Beneath the Surface of Consistency: Exploring Cross-lingual Knowledge Representation Sharing in LLMs

TL;DR

This paper investigates how multilingual large language models represent and share factual knowledge across languages, revealing that script similarity influences sharing and proposing methods to measure and improve this capability.

Contribution

The study introduces a novel methodology to measure cross-lingual knowledge sharing in LLMs and provides empirical insights into factors affecting this sharing, such as script similarity.

Findings

High consistency does not guarantee shared representation.

Script similarity significantly influences knowledge sharing.

Full sharing could improve accuracy by up to 150%."

Abstract

The veracity of a factoid is largely independent of the language it is written in. However, language models are inconsistent in their ability to answer the same factual question across languages. This raises questions about how LLMs represent a given fact across languages. We explore multilingual factual knowledge through two aspects: the model's ability to answer a query consistently across languages, and the ability to ''store'' answers in a shared representation for several languages. We propose a methodology to measure the extent of representation sharing across languages by repurposing knowledge editing methods. We examine LLMs with various multilingual configurations using a new multilingual dataset. We reveal that high consistency does not necessarily imply shared representation, particularly for languages with different scripts. Moreover, we find that script similarity is a dominant factor in representation sharing. Finally, we observe that if LLMs could fully share knowledge across languages, their accuracy in their best-performing language could benefit an increase of up to 150\% on average. These findings highlight the need for improved multilingual knowledge representation in LLMs and suggest a path for the development of more robust and consistent multilingual LLMs.