Page MenuHomePhabricator
Create Task
Maniphest T304459

Add GreynirT2T for Icelandic
Open, MediumPublicFeature
Actions

Description

Name: GreynirT2T
Translation Engine: Tensor2Tensor
Languages: English to Icelandic
Repo: github.com/mideind/GreynirT2T
Maintainer: Miðeind ehf. (Icelandic company)
Type: Neural Machine Translation
Licence: MIT (code), CC-BY-4.0 (corpus)
Requirements: ParIce (corpus) - parice.arnastofnun.is
Demo: Velthyding.is

Problem: Current accuracy causes some is.wikipedia editors not to use ContentTranslation.

Benefit: GreynirT2T claims to be more accurate than our current best translator used by WMF, GoogleTranslate. GreynirT2T is correct 71% of the time, while GoogleTranslate is 63% and Apertium is 50%.

Related Objects
Search...

Event Timeline

Snaevar created this task.Mar 22 2022, 8:56 PM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptMar 22 2022, 8:56 PM
Pginer-WMF triaged this task as Medium priority.Mar 28 2022, 11:06 AM
Pginer-WMF added a parent task: T86700: Add more machine translation services (tracking).
Pginer-WMF removed a project: Language-Team.
Pginer-WMF moved this task from Needs Triage to MT on the ContentTranslation board.
Snaevar added a comment.Apr 7 2022, 6:39 PM

sources for the claims in description: https://acl-bg.org/proceedings/2019/RANLP%202019/pdf/RANLP160.pdf (for GreinirT2T) and https://www.ru.is/faculty/hrafn/students/IndependentStudy_ApertiumIceNLP.pdf (for Aprertium and GoogleTranslate).

Asked for local conseus on //is.wikipedia.org/wiki/Wikipedia:Potturinn.

Pginer-WMF added a subscriber: Pginer-WMF.Apr 13 2022, 9:07 AM

Thanks for bringing this to our attention @Snaevar.
Flores, a new translation system supporting Icelandic, was added to Content Translation. Recently it has been enabled as the default for a month to get further feedback. It would be useful to have a sense on how the quality compares to the previous default (Google) as well as to other options you were sharing, from the perspective of a native speaker. So feel free to share any impressions. Thanks!

Content licensed under Creative Commons Attribution-ShareAlike 3.0 (CC-BY-SA) unless otherwise noted; code licensed under GNU General Public License (GPL) or other open source licenses. By using this site, you agree to the Terms of Use, Privacy Policy, and Code of Conduct. · Wikimedia Foundation · Privacy Policy · Code of Conduct · Terms of Use · Disclaimer · CC-BY-SA · GPL