An agency sends you a 15,000-word job: “MT output ready, just post-edit, $0.04/word.” Is that a reasonable offer? Should you take it? And what does “post-edit” actually mean here - skim it for typos, or basically retranslate?
If you’ve ever stared at an MTPE offer trying to figure out what you’re actually signing up for, this article will clear it up. MTPE has its own terminology, a formal international standard, two distinct quality levels, and a whole set of industry debates that rarely get explained upfront.
What MTPE is, in plain terms¶
MTPE - Machine Translation Post-Editing - is a workflow where a machine translation engine produces a draft first, and a human translator then reviews and corrects it to the required quality level.
That’s the short version. The longer one: MTPE isn’t a single thing. It’s a spectrum ranging from “fix the obvious errors so it’s understandable” to “basically retranslate using the MT output as a rough scaffold.”
The human who does this work is called a post-editor. Their job isn’t to translate from scratch - it’s to evaluate the MT output and bring it up to the quality standard the project requires.
What does an MT engine actually produce? Industry estimates put raw MT accuracy at 70-85% - but this varies wildly by language pair, content type, and which engine you’re using. DeepL tends to score better on European language pairs. Specialized neural MT engines trained on domain-specific data outperform general-purpose ones on technical texts. The upshot: a post-editor is always needed, just more or less intensively depending on the situation.
The LLM vs. NMT comparison goes deeper on why different engines fail differently - which matters a lot for what a post-editor actually encounters.
Two levels: light and full post-editing¶
This is the most important distinction in MTPE, and also the one most commonly glossed over in agency job postings.
Light post-editing (LPE) - the goal is making the translation accurate and understandable, without polishing the style. You fix factual errors, mistranslations, and anything that would confuse the reader. You don’t rewrite awkward sentences if they’re technically correct. You don’t adjust the register or tone unless it’s actively wrong.
Light PE is appropriate for: - Internal documentation that staff will read once - Rapidly obsoleting content (release notes, changelogs, patch descriptions) - User-generated content moderated at scale - Gisting - content where understanding the meaning is enough
Full post-editing (FPE) - the goal is output indistinguishable from a human translation. You fix everything: accuracy, fluency, terminology consistency, register, naturalness. The final text should read as if no machine was ever involved.
Full PE is required for: - Client-facing and published content - Marketing materials and website copy - Legal and contractual documents - Medical documents and patient-facing materials - Any content where a translation error has serious consequences
The problem in practice: agencies often request “light post-editing” but send content that clearly requires full post-editing. This is one of the most persistent frustrations in the profession - which is exactly why spelling out the level in your contract matters before you start work.
As one translator noted on ProZ.com:
MTPE is sold as “just cleaning up,” but when the MT output has wrong terminology, mistranslated numbers, and paragraphs that don’t make sense, you end up rewriting from scratch anyway - except now you’re paid at half the rate.
This isn’t a minority view. In a 2025 survey of 212 freelance translators by GTS Translation, 66% rated MT output as “acceptable but requiring significant edits” - only 13% rated it as usually high quality.
ISO 18587: the standard that defines MTPE¶
In 2017, ISO published the first international standard specifically for machine translation post-editing: ISO 18587:2017 - Translation services - Post-editing of machine translation output - Requirements.
This is the document that gives the industry a shared definition of what MTPE means. It covers: - The two quality levels (light PE and full PE) and what each requires - The competencies a post-editor must have - What the post-editing process should include - How to assess the quality of post-edited output
What ISO 18587 says about post-editor competencies: active bilingual proficiency in both languages, translation competence, knowledge of MT technology and its typical error patterns, domain expertise, and - critically - the ability to edit efficiently without over-editing.
ISO 18587 makes it explicit that post-editing isn’t a junior skill. It requires a professional translator who also understands how MT systems work and fail - so they can identify systematic errors efficiently rather than treating every segment in isolation.
If an agency says their MTPE workflow is “ISO 18587-compliant,” that’s a meaningful statement about process and quality level - not just a marketing badge.
MTPE vs. human translation: the actual numbers¶
The argument you’ll hear from agencies: MTPE is “faster and cheaper with the same quality.” The argument from many translators: “it’s the same work for less money.” Both have truth in them, depending on the content.
