You get a 10,000-word technical manual. The client paid for machine translation plus post-editing - the budget is tight, the deadline is tomorrow. You open the file and realize: this is your first time doing MTPE seriously. Do you fix every slightly awkward phrase? Do you skip sentences that feel “good enough”? How fast should you be going? If you’ve been there, you know the specific paralysis of not having clear rules for the job. This guide is for translators building post-editing competency - and for trainers and agencies trying to do that systematically.
Why post-editing is now a baseline skill¶
According to the GTS 2025 MTPE Survey, 87.93% of professional translators now engage with post-editing work - 47.83% frequently, 40.10% occasionally. Only 12% have never touched MTPE. That’s not a niche anymore - that’s the profession.
The shift happened fast. Five years ago, MTPE was a specialized track mostly for localization insiders and tech companies with huge content volumes. Now it shows up in freelance briefs, agency RFPs, and university translation curricula. The translators who treated it as “something I’ll figure out when the first project comes” are now scrambling.
The economics are clear. MT output quality has gotten good enough that most clients default to it for standard content - user manuals, product descriptions, corporate communications. They need skilled post-editors, not from-scratch translators, for that volume. For translators, refusing to engage with this workflow means working with a shrinking slice of the market.
That said, 84% of translators in Acolad’s 2025 survey expect MTPE to grow while pure human translation demand decreases. The question isn’t whether to develop this skill. It’s how to develop it so you’re actually efficient at it, not just technically capable.
Light vs. full post-editing: not the same job¶
The first thing every new post-editor confuses is treating all MTPE projects the same. There are two distinct modes, and they require genuinely different approaches.
Light post-editing (also called fit-for-purpose post-editing) has one goal: make the MT output understandable and accurate, nothing more. Fix errors that affect meaning - wrong terminology, incorrect numbers, missing segments, obvious mistranslations. Do not touch style. Do not rewrite fluent sentences you’d phrase differently. Do not polish. If a human reader would understand the correct meaning without confusion, move on.
Full post-editing brings the text to publication quality - the output should be indistinguishable from a human translation in terms of fluency, style, and consistency. This means addressing awkward phrasing, unnatural register, inconsistent terminology, and stylistic issues on top of all accuracy errors. It takes significantly more time and should be priced accordingly.
The confusion between these two modes costs translators time and money. Applying full post-editing effort to a light post-editing project means you’re doing unpaid work. Delivering light post-editing quality on a full post-editing contract means unhappy clients and revision requests.
When you get a brief, confirm which mode the client wants. If it says “MTPE” without specification, ask. If they say “make it sound natural and fluent” - that’s full. If they say “just make sure it’s accurate for internal use” - that’s light. Both are legitimate; they’re just different jobs.
What ISO 18587 says a post-editor needs¶
ISO 18587:2017 is the international standard for post-editing of machine translation output. It’s the closest thing the industry has to a formal definition of what this job requires.
The standard identifies five core competencies:
1. Translation competence - the ability to translate between the working language pair. ISO 18587 requires the same baseline qualifications as ISO 17100 - a recognized translation degree or equivalent professional experience. Being a proficient bilingual is not the same as being a translator; post-editing requires proper translation training.
2. Linguistic competence - deep knowledge of both source and target language: grammar, idioms, register, syntax. This is what allows a post-editor to recognize when a sentence is technically correct but sounds wrong in the target language.
3. MT literacy - understanding how machine translation systems work, what types of errors they produce, and how to recognize them. This is the competency most translators lack and most training programs underemphasize. MT literacy is not about knowing the engineering - it’s about knowing the failure modes.
4. Domain expertise - knowledge of the subject matter being translated. A post-editor handling pharmaceutical documentation without life sciences background will miss terminology errors that look plausible but are medically wrong. Domain knowledge is what separates a fast, dangerous post-editor from a competent one.
5. Post-editing skills - the practical ability to work efficiently within MTPE constraints: applying the minimal-edit principle, maintaining consistent output quality under speed pressure, using CAT tools and quality assurance tools correctly.
As researchers Nitzke and Hansen-Schirra note in their analysis of post-editor competencies for Slator:
Post-editors should perceive themselves not only as mere proofreaders of MT output, but as competent language consultants.
