- From: Yves Savourel <ysavourel@enlaso.com>
- Date: Fri, 16 Nov 2012 10:48:57 -0700
- To: <public-multilingualweb-lt-tests@w3.org>
Hi Pablo, > ...regarding LQI would be a valid use case that in a process > of machine translation plus human revision, the editor would be able > to add LQI markup to inform of machine translation errors? > I'm not so sure since this information it's likely not to get > to the MT System developers ever. I tend to agree with you. My guess is that, for a statistical systems, feeding back correction to the MT engine would likely be done through re-training using (for example) a TMX files of all the post-edited segments. Like you said, nothing would prevent to markup the document itself with LQI annotations, but a) there is not really any mechanism in LQI to give a correction (most of the data is describing the issue not providing a machine readable correction), and b) most workflows would probably just use some kind of TMX-based retraining to feedback the corrections to the SMT systems. Maybe things would be different for the Rule-ased MTs, but I doubt it. So making up the MT error is a valid use case for LQI, but it's unlikely that such markup be used by the TM system. It may be used by human though. cheers, -yves
Received on Friday, 16 November 2012 17:49:32 UTC