Here’s what the data actually shows:
| Factor | Human translation | Light MTPE | Full MTPE |
|---|---|---|---|
| Speed | 250-400 words/hour | ~1,000 words/hour | ~700 words/hour |
| Daily output | 2,000-3,000 words | 6,000-8,000 words | 4,500-5,500 words |
| Cost per word | $0.09-$0.35 | $0.03-$0.08 | $0.08-$0.15 |
| Quality level | Depends on translator | Good for internal use | Comparable to human |
Productivity sources: Slator research on post-editing speed, Artlangs 2026 pricing guide.
The productivity numbers matter because they change the economics. A translator doing light MTPE at $0.05/word and 750 words/hour earns $37.50/hour. The same translator doing human translation at $0.15/word and 250 words/hour also earns $37.50/hour. The per-word rate looks lower, but the hourly rate can be identical.
The catch: this only works when MT quality is actually good enough to achieve that throughput. Less common language pairs, specialized content, and poor MT engines can reduce MTPE speeds close to manual translation - at which point the economics don’t work.
For a 50,000-word technical manual, the numbers look like this: - Light MTPE at $0.05/word: ~$2,500 - Full MTPE at $0.12/word: ~$6,000 - Human translation at $0.20/word: ~$10,000
That’s a real cost difference. For high-volume, structured content, MTPE’s case is straightforward. For complex, creative, or sensitive content, the “savings” often evaporate in editing time.
What content works for MTPE (and what doesn’t)¶
Not all text is equally post-editable. This is usually skipped in agency pitches but is the most important factor in whether MTPE actually saves time.
High suitability - MTPE makes sense: - Technical documentation: manuals, specs, how-to guides - structured and repetitive, MT handles it well - Product descriptions and catalogs - formulaic structure, MT does the heavy lifting - Support knowledge bases and FAQs - conversational but repetitive - Internal communications - functional accuracy is enough, stylistic polish isn’t required
Medium suitability - MTPE helps but needs more work: - Website copy - MT needs significant work on tone and cultural adaptation, but usable as a base - Business correspondence - often formal enough for MT to handle, needs style check - Financial reports (internal) - numbers and standard constructions handled well, complex analysis needs attention
Low suitability - MTPE usually isn’t worth it: - Marketing campaign copy: requires cultural adaptation, wordplay, brand voice - MT output often needs a near-complete rewrite - Literary texts: by definition can’t be post-edited to publication quality - Certified/sworn translations: legally require a human translator as the responsible party (more on that here) - Highly specialized texts with no clean MT training data for the specific subject matter
The practical test: if the text is formulaic, repetitive, or technical with established terminology, MTPE saves time. If it’s creative, stylistically nuanced, or requires terminology MT doesn’t know, the savings shrink toward zero fast.
What a post-editor actually does¶
“Fix the machine translation” sounds simple. In practice, post-editing is a distinct professional skill that requires specific habits - including the discipline to know when to stop editing.
A post-editor’s typical workflow:
1. Sample the output before starting. Experienced post-editors spend 5-10 minutes testing a sample to gauge the error rate. If MT quality is too low, it makes more sense to flag it and renegotiate scope - or decline. Taking a full MTPE job and discovering it requires full retranslation is how you lose money and time.
2. Work inside a CAT tool. Professional MTPE almost always happens in a CAT (computer-assisted translation) tool like Trados, memoQ, or Smartcat, where MT output is loaded as a draft. This ensures consistency against existing TMs and glossaries, and allows QA automation at the end. The CAT tools comparison guide covers the integration options for each.
3. First pass: factual accuracy. Fix mistranslations, omissions, wrong numbers or dates, terminological errors. This is the minimum for light PE.
4. Second pass (full PE only): fluency and style. Read the text independently of the source - does it sound natural? Fix calques, awkward syntax, wrong register.
5. Don’t over-edit. This sounds obvious but it’s the #1 time sink in MTPE. ISO 18587 explicitly lists “editing efficiently without over-editing” as a required post-editor competency. Rewriting technically-correct sentences just because you’d have phrased them differently erases the entire productivity advantage.
As the Crowdin MTPE guide puts it:
The goal of post-editing is to achieve the required quality level in minimum time - not to produce the translation you would have written from scratch.
This mindset shift is harder than it sounds for translators trained to produce their best work on every sentence. Post-editing efficiently is a separate skill from translating well.
MTPE rates: what the market pays¶
MTPE rates are a persistent debate on translator forums - because the per-word numbers look lower than human translation, but the economics are more complex than they appear.
