#561 @cert needs data.probability as well as data.certainty

[Martin Holmes]

This arose as part of our work on the standoff proposal, but has more general utility:

Certainty expressed by a human judge is typically imprecise, hence the current values of @cert; but annotation in particular may be done be tools such as natural language parsers, which express their "judgements" regarding (for instance) a word's being a member of a grammatical category in the form of probability (values between 0 and 1). To support this, the @cert attribute should have a datatype which is a choice between data.certainty and data.probability.

[Laurent Romary]

This would be an important support for the standoff work indeed. Is the council deciding on this on the fly during the F2F?

[Martin Holmes]

The FTF is over (although a few of us are still sitting in the hotel lobby doing tickets while we wait to leave). But I think this is a relatively uncontroversial proposal, it doesn't break backwards compatibility, and we can fast-track it to assist the standoff proposal, so I would expect it to go quickly.

[Lou Burnard]

I dont mind this change going forward, though it does reverse a decision taken some time ago by the Council.

[Syd Bauman]

I agree with Lou. However, I didn’t like the previous decision. :-)

[Hugh A. Cayless]

I think we were in generally in agreement with this. I think the original restriction was because we thought humans ought not to be quantifying their opinions of certainty, but if it's the output of an algorithm, I don't think there's any problem. We might want to add a note recommending when to use one or the other method. Assigning to Raff to implement.