W3C

- DRAFT -

MLW-LT WG / ITS IG Call

24 Jul 2013

Agenda

See also: IRC log

Attendees

Present
felix, yves, des, df, jirka, phil, ankit, chriLi, pnietoca, pedro
Regrets
Arle, Declan, Dave, Karl
Chair
Yves
Scribe
fsasaki

Contents


agenda is here http://lists.w3.org/Archives/Public/public-multilingualweb-lt/2013Jul/0040.html

MT Confidence clarification (MLW-LT WG topic)

yves: there was the question wherer MT Confidence should be restricted to self generated score or not
... felix made change after the discussion

result would be here http://www.w3.org/International/multilingualweb/lt/drafts/its20/its20.html#mt-confidence-score-generation-tools

scribe: would you have any comments? Time to look at it? That is the current text

phil: didn't have time to look at the text, but at the thread
... I had in my mind something generated by the MT engine
... but I am happy enought to broaden this, as some people said in the thread
... as long as we are careful about the definition

yves: any other comments?

Yves_, comment from felix on IRC: the change I made was only removing the "self reported" part and adding above note

Yves_: MT Confidence still means the same, but it is not linked only to the MT engine

Yves reading the note in the link above

Yves: the note makes the distinction between the two types of generators for confidence information
... so it seems everybody is ok with the change

yves, where was no tracker issue for this, but I (felix) have added an item in the change log

XLIFF mapping

yves: several entries in XLIFF mapping

<Yves_> http://lists.w3.org/Archives/Public/public-i18n-its-ig/2013Jul/0057.html

yves: mail is about update of LQI and ...

dF: I had dropped of, what did you hear last? You didn't hear what I said about MT Confidence

yves: no - but go ahead

mt confidence again

dF: tools should rather use lq rating than mt confidence for information that is not self generated

phil: one thing that came up in relation to broadening mt confidence
... we don't have a standoff version of mt confidence
... not sure whether that would be an issue

dF: this category doesn't need standoff because it does not need mashing
... if you broaden the definition you open the use case to provide conflicting confidence scores
... we never intended to have standoff because we said this is self confidence
... one should use rating because, it has standoff

yves: it doesn't, only ITS lqi and provenacne have standoff
... what you said is the same for rating
... you can use different engines but there is only one way to mark it up, since there is no standoff for ITS rating

dF: don't you feel that third party conformance is for rating?

yves: I looked at the example of quest
... the value they are getting is the same like the mt confidence self reported
... to me lq rating is much more like a composite index based on many things
... that is much more linked to conformance and human voting

dF: we have one issue or the other
... usage of multiple scores either on confidence or rating

yves: don't think that this is a huge issue
... don't think that these data categories will be used a lot

phil: isn't this tied up with MQM / QTLaunchpad?

yves: LQI is, not sure about rating

phil: will QTLP have a rating?

yves: don't think so
... the question is: if you have a value like quest where should it go

phil: is it a problem to have either mt confidence or lqi?
... until we have lots of use cases
... or do we need a new quality type

yves: looking at declan's feedback he was ok with result of quest as an MT confidence value

dF: quest or other MT metrics are closer to MT confidence
... if you are an MT enginge owner it is breaking of transaction if you allow for overriding self reported confidence with a third party tool

yves: this is processing related

yves_, can I say something?

yves_: people will not read the spec and will just put an MT confidence score

phil: don't feel strongly enough to stand in the way of broadening this

<Yves_> felix: to dF, would current draft be ok?

<Yves_> .. and could you check with MT providers if it's ok this way

<Yves_> .. we need more feedback

<Yves_> .. I am asking you if we can move forward with current version

<Yves_> .. we could reverse the change if needed during PR - it is informative only

<Yves_> .. there is no testing impact

<Yves_> df: it's not substantive

<Yves_> .. so change would be ok

<Yves_> second quaestion: what's the current version

<Yves_> .. self-reported? or not?

<Yves_> felix: not self-reported

<Yves_> .. action for david would be to check whether allowing non self reported MT confidence should be reverted or not

<Yves_> felix: would current version be ok?

