Machine Translation Post-Editing (MTPE)

The industry splits post-editing into two service levels, and the difference decides both the price and what the finished text reads like. The split is codified in ISO 18587, the international standard for post-editing of machine translation output: it defines light post-editing as producing a merely comprehensible and accurate text, and full post-editing as producing output comparable to human translation, and it requires post-editors to be qualified translators rather than casual reviewers. Our two service levels follow those definitions.

Light post-editing makes the machine output accurate and understandable, nothing more. The editor corrects mistranslations, fixes anything misleading, restores dropped negations and numbers, and repairs grammar that blocks comprehension. Awkward but correct phrasing is left alone. The result is fit for internal circulation: meeting notes, support macros, knowledge-base drafts, correspondence where the reader just needs to know what was said.

Full post-editing takes the same output to publishable quality. On top of accuracy, the editor rewrites unnatural phrasing, enforces your terminology and register, fixes consistency across the whole document set, and formats numbers, dates and units for the target market. The result should be indistinguishable from human translation to a normal reader. Product content, published articles and customer-facing help centres sit here.

The honest guidance: if a text is worth publishing under your brand, it is worth full post-editing. If it only needs to be understood internally, light post-editing saves the difference. We quote both levels so you can see the gap before deciding.

The service levels stop being abstract when you see them on one sentence. Here is a typical clause from a German commercial agreement and what each stage actually delivers.

Every quote we give tells you which of these levels your content needs, and the sample assessment shows you the raw quality of your actual material before you commit to either.

MTPE earns its saving when three things line up: high volume, repetitive or formulaic language, and tolerance for occasional stiffness in style. E-commerce catalogues, internal reports, user-generated reviews and support documentation tick those boxes, which is why they dominate real MTPE workloads.

The economics flip when the machine output is poor. An editor who has to re-translate every second sentence works slower than a translator starting fresh, and you pay post-editing rates for translation work. This is the heavy-edit trap, and it is why we assess a sample of the actual machine output before quoting rather than promising a blanket discount.

Some content should never go through this pipeline at all, whatever the volume. Certified documents need a qualified human translator and a signed statement of accuracy. Legal drafting turns on nuance a machine flattens. Medical and safety content carries risk no saving justifies. Marketing copy usually needs transcreation, not correction: editing a machine translation of a slogan costs more than writing one properly. We turn this work away from MTPE and quote the right service instead.

This is the routing we actually apply when quoting. It exists to stop you paying for more than the content needs, and to stop MTPE being sold where it does not belong.

Swipe sideways to see all three columns.

If your project mixes rows, we split it: the catalogue goes through full MTPE while the contract beside it goes to a certified translator, on one invoice.

Neural machine translation quality varies enormously by language pair, and that variation decides whether post-editing is economic.

Swipe sideways to see all three columns.

Arabic adds its own layer: engines are trained mostly on Modern Standard Arabic and stumble on dialect, and right-to-left punctuation and numerals need a human eye. As an agency with deep Arabic expertise we price those realities in rather than discovering them after delivery.

Pasting a contract, an HR case or unreleased product material into a free public translation tool sends that content to someone else's servers, where the terms of service often allow it to be retained and used to improve the service. For anything commercially sensitive, that is an avoidable leak.

A managed MTPE workflow keeps the same speed advantage without the exposure:

• One controlled pipeline, with files deleted from working systems after delivery under our retention policy
• A signed non-disclosure agreement available on request
• Systems assessed and secured under the UK Government Cyber Essentials scheme

If your team is already pasting text into free tools to get by, a structured MTPE arrangement is usually cheaper than you expect and removes the risk you are currently carrying without having priced it.

Post-editing is skilled work, and the difference between a good and bad MTPE supplier is process, not software.

Sample assessment first. Before quoting we run a representative sample through the engine and score the raw output. That tells us, and you, whether light or full post-editing fits and whether the saving is real.

Qualified editors, not casual reviewers. Our post-editors are professional translators in the same language pair and subject area. Spotting a fluent-sounding mistranslation is exactly as hard as translating, which is why fluency alone is not a qualification.

Terminology locked before editing starts. For ongoing work we build a glossary and style guide, and translation memory carries your approved phrasing across batches, which also reduces cost on repeated content over time.

A second QA pass on delivery. Numbers, names, dates, completeness and formatting are checked against source before anything leaves us, the same final gate our human translation work goes through under our ISO 17100 conformant process.

MTPE is priced per source word in two published bands: light post-editing runs £0.03 to £0.05, full post-editing £0.06 to £0.10. Within each band the exact rate depends on the language pair and the measured quality of the raw machine output on your material; difficult pairs or poor raw output push towards the top of the band. Where the sample assessment shows post-editing will not save you money, we say so and quote human translation instead.

Turnaround scales with volume rather than being fixed per document. A from-scratch translator typically produces 2,000 to 2,500 finished words a day; a post-editor working on decent machine output gets through roughly 5,000 to 8,000 words a day on full post-editing and more on light. Large batches land in days rather than weeks, and multiple editors can work in parallel on big projects under a shared glossary.

Repeat work gets cheaper. Translation memory means identical and near-identical segments from earlier batches are not charged at the full rate again, which compounds usefully for catalogues and documentation that update rather than change.

Upload a sample for an assessed quote, or ask us to compare MTPE against full human translation on your actual content so the decision is made on numbers, not promises.