Translation QA

Translation QA, short for translation quality assurance, is the set of checks a translation goes through before anyone relies on it. It asks whether the translation says what the source says, whether it reads as written in the target language, and whether it works where it will be used, which on a website means on the page itself.

The three kinds of check#

Linguistic review is a person who knows both languages reading the translation against the source. Frameworks give that review a shared vocabulary. MQM, the Multidimensional Quality Metrics framework stewarded by the MQM Council, has an open typology of error types, with accuracy, fluency, terminology and style among them, and evaluators mark each issue they find with its type and a severity. Counting issues that way makes two reviews, or two vendors, comparable.

Automated checks catch what people miss when they read for sense. Typical ones compare numbers between source and target, look for text left untranslated, check that required terms were used, and make sure markup and placeholders survived.

In-context review looks at the translation where readers will see it. A label that is correct in a spreadsheet can be cut off on a button, wrap badly in a menu, or sit next to a picture that contradicts it.

A worked example#

This is a worked example. A shop's English banner reads "Free shipping on orders over €50", and its machine translation into Italian came back as "Spedizione gratuita per ordini superiori a 500 €". The sentence is fluent and every word is correctly chosen, so a reviewer reading only the Italian would pass it. A check against the source finds the price changed from 50 to 500, which MQM-style review files under accuracy and any QA process treats as serious, because it changes an offer.

Translation QA in LangRelay#

LangRelay has a check of each kind for a website's translations, and free checks hold a page with untranslated text, the wrong language or a changed price before it goes live.

Relay reviews any page you ask it to, the way a language expert would, and tells you the cost in AI credits before it starts. A Relay review reads one page in one language against the source and records findings. A finding is major when the translation is wrong for a visitor: a changed number, price, date or name, a lost or added negation, missing or added content, text left untranslated, or a broken tag. It is minor when the translation is correct but could read better, such as awkward phrasing, a line in a different register from the rest of the page, or punctuation. A Relay review recommends pass, fail or uncertain; it never approves or changes anything.

The glossary marks a term Missing for any language that has no translation or Keep as source rule for it yet, so gaps show before the next translation runs. In the Visual editor a reviewer reads each page at desktop, tablet and phone widths and edits text in place. After a translation goes live, a delivery check reads the live page and confirms its translated text, lang attribute, canonical, hreflang and robots rules.

Back-translation covers an older check and its limits, and post-editing website copy covers the human review of machine output. The translation quality page shows the controls together.