Translations: indicate when they are out of date with the English version

[hcayless]

A sizable chunk of the text in the reference pages is translated to other languages (mostly French). It would be useful to indicate on the ref pages whether the English text has outdated the translation. It would be possible to determine this based on @versionDate on <desc> and <remarks>.

Moved from TEIC/TEI#1348

[martindholmes]

So, to clarify: every instance where the @versionDate on the English version postdates the @versionDate on another language version, there should be some signal?

That's going to be a scary number of instances. It had better be a rather unobtrusive signal. What do we suggest? If it's text, we'll have to go and get all the translations of that piece of text ahead of time, otherwise it's pointless. If it's an icon or a symbol, it will probably need descriptive text for mouseover, in which case ditto.

[lb42]

I had assumed that all that would be required would be to add a signal in the HTML (say, add class="outofdate" to the relevant block) so that it can be rendered in a different way, e.g. wavy type to indicate uncertainty or something. The stylesheet doesn't need to do more than compare the @versionDate on the chunk it's handling with that of its sibling with xml:lang="en" (assuming that the English one is always up to date, ha ha)

[martindholmes]

It's easy to do that, but unless you know what it means, it's not helpful. I don't know of any generally-accepted visual/design signal for "out-of-date translation", so somewhere, somehow, it has to be explained; and if you're in need of the translation for a gloss or desc, the chances are you'd be equally in need of a translated explanation for the signal.

[lb42]

So what is your suggestion? Insert the phrase "[traduction douteuse/dépassée/périmée]" vel sim? It seems to me that there are already quite a few visual conventions in the way the Glines are displayed on screen (e.g. use of red box for deprecated examples). Or is the suggestion that out of date translations should be silently suppressed/replaced by English or what?

[martindholmes]

I don't have a suggestion; I'm just looking for guidance. The ticket was assigned to me, but I don't know exactly what to do to handle it.

[martindholmes]

Quick note to say that this is problematic with <remarks> because there may be multiple remarks in various languages and it's not clear which remark in (say) Japanese should correspond with which remark in English. So I'm working on <gloss> and <desc> at the moment.

[martindholmes]

First stage is done in commit 863fed8: the odd2lite.xsl process now adds @rend='outOfDateTranslation' to <seg> elements generated from <gloss> and <desc> elements which are dated before the English version. The next stage is to process that @rend attribute into a class in the HTML generation. (Do we care about PDF?)

[martindholmes]

Oops -- I just realized that I'm not supposed to do this ticket, it was just assigned for the Stylesheets coop group meeting. Oh well, I've done a bit of it. Anyone who runs this:

java -jar Utilities/lib/saxon9he.jar -xsl:`pwd`/../../Stylesheets/odds/odd2lite.xsl -s:p5.xml -o:test.xml documentationLanguage=ja lang=ja language=ja doclang=ja

in the P5 directory will get a test.xml file of TEI Lite in Japanese with the seg[@rend='outOfDateTranslation'] elements in it (assuming the path to your local Stylesheets dev checkout is the same as mine, relative to P5). Anyone else in the group want to have a go at figuring out why that doesn't get turned into a hi element with a class in the HTML guidelines output?

Incidentally, I cannot for the life of me tell the difference between the parameters documentationLanguage, lang, language, and doclang. I just kept adding new ones until I got the output I wanted.

[martindholmes]

Hmmph. Now I have to fix the Stylesheets tests. :-(

[martinascholger]

Sorry that I'm just joining the discussion now. I try to summarize for myself:
we need some kind of indicator for outdated translations, which means a) text rendered in some way or b) an icon. Both has to be explained. I personally tend towards an icon before or after the affected text with a tooltip saying "translation out of date". I think it rarely affects <gloss>?, it more frequently affects <desc> and <remarks> (if we find a solution for the allocation problem).
Should we use @corresp for this purpose in the future? I'm not suggesting to add this to every single spec now, but we could make those changes whenever we are working on a specific spec anyway.
And, we must not forget attributes and their values ...
I don't know either if there is a generally-accepted icon for out-of-date-translation. Maybe an icon like this: https://cdn2.iconfinder.com/data/icons/social-productivity-line-art-1/128/history-512.png

[jamescummings]

I like b). Just an icon or asterisk or something afterwards with a hover/title tooltip saying "Translation out of date".

In discussion I suggested that whatever system there is should feed into a method to submit an up-to-date translation. (i.e. goes to an email that is monitored by IFTTT to stick it into a google spreadsheet or some other system). All thought a good idea but that it is a separate ticket... once this one is done.

[martindholmes]

Stylesheets coop group likes the idea of an icon with a tooltip in the HTML output. We need to get translations in all the languages of the explanatory text, and then figure out how to put those into the i18n system, then make the icon and tooltip appear in the text.

[sydb]

Of the 518 constructs with child <remarks>, only 6 have different counts of individual languages:

construct	en	fr	ja	es	de
@degree of att.damaged	1	1	2	1	0
att.editLike	2	1	1	1	0
@unit of att.milestoneUnit	1	2	1	0	0
<expan>	2	1	1	0	2
@type of <list>	2	1	1	0	2
<locus>	2	1	0	0	0

[martindholmes]

Thanks Syd. I can confirm that my code also handles gloss and desc in valItems.

[martindholmes]

Thanks @sydb. In the cases where English has more than the others, I guess there are new ones waiting for translation. In the two cases where Japanese has an extra one, or French has an extra one, we should probably take a look and see what's happening there.

