Translation setup

The translation setup step controls how a project is machine-translated: the MT engine, the translation memories and glossaries attached to each language pair, how TM matches are reused, and the project workflow. Everything here has a sensible default — you can skip the step entirely and adjust later. Its most useful feature: you can run machine translation up front and review the result in the editor before the project starts.

Choose the MT engine

Engine	What it does
Default	The recommended engine, with automatic fallback for reliability
On-Prem	The platform’s in-house engine only
Cloud	The cloud engine only
Google / DeepL	Route to that external engine
Copy Source	Fills targets with the source text unchanged
None	Skips machine translation entirely

See Machine translation & quality for how engine output is scored and routed.

Attach translation memories and glossaries

Each language pair has a summary line — for example “2 TMs · 1 glossary”, or “no TM/glossary attached” — with a TM-leverage preview of how many segments are exact matches, near matches, or new. Expand it to manage the pair:

Translation Memories — attach existing TMs, create a new one in place, or upload one. See Translation memories.
Glossaries — attach existing glossaries, create, or upload. See Glossaries.

Reuse TM matches instead of MT

Two controls per language pair decide when the memory beats the machine:

Reuse exact TM and repeated segments before MT (on by default) — exact matches from attached TMs and repeated source segments are filled from the memory and never sent to machine translation.
Skip MT at TM match ≥ — TM matches scoring at or above this percentage (100, 95, 90, 85, 80, or 75) are filled from the memory instead of machine translation. At 100% — the default — matches that differ only in case or spacing are still reused; lower the threshold to reuse fuzzier matches. The dropdown needs a TM attached to the pair.

Each target also has an Assignee AI budget — a token budget each assignee may spend on AI actions (style re-runs, AI suggestions) on that target. Leave it empty for the automatic default.

Run MT and review

Run MT & review fills every target right away — the project stays unactivated while you look. You see the estimated token usage and confirm before it runs; it uses the project’s MT quota.

Each target row shows its progress and percentage translated. When a target finishes, open it with Review to read — and fix — the results in the editor before deciding the workflow and assigning anyone.

Re-run MT replaces a target’s current fill, including any hand edits made during review, so re-run before people start working.
If new TM matches have appeared since a target was translated, the row says so — re-run MT to apply them.
With Copy Source or None as the engine, the button prepares the targets accordingly (source-filled or empty) instead of running MT.

Choose the workflow

The Workflow dropdown sets the stage sequence for the whole project — for example plain translation, or machine translation post-editing (MTPE). It can be changed freely until a seat is confirmed — so review the MT first, then decide. Per-target workflow overrides are made later, on the assignment step.