search

w3c / Guide

The Guidebook is the collected wisdom of the W3C Group Chairs, team contacts and other contributors.
https://www.w3.org/Guide/
24 stars 43 forks source link

Allow saving generated transcripts responsibly #170

Open michael-n-cooper opened 1 year ago

michael-n-cooper commented 1 year ago

@a11ydoer raised a question that we discussed in the WAI coordination call, about the meeting transcript saving policy. The Guide instructs staff to disallow transcript saving, for important reasons. But we now have some experience with the feature, and in most of my groups it has become routine to turn on auto-transcription as an accessibility and language support. While transcripts have errors, they are generally understandable. Groups have begun to wonder if scribing should be a priority, especially given the participation cost to the scribe in a given meeting. Therefore, we would like to explore incorporating current experience into the transcript saving policy.

I believe many of the concerns we have with saving transcripts could be addressed with policy and guidance. I believe it is still inappropriate to use automated transcripts as a formal meeting record, but I think it is appropriate to allow them to supplement the official record.

There are also social concerns about the use of saved transcripts. While we can't know what would happen in an untried experiment, our social experience with live captioning has been encouraging. People are used to it, and I think the step to saving transcripts is less of a leap than it was.

We look forward to discussing an appropriate way to make use of this new technology.

michael-n-cooper commented 1 year ago

There has been discussion of considerations, solutions, and limitations on the WAI coordination call list. I encourage people to migrate points here but meanwhile am adding that reference.

nigelmegitt commented 1 year ago

As a Chair and frequent scribe, my main concern with saving meeting transcripts is that there is no mechanism for deliberately not recording off the record comments: "Please don't scribe this... [sensitive content deliberately omitted from the log]" etc.

It is a valuable function of meetings that such information can be shared with trust, and a policy of saving transcripts would likely remove or reduce the willingness of participants to do so.

KimPatch commented 1 year ago

Seems like the best use of meeting transcripts would be allowing the scribe to cut-and-paste from the transcript as an aid to doing minutes and then not saving the transcripts. So the scribe would continue to be the gatekeeper to the minutes, but the scribe would also have access to a tool that would help the job go faster and better. (As I mentioned previously, automatic transcripts have mistakes, so I'd strongly suggest that if there is a decision to save transcripts in addition to allowing the scribe to use them in real time, that the audio be saved as well.)

swickr commented 1 year ago

I do think the 2-year-old policy can be revisited. The automated speech-to-text tooling has gotten better and people appear to be using the machine transcription more and more. I do think that (a) permission from all meeting participants must be explicitly obtained and (b) participants should be allowed an opportunity to correct the transcript before it is shared outside of the participants (e.g. to correct horrendous transcription errors and remove any "off the record" comments).

KimPatch commented 1 year ago

Agree that policy should be revisited. Clarifying that my suggestions –

  1. having the ability to cut-and-paste from a transcript be a tool for the scribe and
  2. making sure that if transcripts are saved, the audio is saved as well – were very specific to avoid a key pitfall.

I think it's important to be able to use this tool without leaving people with a bad choice: having to spend spend time correcting a transcript or knowing there's transcript with errors and no way to tell what is an error.

The first choice is uncomfortable and the second takes time even before you get to the logistics of do you allow anyone to correct a transcription error and also how easy it is to introduce errors in correcting…

Keep in mind that "can" and "can't" sound an awful lot alike and are regularly mixed up by automatic transcription – that's the kind of thing a scribe would be likely to catch in real time, but gets confusing later without the audio to go back to.

Again, this is a really good tool that we should use, but how it's used makes all the difference.