A Diverse and Explainable Multi-hop QA Dataset for Bengali Language

Abstract

Bengali is a resource-scare language with a scarcity of quality data sets both in single and multi-hp question answering. In an approach to fill that gap, we want to take a little step by generating a reading comprehension based open-domain multi-hop question answering which will be explainable and diverse. We will generate about 100 passages from news and Wikipedia articles and 500 question-answer pairs. We will maintain the diversity in selecting domains of contexts and also in generating questions and answers. Our data set will be explainable in generating the answer to a given question by providing supporting facts and showing the reasoning chain