MTPE
Machine Translation Post-Editing: the process of having a human translator review and correct machine-translated text to a defined quality standard, rather than translating from scratch.
The honest version
MTPE is not proofreading. The post-editor is responsible for the output (legally and professionally) regardless of what the machine produced. The distinction matters because MTPE is often sold as a cost-reduction mechanism without being honest about where the responsibility lands.
There are two distinct tasks that share the name. Light post-editing fixes only critical errors: mistranslations, omissions, terminology violations. The output may still read awkwardly; that is acceptable. Full post-editing brings the output to the same publishable standard as a human translation. The effort, rates, and what “done” means differ substantially between the two. Conflating them is one of the most reliable ways to produce bad contracts.
The practical question is always: is the raw MT output good enough that editing it is faster than translating from scratch? For good domain match and well-resourced language pairs, often yes. For poor domain match, rare languages, or high-entropy source text, often no.
Why it matters for translation
MTPE is the dominant workflow in professional localization for a reason: volume. When a product ships updates weekly across 30 locales, pure human translation cannot keep pace. MT handles the throughput; post-editing maintains accountability.
The economics only work if the MT output is genuinely useful. A well-configured pipeline (with domain-matched MT, termbase injection, and TM pre-translation) can produce output that a skilled post-editor improves significantly faster than translating from scratch. A generic MT API with no domain configuration often produces output that is harder to fix than to replace.
The choice between LPE and FPE is a content decision, not a cost decision. Internal tools, low-visibility documentation, and draft content are candidates for LPE. Customer-facing copy, legal agreements, regulated content, and anything that will be read by humans who form opinions about your brand are candidates for FPE or pure human translation.
Where it fails
MTPE fails when the MT output quality is lower than the post-editor’s rate assumes.
This is the dirtiest secret in LSP pricing. The industry set MTPE rates based on the assumption that MT output is good: 60–70% usable, requiring light intervention. When the domain is poorly matched or the language pair is under-resourced, the actual edit distance is 80–90%, meaning the post-editor is essentially re-translating but being paid at a rate calculated for light editing. The translator absorbs the quality risk silently, or delivers worse output than full translation would have produced, or both.
The second failure mode is feedback invisibility. In a pure MTPE workflow, the MT system does not learn from corrections. Errors that a post-editor fixes on Tuesday appear again on Wednesday. Without a feedback loop (reviewed segments returning to TM, terminology violations flagged to the termbase), you are running an assembly line, not building a system.
Finally: not all languages benefit equally. High-resource languages (English, French, German, Spanish, Japanese, Chinese) have strong MT coverage. Low-resource languages (many African and Southeast Asian languages, less-studied European languages) have substantially worse MT, which shifts the MTPE calculus toward full human translation for quality-critical content.