That framing matters. The passive “fix what the machine got wrong” mindset produces slow, inconsistent work. The active “I am managing translation quality for this content” mindset produces efficient, professional output.
The 6 mistakes novice post-editors make¶
Understanding what goes wrong is as important as knowing what to do. These six patterns appear in almost every untrained post-editor’s work.
1. Over-editing¶
The perfectionist trap. You know how to translate well, so you rewrite everything that’s not exactly how you’d phrase it. A sentence is fluent, accurate, and contextually correct - but you’d use a different word order. You rewrite it anyway.
This is the most expensive mistake in MTPE. According to experienced post-editors, over-editing is the primary reason post-editors fail to hit speed benchmarks. The rule is simple: if it’s accurate and understandable, don’t touch it. Your personal stylistic preferences are not the quality benchmark.
2. Under-editing due to fatigue¶
The opposite problem, usually appearing later in the document. After 50 segments of careful work, the eye glazes over and starts approving things it shouldn’t. Subtle errors slip through: a negation dropped, a number transposed, a term inconsistently translated.
The fix: work in structured sessions with breaks, not marathon runs. If you’re post-editing for 4+ hours straight, error rates go up. Build your workflow around this.
3. Editing without the source¶
Reviewing only the MT output without checking against the source text at every step. This is what produces the most dangerous errors - fluent-sounding text that doesn’t match the original. Machine translation looks convincing. You need the source open and referenced.
4. Inconsistent terminology¶
Machine translation frequently generates multiple translations for the same term within a single document - “software” becomes both “program” and “application” on different pages, or a product name gets romanticized differently in different segments. A post-editor who doesn’t track terminology will produce an inconsistent final document.
Before starting any MTPE project, set up a termbase or glossary in your CAT tool. Review it and apply it consistently throughout. This is one of the most visible quality markers for professional clients.
5. Missing hallucinations and additions¶
Neural machine translation occasionally generates content that wasn’t in the source: an added clause, a plausible-sounding but incorrect figure, a name that got “completed” incorrectly. These don’t look like errors - they look like correct translation. The only way to catch them is systematic source comparison, not just output review.
6. Skipping the QA pass¶
Post-editing is not proofreading-as-you-go. A final pass through the whole document after all segments are processed is mandatory - both for consistency (does terminology hold throughout?) and for catching errors that became visible only after context accumulated. QA tools built into most CAT tools will flag numbers, omissions, and terminology inconsistencies automatically. Use them.
How to build post-editing competency: a practical curriculum¶
Training someone to post-edit effectively - whether it’s yourself or a team - follows a clear progression. Skipping stages produces post-editors who are technically capable but painfully slow or inconsistent.
Stage 1: Learn your MT engine’s failure modes (week 1-2)¶
Before anything else: understand what the MT system you’re working with gets wrong. Different engines produce different error patterns. Neural MT systems that handle context well still make systematic mistakes - specific error types appear consistently in specific content types.
The fastest way to build this knowledge: take 1,000 words of MT output in your domain, translate the same content yourself from scratch, then compare. Document every error the MT made: what type, in what context, how obvious or subtle. This error taxonomy becomes your detection framework.
Common MT error categories: - Literal translation of idioms (“it’s raining cats and dogs” rendered as actual cats and dogs) - Wrong register (formal source translated too casually or vice versa) - Incorrect disambiguation (the word “bank” translated as financial institution when it means riverbank) - Inconsistent handling of proper names and product terms - Dropped negations in complex clauses - Incorrectly transposed numbers or dates
Stage 2: Internalize the minimal-edit principle (week 2-3)¶
This is the core discipline of professional post-editing and the hardest mindset shift for translators trained to produce perfect text.
The minimal-edit principle: change only what needs to be changed to achieve the required quality level (light or full). Period.
Practical exercise: take MT output you’ve already fully post-edited and score it - mark every edit you made. Then categorize each edit as “required for accuracy,” “required for fluency” (full PE only), or “stylistic preference.” Every edit in the third category was unnecessary and cost time.
Do this exercise regularly until the stylistic preference edits disappear from your workflow. It typically takes 3-4 weeks of conscious practice.
