream up cunning 
visual tricks to show where your nice clean transcription strays onto 
thin ice. Graying out or italicising the text to show that there's some 
kind of uncertainty attached to this passage might be enough. The full 
Leiden-like apparatus that Gabriel talks about would probably scare a 
lot of people away from actually reading the transcripts -- which might 
not be what you want.

Courtney Michael wrote:
> The latter.
> On 7/15/09 11:31 AM, "Daniel Paul O'Donnell" <[log in to unmask]> wr=
> ote:
> Interesting question. I don't know--but am interested in discovering the
> answer. What would the nature of the sic/corr relationship be? I.e.
> self-correction by a speaker or editorial correction of a mistake?
> Courtney Michael wrote:
>> We are looking for an answer to the former question - for examples or
>> models actually.
>> We are displaying transcripts on the web and want to see if there is a
>> standard or generally accepted way to display the corrections.