You get an offer: 80,000 words, MT output ready, $0.04/word for post-editing. The brief says “light PE, just fix the obvious errors.” You do a quick calculation - that’s $3,200 - and accept. Then you open the files.
The MT output is rough. Every third sentence needs structural rework. Three days in, you’ve done 20,000 words. The rate that looked fine on paper is tracking out to $11/hour.
This is the MTPE discount problem. Not “what is the standard discount” - but “what discount can you actually afford to give without destroying your hourly income.” These are very different questions, and agencies benefit from translators not clearly separating them.
Why there’s no standard MTPE discount¶
Every agency has a number. Some offer 50% of your regular rate. Some offer 60%. Some send a rate card listing “light PE: $0.04/word” without explaining what they mean by “light” or what MT engine was used.
ISO 18587:2017 - the international standard for machine translation post-editing - defines quality levels and process requirements, but says nothing about rates. TAUS post-editing guidelines describe what post-editors should do, not what they should earn. The result: MTPE rates are whatever an agency can negotiate with a translator who doesn’t have a strong counter-argument.
According to a 2025 survey of 212 freelance translators by GTS Translation:
Around 50% of respondents do not offer discounts for MTPE work, arguing that post-editing can take as much time as traditional translation.
The translators who don’t discount aren’t wrong. They’re protecting their hourly income in the only situation where it matters: when MT quality is poor. Agencies price MTPE as if productivity gains are guaranteed; experienced translators know MT quality varies enough that those gains often don’t materialize.
What the market actually pays: rate ranges in practice¶
If you do offer MTPE at a discount, here’s the range you’re working in:
| Service type | Typical per-word rate | vs. human translation |
|---|---|---|
| Human translation | $0.15-$0.30/word | baseline |
| Full post-editing (FAPE) | $0.08-$0.15/word | 15-50% below |
| Light post-editing (LAPE) | $0.02-$0.05/word | 70-90% below |
These are USD rates for common European language pairs. Specialized fields (legal, medical, technical) sit at the higher end of each range. High-volume general content sits lower.
The “typical” label hides wide variation driven by:
- Language pair. MT quality for English→German is far better than English→Arabic. Better quality means fewer edits, which can make a steeper discount defensible.
- Content type. Technical documentation with consistent terminology post-edits faster than marketing copy, where MT output is often stylistically unusable even if technically accurate.
- MT engine. A domain-specialized engine trained on millions of words in your field produces better drafts than a generic API call.
- Volume and relationship. Sustained high-volume work at predictable quality can justify lower per-word rates because the overall workflow becomes more efficient.
The table is a starting point, not a formula.
Light vs. full post-editing: completely different discount logic¶
This distinction is what agencies most commonly blur - and what translators most commonly fail to push back on clearly.
Light post-editing (LAPE) per ISO 18587 means correcting only the most serious errors so the text is understandable. You’re not fixing style, flow, or terminology consistency. You’re asking one question per segment: does this convey the correct meaning in a way the reader can understand? If yes - move on, even if it sounds awkward.
At genuine light PE, an experienced post-editor averages around 1,000 words per hour, according to Swiss Global’s 2026 productivity research. Human translation runs 250-400 words per hour. That’s a 2.5-4x productivity multiplier - and with that multiplier, a steep per-word discount actually works:
- Human translation: $0.20/word at 300 words/hour = $60/hour
- Light PE: $0.05/word at 1,000 words/hour = $50/hour
That’s a 75% per-word discount, but only a 17% hourly income reduction. Many translators find that acceptable for good MT output.
Full post-editing (FAPE) targets output comparable to professional human translation - accurate, terminologically correct, stylistically appropriate, and consistent throughout. The effort is much closer to translating from scratch.
At full PE, average throughput is around 700 words per hour - faster than human translation, but the gap is much smaller. The discount math becomes much tighter:
- Full PE: $0.10/word at 700 words/hour = $70/hour (workable)
- Full PE: $0.05/word at 700 words/hour = $35/hour (not workable)
A 70-80% discount is only defensible for genuine light PE at genuine light PE productivity. For full PE, a 30-50% discount from human rates is where the math holds. Agencies offering 60-70% off full PE rates are extracting value from translators who haven’t run these numbers.
