Towards Style Alignment in Cross-Cultural Translation

TL;DR

This paper introduces RASTA, a retrieval-augmented method that improves cross-cultural translation by aligning stylistic and cultural norms, addressing biases and style misalignments in large language models.

Contribution

The paper presents RASTA, a novel approach that leverages learned stylistic concepts to enhance style and cultural norm alignment in LLM translations.

Findings

RASTA reduces style bias in translations.

Improves translation quality for non-Western languages.

Enhances cultural communication in multilingual settings.

Abstract

Successful communication depends on the speaker's intended style (i.e., what the speaker is trying to convey) aligning with the listener's interpreted style (i.e., what the listener perceives). However, cultural differences often lead to misalignment between the two; for example, politeness is often lost in translation. We characterize the ways that LLMs fail to translate style - biasing translations towards neutrality and performing worse in non-Western languages. We mitigate these failures with RASTA (Retrieval-Augmented STylistic Alignment), a method that leverages learned stylistic concepts to encourage LLM translation to appropriately convey cultural communication norms and align style.