Transitioning from translation batches to translation request regular cadences

Context

In preparation for launching the JA site we needed to translate all the content of the website. In order to achieve that we broke down the content into three phases, based on number of page views. We also implemented a new translation workflow with two retrieval steps: first retrieval after AI-translation is complete and second retrieval after human review is completed.

Current Status

  • Phase 1: all the content has been human reviewed
  • Phase 2 + 3: AI translations came back to GitLab. Human review is still pending.
Translation MRs Status
  • Phase 1 human review Translation MRs were merged and used in the staging environment when the internal localization team performed the QA
  • Phase 2 + 3 AI only Translation MRs had some pipeline and linting issues that required investigation and troubleshooting. After all of that is solved, they'll appear in the staging environment.
  • Phase 2 + 3 human review commits will start to flow through as soon as I re-open the translation requests. The Argo key will be the same as the previous Phase 2 + 3 AI only round but Argo will create new Translation MRs with the human commits because the AI only Translation MRs have been merged.

Transition Plan

Step Action Item Related Issue DRI Status Date
1 Deliver Phase 1 resend files gitlab-com/localization/tech-docs-forked-projects/prod/gitlab#1 (moved) Argos Multilingual in progress June 27-June 30
2 Reopen Phase 2 + 3 projects on Phrase #270 (closed) Argos Multilingual not started July 1st
3 Reopen Argo Phase 2 + 3 Argo requests gitlab-com/localization/tech-docs-forked-projects/prod/gitlab#1 (moved) @mjsibanez / Spartan team not started July 2nd
4 Merge of Phase 2 + 3 human reviewed files in Translation MRs Loc engineering team not started ongoing until all files are reviewed
5 1st weekly translation request with updated files to be discussed @mjsibanez not started July 15th
6 Merge of AI translations from 1st translation request with updated files to be discussed Loc Engineering team not started July 16th
7 2nd weekly translation request with updated files @mjsibanez not started July 22nd
8 Merge of AI translations from 2nd translation request with updated files Loc engineering team not started July 23rd

Translation Request Cadence Plan

Moving forward, we'll manage two parallel streams:

  1. Phase 2 + 3 Human Review: Completing human reviews of existing AI translations
  2. Weekly Translation Requests: New/updated content requiring fresh translation

File Filtering Logic

For weekly/biweekly requests, we'll use the Argo Asset Dashboard to identify "updated" files:

  • Files not currently in any ongoing translation request
  • Files modified since their last translation submission

Expected Pattern: Initially, few files will qualify as "updated" since most content is tied up in Phase 2 + 3 requests. As Phase 2 + 3 human reviews complete, more files will become available for regular translation cycles, gradually increasing the size of weekly/biweekly requests.

Translation MRs Merge

For Localization Engineers, this means that any future Translation MR, regardless of where it'll come from they should be ready to merge. I suggest to schedule merges on Wednesdays so that, in the future, we're always as close as possible to production for GitLab's Releases, which happen the third Thursday of the month.

TW Monthly Maintenance tasks

The TW team follows a monthly maintenance task where they introduce content in the source files. These changes will be trickled down to the translated files through the regular translation request cadence workflow. We have yet to oil the regular translation request cadence and the Translation MR review + merge so a hybrid approach should be used. See thread here.

Action Items

  • Argos Multilingual to deliver Phase 1 resend files
  • Argos Multilingual to keep on working on AI-only translation improvements and investigation of previous errors on pipelines and linting
  • Define weekly or biweekly cadence to start with.
  • [ ]
Edited by Maria Jose Salmeron Ibáñez