The math behind a defensible discount¶
Before accepting any MTPE rate, run this calculation:
Step 1: Get a sample. Ask the agency for 500 words of representative MT output - not cherry-picked, from the actual files you’d be working on.
Step 2: Post-edit the sample at normal pace. Time yourself. Divide 500 by the number of minutes spent, multiply by 60. That’s your words-per-hour for this specific job.
Step 3: Calculate your effective hourly rate. Words-per-hour × offered per-word rate = effective hourly income.
If that number is below your minimum acceptable rate - the negotiation isn’t done.
Most agencies will resist providing a sample. That resistance is itself information: if the MT quality were demonstrably good, they’d want you to see it. When an agency won’t share a sample, assume the output needs heavy editing and price accordingly.
Swiss Global’s research shows how dramatically productivity varies by language pair:
| Language pair | MTPE speed vs. human translation |
|---|---|
| English → French | +130% faster |
| English → German | +85% faster |
| English → Polish | +18% faster |
| English → Swedish | -7% (slower) |
For English → Swedish, there is no productivity gain. Any discount from human rates is unjustified - and a translator offering one is simply working cheaper for no reason. This holds for any language pair where your sample shows you moving at translation speed rather than PE speed.
Effort-based pricing: paying for what was actually changed¶
The cleanest solution to the MTPE pricing problem is effort-based pricing - paying based on how much of the MT output was actually modified, rather than on total word count.
As Hunnect’s analysis of fair MTPE pricing describes the leading model:
If 20% of a segment was rewritten, the linguist receives the same amount as for an 80% fuzzy match in the translation memory.
In practice, the tiers look like this:
| Edit level | % of MT segment changed | Rate paid |
|---|---|---|
| Accepted as-is | 0% | 10-25% of translation rate |
| Minor edits | <30% changed | 40-60% of translation rate |
| Major edits | 30-60% changed | 70-80% of translation rate |
| Full rewrite | >60% changed | 90-100% of translation rate |
Most major CAT tools support this: memoQ, Phrase, SDL Trados, and Smartling all generate post-editing reports showing editing distance per segment.
This model protects both sides. The translator is compensated for actual effort, not penalized for good MT quality or unpaid for bad MT quality. The client has a predictable maximum charge (equivalent to full translation cost) and typically pays 30-50% less than that ceiling.
The challenge is getting agencies to agree to it. Many prefer fixed per-word rates because they’re simpler to budget. For high-volume ongoing work, effort-based pricing is worth pushing for - the argument is clear: their maximum exposure equals full translation cost, expected cost is significantly lower, and the translator is fairly compensated regardless of output quality.
The 50% who don’t discount: when they’re right¶
Half the market offering MTPE at full translation rates isn’t stubbornness. There are scenarios where any discount is inappropriate:
Poor MT quality throughout. If you’re rewriting 60-70% of each segment, you’re doing translation work that uses an MT scaffold you didn’t choose and wouldn’t have produced yourself. That’s not faster - in many cases it’s slower, because you’re fighting the output rather than building clean translation from the source text.
Content type that doesn’t suit MT. Literary text, highly creative marketing, legal texts requiring nuanced argument, certified or sworn translations - MT performance on these ranges from poor to actively misleading. For sworn translation specifically, certifying a document you know was heavily MT-generated creates professional and legal exposure.
“Light PE” brief for full PE work. The most common MTPE project problem: a brief that says light PE for content that genuinely needs full PE quality. You do full PE effort at light PE rates.
No information on the MT system. If the agency doesn’t know what engine was used, they haven’t evaluated whether the output quality justifies the discount level they’re offering.
In these cases, charging full translation rates is the accurate response. MTPE discounting only makes sense when there’s a genuine, demonstrable productivity advantage. When that advantage doesn’t exist, the discount is a pay cut presented as a workflow arrangement.
Negotiating MTPE rates with agencies¶
When an agency sends a rate card, here’s the practical response sequence:
Ask about the MT system. “What engine was used for this language pair? Do you have quality benchmark data or BLEU scores?” Agencies doing serious MTPE have this information. Those that don’t are guessing at quality - which means so are you if you accept a rate without seeing a sample.
