Cross-Lingual Sentiment Analysis in Literary Translation: A Case Study of the Novel Crystal Boys

Sentiment analysis, which is a subdomain of natural language processing (NLP), has attracted significant interest due to its extensive applications. However, its utilization in literary works, particularly in the context of translated novels, remains largely uncharted territory. This research introduces an innovative approach to exploring the emotional landscape of literary texts across different languages through the automated classification of emotion. This approach is applied to the translation of the novel Crystal Boys (孽子), which is a seminal work in lesbian, gay, bisexual and transgender (LGBT) literature authored by Pai Hsien-Yung (白先勇). The research employs a variant of Bidirectional Encoder Representations from Transformers (BERT), which is a multilingual language model, and fine-tunes it on the GoEmotions dataset. This dataset, which includes examples from 28 categories of emotion, is extended to include Chinese text via machine translation. The model is trained using a process with two stages: model training and sentiment analysis. The model was initially trained on the GoEmotions dataset, whose content was in English and machine-translated Chinese and represented a cross-language corpus. These data were divided into labeled and unlabeled sets to train and refine the model, respectively. The resulting model, which was capable of identifying a wide range of emotions in both English and Chinese text, was applied to the text of Crystal Boys. The sentiment analysis showed that the emotional profiles significantly varied between original Chinese novel and its English translation. The results provided a unique perspective on the emotional dynamics within the novel and highlight the potential for automated emotion classification to analyze literary works.