Re: [ISSUE-1] ITS2.0-to-XLIFF1.2 mapping: updates to LQR, mtConfidence and text analysis

Hi Dave, all,

> 1)
> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#Text_Analysis
> note: recommends only in-line use using xlf:mrk with mtype="phrase"
> as I recall this is unchanged from Bled.

Yes, I think that was the conclusion.


> 2)
> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#Localization_Quality_Rating
> Where there is an open question on whether its appropriate to use 
> in-line - we didn't resolve this when we discussed in on the mlw-lt list 
> earlier in the year - see thread starting at:
> http://lists.w3.org/Archives/Public/public-multilingualweb-lt/2013May/0095.html

I don't have a strong opinion either way for inline LQR mapping.
But if we don't have inline mapping for LQR, the question is what do we do when we find one in the original document? We just ignore
it and don't map it?

Note that the specification simply says "...is used to express an overall measurement of the localization quality of a document or
an item in a document." And "item" could be any element, including a span-like one.


> 3)
> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#MT_Confidence
> This has been reworked to reflect the outcomes of the discussion in the
> first day at Bled, i.e. _not_ to overload existing XLIFF origin and 
> match-quality attributes as previously proposed.
> I also added some guideance on option for annotatorRef when multiple
> engines provide confidence scores.

While I was fixing examples there I was taken by some fever of simplification and edited most of that text. (Sorry Dave. But I can
put it back as needed). Actually the part about using annotatorsRef on different alt-trans was useful and I'll put it back.

But there was a part about what to do when copying an alt-trans translation into target:

"In addition, if the content of an alt-trans target element is copied verbatim to the target element of a trans-unit, i.e. no
post-editing is conducted on the MT translation, then the confidence value can be copied to a its:mtConfidence for the target
element in the trans-unit."

I think those type of processing expectation/requirements must not be there. The document provides a mapping it shouldn't tell
people how to perform operations.



> Also there was note indicating that we didn't think inline confidence
> scores were appropriate. Are we happy to rule this out - Declan, Ankit,
> Pedro? I've certainly seen some acadmic paper where differential word or
> phrase confidence scores within a single segement are displayed to the
> posteditor as a guide.

I think (especially based on my better understanding of what MT Confidence means now) it's unlikely to be used lower than the
segment level.

Cheers,
-yves

Received on Tuesday, 23 July 2013 06:54:21 UTC