Crosslingual Capabilities and Knowledge Barriers in Multilingual Large Language Models
This addresses the problem of crosslingual knowledge barriers in LLMs for AI researchers and developers, representing an incremental improvement through fine-tuning methods.
The study evaluated state-of-the-art multilingual large language models (LLMs) on crosslingual tasks, finding they struggle with deeper knowledge transfer across languages, and proposed fine-tuning on mixed-language data to reduce these gaps effectively.
Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora. But can these models relate corresponding concepts across languages, i.e., be crosslingual? This study evaluates state-of-the-art LLMs on inherently crosslingual tasks. We observe that while these models show promising surface-level crosslingual abilities on machine translation and embedding space analyses, they struggle with deeper crosslingual knowledge transfer, revealing a crosslingual knowledge barrier in both general (MMLU benchmark) and domain-specific (Harry Potter quiz and TOFU benchmark) contexts. Since simple inference-time mitigation methods offer only limited improvement, we propose fine-tuning of LLMs on mixed-language data, which effectively reduces these gaps, even when using out-of-domain datasets like WikiText. Our findings suggest the need for explicit optimization to unlock the full crosslingual potential of LLMs. Our code is publicly available at https://github.com/google-research/crosslingual-knowledge-barriers.