Raw machine output.Machine speed, in real Vietnamese.
I'm Dao Huy (Lucas), a native Vietnamese translator. The machine drafts, then I take a red pen to it: catching the false friends, terminology slips and wrong register it leaves behind, line by line.
Phiên đăng nhập đã hết hạn sử dụnghết hiệu lực
Mở khóaĐánh thức trọn vẹn tiềm năng của đội nhóm bạnđội ngũ
Người bán không chịu trách nhiệmchịu trách nhiệm pháp lý đối với thiệt hại gây hậu quảmang tính hệ quả
Machine translation post-editing is a machine draft, finished by a human.
MTPE means taking raw output from a neural machine translation engine, DeepL, Google, Microsoft or a GPT-based model, and editing it to the standard a human translator would publish. You get the speed of the machine and the judgement of a native Vietnamese linguist on the same text.
See where the machine slips, and how I fix it.
Pick a content type, read the raw machine output, then reveal the post-edit. The machine is fast and often close, but "close" in legal, marketing or UI copy is exactly where trust leaks. That judgement is what post-editing adds.
Machine translation is not the enemy. Shipping it unread is. Post-editing is where machine speed becomes something you can put your brand on.
For more than seven years I have translated and localized for global brands and the agencies that serve them. MTPE is a distinct discipline: not retranslating from scratch, and not rubber-stamping the machine, but reading the source and the output together and correcting exactly what the machine got wrong, at a pace that still beats full human translation.
I work in Trados, memoQ, Smartcat and Crowdin, so your translation memory and glossary drive every decision, and the same term reads the same way from the first screen to the last footnote. You choose the level: full post-editing for publish-ready quality, or light post-editing when you just need accurate meaning fast.
The errors a machine leaves are the ones that cost you.
Neural MT is fluent, which is exactly the trap: it reads smoothly while quietly saying the wrong thing. These are the failure modes I catch on every job.
Terminology drift
The machine translates the same term three different ways across a file. I lock it to your glossary so it reads identically everywhere.
False friends
"Unlock", "execute", "table" land literally and wrong. I render the intended meaning, not the surface word.
Register slips
Legal text reading casual, marketing reading like a manual. I lift or soften the tone to match the audience.
Hallucinations & drops
Added words, skipped negations, numbers and units quietly changed. I check the output against the source line by line.
Full or light, on the content that matters.
Full post-editing
Publish-ready quality: accuracy, terminology, register and flow all brought to human standard. For content that carries your brand.
Light post-editing
Accurate, understandable meaning, fast and lean. For internal docs, high-volume support content and gisting.
Bilingual review & error correction
Source-against-target QA on an existing translation or MT batch, with a clear log of what changed and why.
Terminology & style consistency
Glossary and TM enforcement across large projects so the same term never reads two ways.
High-volume content
E-commerce catalogues, support centers and UI strings, where consistency at scale is the whole game.
Neural-MT & AI output cleanup
Output from DeepL, Google, GPT and custom engines, cleaned to natural Vietnamese a human would actually write.
The content I post-edit into Vietnamese.
Legal & compliance
Contracts, NDAs, terms of service, court and immigration documents. Post-edited so "liable", "indemnify" and "consequential" carry their exact Vietnamese legal sense.
Marketing & transcreation
Landing pages, ads, email and brand copy. Machine literalism rewritten into Vietnamese that persuades, not just translates.
App & UI localization
Software strings, error messages, onboarding and microcopy, post-edited to fit length limits and the register users expect.
Technical & IT
Manuals, knowledge bases, API docs and specs, where one wrong term can break a whole procedure.
E-commerce & product
Catalogues, product titles and descriptions at scale, kept consistent across thousands of SKUs.
Subtitles & media
Subtitle and e-learning machine output, post-edited for timing, reading speed and natural spoken Vietnamese.
Four steps, machine speed with a human gate.
Set up TM & glossary
Load your translation memory and term base into the CAT tool so the machine and I work from the same source of truth.
Read source against output
Line by line, checking meaning, numbers, negations and intent, not just how the Vietnamese reads.
Post-edit to the chosen level
Full for publish-ready, light for fast gist, fixing terminology, false friends and register as I go.
Consistency pass & deliver
A dedicated sweep for defined terms, units and formatting, delivered in your CAT format ready to ship.
What clients say about the post-edited result.
