Examining emotions in English and translated Chinese children’s literature : a bilingual emotion detection model based on LLMs

Abstract

This study investigates the Chinese-English bilingual emotion detection within the context of children’s literature. The study utilizes a parallel corpus of classical Chinese-English children’s literature and compiles a bilingual dataset of emotionally-labelled text. The dataset is then leveraged to fine-tune and evaluate the performance of various Large Language Models (LLMs). The results indicate that the GPT-4o model outperforms alternative LLMs, achieving an F1 Micro score of 0.779 and an F1 Macro score of 0.764 on the evaluation task. These findings substantiate the viability of cross-lingual emotion detection within this domain and underscore the importance of selecting appropriate pre-training techniques. Furthermore, this study addresses specific cross-cultural challenges inherent in bilingual emotion detection, elucidating the complexities posed by language-specific and culturally bound emotional expressions. This study contributes to the expanding body of literature on emotion recognition in multilingual contexts, particularly in relation to the analysis of affective content in cross-cultural translated children’s literature, and provides insights for future investigations in this field.