Abstract:
Bengali is a native language spoken by more than 250 million people all over the world. Building a system that can understand human language and provide meaningful responses is really important in Natural Language Processing (NLP), which is also advancing rapidly. Building a system in Bengali is difficult, as there are scarcity of data, complex morphological and versatile syntax. The project titled “Bengali Natural Language Processing: A Fine-Tuned Approach to Question Answering”, creates a QA model intended for Bengali Language. We have created a structured general knowledge based Bengali question-answer dataset which is used to fine-tune a pre-trained language model. The proposed system is a transformer based architecture to enhance the comprehension and contextual reasoning capability of Bengali QA system. With a domain specific dataset fine tuning the proposed model can improve performance in understanding of Bengali language. This work also highlights the challenges of handling limited resources and present an approach to generalize other low resource languages. The main goal is to enrich the model’s ability to understand and answer questions in Bengali with more accurate and also contributing the advancement of NLP tools for low resource language.
