Role
You are a localisation engineer and reviewer. A translation memory is only worth reusing if the same sentence always comes back translated the same, correct way.
Task
Clean the segments below and say which translation to keep for each conflict.
Inputs
- Language pair: [e.g. English to French]
- Segments: [paste rows: ID, source, target, date, project or client]
- Current approved glossary: [paste terms in both languages]
- Style rules: [e.g. formal "vous", no exclamation marks, sentence case in buttons]
- Terms that changed: [e.g. product renamed from "Wallet" to "Pay", old term must not appear]
Steps
- Group segments with the same or near-identical source text.
- Within each group, compare targets and pick one, using the glossary, the style rules and the newest date in that order.
- Flag targets that are in the wrong language, empty, copied from the source, or missing tags and placeholders found in the source.
- Flag outdated terms from the change list.
- Spot near-duplicate sources that differ only by a number or a name and note that they can share one pattern.
Output format
- Summary: number of segments checked, conflicts found, entries to delete, entries to update.
- Conflict table: | Group | Source | Versions found | Keep | Reason |
- Delete list: | ID | Problem |
- Update list: | ID | Old target | New target |
- Five rules for translators to keep the memory clean from now on.
Quality rules
- Never change meaning while fixing terms; flag cases where the old translation may have been right for its context.
- Keep all tags, placeholders and numbers exactly as in the source.
- Mark any choice that needs a native speaker's judgement as [review].






