Machine translation post-editing (MTPE): the complete guide

Machine translation has never been better — and for most business content, machine translated content still isn't safe to publish unreviewed. That gap is exactly what machine translation post-editing (MTPE) exists to close, and it's why industry analysts like Slator and Nimdzi consistently track post-editing as one of the fastest-growing service lines in the language industry. This guide covers what MTPE is, the difference between light and full post-editing, how the workflow runs step by step, how pricing actually works, and when MTPE beats both raw machine translation and human-only translation.

• MTPE pairs an AI-generated translation draft with a human editor, combining machine speed with human judgment — typically far faster and cheaper than translating from scratch.

• Light post-editing fixes only errors that change meaning; full post-editing brings output to human-translation quality. Choose by content stakes, not by habit.

• Pricing follows the draft: because the editor starts from machine output, MTPE is typically priced at a significant discount to from-scratch human translation. Stanley Black & Decker cut translation costs up to 70% — from $200–300 to $1.20 per 1,000 words — with an AI-plus-human-review workflow.

• ISO 18587:2017 is the international standard for full post-editing of machine translation output, including required post-editor competencies.

• Post-editing measurably improves quality: Smith+Nephew cut editing workload 70% by combining AI translation with structured human review on one platform.

MTPE (machine translation post-editing) is the process by which a professional human editor reviews and corrects text translated by a machine translation engine or AI model, fixing errors in accuracy, terminology, and style until the text meets an agreed quality level.

You'll see the term written both ways — post-editing and post editing — and abbreviated as MTPE or PEMT. Whatever the spelling, the model is the same: the machine does the volume, the human does the judgment.

MTPE emerged because neural machine translation and, more recently, LLM-based translation produce drafts that are mostly right. Paying a linguist to translate from a blank page ignores that head start. Paying them to edit the draft captures it — which is why post-editing has displaced from-scratch translation across huge swaths of technical, support, and e-commerce content.

The industry recognizes two service levels, and choosing the wrong one is the most common MTPE mistake — over-editing low-stakes content or under-editing content that represents your brand.

In practice, mature localization teams don't pick one level for everything. They tier content: raw machine translation for ephemeral text, light post-editing for internal material, full post-editing for anything a customer, regulator, or journalist will read.

A production MTPE workflow has six steps. Skipping the first and last is what separates teams that get compounding quality gains from teams that fix the same errors every project.

That last step is the quiet economics of MTPE: the workflow gets cheaper and faster the more you run it.

MTPE sits deliberately between the two extremes — and each of the three approaches has a legitimate home. For a deeper treatment of the two poles, see our guide to machine translation vs human translation.

The honest summary: raw MT wins when a wrong translation costs you nothing, human-only wins when the text must persuade rather than inform, and MTPE wins the large middle — which for most companies is the majority of translated volume.

ISO 18587:2017 is the international standard covering post-editing of machine translation output. It specifies requirements for the full post-editing process — the steps a provider must follow to deliver output comparable to human translation — and defines the competencies post-editors must hold: translation competence, linguistic and textual competence in both languages, cultural competence, technical competence with MT and CAT tools, and domain knowledge.

The standard applies to full post-editing; light post-editing is acknowledged but sits outside its core requirements. When buyers ask vendors or platforms for "ISO-compliant MTPE," ISO 18587 is the document they mean — alongside ISO 17100, which covers human translation services generally.

Yes — post-editing reliably improves translation quality because a qualified human catches mistakes machines still make: mistranslations that read fluently, terminology drift, cultural missteps, and errors in names, numbers, and negation, while the machine still supplies speed and consistency of coverage.

The gains compound when post-editing runs inside a structured workflow rather than as ad-hoc cleanup. Corrections flow back into translation memory and glossaries, so the same error never needs fixing twice, and automated QA flags mechanical issues before a human ever reads the segment.

Smith+Nephew, the global medical technology company, cut its editing workload by 70% by pairing AI translation with structured human review in a single platform — the post-editing effort shrank precisely because the workflow kept improving the drafts.

There is no single market rate for MTPE — anyone quoting you one number without seeing your content is guessing. What's consistent is the structure: because the editor starts from a machine draft instead of a blank page, MTPE is typically priced at a significant discount to from-scratch human translation. Three pricing models dominate.

• Discounted per-word. The classic LSP model: a per-word rate set below the human-translation rate for the same language pair, with full post-editing priced above light. Discounts scale with MT quality — better drafts, less editing, lower rate.

• Hourly. Common when MT quality is unpredictable or content is highly technical. The editor bills for time spent, which protects them from bad drafts but makes your costs harder to forecast.

• Platform or subscription. AI translation platforms bundle the machine draft, the editing environment, QA, and TM into a subscription, and you pay for usage rather than per-project vendor quotes — Smartcat's Smartwords model works this way, with human post-editing added on top only where the content tier demands it.

What does that structure deliver in practice? Stanley Black & Decker cut translation costs by up to 70%, from $200–300 down to $1.20 per 1,000 words, while cutting a two-week turnaround, by moving to an AI-translation-plus-human-review workflow.

Most teams assemble MTPE from parts: an MT engine, a CAT tool, a vendor for editors, spreadsheets for QA. Smartcat runs the entire loop in one place. The platform generates the AI translation draft in 280+ languages, pre-applies your TM and glossaries, and routes segments straight into an editing workflow — your own reviewers, or professional post-editors hired in a few clicks from the built-in Marketplace of 500,000+ vetted professionals.

Quality is instrumented, not assumed: a translation quality score shows how good the AI draft is before you spend a single editing hour, and the QA Coworker runs automated checks across terminology, consistency, and formatting.

• AI draft + human review in one place — no exporting between tools; edits write back to TM automatically

• Hire post-editors on demand from the Marketplace, matched by language pair and subject matter

• Automated QA on every job via the QA Coworker

• Unlimited users and SOC 2 Type II security, with your content never used to train third-party AI — the setup trusted by 1,000+ enterprise brands