[martinascholger]

@martindholmes I wanted to run this test, but can't find text.xml
java -jar Utilities/lib/saxon9he.jar -xsl:pwd/../../Stylesheets/odds/odd2lite.xsl -s:p5.xml -o:test.xml documentationLanguage=ja lang=ja language=ja doclang=ja
Or did you already fix the issue with hi ?

[martindholmes]

@martinascholger You'd be running this in order to create "test.xml". It assumes you have p5.xml already in your TEI/P5 folder, and that the Stylesheets repo is in the same folder as the TEI repo. What happens when you run it?

[martinascholger]

Do we agree to say "Translation out of date" in the tooltip? We could start to collect the translations then.
ko, zh-TW, ja, fr, es, it, de

[martindholmes]

That works for me.

[martinascholger]

Ok, we need translations of the explanatory text "Translation out of date" in the tooltip.
@lb42, @raffazizzi could you please help us with this and add translations?

language	info text
en	Translation out of date
de	Übersetzung veraltet
ko
zh-TW
ja
fr	Traduction périmée
es
it

[lb42]

Traduction périmée

[lb42]

Otoh the translation is not necessarily wrong, even if it's "out of date". Itmay just lack some nuance added in the English. I don't think anything is edited for the first time in any non English language is it,?

[lb42]

Sorry @martindholmes, but the evidence does not support your assertion. I ran the attached stylesheet langCounts.txtagainst the current version of p5subset to look at "out of date" desc elements in French (there are 351 of them) in comparison with the "correct" English version. I got bored eyeballing the results, but well over 70% of those I checked were perfectly OK. Try it for yourself.

[martindholmes]

@lb42 That suggests one of two things: either the dates on the English translations were changed without their contents being changed, or the French translations were updated without their version date being updated. I think the latter is more plausible.

Nevertheless, you can't know without checking, can you? So they all have to be looked at.

[lb42]

Speculation as to whether contents were changed without dates being changed could be resolved by attention to the version history for the file I suppose, but I don't think that's what's happening here. My speculation is that @versionDate changed when any change was made to the specification concerned, whether or not it invalidated existing descriptions.

[martindholmes]

If people are arbitrarily changing the @versionDate on the English <gloss> or <desc> when they didn't actually change the <gloss> or <desc> itself, then we have a problem. Why would anyone do that?

[lb42]

My suspicions focus on commit 087964ad7e7b3b369368e4974d7481d0f7ddbef5, which is the one that introduced @versionDate values for english language descs (previously missing). Who knows where the date that was introduced there came from?

[lb42]

FWIW, here are the counts for elements which have been translated in the current p5subset
I'll do another table for the allegedly out of date ones later. This shows that there's still some way to go on German, and that the French are excessively fond of glosses.

element	en	de	fr	ja	ko	es	it	zh-TW	nolang
DESC	1808	292	1471	1410	1374	1396	1417	1287	24
GLOSS	530	209	680	10	460	488	478	306	3
REMARKS	523	87	391	367	0	69	0	0	0

[martindholmes]

How does French have more glosses than English?

[lb42]

You may well ask....

[martindholmes]

Stylesheets group believes the manifestation of this should be an asterisk with a mouseover explanation. We need to strike a balance between worrying people and soliciting corrections. Council will try to decide on a method to allow people to provide a proposed correction. Should the English equivalent be available to the reader of the non-English version of the Guidelines somehow, so that they can more easily provide a translation?

[ebeshero]

See TEIC/TEI#1780 : which should be an agenda item for next Council meeting.

[martindholmes]

Counts of out-of-date items versus total items for each language:

GI	en	de	fr	ja	ko	es	it	zh-TW
gloss	0 / 530	6 / 209	121 / 680	0 / 10	17 / 460	104 / 488	63 / 478	101 / 306
desc	0 / 1800	22 / 292	350 / 1449	250 / 1389	268 / 1374	409 / 1396	652 / 1396	603 / 1266
remarks	2 / 523	1 / 87	147 / 391	128 / 367	0 / 0	26 / 69	0 / 0	0 / 0

(Syd and I generated this.)

[martinascholger]

@sydb and @martindholmes: since you already have a script, could you please let me know which German gloss/desc/remarks are out-of-date? Since there aren't that many, I'd update them.

[sydb]

OK. (@martindholmes, if you want to send me a copy of the script, I can modify for German-only and say which ones; @martinascholger: I will try to do that tomorrow night your time)

[lb42]

To answer Martin's question of a while ago (why does French have more glosses than English?) -- gloss is only (should only be) provided if the meaning of an element's GI is not apparent from its form, i.e. we don't explain that the tag "editor" is so called because it contains an editor, though we do explain that the tag "fs" contains a feature structure. The French translators, being French and logical, reasoned that therefore they should provide a literal translation of the GI for each element, and tag it as gloss, since (e.g.) "editor" is not a French word (well it is nearly but it means something different).

[martindholmes]

Here it is (quick and dirty as usual).
outOfDate.xsl.zip

[sydb]

Just sent the results to @martinascholger, a mere 24 hours late.

[martinascholger]

Thanks for the list @sydb. I had a look at the translations today. Some of the German glosses aren't out of date, although they have an older date than the English gloss. This is because the spelling has been changed to lower-case in some of the English glosses, which doesn't make a difference for the German translations. How should we deal with that? For practical reasons it would make sense to change the date for the German translations, but that would distort the history.

[martinascholger]

Updated German translations with TEIC/TEI@1e148284d78b6c52a526f3642769d3b36cf6bc02U

[hcayless]

I contend this is done.