@inproceedings{sun-etal-2026-idioms,
title = "Idioms Understood, Yet Translated Literally: Diagnosing Literal Translation Bias in Multilingual {LLM}s",
author = "Sun, Yanming and
Zhan, Runzhe and
Chao, Yu and
Ma, Jingkun and
Wang, Shuo and
Chao, Lidia S. and
Wong, Derek F.",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2619/",
pages = "59506--59529",
ISBN = "979-8-89176-459-0",
abstract = "Large language models persistently produce literal translations of idiomatic expressions, but the source of this failure remains unclear. We introduce **IdiomDIT**, a diagnostic framework that jointly evaluates idiom **D**etection, idiom **I**nterpretation, and idiomatic **T**ranslation, attributing each error to a specific source. Across six language pairs and five models spanning 4B to 70B, IdiomDIT shows that literal translation error (LTE) can persist even after successful detection and interpretation, a know-but-error pattern we term Literal Translation Bias (LTB). This pattern is direction-specific: LTB dominates some directions, while others show greater knowledge deficits, likely reflecting lower idiom exposure in pretraining. We then use linear probing to test whether LTB is already reflected in the model{'}s pre-generation hidden state, recovering per-language literal translation directions with high accuracy, and ablating this direction reduces LTE. However, once we ensure no idiom{'}s data leaks across training and test, and subtract out each idiom{'}s mean hidden state before probing, accuracy collapses, confirming the probe was largely detecting which idiom it is looking at rather than a genuine bias signal. A matched random direction also reduces LTE just as much as the identified one. We release these controls as a reusable protocol for testing whether such hidden-state attribution claims in machine translation are genuine or confounded. The code is publicly available at https://github.com/NLP2CT/IdiomDIT."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sun-etal-2026-idioms">
<titleInfo>
<title>Idioms Understood, Yet Translated Literally: Diagnosing Literal Translation Bias in Multilingual LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yanming</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Runzhe</namePart>
<namePart type="family">Zhan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Chao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jingkun</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuo</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lidia</namePart>
<namePart type="given">S</namePart>
<namePart type="family">Chao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Derek</namePart>
<namePart type="given">F</namePart>
<namePart type="family">Wong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Large language models persistently produce literal translations of idiomatic expressions, but the source of this failure remains unclear. We introduce **IdiomDIT**, a diagnostic framework that jointly evaluates idiom **D**etection, idiom **I**nterpretation, and idiomatic **T**ranslation, attributing each error to a specific source. Across six language pairs and five models spanning 4B to 70B, IdiomDIT shows that literal translation error (LTE) can persist even after successful detection and interpretation, a know-but-error pattern we term Literal Translation Bias (LTB). This pattern is direction-specific: LTB dominates some directions, while others show greater knowledge deficits, likely reflecting lower idiom exposure in pretraining. We then use linear probing to test whether LTB is already reflected in the model’s pre-generation hidden state, recovering per-language literal translation directions with high accuracy, and ablating this direction reduces LTE. However, once we ensure no idiom’s data leaks across training and test, and subtract out each idiom’s mean hidden state before probing, accuracy collapses, confirming the probe was largely detecting which idiom it is looking at rather than a genuine bias signal. A matched random direction also reduces LTE just as much as the identified one. We release these controls as a reusable protocol for testing whether such hidden-state attribution claims in machine translation are genuine or confounded. The code is publicly available at https://github.com/NLP2CT/IdiomDIT.</abstract>
<identifier type="citekey">sun-etal-2026-idioms</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2619/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>59506</start>
<end>59529</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Idioms Understood, Yet Translated Literally: Diagnosing Literal Translation Bias in Multilingual LLMs
%A Sun, Yanming
%A Zhan, Runzhe
%A Chao, Yu
%A Ma, Jingkun
%A Wang, Shuo
%A Chao, Lidia S.
%A Wong, Derek F.
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F sun-etal-2026-idioms
%X Large language models persistently produce literal translations of idiomatic expressions, but the source of this failure remains unclear. We introduce **IdiomDIT**, a diagnostic framework that jointly evaluates idiom **D**etection, idiom **I**nterpretation, and idiomatic **T**ranslation, attributing each error to a specific source. Across six language pairs and five models spanning 4B to 70B, IdiomDIT shows that literal translation error (LTE) can persist even after successful detection and interpretation, a know-but-error pattern we term Literal Translation Bias (LTB). This pattern is direction-specific: LTB dominates some directions, while others show greater knowledge deficits, likely reflecting lower idiom exposure in pretraining. We then use linear probing to test whether LTB is already reflected in the model’s pre-generation hidden state, recovering per-language literal translation directions with high accuracy, and ablating this direction reduces LTE. However, once we ensure no idiom’s data leaks across training and test, and subtract out each idiom’s mean hidden state before probing, accuracy collapses, confirming the probe was largely detecting which idiom it is looking at rather than a genuine bias signal. A matched random direction also reduces LTE just as much as the identified one. We release these controls as a reusable protocol for testing whether such hidden-state attribution claims in machine translation are genuine or confounded. The code is publicly available at https://github.com/NLP2CT/IdiomDIT.
%U https://aclanthology.org/2026.emnlp-main.2619/
%P 59506-59529
Markdown (Informal)
[Idioms Understood, Yet Translated Literally: Diagnosing Literal Translation Bias in Multilingual LLMs](https://aclanthology.org/2026.emnlp-main.2619/) (Sun et al., EMNLP 2026)
ACL
- Yanming Sun, Runzhe Zhan, Yu Chao, Jingkun Ma, Shuo Wang, Lidia S. Chao, and Derek F. Wong. 2026. Idioms Understood, Yet Translated Literally: Diagnosing Literal Translation Bias in Multilingual LLMs. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 59506–59529, Budapest, Hungary. Association for Computational Linguistics.