MT post-editing (MTPE)
A human checks machine-translated text against the source, fixes meaning and terminology, and brings it up to a defined quality level.
What it is
MT post-editing is human revision of machine-translated output against the source. Light post-editing aims for understandable, usable text; full post-editing aims for the quality expected from human translation. ISO 18587 describes requirements for full post-editing work and the people doing it.
Reach for MTPE when machine translation gives enough useful draft value to beat translating from scratch, and define the target quality before work begins. The editor checks meaning, omissions, terminology, tone, grammar, formatting, tags, numbers, and placeholders, then reviews the result in the product.
Gotcha: fluent output can hide a reversed condition, missing negation, or invented detail. A polished target sentence that says 'you can cancel anytime' where the source said 'you cannot cancel' reads perfectly and is a refund liability. Editors must compare with the source rather than merely making the target sound natural. Measure real editing effort by content type and locale instead of assuming machine translation always saves time.
Ask AI for it
Create an MT post-editing workflow that sends source segments to the DeepL API, stores the raw machine output separately, and assigns it to a human editor in the TMS with source text, translation context, and the terminology database visible. Require checks for omissions, negation, numbers, names, ICU MessageFormat placeholders, tags, tone, and approved terms. Mark each job as light or full post-editing before assignment, record edit distance and review time, and publish only after localization QA approval.