Turned our raw MT into Vietnamese that actually reads right, kept every product term consistent, and the turnaround was excellent. Exactly what MTPE should be.
We send large batches and the terminology never drifts. The glossary discipline alone is worth it. Fast, reliable, will keep sending work.
Top-Rated Plus for a reason. Caught a flipped negation our previous vendor shipped. That single catch paid for the whole project.
Send your MT batch, get a quote in 10 minutes.
Rates are USD per source word and depend on the raw MT quality and the level of post-editing. Poor MT may be quoted closer to translation rates; clean MT costs less. Large and ongoing programmes are quoted as a package, and the rate is negotiable.
- Terminology check & QA on every file
- Revisions until you are satisfied
- Delivery in your original format
- NDA signed before any document
Start a project
⚡ Reply in < 10 minutesRequest received!
I'll review the MT and send a detailed quote within 10 minutes (during business hours).
Frequently asked questions.
Full post-editing brings the machine output to publish-ready human quality: accuracy, terminology, register and flow. Light post-editing fixes only what blocks understanding, for fast, accurate meaning on internal or high-volume content. I help you pick the right level for each batch.
Post-editing is cheaper than full translation because the machine does the first pass. Light PE runs $0.03-0.04/word, full PE $0.04-0.06/word, review/QA $0.02-0.04/word. Very poor MT may be quoted closer to translation rates, because fixing bad output can take as long as starting over.
Any output: DeepL, Google, Microsoft, GPT-based and custom engines. I work in Trados, memoQ, Smartcat and Crowdin, using your translation memory and glossary so terminology stays consistent across the whole project.
Yes, that is the core of the service. I enforce your glossary and translation memory so the same term, product or clause reads identically across thousands of strings and across every future update.
About 4,000+ words per day, faster than full translation since the machine handles the first pass. Same-day rush under 24 hours is available for a 25-50% surcharge.
English → Vietnamese (my strongest), Chinese → Vietnamese, French → Vietnamese and Vietnamese → English. Vietnamese is my native language; English at IELTS 7.0, Chinese at HSK 5.
MTPE is the process of taking raw output from a machine translation engine and editing it to human quality. Instead of translating from scratch, I read the source and the machine's Vietnamese together and correct what the machine got wrong, at a pace that beats full human translation.
Yes. Output from GPT-based and other LLM engines is fluent but invents terminology, drifts in register and sometimes adds or drops meaning. I post-edit AI output the same way I do neural MT: against the source, with your glossary, into Vietnamese a native speaker would actually write.
With full post-editing, yes, the published result reads as native Vietnamese. Light post-editing is a deliberate trade-off: accurate meaning, faster and cheaper, but not styled for publication. I tell you honestly which level a given batch needs.
I work in Word, Excel, PowerPoint, PDF, JSON, XLIFF, .strings, CSV and CAT-tool packages, and deliver in your original format. An NDA is signed before any document changes hands, and your files are kept strictly confidential.
MTPE, in plain terms.
- MTPE
- Machine translation post-editing: editing raw machine output to human quality, the service this whole page is about.
- Full post-editing
- Editing to publish-ready quality, indistinguishable from human translation: accuracy, terminology, register and flow.
- Light post-editing
- Editing only what blocks understanding, for fast, accurate meaning on internal or high-volume content.
- Neural MT (NMT)
- Modern machine translation from neural networks (DeepL, Google, Microsoft). Fluent, but still makes terminology and nuance errors.
- Raw MT
- Unedited machine output, straight from the engine, before any human touches it.
- PEMT
- Post-editing of machine translation, the same discipline as MTPE.
- Translation memory (TM)
- A database of your past approved translations, reused so wording stays consistent across projects.
- Glossary / termbase
- Your approved term list, enforced so a product, feature or clause always reads the same way.
- MQM
- Multidimensional Quality Metrics, an industry framework for scoring translation errors by type and severity.
- CAT tools
- Computer-assisted translation tools (Trados, memoQ, Smartcat, Crowdin) that drive TM and glossary during post-editing.
- False friend
- A word that looks like a correct translation but means something else, a classic machine-translation trap.
- Register
- The tone and formality of the text. Machines often miss it; legal must read legal, marketing must read human.
Send your MT batch. I'll tell you the level and the price.
A few thousand strings or a million-word programme, I post-edit it into Vietnamese you can ship with confidence. Reply within 10 minutes.