A 50,000-word technical manual lands in your queue on Monday. The client wants it by Thursday. Your in-house capacity is maxed. Two options: MTPE at roughly half the cost and 2x the throughput, or full human translation at full rate, full timeline. Which do you commit to - and how do you justify it to the client?
That decision is what this guide is for. Not theory - a practical framework you can apply project by project.
What MTPE actually means (and what it doesn’t)¶
MTPE - machine translation post-editing - is a workflow where an MT engine produces a raw draft and a human linguist edits it to the required quality level. The human is still doing real translation work: correcting mistranslations, fixing terminology, adjusting register, making the output read like a human wrote it.
What MTPE is not: it’s not “let AI handle it and do a quick read-through.” The hybrid AI + human workflow that actually produces quality requires structured post-editing methodology - not a casual proofread.
ISO 18587:2017 - the international standard for MTPE - defines two levels:
Full post-editing (FPE): output must meet the same quality bar as human translation - accurate, natural, terminologically consistent, stylistically appropriate. Suitable for publication. This is the level agencies guarantee to clients when they quote MTPE as a service.
Light post-editing (LPE): output is understandable and factually correct, but stylistic quality is not guaranteed. Suitable for internal use, gisting, or content with a short shelf life. Not for client-facing published materials.
This distinction matters enormously for pricing, translator expectations, and what you can contractually commit to.
The four factors that drive the decision¶
Every project decision comes down to these four variables. Get all four right and the choice is obvious.
1. Content type¶
This is the strongest predictor of MTPE success or failure. MT engines perform well on formulaic, repetitive, domain-specific text with predictable syntax. They fail on text where meaning depends on cultural context, idiomatic expression, or creative intent.
MTPE works well: - Technical documentation: user manuals, product datasheets, API docs - Product descriptions and e-commerce catalogs - Knowledge base articles and FAQs - Internal communications and corporate reports - Software UI strings and error messages - Standard legal clauses that repeat identically across contracts
Human translation required: - Legal contracts with jurisdiction-specific nuances - Certified translations (passports, diplomas, birth certificates) - Medical patient-facing materials and clinical trial documents - Marketing campaigns, slogans, brand copy - Literary and creative content - Regulatory filings and compliance documents
As Tomedes notes in their quality tier framework, the key question isn’t “is this content important?” - it’s “does this content depend on cultural nuance, idiomatic expression, or legal precision that MT systematically fails at?” Technical documentation can be mission-critical but still be MTPE-suitable. A slogan for a new product launch can be three words but absolutely requires human translation.
Per ISO 18587, there are content types where post-editing cannot produce acceptable output regardless of the editor’s skill - where the MT draft introduces structural and cultural errors that post-editing cannot fully correct without effectively retranslating the text.
2. Language pair¶
MT quality is not uniform across language pairs. A decision that’s clearly MTPE for EN→FR might be clearly human translation for EN→KO or EN→AR.
SwissGlobal’s 2026 productivity analysis documented the range across pairs: - English → French: +130% speed gain with MTPE - English → Polish: +18% speed gain - English → Swedish: -7% (slower than human translation)
That Swedish data point isn’t a typo. For some language pairs, MT output quality is poor enough that post-editing takes longer than translating from scratch - the post-editor spends more time parsing and correcting the MT draft than they would have just translating. In those cases, MTPE is a net loss.
Before committing any language pair to an MTPE workflow, run a pilot on 5,000-10,000 words and measure actual post-editing speed against the translator’s human translation rate. Don’t assume MT is always faster.
3. Volume and timeline¶
MTPE’s throughput advantage compounds at scale. A linguist doing full post-editing processes 3,000-5,600 words/day vs 2,000-2,500 words/day for unaided human translation. Light post-editing goes to 4,000-8,000 words/day.
The EU Translation Centre’s RFP documented by Slator set these benchmarks for a full-time linguist: - Human translation: ~500 standard pages per 6 months (~10 pages/day) - Full post-editing: ~800 pages per 6 months (~15 pages/day) - Light post-editing: up to ~20 pages/day
Post-editing trainer Enrico Antonio Mion puts it in word counts:
“700 words per hour for full post-edit and 1,000 words per hour for light post-edit - a total of 5,600 or 8,000 words per day respectively.”
For a 20,000-word technical manual with a 3-day deadline: human translation might need 8-10 linguist-days, MTPE might take 4-6 days. If you have one translator allocated, MTPE is the only realistic path. For a 2,000-word marketing brochure: the volume argument disappears, and the content type argument takes over.
4. Risk and accountability¶
What happens if the translation contains an error? For internal documentation, the downside is minor. For a published pharmaceutical instruction leaflet, a medical device manual, or a legal contract, the downside is regulatory non-compliance, liability, or worse.
