Streamlining Localization with Weblate's Auto‑Translation: Enable, Run, and Review

Problem: Manual Translation Bottlenecks

In many projects, translators must review hundreds or thousands of strings each release. This manual effort slows release cycles and inflates costs. Weblate offers a built‑in machine‑translation engine that can automatically propagate translations across all components of a project, but it must be enabled and monitored carefully.

Thesis: Auto‑Translation Can Accelerate Localization If Configured Correctly

When the "Translate all strings using machine translation" option is turned on, Weblate pulls suggestions from the configured MT service (Google Translate, DeepL, etc.). The feature respects existing translations, only fills untranslated strings, and can be scheduled or triggered via CI. However, automatic output is never perfect; a post‑edit review is essential.

1. Understanding the Feature

• Scope: Applies to all strings in a component that are still marked as untranslated.
• Overwrite flag: If set, existing translations are replaced by MT output; otherwise, MT suggestions appear as new translations awaiting review.
• Permissions: Only project administrators can enable the feature and schedule jobs.
• Performance: Small‑to‑medium projects see negligible latency; large repositories may experience noticeable batch processing time.

2. Configuring Auto‑Translation

1. Navigate to Projects → {project} → Components → {component} in the Weblate UI.
2. Under the "Translation" tab, locate the Auto‑translation section.
3. Check Translate all strings using machine translation.
   • Choose the MT service from the dropdown (e.g., DeepL).
   • Optionally enable Overwrite existing translations if you want MT to replace all current values.
4. Save changes.

At this point, Weblate will queue a one‑time job that runs automatically on the next background‑task cycle (usually every 5 minutes). You can trigger an immediate run via the API:

curl -X POST \
  -H "Authorization: Token {api_token}" \
  https://{weblate_host}/api/components/{component_id}/jobs/translate/

Replace {api_token} with a user token that has admin rights, {weblate_host} with your Weblate instance URL, and {component_id} with the numeric ID of the component (visible in the component URL).

3. Running and Monitoring Jobs

After enabling, the job appears under Manage → Background tasks. Each task shows:

• Start time and duration
• Number of strings processed
• Success / error status

To verify that the job ran successfully, look for a status of Completed and a count that matches the number of untranslated strings. If errors appear (e.g., MT service unavailable), check the service credentials and API limits.

4. Post‑Edit Review and Limitations

MT output is only a first draft. The safest workflow is:

1. Open the component’s Strings view.
2. Filter by State: Untranslated to see newly generated translations.
3. Use the review panel to edit, add notes, or mark as Approved.
4. Once all strings are approved, publish the component.

Trade‑offs:

• Accuracy: MT can misinterpret context or terminology; human review is mandatory.
• Terminology drift: Overwrite mode can erase custom terminology that translators had already added.
• Cost: Some MT services charge per character; monitor usage to avoid unexpected bills.
• Performance: Large components may queue many jobs, leading to longer wait times. Consider splitting components or running jobs during off‑peak hours.

Actionable Checklist

• Enable auto‑translation only in a staging environment first.
• Configure the MT service credentials in Settings → MT services.
• Schedule a test job and verify the task logs under Manage → Background tasks.
• Review at least 10% of the generated strings manually before approving the entire component.
• Set up a CI step that triggers the translation job after each merge to main (e.g., a GitHub Action that calls the API).
• Monitor MT usage via the API or the service’s dashboard to keep costs in check.

By following this workflow, teams can significantly reduce manual translation effort while maintaining quality control. The key is to treat auto‑translation as a first‑pass tool, not a final solution.