deep translate
Talk to us

QA agent

Every segment checked before an editor sees it.

The DeepTranslate QA agent is a language-specific checker that scores every translated segment against the source. Weak segments go back through the engine automatically, and only what still needs a human reaches your editors.

Last updated 19 October 2026

How does the QA agent work?

  1. 01

    Translate

    The engine produces a first version, with your glossary applied.

  2. 02

    Score

    The QA agent rates each segment against the source.

  3. 03

    Re-run

    Segments below the threshold go back through the engine.

  4. 04

    Flag

    What still needs a human goes to your editors.

  5. 05

    Learn

    Editor fixes improve the engine and the glossary.

What does the QA agent check?

Meaning kept

Catches output that drifts away from the source.

Names and numbers

Flags changed figures, dates and names.

Language leaks

Stops the source language slipping into the target.

House terms

Confirms your glossary terms are used.

Why do fluent translations still fail?

Generic models can sound fluent and still be wrong. Research on low-resource pairs found that about 1 in 10 machine translations detach from the source entirely (Guerreiro et al., TACL 2023).

A fluent wrong subtitle is worse than a slow right one. The QA agent is there so you never have to choose.

Frequently asked questions

What is a translation QA agent?

A QA agent is a language-specific checker that scores each translated segment against the source, sends weak segments back through the engine, and flags what still needs a human.

Does the QA agent replace editors?

No. It filters, so editors spend their time on the segments that need them.