CodeMixBench: Evaluating Code-Mixing Capabilities of LLMs Across 18 Languages

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yang, Yilun, Chai, Yekun
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908523808423936
author Yang, Yilun
Chai, Yekun
author_facet Yang, Yilun
Chai, Yekun
contents Code-mixing, the practice of switching between languages within a conversation, poses unique challenges for traditional NLP. Existing benchmarks are limited by their narrow language pairs and tasks, failing to adequately assess large language models' (LLMs) code-mixing abilities. Despite the recognized importance of code-mixing for multilingual users, research on LLMs in this context remains sparse. Additionally, current techniques for synthesizing code-mixed data are underdeveloped to generate code-mixing. In response, we introduce CodeMixBench, a comprehensive benchmark covering eight tasks, including three specific to LLMs and five traditional NLP tasks, and 18 languages across seven language families. We also propose a new method for generating large-scale synthetic code-mixed texts by combining word substitution with GPT-4 prompting. Our evaluation reveals consistent underperformance of LLMs on code-mixed datasets involving different language families. Enhancements in training data size, model scale, and few-shot learning could improve their performance. The code and dataset are available at https://github.com/Jeromeyluck/CodeMixBench.
format Preprint
id arxiv_https___arxiv_org_abs_2507_18791
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CodeMixBench: Evaluating Code-Mixing Capabilities of LLMs Across 18 Languages
Yang, Yilun
Chai, Yekun
Computation and Language
Code-mixing, the practice of switching between languages within a conversation, poses unique challenges for traditional NLP. Existing benchmarks are limited by their narrow language pairs and tasks, failing to adequately assess large language models' (LLMs) code-mixing abilities. Despite the recognized importance of code-mixing for multilingual users, research on LLMs in this context remains sparse. Additionally, current techniques for synthesizing code-mixed data are underdeveloped to generate code-mixing. In response, we introduce CodeMixBench, a comprehensive benchmark covering eight tasks, including three specific to LLMs and five traditional NLP tasks, and 18 languages across seven language families. We also propose a new method for generating large-scale synthetic code-mixed texts by combining word substitution with GPT-4 prompting. Our evaluation reveals consistent underperformance of LLMs on code-mixed datasets involving different language families. Enhancements in training data size, model scale, and few-shot learning could improve their performance. The code and dataset are available at https://github.com/Jeromeyluck/CodeMixBench.
title CodeMixBench: Evaluating Code-Mixing Capabilities of LLMs Across 18 Languages
topic Computation and Language
url https://arxiv.org/abs/2507.18791