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

Abstract

The study of emotions within the language sciences has been an area of scholarly interest since the 1880s. Emotion analysis in this field primarily examines the expression of emotions in various texts, encompassing a broad spectrum from online commentary to classical literature. Recent years have seen an increased emphasis on the detection and analysis of emotions within children's literature. This burgeoning interest is motivated by the recognition that a deeper understanding of the emotional layers embedded in children's stories can greatly enhance the insights of educators and caregivers into the emotional development and experiences conveyed through these narratives. While the majority of research in this field has concentrated on the analysis of emotion in monolingual datasets, efforts to explore emotion within bilingual contexts, such as in translated children’s literature, are relatively rare. To address this gap, this paper firstly compiles a bilingual Chinese-English dataset of emotions from a parallel Chinese-English classical children’s literature corpus. Then, the dataset is fine-tuned and evaluated on different Large Language Models (LLMs). The fine-tuning results indicate that the GPT-3.5-turbo model surpasses other language models, reaching its best performance with an F1 of 0.869. This performance denotes not only the feasibility of Chinese-English bilingual emotion detection, but also the applicability of this modelled dataset for future Chinese-English emotion detection tasks.