When English Isn't the Best Teacher: Source Language Effects in Cross-Lingual In-Context Learning

2026-06-16

AI Summary

This study investigates the impact of source language selection on cross-lingual transfer performance in few-shot in-context learning (ICL), challenging conventional assumptions from the fine-tuning paradigm that prioritize linguistic similarity and data scale. Through systematic evaluation across seven tasks, six models, and typologically diverse languages, complemented by analyses of cross-lingual task performance and language interference in generation, the work reveals distinct transfer dynamics unique to ICL. The findings demonstrate that English is not universally optimal as a source language and that the influence of linguistic distance and data quantity operates differently than in fine-tuning settings. Building on these insights, the study proposes a novel heuristic for source language selection tailored to ICL, offering both theoretical grounding and practical guidance for cross-lingual few-shot learning.

Abstract

Cross-lingual transfer in multilingual NLP has been widely explored in supervised fine-tuning contexts, where factors like data availability and linguistic similarity largely determine transfer quality. As the field shifts toward few-shot In-Context Learning (ICL), it is often presumed that insights from fine-tuning carry over unchanged. Yet this assumption has not been rigorously evaluated, leaving open the question of how to choose source languages for cross-lingual ICL. We conduct a broad empirical study of cross-lingual transfer in ICL spanning seven tasks, six models, and a typologically diverse set of languages. We further analyze language confusion, a key obstacle for generative tasks in cross-lingual ICL. Our results show that conventional fine-tuning-based expectations do not consistently apply in the ICL regime and point to alternative heuristics for selecting source languages effectively.

Problem

• cross-lingual transfer
• In-Context Learning
• source language selection
• multilingual NLP
• language confusion

Innovation

• In-Context Learning
• Cross-lingual Transfer
• Source Language Selection
• Language Confusion
• Multilingual NLP