<scribe> meeting: MLW-LT
rsagent, make log public
<scribe> scribe: Arle
<daveL> Hi jirka, we changed it last minute because enought WG topic have come up and we should prioritse minuting WG outcome
<omstefanov> Jirka: There was an email saying because of MLW-LT business that would be default, with IG page linking to it
<daveL> apologies: jorge, felix, yves,
Dave: We have a few MLW-LT working group topics.
... Felix proposed some minor changes.
Dave: No objections
Arle: Seem like reasonable clarifications.
Dave: Any objection to the proposal from Felix.
... There was no perfect solution, but we tried to work with the relevant people to address this in the future, rather than changing the spec now. JSON is one possibility.
... Some people have responded favorably on the email list.
Olaf: No win-win is possible. So we stick with what we
have and look for 2.x or 3 for better solutions.
... No one objected.
Dave: I've already set up in the IG tracker something for requirements not addressed in 2.0. This could go there.
<scribe> ACTION: dlewis6 to raise issue of addressing script topic for 2.x in IG.
<daveL> arle: agrees with yves that a confidence score could be generated separately from mt engine itself
<daveL> phil: how does confidence score relate to conformance score
<daveL> arle: yes there are relations
<daveL> olaf: but it seems unclear if we mean an advance score before fact or after fact
<daveL> ... and there is a need to have multiple instance of MT confidence
<daveL> arle: usually confidence score is returned after the target
Dave: I can ask internally about this for those working on these topics. Isn't QE a source measure looking at source and training data?
<daveL> arle: for quality estimation with lucia specia this is definitely an assessment of the translation
<daveL> dave: in this case would it be the same of LQI conformance type
<daveL> arle: but isn't this just a switch
<daveL> phil: but it does give a score
<daveL> ... and has the advantage that it provides a standoff for multiple conformance annotaitons
<omstefanov> we also should make sure we get Serge Gladkoff and Yves' comments on whatever final text we develop.
<daveL> ACTION: dlewis6 to consult with MT expert in CNGL for clarification
<daveL> ankit can you join the audio
<Ankit> just a sec,,
Christian: What is the goal of discussion? To address
ITS 2.0 or 2.x?
... We just need some clarity
Ankit: ITS could use MT confidence to address both traditional MT confidence and QE.
Dave: The specific issue was whether it is
self-reported (the spec talks about an MT engine self-reporting a score). If there is
no bitext, that is then a quality estimation. We could have multiple, so we can't
handle the issue of multiples in the current set-up.
... There is also the conformance item in LQI, which could be used.
<daveL> arle: using LQI does add some implementation weight
<daveL> arle: last week there was a call on how to address the MQM topic
<daveL> ... two separable issue to be addressed
<daveL> ... one is the mapping between MQM and LQI issue categories
<daveL> .. this is fairly well solved with an exisitng mapping, though with some info lost in both directions
<daveL> .. these can be dealt with in IG and documents
<daveL> ... second issue is how to maintain the MQM work beyond the lifespan of the supporting project (QT-launchpad)
<daveL> ... two possible homes are CRISP as GALA, or ITS IG
<daveL> .. concensus in call was initially to do it in CRISP, but subsequently, GALA expressed preference to run it under ITS IG, led by DFKI initially
<omstefanov> Arle: what is the deadline (end of life) of the EU project now developing MQM?
<daveL> Arle: there are two different tasks - maintaining MQM category list, the other is the way to represent quality metrics with a defined reference to the quality profile
<daveL> ... so the latter is a fairly big technical task
<daveL> Phil: asks if project is done under GALA, is GALA membership required?
<daveL> arle: no, keen to avoid and cost barrier and CRISP is happy to work this way
<daveL> ... then once there was consensus moving forward to a more formal standard, at W3C or elsewhere
<omstefanov> which wiki?
<daveL> arle: on the IG wiki
<daveL> phil: would be interest to be involved
<daveL> .. but membership fee an issue
<scribe> ACTION: Arle to add MQM products on the IG tracker
<Pedro> We are member of GALA, if helps.
<daveL> olaf: when does the current MQM proect end
<daveL> arle: june next year
<daveL> ... looking for further funding already
Dave: Yves isn't here so we don't have time to address the XLIFF/ITS mapping.
Dave: One small part for Phil.
... I was putting LQR (rating) into the new mapping page on the IG page. Would LQR ever apply inline or only structural?
Phil: It is more structural.
... But that led to confusion, but when we added voting, that adding confusion. When we discussed voting, it led to the requirement for inline.
Arle: We might want to note as best practice that you should use the biggest scope possible and keep it structural when possible.
<daveL> action; dlewis6 to add LQR inline for voting and best practice note of using widest scope for this anntioan where appropriate
<daveL> ACTION: dlewis6 to add LQR inline for voting and best practice note of using widest scope for this anntioan where appropriate
<daveL> Topic; IG wiki and action item tracker set
Dave: I'm leaving deliverables in the WG and moving community building, etc. to the IG.
<omstefanov> Dave: what would be good is to get an email (and post to BOTH wikis) is to have a table of which things relate to which in both groups, with links to both the WG and IG pages.
Dave: I have separate products in the tracking to deal
with XLIFF mapping, ontology, requirements, etc.
... If other ones come up, we can add new products to associate actions and issues with them. We will only bring up ones associated with a product.
... Any comments?
Arle: Good way to do it and the products will help keep us focused.
Dave: I'll talk to Felix about what to do with deliverables that could go in either the WG or IG.
<omstefanov> happy holidays, Dave.
<omstefanov> I'm off next wednesday, too.
<Pedro> happy holidays!