Explainable Machine Translation for Intelligent E-Learning of Social Studies

Social Studies (SST) fosters civic knowledge of topics such as crime, economic development, and health, allowing students to make informed decisions in a culturally diverse society. Although SST is the subject with the highest performance in the Primary Leaving Examination (PLE), the Ministry of Education reported that pupils who fail do so because they are unable to interpret the questions or respond to them in a timely manner. This is due in part to a linguistic barrier, as children in areas such as rural Central Uganda speak Luganda and only utilize English at school, which affects learning. This prevents them from properly learning to understand the dialogue and exercises provided in English. Machine Translation (MT) is a powerful tool for overcoming language barriers, but the black-box structure of most MT models makes understanding the reasoning behind the translations challenging. Therefore, this paper illustrates the application of Explainable Artificial Intelli-gence(XAI) to support the translation of SST notes from Luganda to English and vice versa. Two transformer models(opus-mt-en-lg and opus-mt-lg-en models) were trained and evaluated using the BiLingual Evaluation Understudy (BLEU) score. The English-Luganda model obtained a 21.43% BLEU score while the Luganda-English model obtained a 16.87% BLEU score, which was higher than the model reference score. This work can be used as an innovative approach to Intelligent E-Learning to overcome barriers to traditional classroom-based language instruction, such as geographical distance and lack of qualified teachers, and can be used to reach a wider audience of children.