Cross-Lingual Transfer of Cultural Knowledge: An Asymmetric Phenomenon

2025-06-02

AI Summary

This study identifies a systematic asymmetry in multilingual cultural knowledge transfer among large language models (LLMs): bidirectional transfer occurs between English and high-resource non-English languages, whereas low-resource languages exhibit only unidirectional transfer toward English. We formally define this asymmetry and propose a frequency-driven, interpretable hypothesis—positing that word-frequency distributions in pretraining corpora govern cross-lingual cultural knowledge migration. To test it, we develop a transparent, controllable cultural transfer analysis framework grounded in training-data provenance. Using cross-lingual knowledge probing, corpus-level frequency statistics, and a multilingual benchmark covering four non-Anglo cultural domains, we empirically validate the hypothesis. Quantitative evaluation across 12 language pairs reveals statistically significant asymmetry in seven pairs. Our work establishes a theoretical foundation for culturally equitable multilingual model design and delivers an interpretable diagnostic toolkit for analyzing cultural knowledge transfer.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Knowledge Representation Languages

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Despite substantial research efforts evaluating how well large language models~(LLMs) handle global cultural diversity, the mechanisms behind their cultural knowledge acquisition, particularly in multilingual settings, remain unclear. We study this question by investigating how cultural knowledge transfers across languages during language adaptation of LLMs. We introduce an interpretable framework for studying this transfer, ensuring training data transparency and controlling transfer effects. Through a study of four non-Anglophonic cultures, we observe bidirectional cultural transfer between English and other high-resource languages, while low-resource languages primarily transfer knowledge to English with limited reverse flow. To explain this asymmetric phenomenon, we propose a frequency-based hypothesis: cultural knowledge appearing more frequently in the pretraining data transfers more easily, which is supported by empirical analysis of the training corpora.

Problem

• Mechanisms of cultural knowledge transfer in multilingual LLMs unclear
• Asymmetric cultural transfer between high and low-resource languages
• Frequency-based hypothesis explains uneven cultural knowledge distribution

Innovation

• Interpretable framework for cultural transfer study
• Bidirectional transfer in high-resource languages
• Frequency-based hypothesis for asymmetric transfer