Page MenuHomePhabricator

Inconsistent link adaptation in nn-nb translations
Closed, ResolvedPublic

Description

Not all links are getting adapted correctly when translating from nn to nb, id to ms and ms to id. The problems that have been observed and replicated are:

  1. On adding translation into a section (mostly with the first section), the links are initially adapted in the target and then removed.
  2. After selecting a text in the target that previously displayed a link, the link card is not displayed (inconsistent; sometimes displayed)
  3. Clicking on the source link automatically adds the link in the target (not consistent)
  4. Clicking on the source link turns the source link red. No effect on the target side (not consistent)

Event Timeline

Arrbee raised the priority of this task from to High.
Arrbee updated the task description. (Show Details)
Arrbee updated the task description. (Show Details)
Arrbee set Security to None.
Arrbee added subscribers: Aklapper, Arrbee.

I think this may be related to annotation mapping rather than link adaptation. I tried translating something from EN to CA with link adaptation commented out and saw that several link tags were not properly mapped over to the translation html from the source html. They just disappeared. Because link adaptation works on the target html produced by the annotation mapping algorithm rather than on the source html, if the link tags are not mapped over to the target the links cannot be adapted.

I have also noticed that if you have a link with text such as "Han Fei" and this link is preceded in the same segment by the word "Han" (no link) the link tag is mapped so that it goes around the preceding "Han" rather than around "Han Fei."

In general, how well links (and I'm assuming other tags) are mapped seems to vary significantly by language pair. For ES to CA it is pretty good, but for other languages there are a few issues.

As for the other link adaptation issues mentioned here, these are being addressed as part of the red links patch I am working on. Once that is merged we should have a more consistent link adaptation experience.

For ES to CA, here is a sample:

Special:ContentTranslation?page=Mickey+Mouse&from=es&to=ca&targettitle=Mickey+Mouse+CA

First paragraph: The link for wiki/Oswald,_el_conejo_afortunado is not copied over to the translation column.

Another. EN -> ES

Special:ContentTranslation?page=Lemming&from=en&to=es&targettitle=Lemming+ES

Arvicolinae is not adapted even though article exists.

We should retest this after the red links patch is merged.

https://gerrit.wikimedia.org/r/181374

The error for nn -> nb was caused due to the use of the language code 'nb' instead of 'no'. This redirect used on the Norwegian (Bokmål) wiki was creating link redirects which were incorrect. This has been fixed and 'no' will be used as the language code for the target wiki in Content Translation.

The issues for id-ms have been moved to T85919.

Arrbee renamed this task from Inconsistent link adaptation in nn-nb and id<->ms translations to Inconsistent link adaptation in nn-nb translations.Jan 6 2015, 2:55 PM