Transformers in Bangla Text: Advancing MCQ and Answer Generation Methods

Multiple-choice questions (MCQs) serve as a vital tool in educational assessments, not only in competitive exams but also in technology-driven learning environments. However, automating the generation of high-quality MCQs from Bangla text presents a complex challenge due to the language’s intricate grammar, contextual dependencies, and limited resources for natural language processing. While significant progress has been made in automatic MCQ generation for other languages, no substantial work has addressed Bangla, leaving a critical gap in the literature. In this research, we present a novel approach for generating MCQs along with correct answers from Bangla text using transformer-based models, trained on a dataset of 10,135 MCQs from 2,024 passages. Among various models, the fine-tuned Llama-3.1-8B model showcased superior efficacy, achieving success rate of 93.5% in generating relevant MCQs and 100% accuracy in generating the correct answers demonstrating key features such as contextual integrity, syntactic correctness, and relevance of both questions and distractors. This approach outperforms existing models in other languages, providing a robust solution for MCQ generation in Bangla text.