Request a sample before committing. “I’d like to evaluate 500-1,000 words of MT output to assess the editing effort level before agreeing to a rate.” This is standard professional practice. A scope assessment before commitment is reasonable for any professional engagement.
Present a rate range tied to quality. “Based on my experience with this language pair: if the MT quality shows 70%+ of segments usable with minor edits, I can work at $0.06/word. If it’s heavy editing throughout, my rate is $0.12/word. Can we agree on a sample assessment to determine which tier applies?”
Offer effort-based pricing as an alternative. “Rather than a fixed per-word rate, consider effort-based pricing - I track editing distance per segment and invoice based on actual edits made. Your maximum exposure equals the full translation rate; your expected cost is typically 30-50% below that.” Some agencies will decline. Some will find this compelling because it aligns incentives and gives them a cost ceiling.
Name a minimum floor. Below a certain per-word rate you won’t go, regardless of how the PE level is classified. Your floor = your minimum acceptable hourly rate divided by your minimum expected words-per-hour for this job type.
Red flags in MTPE offers¶
These signal an agency that hasn’t thought carefully about MTPE pricing:
Rate set before you’ve seen the files. “$0.04/word for all post-editing” regardless of MT engine, content type, or language pair means the rate wasn’t derived from any quality assessment - it was set to match a budget.
“Light PE” for anything client-facing. Light PE produces understandable text, not publishable text. If the output goes to clients, into marketing materials, published documents, or anything with the client’s name on it - you’re being asked for full PE quality at light PE rates.
No definition of “acceptable quality.” What does the finished post-edited text need to achieve? Without a defined standard, disputes about whether you met the brief are inevitable.
Pushing for 50%+ discount on full PE work. At 50% off a $0.20/word rate, you’re at $0.10/word for work running at ~700 words/hour - $70/hour before expenses, still workable. But agencies asking for 60-70% off full PE are at $0.06-0.08/word - $42-56/hour - with zero margin for bad MT output days.
“It’s all consistent content, should be fast.” Consistency doesn’t guarantee MT quality. A 50,000-word project with highly repetitive content from a well-trained engine can be fast. The same volume in a technical domain with variable MT quality can be a nightmare. Consistency in source content doesn’t predict quality in MT output.
FAQ¶
What is a fair MTPE discount from full human translation rates?¶
For genuine light post-editing with good MT quality: 70-80% below human rates can be fair if you’re achieving 800-1,000 words per hour. For full post-editing: 15-40% below human rates is the defensible range - corresponding to $0.08-0.15/word if your human rate is around $0.20/word. Anything above 50% discount on full PE work is hard to justify without exceptional volume and workflow efficiencies.
What percentage of freelance translators discount their MTPE rates?¶
According to a 2025 GTS survey of 212 freelancers, approximately 50% do not offer any discount for MTPE. Among those who do discount, the most common range is 10-30% below their standard translation rate.
How does the language pair affect MTPE discount decisions?¶
Significantly. Swiss Global’s 2026 research shows MTPE is 130% faster than human translation for English-French, but -7% for English-Swedish (i.e., actually slower). For language pairs where MT quality is poor, any discount from human rates is unjustified by the data.
Should I charge the same for MTPE as for human translation?¶
If the MT quality is poor enough that you’re essentially retranslating, yes. The test is your words-per-hour on a real sample: if you’re moving at translation speed (200-300 words/hour) rather than post-editing speed (700-1,000 words/hour), there’s no productivity gain to share with the client through a lower rate.
How does effort-based MTPE pricing work in practice?¶
Your CAT tool tracks what percentage of each MT segment was changed. Segments accepted as-is are paid at a fraction of the translation rate (typically 10-25%). Segments with major changes are paid at 70-90% of translation rate. Fully rewritten segments are paid at 100%. Tools like memoQ, Phrase, and Trados generate post-editing distance reports that support this billing model.
What’s the minimum MTPE rate I should accept?¶
Calculate from hourly income down. Decide your minimum acceptable hourly rate (for example $45/hour). Estimate your realistic words-per-hour for this specific MT output quality (for example 800 words/hour for good light PE). Divide: $45 / 800 = $0.056/word minimum. Below that number, you’re below your floor regardless of how the agency frames the offer.