Do Multilingual Language Models Think Better in English?

TL;DR

This paper introduces self-translate, a method enabling multilingual language models to translate and infer within the model itself, significantly improving performance over direct inference in non-English languages.

Contribution

The paper presents self-translate, a novel approach that eliminates external translation systems by using the model's own few-shot translation capabilities, enhancing multilingual inference.

Findings

Self-translate outperforms direct inference across 5 tasks.

Language models struggle to leverage their full multilingual potential without self-translation.

Self-translate demonstrates consistent performance improvements.

Abstract

Translate-test is a popular technique to improve the performance of multilingual language models. This approach works by translating the input into English using an external machine translation system, and running inference over the translated input. However, these improvements can be attributed to the use of a separate translation system, which is typically trained on large amounts of parallel data not seen by the language model. In this work, we introduce a new approach called self-translate, which overcomes the need of an external translation system by leveraging the few-shot translation capabilities of multilingual language models. Experiments over 5 tasks show that self-translate consistently outperforms direct inference, demonstrating that language models are unable to leverage their full multilingual potential when prompted in non-English languages. Our code is available at https://github.com/juletx/self-translate.