Do Multilingual LLMs Think In English?
This reveals a hidden bias in multilingual LLMs that could affect users relying on non-English reasoning, making it an incremental but important insight for AI transparency.
The paper demonstrates that multilingual large language models (LLMs) internally process key decisions in a representation space closest to English, regardless of input or output language, as shown through logit lens analysis and activation steering experiments.
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.