Bilingual Sentiment and Emotion Analysis: A Multi-Task Learning Framework for Bengali and English

Multi-task learning (MTL), a powerful paradigm in the field of machine learning enables us to learn and handle multiple different tasks simultaneously.Numerous advantages and novel approaches of MTL inspire us to analyze how MTL performs for the English language as well as for low-resourced languages such as Bengali.This paper proposes a fusion-based bilingual MTL framework for sentiment and emotion classification in Bengali and English languages with the help of transformer-based multilingual BERT, MuRIL and IndicBERT models.By training the MTL frameworks with around 32.5K Bengali and English texts acquired from publicly available datasets, our fusion-based best-performing MTL framework achieves a macro F1 score of 71.14 and 38.92 for sentiment and emotion classification in the Bengali language, and 67.09 and 83.48 for sentiment and emotion classification in the English language.