Risk tolerance maps directly to the choice: - Internal / low-stakes / short shelf life → MTPE (light or full) - Client-facing / medium stakes / published → full MTPE only (ISO 18587 standard) - Legal / medical / certified / high stakes → human translation only
Cost comparison: the actual numbers¶
Here’s what the market charges in 2026:
| Service | Per-word rate | 100K word project |
|---|---|---|
| Light post-editing (LPE) | $0.03-0.08 | $3,000-8,000 |
| Full post-editing (FPE) | $0.08-0.15 | $8,000-15,000 |
| Human translation (standard) | $0.15-0.30 | $15,000-30,000 |
| Human translation (legal/medical) | $0.20-0.40+ | $20,000-40,000+ |
Source: Artlangs MTPE Rate Guide 2025, industry benchmarks.
The 50% saving at full MTPE vs human translation is real - but it comes with caveats.
The pricing pressure problem. A GTS Translation survey of 212 freelance translators in 2025 found that 66% consider MT output “acceptable but requires significant edits.” If you’re quoting MTPE rates but your linguists are doing full post-editing effort, their effective hourly rate drops while workload stays the same. That creates quality risk downstream.
Some agencies are shifting to hourly pricing for MTPE (averaging around CHF 90/hour in Switzerland per SwissGlobal data) to better reflect actual cognitive effort. This removes the incentive for translators to rush and fixes the rate mismatch.
The MT quality variability problem. A low-quality MT draft means post-editing approaches retranslation effort at post-editing rates. The Nimdzi MTPE Efficiency Gap report calls out exactly this: agencies that route content through MTPE without assessing MT quality often end up paying near-human-translation effort at MTPE rates, burning translator goodwill in the process.
Productivity comparison¶
| Workflow | Words/day | vs. human baseline |
|---|---|---|
| Unaided human translation | 2,000-2,500 | Baseline |
| Full post-editing (FPE) | 3,000-5,600 | +50 to +140% |
| Light post-editing (LPE) | 4,000-8,000 | +100 to +220% |
These are averages. A 2025 Finnish study of 908 segments found GenAI post-editing was 14% faster than translating from scratch - but with individual variation ranging from -2% to +102% across translators, reflecting both translator skill with MTPE and MT output quality for the specific domain and language pair.
SwissGlobal’s analysis also found MTPE reduces keystrokes by 59% and cognitive pauses by 63% compared to human translation - for language pairs and content types where MT performs well. The emphasis on that last clause matters: these gains are not universal.
The content suitability matrix¶
Use this as a quick reference at the project intake stage:
| Content type | MTPE suitable? | Level | Notes |
|---|---|---|---|
| Technical manuals | Yes | FPE | Works best with domain TM loaded |
| Product descriptions | Yes | LPE or FPE | High repetition = reliable MT |
| Software UI strings | Yes | FPE | Terminology management critical |
| Internal reports | Yes | LPE | Gist quality acceptable |
| FAQs and knowledge base | Yes | FPE | Formulaic structure = reliable MT |
| Marketing copy | No | Human only | Tone, idiom, cultural fit not MT-reliable |
| Legal contracts | No | Human only | Jurisdictional nuance cannot be post-edited |
| Certified translations | No | Human only | Accountability is personal, not MT-based |
| Medical (patient-facing) | No | Human only | Safety-critical; errors have consequences |
| Literary content | No | Human only | MT destroys style and voice |
| Financial regulatory filings | No | Human only | Compliance risk too high |
| Repetitive legal boilerplate | Sometimes | FPE | Assess MT quality per language pair first |
A five-question intake checklist¶
Your PMs can apply this in under two minutes at project intake:
1. What is the content type? If it falls in the “No / Human only” category → stop here, assign to human translation.
2. What is the language pair, and have we tested MT quality for it? Untested pair → run a 2,000-word pilot first. Tested and poor MT quality → human translation.
3. What is the risk if the translation has an error? Client-facing published material with any legal, medical, or compliance dimension → human translation. Internal / informational → MTPE eligible.
4. What volume and timeline are we working with? Under 3,000 words → the throughput advantage disappears, human translation may be equally fast. Over 10,000 words with a tight deadline → MTPE likely necessary.
5. Does the client’s brief specify quality requirements? Certified or notarized translation required, ISO 17100 compliance stated, or “human-authored output” specified → human translation regardless of content type.
Where AI translation tools fit in agency workflows¶
For agencies handling large volumes of formatted documents - technical manuals in DOCX, reports in PDF, product datasheets - the friction in MTPE workflows is often about the files themselves: stripping formatting, reassembling translated output, preserving tables and layouts through the MT stage.
AI document translation platforms like ChatsControl handle this file-handling layer: upload a DOCX or PDF, get back a formatted translated draft with the original layout preserved, then route it to your post-editor directly. The linguist focuses on the post-editing; the platform handles the formatting preservation. The output still requires human review - this is a tool for the MT draft generation step, not a replacement for the post-editor.