Current market rates (2026):
| Type | Per-word rate | Notes |
|---|---|---|
| Light post-editing | $0.03-$0.08 | Wide range by language pair and MT quality |
| Full post-editing | $0.08-$0.15 | Some translators price at full human rate |
| Human translation | $0.09-$0.35 | Depends on pair, specialization, and market |
Source: Artlangs 2026 MTPE pricing guide.
Three things to know about these numbers:
Rate ≠ income. $0.05/word MTPE at 750 words/hour = $37.50/hour. $0.15/word human translation at 250 words/hour = $37.50/hour. The per-word rate is lower, but hourly income can be the same - if MT quality is good enough to achieve that throughput.
Language pairs change everything. Less common pairs have worse MT quality, meaning more editing time per word. The $0.05/word rate that makes sense for English-German may be unprofitable for the same content in English-Ukrainian or Arabic-Portuguese.
Many translators refuse to discount. According to the GTS 2025 survey, about 50% of translators who do MTPE don’t offer a per-word discount compared to their standard translation rate, arguing the editing effort is equivalent. Those who do offer discounts typically reduce rates by 10-30%.
The most useful thing you can do before accepting an MTPE job: request a 500-word sample and post-edit it. 30 minutes will tell you more about the actual effort required than any rate discussion.
How widespread MTPE has become¶
MTPE isn’t a niche workflow anymore. According to industry data from Polilingua, MTPE adoption grew from 26% of projects in 2022 to nearly 46% in 2024 - a 75% increase in two years. Among top LSPs (language service providers), adoption has reached 91%.
The GTS 2025 survey found that 87.93% of freelance translators already engage with MTPE regularly - only 12% have never worked on it. And 38.68% expect MTPE to dominate the industry going forward.
What the same survey found about job market impact: 37.74% believe AI/MTPE has significantly reduced freelance opportunities, and 42.45% say it has had some negative impact. The translators reporting the most pricing resilience are specialists in legal, medical, and technical fields - where MT quality is lowest and human expertise matters most.
The practical takeaway: knowing how to post-edit efficiently isn’t optional anymore if you work with agencies or LSPs that handle volume. The translators doing well with it are those who treat MTPE as a distinct skill with its own workflow - not just “translation but with a head start.”
For a deeper look at how this fits into a full hybrid AI + human workflow, including tools, prompt strategies, and billing models, that article covers the complete picture.
FAQ¶
What does MTPE mean in translation?¶
MTPE (Machine Translation Post-Editing) means a human translator reviews and corrects the output of a machine translation engine to bring it to the required quality level. The post-editor doesn’t translate from scratch - they work with the MT draft, fixing errors and improving fluency.
What’s the difference between MTPE and proofreading?¶
Proofreading checks a human-produced translation for errors. MTPE starts with machine-generated text and can require anything from light corrections to near-complete rewriting. MTPE requires active bilingual proficiency, not just reading-level knowledge of the target language.
Is MTPE faster than human translation?¶
Yes, significantly - under the right conditions. Post-editors process 700-1,000 words/hour versus 250-400 words/hour for translation from scratch. But this depends heavily on MT quality for the specific language pair and content type. Poor MT output can reduce MTPE to the speed of manual translation, making it uneconomical.
Do MTPE rates pay fairly?¶
It depends on how you calculate. Rates are lower per word ($0.03-$0.15) than human translation ($0.09-$0.35), but higher throughput can maintain similar hourly income. About half of professional translators refuse to discount their rates for full post-editing, arguing the effort is comparable to translating from scratch.
What skills does a post-editor need?¶
According to ISO 18587: full bilingual proficiency in both languages, translation competence, knowledge of MT technology and its typical failure modes, domain expertise, and the discipline to edit efficiently without over-editing technically-correct segments.
Can MTPE be used for certified translation?¶
Generally no. Certified (sworn) translation requires the human translator to be legally responsible for the output. MT can be used as a reference tool, but the certified translation must be the translator’s own work. Policies vary by country and certifying authority.
Sources¶
- GTS Translation: The State of MTPE in 2025 - What Translators Think
- Artlangs: How Much Does MTPE Cost - Transparent Pricing Guide for 2026
- Slator: How Fast Can You Post-Edit Machine Translation?
- Polilingua: MTPE Adoption Surges 75%
- ProZ.com Forum: What is the best rate for MTPE?
- ProZ.com Forum: MTPE is a dirty word
- Crowdin Blog: Machine Translation Post-Editing Complete Guide
- Argo Translation: Full and Light MTPE - What’s the Difference?