Machine Translation Post-Editing: A Guide to Combining AI and Human Translation

What Is Machine Translation Post-Editing?

Machine translation post-editing is the process of having a professional linguist review, correct, and refine text that a machine translation engine has already produced, rather than translating the document from scratch. A company gets the speed of automated translation and the judgment of human review, instead of publishing raw machine output or paying for a fully manual translation on every project. For a Bay Area company weighing whether AI translation tools are ready to use on their own, post-editing is usually the answer that protects both budget and quality.

Post-editing sits between two extremes. On one end, a company can run text through a machine translation engine and publish it as-is, accepting whatever errors come along for a fast, low-cost result. On the other end, a human translator can translate every sentence from the source language with no machine assistance at all, which tends to cost more and take longer. Post-editing borrows the speed of the first approach and the judgment of the second: an engine produces a first draft, and a trained linguist checks it against the source text, fixes what the engine got wrong, and brings the tone and terminology in line with how the company actually communicates.

Light Post-Editing vs. Full Post-Editing

Not every project needs the same level of review, and treating all machine-translated content the same way usually wastes either time or quality. Light post-editing focuses on accuracy: the post-editor fixes mistranslations, missing meaning, and anything that would embarrass or confuse a reader, but leaves awkward phrasing alone as long as the sentence is understandable. It works well for internal documents, support tickets, or large volumes of product data where speed matters more than polish.

Full post-editing goes further. The post-editor also smooths out sentence structure, adjusts tone, and brings the text as close as possible to what a skilled human translator would have produced from the start. A style guide and glossary matter more here, since the goal is a document a reader cannot tell was ever machine-translated. Full post-editing is the right choice for anything a customer, regulator, or business partner will actually read closely, including marketing pages, contracts, and product documentation.

Where Raw Machine Translation Still Falls Short

Modern machine translation engines handle straightforward sentences well, and the technology keeps improving. The gaps that remain are not about missing words. A modern engine translates every word it sees. The trouble shows up in nuance: idioms translated literally, a formal register swapped for a casual one, and industry-specific terminology rendered as a generic dictionary equivalent instead of the term a company's customers actually recognize. A phrase that reads naturally in English can come out stiff, overly formal, or simply strange once an engine has translated it word by word. See our full AI vs. human translation comparison for more of these gaps side by side.

Cultural context causes similar trouble. A marketing phrase built around a pun, a sports reference, or a seasonal idiom may translate accurately on the word level while missing the point entirely for a reader in another country. A post-editor who is a native speaker of the target language, and who lives in or near that market, catches these problems immediately, because he or she reads the sentence the way an actual customer would.

The ISO 18587 Standard for Post-Editing Quality

Because "post-editing" can mean very different things depending on who is doing it, the translation industry uses ISO 18587:2017, the international standard for post-editing of machine translation output, as a shared reference point. The standard lays out requirements for the post-editing process itself and for the post-editor's competence, covering steps like checking terminology against an approved glossary, confirming that no meaning was added or dropped, and verifying that formatting and numbers carry over correctly from the source document.

A company does not need to memorize the standard to benefit from it. The practical takeaway is simpler: when a provider says a project was post-edited, it is worth asking whether the process followed a defined standard like ISO 18587, or whether "post-edited" simply means someone skimmed the output before sending it back. The difference shows up in the final quality.

When Machine Translation Post-Editing Makes Sense (and When It Doesn't)

Post-editing tends to work best for content that is high in volume and lower in stakes: product catalogs, internal knowledge base articles, user-generated reviews, and routine correspondence all benefit from the speed of machine translation paired with a human check. The savings on a large volume of repetitive content can be substantial without a meaningful drop in quality, especially with light post-editing.

Some content is a poor fit for a machine-translation starting point, no matter how good the post-editing is:

• Legal and regulatory documents. Contracts, immigration paperwork, and government filings often need a certified human translation, and a court or agency may not accept machine-assisted output at all.
• Medical and clinical content. Patient instructions, informed consent forms, and clinical trial materials carry too much risk for an engine's first draft to be the foundation.
• Marketing and brand messaging. A slogan or campaign built around wordplay usually needs to be written for the target market from the start, a process closer to transcreation than post-editing.
• Anything requiring certification. A certified document translation for a diploma, birth certificate, or court document needs to come from a qualified human translator who can sign and stamp it.

A good rule of thumb: the more a mistake would cost — legally, medically, or reputationally — the less appropriate a machine-translation starting point becomes, regardless of how much post-editing follows.

Building a Post-Editing Workflow That Protects Quality

A machine translation post-editing project works best when the groundwork happens before the first sentence is translated. A locked terminology glossary keeps the engine and the post-editor using the same term for the same concept every time, instead of drifting between synonyms across a long document. A multilingual style guide does the same for tone, formatting, and the small style choices (how numbers, dates, and product names are handled) that make a document feel consistent from page to page.

Translation memory adds another layer of consistency: previously approved translations feed back into future projects, so a phrase translated correctly once does not need to be re-approved every time it appears again. A second linguist, separate from the person who did the editing, reviews the post-editor's work and catches mistakes a single reviewer might miss, the same quality-assurance step a fully human translation project would use.

Choosing a Machine Translation Post-Editing Partner

Look for a provider that treats post-editing as a defined process, not an afterthought bolted onto a machine translation subscription. Ask what level of post-editing is included by default, whether the post-editors are subject-matter specialists in a client's industry — such as Auerbach's technical translation services team for engineering and manufacturing content — and whether a second linguist reviews the work before delivery.

Auerbach International has combined AI-accelerated translation with master's-level human linguists since 1990, in 120-plus languages. Our post-editors follow a defined glossary and style guide for every client, and every project gets a second review before delivery. If a growing volume of content has your team weighing machine translation against a fully human project, request a free quote from Auerbach International and let us know your content type, volume, and languages so we can recommend the right level of post-editing.

Frequently Asked Questions

What is the difference between machine translation and machine translation post-editing?

Machine translation is the raw output an engine produces on its own. Machine translation post-editing adds a trained linguist who reviews that output against the source text, corrects errors, and adjusts tone and terminology before the content is used.

How much does machine translation post-editing cost compared to full human translation?

Light post-editing typically costs less than a translation done entirely by hand, since the linguist is reviewing and correcting rather than translating every sentence from scratch. Full post-editing costs more than light post-editing but usually still less than a project with no machine translation involved at all, though the exact savings depend on content type, volume, and language pair.

Can machine translation post-editing be used for certified translations?

No. A certified translation for a diploma, birth certificate, or court filing needs to be completed and signed by a qualified human translator, and many agencies will not accept a machine-assisted document even after post-editing.

What is the difference between light and full post-editing?

Light post-editing corrects factual and meaning errors but leaves readable, awkward phrasing alone. Full post-editing also smooths sentence structure and tone so the finished text reads as if a human translator wrote it from the start.

Does post-editing work for every language pair?

Machine translation quality varies by language pair, and it tends to be strongest between widely used languages with large amounts of training data. A post-editor working in a lower-resource language may need to correct more of the engine's output, which can shift a project toward full post-editing or fully human translation.

Who decides whether a project needs post-editing or full human translation?

A translation provider should recommend the right approach based on the content's purpose, audience, and risk, not default to whichever option is fastest. A reputable provider will ask about a document's intended use before quoting a machine-assisted or fully human approach.