Worth considering for agencies doing high-volume document translation where file handling itself creates bottlenecks. Not suitable for certified translation or content requiring human authorship accountability.
Running a pilot before committing¶
Any new language pair, content type, or MT engine combination should be piloted before you quote MTPE rates at scale.
- Select 5,000-10,000 words representative of the actual content
- Generate MT output using your chosen engine
- Have a trusted linguist post-edit and track actual hours
- Calculate effective words/hour and compare against their human translation rate
- If MTPE productivity gain is under 20%, reassess whether MTPE is worth quoting
A pilot costs a few hours of linguist time. Committing 100,000 words to MTPE without a pilot and discovering MT quality is poor costs you a project, a client relationship, and a translator.
Quality standard: ISO 18587¶
ISO 18587:2017 governs full post-editing. If you’re offering MTPE as a formal service, this standard defines the baseline:
- Post-editors must have the same language pair competencies as translators
- The post-editor is responsible for the final output quality
- Output must be accurate, understandable, and meet the same grammar/syntax/terminology bar as human translation
When clients ask “is this the same quality as human translation?” - the honest answer is yes for full MTPE under ISO 18587, and no for light post-editing.
As noted in SwissGlobal’s approach, agencies using ISO 17100, 18587, 9001, and 27001 certifications deploy MTPE selectively - only for language pairs and content types where it demonstrably improves efficiency, not as a blanket cost-cutting measure applied to all projects.
When MTPE is the wrong call even when it looks viable¶
The GTS Translation 2025 survey found 21.74% of translators view MT output as requiring extensive rework - effectively retranslation at a lower rate. That’s nearly one in four MTPE projects where the efficiency case breaks down.
Watch for these signals:
- Poor pilot results: post-editing over 60 minutes per 1,000 words = retranslation territory
- No TM or glossary loaded: MT without terminology management produces inconsistent output
- “Internal only” that isn’t really internal: content that gets shared with clients needs publication-quality output
- The client will see it: any direct client exposure means full post-editing minimum, not light
- Untested language pair: never assume a pair performs like EN→FR without testing it
FAQ¶
How much does MTPE cost compared to human translation?¶
MTPE runs 40-60% cheaper. Light post-editing: $0.03-0.08/word. Full post-editing: $0.08-0.15/word. Human translation: $0.15-0.30/word, more for legal/medical. On a 100,000-word project, that’s roughly $10,000 for full MTPE vs $20,000 for human translation.
What’s the difference between light and full post-editing?¶
Light post-editing fixes major meaning errors so text is understandable - stylistic issues and minor errors may remain. Suitable for internal use only. Full post-editing produces publication-standard output: accurate, natural, terminologically consistent. ISO 18587:2017 governs full post-editing, not light.
Can MTPE be used for legal translation?¶
Standard contracts: not recommended. Jurisdiction-specific nuances and terminology with legal force require human expertise. Repetitive boilerplate clauses can sometimes work under FPE with a specialized MT engine and strong terminology management - but only after a pilot assessment.
Certified translations (for courts, immigration, academic institutions): human translation only. Post-editors cannot certify MT output as their own work under most jurisdictional frameworks.
How do you evaluate whether an MT engine is good enough?¶
Run a pilot on representative content for the specific language pair. Measure: post-editing speed (words/hour), error count per 1,000 words, and whether errors are systematic (structural, meaning MT isn’t domain-adapted) or random (surface-level). Use a structured MT quality evaluation to structure the assessment before committing.
Is MTPE adoption growing?¶
Yes. Nimdzi research documents adoption growing from 26% in 2022 to 46% in 2024 - a 75% increase in two years. A GTS Translation survey of 212 freelancers in 2025 found 87.93% now work on MTPE projects. Among top-tier LSPs, adoption is at 91%.
The same GTS survey found 80% of translators believe MTPE has put downward pressure on pricing expectations - a structural challenge agencies need to manage with transparent pricing and fair compensation models.
What CAT tools support MTPE workflows?¶
All major CAT tools integrate MT engines for MTPE workflows. Trados, MemoQ, and Smartcat all support MT integration, change tracking from MT output, and post-editing productivity metrics. MemoQ’s built-in collaboration features make it particularly strong for agency workflows where multiple post-editors work on one project simultaneously.
Sources¶
- SwissGlobal - Translation productivity 2026: human translation vs MTPE
- GTS Translation - The State of MTPE in 2025: What Translators Think
- Artlangs - MTPE Rates 2025: Cost-Effective Translation for High Volumes
- Slator - How Fast Should You Post-edit Machine Translation? (EU Translation Centre)
- ISO 18587:2017 - Requirements for post-editing of machine translation output
- Nimdzi - The MTPE Efficiency Gap
- Tomedes - Human translation, MTPE, or raw AI: which quality tier does your content need?