<Yves_> df: either is fine

<Yves_> felix: then let's use the current one

<Yves_> .. and get an action for david to gather the feedback from MT engine people

<scribe> ACTION: dfilip to get feedback on the MT Confidence broadening during PR - due 29 August [recorded in http://www.w3.org/2013/07/24-mlw-lt-minutes.html#action03]

<trackbot> Created ACTION-559 - get feedback on the MT Confidence broadening during PR [on David Filip - due 2013-08-29].

XLIFF mapping

<Yves_> http://lists.w3.org/Archives/Public/public-i18n-its-ig/2013Jul/0057.html

<Yves_> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#Text_Analysis

yves: first on text analysis

yves introducing on text analysis mapping

yves: any comments / objections against the mapping?

no comments

yves: now lq rating

<Yves_> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#Localization_Quality_Rating

yves: no resolution on that so far - also question on phil
... do we want to allow granularity for lq rating
... or have it only on the xliff target level
... the ITS 2.0 definition does not say "you should or should not go on the word level"

phil: discussed that last week

des: what would be the case against this?

dF: in source content, no a sub sentence level
... lq rating can appear. So we shouldn't prevent that in XLIFF
... we can discourage it, but it shoud be allowed as a mapping

yves: why would we even discourage it?

dF: it would compete with terminology information

yves: disagree - in term confidence is avail., it is very different from a vote

des: you could use it to mark up terminology

dF: you would not mark terminology but terminology errors
... that is different - you could still have terminology markup

yves: confidence in terminology does not say "it is a good translation"

dF: so we drop the language about local level

yves: agree - will make the change
... there is not a lot to change anyway
... now MT confidence again

<Yves_> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#MT_Confidence

yves: now the XLIFF mapping topic for MT confidence
... we had a discussion about origion XLIFF attribute
... we discussed that in bled
... we decided not to overload the entry
... this linked to the question we had before
... when multiple engines have a score, what to do
... there is no resolution, having standoff here would be a major change

christian: this would be something we would put on our lists of enhancement requests for ITS2?

yves: exactly
... how to put that on the list

felix: would put that on the ITS IG wiki for the time being

will have a pointer to the location soon

<Yves_> http://www.w3.org/International/its/wiki/XLIFF_1.2_Mapping#Locale_Filter

yves: now locale filter mapping

hi yves, all, I would propose to put the issues and proposed features here http://www.w3.org/International/its/wiki/IssuesAndProposedFeatures . Currently it is focused on ITS 1.0, but we can update it with proposals for ITS 2.0

yves describing the locale filter mapping

yves: nobody has made a comment no this one so far

dF: wonder - there might be more for XLIFF
... we should think of valid transformations
... what should be guidance to get from one to the other
... ITS information is not longer available
... the extractor with extended information - what are you doing

yves: you merge the target

david: it no longer contains the extended information

yves: why would you need the extended information?

david: you expect to merge back into multiple languages
... if the extractor does not define a target, the LSP still will need to define a target

yves: if they understand ITS - otherwise you cannot process the file

pedro joining the meeting

discussion about xliff - ITS extraction and merging

yves: think that has not to do with the mapping, it is an XLIFF problem

<Yves_> ACTION: davidF to draft guidance for second processing of Locale Filter when target is not defined in first processor in wiki [recorded in http://www.w3.org/2013/07/24-mlw-lt-minutes.html#action05]

Action is https://www.w3.org/International/its/ig/track/actions/1

yves: one question on termInfoRef is still open
... we both came to an agreement on that

<Yves_> http://lists.w3.org/Archives/Public/public-i18n-its-ig/2013Jul/0022.html

felix: I agreeded with the proposal to say that there is no hard wired definition what is available then one resolves termInfoRef

yves: correct.
... if you have disagreement comment on that asap
... otherwise no other mapping issues currently

aob

<pnietoca> thank you bye!

dF: next week we may be able to use this gotomeeting again

yves: thanks all, bye

adjourned

Summary of Action Items

[NEW] ACTION: davifF to draft guidance for second processing of Locale Filter when target is not defined in first processor in wiki [recorded in http://www.w3.org/2013/07/24-mlw-lt-minutes.html#action04]
[NEW] ACTION: dfilip to get feedback on the MT Confidence broadening during PR - due 29 August [recorded in http://www.w3.org/2013/07/24-mlw-lt-minutes.html#action03]
 
[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.138 (CVS log)
$Date: 2013-07-24 13:14:26 $