Stage 3: Build speed deliberately (week 3-6)¶
Post-editing speed is measured in words per hour (WPH). Benchmarks vary by content type and quality level, but general targets:
| Content type | Light PE target | Full PE target |
|---|---|---|
| Technical documentation | 2,000-3,000 WPH | 1,200-2,000 WPH |
| General content | 2,500-4,000 WPH | 1,500-2,500 WPH |
| Marketing/creative | 1,500-2,500 WPH | 800-1,500 WPH |
For comparison: human translation from scratch typically averages 250-400 WPH, occasionally 600-700 WPH for experienced translators in familiar domains.
Don’t try to hit speed targets from day one. Work at a pace where you’re making correct decisions, track your WPH at the end of each session, and watch it increase naturally over weeks. Forcing speed before the error-detection instinct is solid produces low-quality output.
Stage 4: CAT tool integration (parallel to stages 1-3)¶
Effective post-editing without CAT tool proficiency is barely possible at professional volume. Trados, memoQ, Phrase (Memsource), and similar platforms present MT output segment by segment, track your edits, apply TM leverage where available, and run QA checks automatically.
If you don’t already have a CAT tool workflow: pick one platform and learn it properly before taking serious MTPE work. Most have free tiers or trial periods. The learning curve is 1-2 weeks to basic competency, 1-2 months to fluency.
Key CAT tool features specifically for MTPE: - MT integration (feeding segments to your MT engine directly from the platform) - Edit distance tracking (shows how much you changed each segment - useful for billing and quality review) - Terminology management and inline term verification - QA checks: numbers, omissions, tag consistency, terminology
Stage 5: QA and consistency review (ongoing)¶
No MTPE job is complete without a final QA pass. In practice this means: - Running the CAT tool’s built-in QA check - Reviewing terminology consistency across the whole document - Reading the full target text at normal reading speed - does it read naturally? - Spot-checking 10-15% of segments against the source for accuracy
For full post-editing, this final review is especially important because stylistic consistency only becomes visible at the document level.
Training programs that actually deliver¶
Several certified training options exist. Here’s an honest overview:
ProZ.com Certified MTPE Training - self-paced program focused on practical post-editing skills. Includes assessment and certification. Good for freelancers who need a recognized credential. Widely referenced by agency clients who want verified MTPE competency.
RWS/Trados Linguistic AI Certification - covers MTPE fundamentals plus industry-specific topics (gender bias in MT, quality estimation tools). Credential is recognized specifically in the Trados ecosystem and among LSPs who use SDL/RWS tools. Strong on practical workflow.
NCI Intro to Machine Translation Post-Editing - instructor-led course from the National Center for Interpretation. Language-neutral (works for any language pair). Features live practice sessions which are especially valuable for developing speed instinct.
Meridian Linguistics Machine Translation Post-Editing - designed for translators new to post-editing, progresses from basics to professional-level techniques. Certificate on completion.
GALA Academy - industry association (GALA’s Localization Academy) runs MTPE-focused modules for both linguists and project managers. More focused on workflow and project setup than on individual editing skills.
For any of these, completion alone isn’t the goal. The credential signals willingness to learn; the actual competency comes from practice volume after the course.
Running MTPE training inside an agency or team¶
If you’re training translators at an agency or managing a team that needs to move to post-editing workflow, formal courses are only part of the solution. The other part is structured internal practice.
Parallel translation exercise: give trainees the same source document, have them both translate from scratch and post-edit the MT output, then compare the results. The comparison reveals where MT performs well (saving time) and where it fails (requiring more effort than translating from scratch). This calibrates realistic expectations on both sides.
Error annotation sessions: trainees post-edit a document and then annotate each edit with the error type. Regular group review of annotated edits builds shared vocabulary and consistent quality standards. It’s also the fastest way to identify systematic weak spots in individual trainees.
Blind quality review: give completed post-editing work to a senior translator who reviews it without knowing which segments were edited and which were accepted as-is. This measures whether necessary edits are being caught and whether unnecessary edits are being made.
Speed benchmarking: track WPH per trainee per session. Share the benchmarks transparently. Most post-editors improve fastest when they can see their own progress over time.
As noted in the GTS 2025 survey, approximately 50% of professional translators resist offering MTPE discounts, arguing the effort required is often underestimated. Proper training is partly what closes this gap - a well-trained post-editor working efficiently genuinely does earn comparable hourly rates to full translation, while an untrained post-editor applying full translation effort to MTPE work gets paid less for the same hours.
