Do Multilingual LLMs Think In English?

TL;DR

Multilingual large language models predominantly perform core reasoning and decision-making in an English-centric representation space, regardless of the input language, revealing an English bias in their internal processing.

Contribution

This paper uncovers that multilingual LLMs rely on English-like representations for reasoning, demonstrated through internal analysis and activation steering experiments, highlighting an implicit English bias.

Findings

LLMs emit English-like representations before translating to target languages.

Activation steering is more effective when computed in English.

Multilingual LLMs perform key reasoning steps in an English-shaped representation space.

Abstract

Large language models (LLMs) have multilingual capabilities and can solve tasks across various languages. However, we show that current LLMs make key decisions in a representation space closest to English, regardless of their input and output languages. Exploring the internal representations with a logit lens for sentences in French, German, Dutch, and Mandarin, we show that the LLM first emits representations close to English for semantically-loaded words before translating them into the target language. We further show that activation steering in these LLMs is more effective when the steering vectors are computed in English rather than in the language of the inputs and outputs. This suggests that multilingual LLMs perform key reasoning steps in a representation that is heavily shaped by English in a way that is not transparent to system users.