Machine translation has gotten genuinely good. Used well, MT plus human post-editing (MTPE) can cut cost and turnaround meaningfully on the right content. Used badly — on the wrong content, at the wrong tier — it produces text that's fast, cheap, and quietly wrong in ways that cost you more than starting from scratch.
The trick isn't "MT or humans". It's matching the level of human effort to what the content actually needs.
The two tiers, in plain terms
Light post-editing fixes only what blocks understanding: real mistranslations, grammar that breaks meaning, terminology that's flat wrong. The goal is "accurate and clear", not polished. Good for high-volume, low-visibility, short-shelf-life text.
Full post-editing aims for output indistinguishable from human translation: correct, fluent, on-brand, consistent terminology, natural rhythm. It costs more and saves less time — but it's the only honest option for anything customer-facing.
Where MTPE genuinely pays off
- Support content at scale — knowledge-base articles, FAQs, ticket macros. High volume, factual, forgiving.
- Internal documentation — process docs, training material read by staff, not customers.
- Product catalogs with repetitive, structured descriptions.
- User-generated content you want to make searchable or roughly understandable.
On this kind of text, with a tuned engine and a good glossary, MTPE can deliver real savings versus full human translation — without a quality cliff.
Where it burns you
- Marketing and brand copy. MT translates words, not intent. Slogans, taglines and campaigns need transcreation — a human rewriting for effect. MT here reads as "translated", which is exactly the impression you're paying to avoid.
- Legal, medical and financial. A subtle mistranslation in a contract clause, a dosage instruction or a disclosure isn't a style issue — it's liability. These demand specialist human translation and review, full stop.
- Anything with heavy cultural nuance, humor, or Asian honorific/politeness systems, where the right word depends on who's speaking to whom.
- Tiny, high-stakes strings. A homepage headline or an app's primary CTA is 5 words seen by everyone. The "savings" from MT here are rounding error; the risk isn't.
Things that decide whether MTPE works at all
Engine choice and training
A generic engine and a domain-tuned one produce very different drafts. Feeding the engine your translation memory and glossary up front reduces the editing burden — that's where most of the real saving comes from.
A real glossary and do-not-translate list
MT will happily translate your product names, feature names and UI labels into nonsense. A locked termbase and a do-not-translate list keep them intact.
Post-editors who are translators
Light editing of fluent-but-wrong MT is a specific skill — it's often harder than translating from scratch, because the polished surface hides the errors. Use linguists trained in MTPE, not generalists.
How we approach it
We don't pick MTPE or human translation by default — we pick it per content type. Support and docs may run light or full post-editing; product UI and marketing stay human; regulated content is human plus a second-linguist review. The point of MT is to spend your budget where it changes the reader's experience, and to stop pretending a cheap draft is the same as a finished one.