“Post-editing can take as much time as traditional translation,” said approximately half of the respondents who resist discounting - GTS 2025 MTPE Survey.
The translators for whom this is true are almost always either: working with low-quality MT output where post-editing genuinely isn’t worth it, or post-editing with full translation mindset instead of MTPE-specific workflow. Both are training problems with training solutions.
FAQ¶
How long does it take to train a translator to post-edit effectively?¶
Foundational training - error taxonomy, workflow, tool setup - takes 2-4 weeks. Real proficiency, meaning fast error detection without over-editing, builds over 3-6 months of regular practice with actual MT output. The skill compounds quickly: the more domain-specific MT error patterns you recognize, the faster your detection instinct becomes.
Do post-editors earn less than translators per word?¶
MTPE rates are typically $0.05-$0.15 per word, compared to $0.15-$0.30 for full human translation. However, experienced post-editors process 2-3x more words per hour than translators working from scratch. The hourly income for a well-trained, fast post-editor is often comparable to full translation rates - sometimes better on high-volume projects where MT quality is consistent.
What qualifications do you need to be a post-editor under ISO 18587?¶
ISO 18587 requires the same base qualifications as ISO 17100 for translators: a recognized degree in translation, linguistics, or language studies, or equivalent professional experience. Additionally, demonstrated MT literacy and post-editing-specific competencies are required. A translation degree alone doesn’t qualify someone for post-editing - the MT literacy and domain knowledge components must be demonstrably present.
What’s the difference between light and full post-editing?¶
Light post-editing (fit-for-purpose) corrects errors that affect understanding and accuracy - wrong content, omissions, factual errors, formatting problems - while leaving stylistic imperfections. The goal is a text that communicates correctly, not one that sounds polished. Full post-editing brings the text to publication quality: fluency, natural register, consistent style, and terminology consistency on top of all accuracy corrections.
Which MTPE training program is most recognized by agencies?¶
RWS/Trados Linguistic AI Certification and ProZ.com’s Certified MTPE program are the most widely cited by agency clients vetting post-editors. For clients who care about standards compliance, ISO 18587 alignment is the actual benchmark - certifications that explicitly reference the standard carry the most weight. NCI certification is particularly valued in the US market.
Can translators without CAT tool experience do post-editing?¶
Not efficiently at professional volume. Post-editing almost always runs inside a CAT tool (Trados, memoQ, Phrase) where MT output is presented segment by segment, TM leverage is applied automatically, and QA checks run on the whole document at the end. CAT tool proficiency is effectively a prerequisite. If someone needs to build both skills simultaneously, prioritize CAT fundamentals in the first two weeks - the post-editing techniques layer on top more naturally.
What’s the biggest mistake agencies make when setting up MTPE workflows?¶
Assuming translators can post-edit effectively without training. The research consistently shows that translation competency doesn’t automatically transfer to post-editing efficiency. Untrained translators applied to MTPE projects either over-edit (applying full translation effort, defeating the time and cost savings) or under-edit (passing quality problems through). Training that specifically addresses the minimal-edit principle and MT error recognition is what makes MTPE workflows actually deliver their promised efficiency.
Sources¶
- The State of Machine Translation Post-Editing (MTPE) in 2025 - GTS Blog - comprehensive 2025 survey of translator attitudes and MTPE adoption rates
- ISO 18587: Professional Post-Editing of Machine Translation - the international standard defining post-editor competencies and process requirements
- Post-Editors, MT Engineers, PE Consultants - Here Are the Skills You Need - Slator - competency analysis based on Nitzke and Hansen-Schirra research
- MTPE Rates 2025: Cost-Effective Translation for High Volumes - Artlangs - 2025 MTPE pricing data and benchmarks
- AI in Translation: Key Findings from Acolad’s 2025 Translators Survey - 2025 survey on translator expectations around AI and MTPE demand
- Common Mistakes in Machine Translation Post-Editing - Quicksilver Translate - analysis of frequent post-editor errors and how to prevent them
- RWS Linguistic AI Certification Training - Trados - industry certification for post-editing professionals
- ProZ.com MTPE Training Topic - training resources and certification on the largest professional translator platform
- NCI Intro to Machine Translation Post-Editing - instructor-led MTPE training from the National Center for Interpretation