MTPE Calculator: Machine Translation Post-Editing Cost, Time and Savings vs Human Translation
Compare human translation with machine translation plus post-editing (MTPE): cost per word and in total, post-editing hours and days, savings, break-even MT quality and the post-editor’s effective hourly pay, from word count, MT quality, light or full post-editing, hourly or discounted per-word pricing, engine fees and sampled QA.
How it works
Post-editing speed (words/hour) from the edit distance: about 900 at 10%, 700 at 20%, 480 at 35% and 320 at 55% for full PE, × 1.4 for light PE (or your measured speed) · Hours = words ÷ speed · Post-editing cost = hours × hourly rate, or words × human rate × (1 − discount) · MT engine = words × characters per word ÷ 1,000,000 × price per million characters · QA = words × QA share × human rate × QA review rate · Saving = 1 − MTPE total ÷ (words × human rate)
Machine translation post-editing (MTPE) means a professional linguist corrects machine output instead of translating from scratch. It saves money only when the output needs little change, because the editor’s speed falls as the share of words to fix (the edit distance) rises. Published measurements put full post-editing of typical neural MT at about 700 words an hour (studies range from roughly 400 to over 1,100) and light post-editing at about 1,000, against 2,000–3,000 words a day for fresh human translation. The calculator maps your MT quality to an edit distance and a speed (or uses the speed you measured in a pilot), then prices the work either by the hour or as a per-word discount off the human rate, the two common models in the industry. To the post-editing cost it adds the MT engine fee (Google Cloud Translation lists $20 per million characters; about 6 characters per English word including spaces) and a sampled human QA review, then compares the total with human translation at your per-word rate and throughput. The break-even line shows how fast post-editing must be, and how few words the editor may change, before MTPE stops saving money; in per-word mode it shows instead what the post-editor effectively earns per hour, which matters for fair rates. Light post-editing (accurate but not stylistically polished) suits internal and high-volume content; full post-editing aims at publishable quality, as defined in ISO 18587.
Worked example
With the defaults, 50,000 words of general business content (about 20% of words changed) are fully post-edited at about 700 words an hour: 71.4 hours, or 8.9 days of 8 hours, costing $3,214 at $45 an hour. The MT engine costs 50,000 × 6 ÷ 1,000,000 × $20 = $6.00 and a 10% QA sample at 30% of the $0.12 rate costs $180, for an MTPE total of $3,400 ($0.068 per word). Human translation would cost 50,000 × $0.12 = $6,000 and take 20 days at 2,500 words a day, so MTPE saves $2,600 (43%). It stays cheaper while editors manage more than 387 words an hour, which corresponds to changing fewer than about 47% of the words.
Sources: Slator: How fast can you post-edit machine translation? (published speeds, 2022) · Slator: How fast should you post-edit machine translation? EU Translation Centre tender and MTPE discounts · GitLab Translate: MT post-editing, light vs full PE speeds · Google Cloud Translation pricing ($20 per million characters)
Frequently asked questions
How fast is machine translation post-editing?
Published figures put full post-editing at about 700 words per hour (roughly 5,600 words in an 8-hour day) and light post-editing at about 1,000 words per hour, compared with 2,000–3,000 words a day for human translation. Speed depends on MT quality, language pair, subject matter and the editor’s experience with the engine.
How much cheaper is MTPE than human translation?
Per-word MTPE is commonly priced 30–40% below the full translation rate. With hourly pricing the saving depends on speed: at about 700 words an hour and $45 an hour, post-editing costs about $0.064 per word against $0.12 for translation. With poor MT the saving disappears.
What is the difference between light and full post-editing?
Light post-editing fixes meaning and serious errors so the text is accurate and understandable, accepting machine-like style. Full post-editing produces text equivalent to human translation, with correct terminology, style and fluency. ISO 18587 defines both; full post-editing takes roughly 40% longer.
How do I measure edit distance?
Run a pilot of 1,000–2,000 words. CAT tools report the post-edit distance (or TER) between raw MT and the final text. Below about 20% is excellent, 20–35% is workable and above about 40% usually means MTPE is not worth it.
Should post-editors be paid per word or per hour?
Hourly pay, or per-word rates set from a measured pilot, is fairer because the effort depends on MT quality the linguist does not control. Flat per-word MTPE discounts of 30–50% are common but contentious; check the effective hourly pay the calculator shows.
Related calculators
Industry-typical rates and productivities; agencies, language pairs, subject matter, MT engines and file formats vary widely, so treat the result as a budgeting baseline, not a quote. Measure your own post-editing speed on a pilot before committing.