What this is
Post-editing means a qualified human takes machine translation output and brings it up to an agreed standard. It comes in two levels, and the difference between them is a commercial question as much as a linguistic one.
| | Light post-editing | Full post-editing | |—-|—-|—-| | Goal | Correct, comprehensible, accurate | Reads as though written in the target language | | We fix | Mistranslation, omission, addition, terminology errors, anything that misleads | All of that, plus fluency, register, style, consistency, natural phrasing | | We leave | Awkward but correct phrasing, machine-ish rhythm | Nothing that a native reader would notice | | Right for | Internal documentation, support content, high-volume material with a short shelf life | Anything a customer, a regulator or a court will read |
Most disputes in this industry come from one side quoting light and the other expecting full. We agree which one you are buying before the file moves, in writing.
What we actually check for
Machine output fails in patterned ways, and knowing the patterns is most of the job.
- Silent omission. A clause that simply is not there in the target. Fluent output hides this better than clumsy output does.
- Terminology drift. The same source term rendered three ways across a 200-page document set, each one defensible in isolation.
- Register and formality. Especially in languages that grammatically encode the relationship between speaker and reader.
- Numbers, units and locale. Decimal separators, date formats, measurement conversions, currency. Cheap errors with expensive consequences.
- Tags, placeholders and layout. Broken markup, variables translated by mistake, right-to-left text that breaks around Latin brand names.
- Segment-boundary damage. Sentences the engine handled correctly in isolation and incorrectly in sequence.
- False friends and confident nonsense. The category that grows as engines become more fluent.
How we price it
This is where we differ from most of the market, and it is deliberate.
We price by revision depth, not by a flat per-word discount. A per-word post-editing rate is a bet that machine output quality will be constant. It never is. The effort gap between a clean segment and a badly hallucinated one is enormous, and under flat pricing you pay the same for both — which means one of us is being overcharged and the other is subsidising.
- Light revision — base rate
- Standard revision — 1.5×
- Comprehensive revision — 2.5×
We ask for a quality floor in the contract. Below an agreed quality threshold, a segment reverts to translation rates. When machine output needs rewriting from scratch, the work is translation, and it should be paid as translation. This protects both sides: you are not paying translation rates for content the engine handled well, and we are not absorbing the cost of a bad engine day.
We treat blanket discounts as a question, not a given. The largest study of post-editing productivity ever published — 90 million words, 879 linguists, two and a half years — found post-editing was 130% faster than translating from scratch in French and 7% slower in Swedish. A single uniform discount cannot be correct across both. Ask us for the reasoning on your specific pairs; we track our own throughput by language pair and content type, and we will show you the numbers rather than argue from assertion.
What you get back
- The post-edited file, in your tool, your memory, your termbase. A structured error report. What the engine got wrong, categorised by type and severity using the MQM framework. Over time this tells you which content types and which pairs your engine is failing on — which is worth more than the file. A named reviewer with qualifications on record. Documented training, subject expertise and years in domain. Available on request. A second pass by a different person. Post-editing and review are never the same linguist.
Who this is for
Language service providers who need overflow or ongoing capacity in difficult pairs and need to hand the result to their own client without re-checking it. We work as a subcontractor and we are comfortable being invisible.
Companies post-editing in-house who have hit the limit of what generalist reviewers can catch, particularly in Arabic, Japanese, Korean and Chinese, and particularly in regulated content where a generalist post-editor is not sufficient.
Common questions
Do you work in our CAT tool?
Yes. Trados, memoQ, Phrase and Smartcat are all normal for us, and we work in the client's environment and translation memory rather than asking you to move to ours.
Can you handle our terminology and style guide?
Yes, and we would rather have them before we start than reverse-engineer them from feedback. If you have a termbase, send it; if you do not, we will flag the terms that need a decision.
Do you offer light post-editing?
Yes, defined properly: correct what is wrong or incomprehensible, do not polish. The failure mode we see most often is light post-editing quoted and full post-editing expected. We will agree which one you are buying in writing.
What are your turnaround times?
It depends on pair, content type and volume. Tell us the deadline in the first email and we will tell you straight away whether it is realistic.