Table of Contents
Machine Translation Post-Editing (MTPE): A Complete Guide
Machine Translation Post-Editing (MTPE) is a translation workflow where a human linguist reviews and corrects machine-translated output. Rather than translating from scratch, the linguist starts with a machine-generated draft and edits it to meet the required quality standard.
This approach has become a standard practice in the localization industry as neural machine translation (NMT) engines have improved significantly. MTPE allows teams to ship translations faster while maintaining quality — but only when applied to the right content types with the right process.
What Is MTPE?
MTPE stands for Machine Translation Post-Editing. The process works in three stages:
- Machine translation — Source text is translated automatically by an MT engine (Google Translate, DeepL, Amazon Translate, or similar).
- Human post-editing — A professional linguist reviews the MT output, correcting errors in grammar, terminology, meaning, and style.
- Quality assurance — The edited translation undergoes final checks for consistency, formatting, and completeness.
The key distinction from traditional human translation is that the linguist works with a machine-generated draft rather than starting from a blank page.
Light Post-Editing vs. Full Post-Editing
The localization industry recognizes two levels of MTPE, each suited to different content types and quality requirements.
Light Post-Editing (LPE)
Light post-editing focuses on making the translation understandable and accurate without polishing it to publication quality. The linguist corrects:
- Factual errors and mistranslations
- Offensive or inappropriate content
- Critical grammar errors that impede comprehension
Light post-editing does not typically address:
- Stylistic preferences
- Minor grammar issues that don't affect meaning
- Terminology consistency beyond critical terms
Best for: Internal documentation, knowledge base articles, support tickets, user-generated content, and any content where speed matters more than polish.
Full Post-Editing (FPE)
Full post-editing produces output that is indistinguishable from human translation. The linguist addresses:
- All grammar and syntax errors
- Terminology consistency with glossaries and style guides
- Natural-sounding phrasing in the target language
- Cultural adaptation and locale-specific conventions
- Formatting and punctuation
Best for: Marketing copy, product UI strings, legal documents, published content, and any customer-facing material where brand quality matters.
ISO 18587: The Industry Standard
ISO 18587:2017 is the international standard for post-editing of machine translation output. Published by the International Organization for Standardization, it defines:
- Competencies required for post-editors (bilingual proficiency, MT literacy, domain knowledge)
- Process requirements for both light and full post-editing
- Quality expectations for each post-editing level
Key requirements from the standard include:
- Post-editors must be proficient in both source and target languages
- Post-editors should understand how MT engines work and their common error patterns
- The client must specify the post-editing level (light or full) before work begins
- Pre-editing of source content can improve MT output quality
When to Use MTPE
MTPE works well when:
- Volume is high — Thousands of words need translation within tight deadlines
- Content is repetitive — Technical documentation, product descriptions, or support articles with predictable patterns
- Language pairs are well-supported — MT engines perform best on high-resource language pairs (e.g., English to Spanish, French, German)
- Terminology is established — A glossary exists and can be applied to MT output during post-editing
MTPE may not be the best choice when:
- Creative content is involved — Marketing slogans, brand messaging, and transcreation require human creativity from the start
- Language pairs have limited MT support — Low-resource languages may produce MT output that requires more editing than translating from scratch
- Legal liability is high — Contracts, regulatory filings, and medical content may require certified human translation
MTPE in Modern Localization Platforms
Translation management systems (TMS) increasingly integrate MT engines directly into their workflows. This allows linguists to:
- Receive MT suggestions alongside translation memory (TM) matches
- Apply MT only when no high-quality TM match exists
- Track post-editing distance (how much the linguist changed the MT output) to measure MT quality over time
Post-editing distance — measured as edit distance between MT output and the final translation — provides valuable data for improving MT engines and calibrating expectations.
Frequently Asked Questions
What is the difference between MTPE and human translation?
In human translation, the linguist translates from scratch. In MTPE, the linguist starts with machine-translated output and edits it. The final quality of full post-editing should be comparable to human translation, but the process is typically faster.
How much faster is MTPE compared to human translation?
Speed improvements depend on content type, language pair, and MT engine quality. Industry reports suggest full post-editing can be 30-60% faster than human translation for well-suited content types, while light post-editing can be even faster.
Does MTPE reduce translation costs?
Generally yes, because linguists can process more words per hour when post-editing compared to translating from scratch. However, the cost reduction depends on MT output quality — poor MT output can negate any savings.
What skills does a post-editor need?
According to ISO 18587, post-editors need bilingual proficiency, understanding of MT technology and its limitations, domain expertise, and the ability to edit efficiently without over-editing (especially for light post-editing).
