Skip to main content

Translation memory and termbase

Two levels: project assets and libraries​

  • Project TM / project glossary — entries scoped to a single project, built up as segments are approved.
  • Libraries — standalone Translation Memories and Term Bases (typically per client or domain) managed from the TM and TB pages in the sidebar, attachable to any number of projects.

When a library is attached to a project, its entries participate in translation: TM provides exact and fuzzy matches, TB enforces terminology in AI prompts. Note that Editor, Linguist, and QA roles have read-only TM/TB access — creating and editing libraries is for Admins and PMs.

Translation Memories​

Translation memory browser

Create Translation Memory with a name, description, language pair, and a Fuzzy Match Threshold (%) — the minimum similarity for fuzzy matches from this TM. You can import a TMX file at creation, or add entries later.

Each entry carries its source, target, quality score, and usage count, so you can see which entries actually earn their keep. Entries are edited directly in the browser — search, fix a target, save — alongside statistics, duplicate detection, and export (TMX or Excel).

Match performance​

Fuzzy matching runs in two stages: an inverted index retrieves candidates near-instantly, then fine-grained scoring ranks them. Lookups typically return in around 9 ms even on large TMs — matches appear as you move between segments, with no waiting.

Matching is script-aware: for Chinese, Japanese, Korean, and Southeast Asian languages, similarity is computed on characters rather than space-separated words, so fuzzy matches work properly where "words" aren't delimited.

Cross-language lookup​

TM and TB lookups can search across language pairs, not just the project's own: if you've translated a sentence into German, that entry can surface as reference context when translating the same source into French. Enable it per translation run (Include cross-language matches in the pre-translate dialog) and as an editor default (Show cross-language matches by default in QA Settings).

For organisations using Phrase alongside LanguageOps, a TM can be compared and synced with a Phrase TM — the diff view classifies entries as identical, changed, only in Phrase, or new to upload, so neither side silently diverges.

Term Bases​

Create Term Base with languages and a domain. Each term entry holds the source and target terms plus optional definition, context, domain, notes, and tags.

Terms are edited inline in the browser too — open an entry, adjust the target term or its metadata, and save. Two provenance markers help with curation:

  • Manual Entry vs Auto-Extracted — terms harvested automatically (e.g. during document-mode translation) are marked so you know what needs review.
  • Verified vs Uncertain reliability — only promote terms to Verified once a linguist has confirmed them.

Term bases can be Active or Archived, and import/export as TBX or Excel.

Alignment​

If you have existing translated document pairs but no TM, the Alignment tool pairs source and target documents sentence by sentence to bootstrap a TM from past work. Upload the file pair, get an estimate, and review the result — each pair gets a confidence score, and problem sentences are flagged so you can fix and re-run alignment. Past jobs are kept for reference.