Note: Towards Devising an Efficient VQA in the Bengali Language

Abstract

Designing and implementing visual question answering tasks using Bengali datasets and native VQA based smart systems are important, as a huge number of people speak in Bengali who are relatively less advanced to technology adoption due to the language barrier. The important designing and implementing tasks are little explored in the literature. Therefore, we attempt to investigate the tasks in depth in this study. To do so, we follow a step-by-step procedure for overcoming different barriers encountered while adopting datasets as well as creating our own Bengali CLEVR and Bengali VQA. We perform different sets of experiments to demonstrate the efficacy of our proposed approach. Various VQA-based smart systems for Bengali speakers covering virtual doctors, navigation systems, smart glasses for the visually-impaired people, and so on can be benefited from this study through making the applications usable and understandable to those who are not fluent in a foreign language such as English.