Translation quality and post-editing

Machine translation produces text quickly, but someone still has to decide whether that text can be published. Quality and post-editing tools help make that decision faster and more consistently. Post-editing is the work of a linguist correcting machine output. Quality estimation is a model that predicts, without a reference translation, how likely each segment is to be wrong. QA checkers run rule-based tests for problems such as missing numbers, broken tags and terms that do not match the glossary.

The tools in this category approach the problem from different sides. ModelFront predicts which machine translated segments are safe to approve as they are, so linguists only see the risky ones. Intento routes text to many engines and models and helps compare them. Xbench runs QA checks on bilingual files from most CAT tools. ContentQuo manages human quality reviews, with error categories and scorecards for vendors. Custom.MT trains and evaluates engines on a company's own data. Unbabel and Lilt combine machine translation with human editing, so the quality step is part of the service.

These tools are only as good as the standard you hold them to. Before using any score to skip human review, agree on what counts as an error for each type of content. A wrong number in a price list, a mistranslated safety warning and an awkward sentence in a blog post are not the same risk. Error typologies such as MQM give reviewers a shared vocabulary for severity.

Featured AI translation tools: translation quality and post-editing

SponsoredTranslation quality and post-editing This position is open Put your tool at the top of the translation quality and post-editing page. Paid spots are always labeled Sponsored. From $99 a month. Get featured

What a featured listing costs · from $99 a month

7 AI translation tools in translation quality and post-editing

ModelFront

Quality prediction for machine translation that flags which segments can be approved without human review.

Developer API

Intento

Routes content to many machine translation engines and models, with evaluation tools for choosing between them.

Developer API

Unbabel

Machine translation with quality estimation and human post-editing, aimed at customer service and support content.

Developer API
Human translators or interpreters available

Lilt

Adaptive translation platform where the model updates as linguists edit, offered as software and as a managed service.

Developer API
Human translators or interpreters available

ContentQuo

Translation quality management for buyers and vendors, covering linguistic review, error typologies and scorecards.

Xbench

by ApSIC

QA tool that checks bilingual files for terminology, consistency, numbers and tags across many CAT formats.

Custom.MT

Service that trains and evaluates machine translation and language models on a company's own translation data.

Test quality estimation on your own content before trusting it. Take a batch that humans have already reviewed, run it through the tool, and see how many of the real errors it would have caught and how many good segments it would have flagged. Set thresholds per language pair and content type, and keep sampling approved segments after launch to catch drift.

Rule-based QA is cheaper and should run on everything. Checking numbers, tags, punctuation, glossary terms and consistency catches many visible errors before a reviewer sees them, whether the translation came from a machine or a person.

Feed what reviewers find back into the process. Corrections should update translation memory and glossaries, recurring engine errors should inform which engine handles which language pair, and vendor scores should come up in regular reviews with the people doing the work. Quality tools save the most time when their findings change what happens upstream, so the same mistake does not reach a reviewer twice.

Questions about translation quality and post-editing

What is machine translation post-editing?

It is a linguist correcting machine output instead of translating from scratch. Light post-editing fixes errors that change meaning. Full post-editing also fixes style and fluency so the text reads as if a person wrote it. Agree on which level you need before the work starts.

What is quality estimation?

Quality estimation uses a model to predict how good a machine translated segment is, without a human reference to compare against. Teams use it to send low-risk segments straight through and route uncertain ones to a reviewer. Test it on your own content before relying on it.

How is a QA checker different from quality estimation?

A QA checker runs fixed rules, such as matching numbers, tags and glossary terms between source and target. Quality estimation judges meaning and fluency with a trained model. They catch different errors, so many teams use both.

How do I measure translation quality across vendors?

Review a sample from each vendor using the same error categories and severity levels, often based on a framework such as MQM. Tools such as ContentQuo help run these reviews and keep scores over time, so you compare vendors on evidence instead of complaints.