Machine translation quality systems often favor literal idiom translations
The Lay of the Land in Idiom-Translation Quality Estimation
Computation and Language
Summary
Idioms are tricky for automatic translators because they don't translate literally. The authors built a new dataset comparing literal and idiomatic translations of English idioms in nine languages. They tested 21 quality scoring systems and found most prefer literal translations over correct idiomatic ones. Only a few newer, larger models correctly rate idioms higher and can detect subtle differences in how idiomatic a translation is. This shows common systems might mistakenly reward bad, literal translations in texts with many idioms.
What this means in practice
- •For machine translation developers: Improve quality estimation tools to better assess idiomatic translations in multilingual systems.
- •For content localization teams: Avoid relying on standard quality scores that may favor literal translations when localizing idiomatic content for multiple languages.
Authors
Malik Marmonier, Benoît Sagot, Rachel Bawden
Abstract
Idioms have long been a challenge for machine translation, and while large language models (LLMs) have proven increasingly adept at handling them, whether quality estimation (QE) systems have kept pace and can reliably score idiomatic renderings of these expressions above faulty literal ones remains unclear. We introduce the first parallel dataset pairing idiom-bearing English sentences with both literal and idiomatic translations across nine target languages, augmented into multi-sentence segments and a graded literality continuum. Using this dataset, we benchmark eleven conventional QE metrics from the COMET and MetricX families and ten LLMs from the Qwen-3, Gemma-4 and OLMo-3 families used as GEMBA-style judges. We find that most systems, especially older metrics and smaller models, display a marked literality bias, rating literal translations of idioms above idiomatic ones. Higher parameter counts and newer training sets appear to reduce this bias to an extent; nevertheless, of all the benchmarked systems, only MetricX-24-Hybrid (XL, XXL) and Gemma-4-31B consistently rate idiomatic translations of idioms above literal ones, and only Gemma-4-31B reliably ranks intermediate degrees of idiomaticity and literality. Our findings caution against deploying standard QE metrics in domains rich in figurative language, where higher scores might be rewarding undesirable literalism.