A pair of words in two different languages that look or sound similar but have different meanings, creating a risk of mistranslation when their resemblance is mistaken for shared meaning.
The term comes from the French phrase faux amis, meaning “false friends,” first used in linguistics in the 1920s by French linguists Maxime Koessler and Jules Derocquigny. The idea is that these words appear friendly and familiar, like a word you already know, but they will mislead you if you trust them.
False friends arise in two ways. Some share a common ancestor language but drifted apart in meaning over centuries of separate use. Others are purely coincidental, two unrelated languages happened to develop similar-sounding words that mean entirely different things. Either way, the practical problem for translators and localization teams is the same: a word that looks like it needs no translation actually does.
In everyday conversation, a false friend might produce an embarrassing moment. In a localized product, it can produce a serious error that reaches thousands of users before anyone notices.
A translator under time pressure, or an MT engine without sufficient context, may see a familiar-looking word and carry it across without checking. The result is a translation that reads fluently but says something unintended — or worse, something offensive. For example, the Spanish word embarazada looks like “embarrassed” in English but means “pregnant.” The German word Gift looks like the English word for a present but means “poison.” The English word “sympathy” and the German Sympathie sound nearly identical but carry different weight: in English it implies pity or sorrow, while in German it simply means liking someone.
These errors are particularly hard to catch in review because the translated text often reads naturally in the target language. Nothing looks broken. The mistake is in the meaning, not the grammar.
The most effective protection is prevention at the source. Adding known false friend pairs to a project glossary with clear approved translations, and flagging them for translators in the CAT tool, ensures that the correct meaning is applied consistently regardless of who is doing the translation or whether MT is involved.
Style guides and translator briefings should also call out high-risk false friends for specific language pairs, especially in domains where mistranslation has consequences such as medical, legal, or safety-related content in particular.
QA tools that check against approved glossary terms can catch cases where a false friend has been used instead of the approved translation before the content goes live.