Multilingual Language Models are not Multicultural: A Case Study in Emotion

TL;DR

This paper investigates whether multilingual language models capture cultural variations in emotional expression and finds they predominantly reflect Western norms, highlighting a gap in culturally sensitive AI.

Contribution

The study reveals that current multilingual LMs are Anglocentric and do not effectively learn culturally appropriate emotional nuances, suggesting the need for improved cultural sensitivity.

Findings

Embeddings from multilingual LMs are Anglocentric.

Generative LMs like ChatGPT reflect Western emotional norms.

Multilingual LMs fail to capture cultural variations in emotion.

Abstract

Emotions are experienced and expressed differently across the world. In order to use Large Language Models (LMs) for multilingual tasks that require emotional sensitivity, LMs must reflect this cultural variation in emotion. In this study, we investigate whether the widely-used multilingual LMs in 2023 reflect differences in emotional expressions across cultures and languages. We find that embeddings obtained from LMs (e.g., XLM-RoBERTa) are Anglocentric, and generative LMs (e.g., ChatGPT) reflect Western norms, even when responding to prompts in other languages. Our results show that multilingual LMs do not successfully learn the culturally appropriate nuances of emotion and we highlight possible research directions towards correcting this.