Re: [ISSUE-132] - New edits for LQI

Yes, that example from Phil shows the sort of thing that might get flagged as somehow "weird" when compared to a corpus. Unfortunately the example has to be a little weird to make sense. It is more likely that you'd use this when a system finds something where some terms and phrases don't match what is expected. It took me a while to grasp this one, but it's basically the same sort of thing that the University of Sheffield is doing with QuEST to automatically estimate quality of MT output without having a reference translation: it takes a database of known good translations and then looks, statistically, at a bunch of measures of similarity between the translation and the training data to see if the output looks like it "belongs" in that training data. If it doesn't fit for some reason, it gets a score that indicates low estimated quality.

-Arle


On 2013 Aug 29, at 10:23 , "Pablo Nieto Caride" <pablo.nieto@linguaserve.com> wrote:

> Hi Arle,
> 
> After a quick look everything looks good and thorough to me, anyway I'll
> review it again more carefully in case I missed something important.
> 
> However, in the first document (LQI Types), the example of the value
> non-conformance says "The harbour connected which to printer is busy or
> configared not properly.",  I imagine it's on purpose as explained in the
> note, isn't it?
> 
> Cheers,
> Pablo.
> 
> <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
> 
> Hi all,
> 
> Attached are the latest versions of the LQI sections, including a new,
> non-normative appendix. For the first two files, Word track changes indicate
> what has changed. For the third, all of the content is new except for the
> part with changes tracked, which was moved from the main description.
> 
> Please send any feedback by September 3, if possible, as that will give me
> two days to work it in before handing off for implementation in the spec.
> 
> -Arle
> 
>

Received on Thursday, 29 August 2013 08:27:38 UTC