Article

How Translation Quality Assurance Works

Translation quality assurance is the set of checks that catch errors before a translation is delivered: the translator's self-check, revision by a second linguist against the source, automated checks for numbers and terminology, and, where needed, a review by a subject expert and a final proofread of the layout. Quality is then measured by counting and grading errors.

Hourglass with red sand

Key takeaways

  1. Professional translation QA has several layers: self-check, bilingual revision by a second linguist, automated QA, optional expert review and a final proofread.
  2. The ATC notes that ISO 17100 prescribes revision of each translation by a second person.
  3. ISO 5060 gives guidance on evaluating translation output using error types and penalty points to produce an error score and a quality rating.
  4. Automated QA tools catch mechanical errors such as numbers, tags and terminology, but not mistranslations that read fluently.
  5. Buyer-side review works best when reviewers are briefed, use the glossary and separate errors from preferences.

This guide is for buyers who want to understand what happens to a translation between the translator's first draft and delivery, and how to judge whether the result is good. It covers the QA steps inside a professional workflow, how translation quality is measured, and how your own reviewers can help rather than slow things down.

It focuses on process. What ISO certificates prove about that process is covered in ISO 17100 vs ISO 9001.

What is translation quality assurance?

Translation quality assurance (QA) is the combination of human checks and automated checks that a provider runs to make sure a translation is accurate, complete, consistent and fit for its purpose before delivery. It is different from quality evaluation, which measures the quality of a finished translation, for example by scoring a sample.

What are the steps in translation quality assurance?

A professional translation passes through several distinct checks, each designed to catch a different type of error. The terms below are used widely in the industry, though providers sometimes use them loosely, so ask which steps a quote includes.

  1. Check (self-revision). The translator re-reads their own translation against the source before handing it on.
  2. Revision. A second qualified linguist compares the translation with the source sentence by sentence, checking meaning, omissions, terminology and style.
  3. Automated QA. Software checks numbers, dates, tags, punctuation, untranslated segments and glossary terms.
  4. Review. A subject-matter expert, such as a clinician or engineer, reads the target text alone for domain accuracy and suitability.
  5. Proofreading of the final layout. The formatted document, website or app is checked for line breaks, truncation and typography.

The revision step is the core safeguard. The Association of Translation Companies (ATC) notes in its article on the revision of ISO 17100 that “ISO 17100 prescribes that a second person should revise the translation to ensure quality assurance.”

What does each QA step catch, and what does it miss?

No single check catches every error, which is why professional QA layers them. The table shows the typical strengths and blind spots.

QA stepCatchesMisses
Translator self-checkTypos, obvious slips, missed sentencesThe translator's own misunderstandings
Bilingual revisionMistranslations, omissions, wrong terms, styleLayout problems; some domain-specific errors
Automated QANumber and date mismatches, tag errors, glossary breaches, double spacesFluent but wrong meaning
Expert reviewDomain accuracy, clinical or technical suitabilityOmissions, because the source is not compared
Layout proofreadTruncated text, broken hyphenation, wrong charactersMeaning errors in the text itself

How is translation quality measured?

Translation quality is measured most reliably by counting errors, classifying them by type and weighting them by severity. ISO 5060:2024 gives guidance on this; its ISO catalogue entry says it “focuses on an analytic translation evaluation approach using error types and penalty points configured to produce an error score and a quality rating.” The standard applies to human translation, post-edited machine translation and unedited machine translation output.

A widely used open framework is Multidimensional Quality Metrics (MQM), which describes itself as “a framework for analytic Translation Quality Evaluation” and uses a hierarchical catalogue of error types such as accuracy, fluency, terminology and style. A simplified severity scale looks like this:

  • Critical: could cause harm, legal exposure or a product recall, for example a wrong dosage or a reversed safety instruction.
  • Major: changes the meaning or would confuse the reader.
  • Minor: a small grammar, punctuation or style issue that does not affect meaning.
  • Preferential: a reviewer's preference rather than an error; it should be recorded as a style-guide decision, not counted against the translator.

What role should the buyer play in translation quality?

The buyer's contribution to quality happens mostly before translation starts and at the review stage. The Institute of Translation and Interpreting (ITI) observes in its guide Translation: Getting it Right that “No one reads your texts more carefully than your translator”, so translator queries often expose ambiguities in the source that you can fix.

Useful buyer actions are:

  • supply reference material, a glossary and a style guide (see translation glossaries and style guides);
  • answer translator queries quickly and share the answers with all languages;
  • brief in-country reviewers on scope, timing and the difference between errors and preferences;
  • send reviewer changes back to the provider so the translation memory is updated.

Common mistakes in translation quality assurance

The most common mistake is assuming that “translated and proofread” means a second linguist compared the translation with the source. Others to avoid:

  • Relying on monolingual review only. A reader without the source cannot see omissions.
  • Using an untrained colleague as reviewer. Preferential edits can introduce errors and inconsistency.
  • Skipping the layout check. Truncated buttons and broken characters appear after the text leaves the translator.
  • Judging quality by a single sentence. Use a sample and a severity scale, not impressions.
  • Not feeding back. Errors not reported to the provider recur in the next project.

QA also depends on good inputs; see how to prepare your files for translation. For the full selection process, read how to choose a translation agency.

How TranslationWorks can help

TranslationWorks is a UK-based translation agency that provides professional human translation by native-speaking linguists in 150+ languages and is registered to ISO 9001 for quality management and ISO 27001 for information security. TranslationWorks is certified to ISO 17100, the international standard for translation services. When you request a quote, tell us how the translation will be used, whether it will be published or submitted, and whether you have in-country reviewers, so the quality steps can be agreed with you at the outset. Our translation services cover legal, medical, technical, e-learning and marketing content, with sector expertise such as medical and pharmaceutical translation. Request a free quote within 24 hours.

Need a translation that will be published or submitted?

Tell us how the translation will be used and who will review it, and we will agree the quality steps in your free quote.

Keep reading

More from TranslationWorks

Related reading

Get a quote Call