ISSUE-144: VERY LOW INTEREST: the view of confidence in EMMA is incomplete and restrictive
VERY LOW INTEREST: the view of confidence in EMMA is incomplete and restrictive
- State:
- OPEN
- Product:
- EMMA 1.1
- Raised by:
- Michael Bodell
- Opened on:
- 2010-06-15
- Description:
- Somerset f2f http://lists.w3.org/Archives/Member/w3c-mmi-wg/2010Apr/0034.html, Michael's issue 7.For example, some engines create multiple confidences for each input (e.g. SAPI). Three issues: need multiple confidences, (and then what does the sort order mean?) need a way to express differences, and does the scale need to be 0.0-1.0?
- Related Actions Items:
- No related actions
- Related emails:
- [emma] minutes June 28, 2012 (from dahl@conversational-technologies.com on 2012-06-28)
- [emma] June 21, 2012 minutes (from dahl@conversational-technologies.com on 2012-06-21)
- [emma] minutes from 060211 call (from johnston@research.att.com on 2011-06-02)
- ISSUE-144: the view of confidence in EMMA is incomplete and restrictive [EMMA] (from sysbot+tracker@w3.org on 2010-06-15)
Related notes:
(from June 2011 f2f) We would like to find a way to allow sophisticated IM's to take multiple confidences into account as well as for simpler IM's to use a simpler idea of confidence. It seems clear that we can't enumerate all the possible types of confidences, because they might be proprietary or they might only apply to certain modalities, so we need some kind of extensibility mechanism. With multiple confidences, what's the order? The most likely result might not be the most confident result.
Deborah Dahl, 20 Jun 2011, 21:58:29Discussed during EMMA call June 14. These other concepts (e.g. some kind of "score" or "rank") are so engine-specific that it's hard to see how to standardize. Is there something in between an fully-standardized tag and "throw everything in "info"? Could have a vendor-specific info with your specific confidence information. Probably won't work on this.
Deborah Dahl, 14 Jun 2012, 14:30:56discussed during June 28 call, Michael will add clarification text, but we won't add anything normative.
Deborah Dahl, 28 Jun 2012, 14:15:25Display change log
