Leveraging Weblate’s Automatic Translation Memory Suggestions Across Components

Why Automatic TM Suggestions Matter

When a project grows to dozens of components—think i18n files for a website, mobile app, and command‑line tool—repeating the same phrase can become a maintenance nightmare. Weblate’s translation memory (TM) stores every approved translation and can surface it as a suggestion when a translator edits a string. By turning this feature on per component, teams can cut repetitive work by up to 30‑40% while keeping consistency.

Configuring the Feature

1. Enable Automatic Suggestions in the UI

1. Navigate to Project > Components in the Weblate admin console.
2. Select the component you want to configure.
3. Open the Suggestions tab.
4. Check Automatic suggestions from translation memory and click Save.

When this option is active, Weblate will query the project‑wide TM for every string the translator edits.

2. Fine‑Tune the Matching Threshold

The TM lookup uses a fuzzy‑matching algorithm. By default, a match must be at least 70% similar. You can lower this to 60% for more aggressive suggestions, but be wary of noise.

# project.conf
[translations]
# Minimum similarity percentage for TM suggestions
suggestion_min_similarity = 60

Place this in the component’s project.conf file and reload Weblate.

How the Suggestion Engine Works

When a translator opens a string for editing, Weblate performs the following steps:

• Queries the TM database for entries with the same target language and a similarity score above the threshold.
• Sorts results by score and recency.
• Displays the top match inline in the editor, labeled Suggested.
• Allows the translator to accept, edit, or ignore the suggestion.

Locked or protected strings are exempt from suggestions to preserve workflow integrity.

Concrete Example

Assume you have a project with two components: website and cli. You want the CLI component to benefit from translations created in the website component.

1. Create a TM entry in the website component:
   • String: Hello, world!
   • Translation (Spanish): ¡Hola, mundo!
   • Mark as approved.
2. Enable suggestions in the CLI component as described above.
3. Open the CLI component’s translation editor and add a new string: Hello, world!.
4. Weblate now shows ¡Hola, mundo! as a suggestion. The translator can accept it with a single click.

   Verification steps:

   1. Check the Suggestions tab of the CLI component to confirm the TM section lists the website entry.
   2. In the editor, confirm the suggestion appears in the Suggested section above the editor area.
   3. After acceptance, the translation is stored in the CLI component’s TM, making it available for future components.

   Trade‑offs and Limitations

   • Context mismatch: A phrase used in a web UI may have a different nuance than in a CLI. Automatic suggestions do not know the component’s domain, so translators should still review the suggestion.
   • Noise from low‑confidence matches: Lowering the similarity threshold increases suggestion volume but also the risk of incorrect translations. Monitor the acceptance rate and adjust accordingly.
   • Cross‑language leakage: If a component is available in multiple languages, make sure the TM is scoped per target language; otherwise, suggestions from unrelated locales can surface.

   Best Practices for Production Use

   • Enable suggestions only on components that share a common terminology base.
   • Use the suggestion_min_similarity setting to tune aggressiveness per component.
   • Regularly audit the TM for outdated or conflicting entries.
   • Leverage Weblate’s TM export feature to back up the memory before major changes.

   Actionable Next Steps

   1. Pick a component that frequently repeats strings across the project.
   2. Enable automatic suggestions via the UI.
   3. Add a high‑confidence TM entry in a related component.
   4. Verify the suggestion appears and is accepted.
   5. Monitor the acceptance rate in the Suggestions report to fine‑tune the threshold.

   By following these steps, teams can harness Weblate’s TM suggestions to accelerate localisation while keeping quality control in place.