Lost in Translation? LLMs, Education, and Linguistic Fairness

Recent advancements in Large Language Models (LLMs) have demonstrated remarkable capabilities, particularly in English prompts. However, lingering questions persist regarding these models’ ability to grasp the subtleties of language beyond mere rule recognition.As these models are increasingly used across various settings, particularly by students and trainees worldwide, a key concern is their ability to understand and accurately translate nuanced English, especially in tasks like translation and academic assistance, where English often dominates as the primary language. In this study, we collaborated as a group of researchers native to diverse linguistic backgrounds to assess LLMs’ aptitude and challenges in translating English idioms and context-rich phrases into 25 languages in a cross-cultural context. Our focus included idioms, gendered sentences, and contextually ambiguous cases. We identified significant patterns in LLMs’ performance on idiomatic expressions across languages. These patterns show substantial disparities and gender bias in translations, particularly disadvantaging low-resource languages. These findings raise concerns about these models worsening educational inequities and call for targeted improvements to ensure fairness and inclusivity. By revealing both limitations and strengths, our work emphasizes the need for a deeper understanding of language nuances to support multiculturalism, expand access to quality models for non-English speakers, and drive equitable progress in education and beyond.

Authors

Institutions

Publication Details

Published
2025-03-15
DOI
https://doi.org/10.1109/isec64801.2025.11147325
Primary Topic
Legal Language and Interpretation
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Lost in Translation? LLMs, Education, and Linguistic Fairness

I. L. Dmitrieva, Ashwin De Silva, Amir Hossein Daraie, Eva Yezerets et al.
Legal Language and Interpretation
article

Lost in Translation? LLMs, Education, and Linguistic Fairness

I. L. Dmitrieva, Ashwin De Silva, Amir Hossein Daraie, Eva Yezerets, Emma J. Jones, Adrián Perez Diéguez, Alice Amitrano, Kun Liang, Silja Sormunen, Alan C. Love, Noga Mudrik, Xiaoxuan Lei, Claire Lemaitre, Carolyna Yamamoto Alves Pinto, Lou C. Kohler Voinov, Jerry Huang, Jayanta Dey, Liang Xiang, Anh Duong Vo, Dhwani Vaishnav, Helen Yezerets, Ram Dyuthi Sristi, Ruthie Yezerets, Vatanak Lavan, Alekhya Dronavalli
article en

Abstract

Recent advancements in Large Language Models (LLMs) have demonstrated remarkable capabilities, particularly in English prompts. However, lingering questions persist regarding these models’ ability to grasp the subtleties of language beyond mere rule recognition.As these models are increasingly used across various settings, particularly by students and trainees worldwide, a key concern is their ability to understand and accurately translate nuanced English, especially in tasks like translation and academic assistance, where English often dominates as the primary language. In this study, we collaborated as a group of researchers native to diverse linguistic backgrounds to assess LLMs’ aptitude and challenges in translating English idioms and context-rich phrases into 25 languages in a cross-cultural context. Our focus included idioms, gendered sentences, and contextually ambiguous cases. We identified significant patterns in LLMs’ performance on idiomatic expressions across languages. These patterns show substantial disparities and gender bias in translations, particularly disadvantaging low-resource languages. These findings raise concerns about these models worsening educational inequities and call for targeted improvements to ensure fairness and inclusivity. By revealing both limitations and strengths, our work emphasizes the need for a deeper understanding of language nuances to support multiculturalism, expand access to quality models for non-English speakers, and drive equitable progress in education and beyond.

SIL International (US), Johns Hopkins University (US), Lawrence Berkeley National Laboratory (US), American University (US), California State University Los Angeles (US), United States Pharmacopeial Convention (US), ETH Zurich (CH), Paramo (Czechia) (CZ), The University of Texas at San Antonio (US), McGill University (CA), Indiana University (US), Université de Montréal (CA), Rice University (US), Tsinghua University (CN), Aalto University (FI)
Openalex Percentile: Top 63%
Legal Language and Interpretation
DOISourceOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.