Automated essay scoring for Bangla

Abstract

Grading papers is one of the most basic everyday tasks carried out in various manners, but the element of complexity always manages to find its way. It is a rigorous task to grade hundreds of papers. Still, the concept of automation has made the job easier as the process decreases the risk of error in checking the papers while simplifying the lives of the teachers. Now, in the case of the English language, this simpleness has already been achieved. However, reaching an equivalent level of sophistication in the context of grading papers in Bangla is still an ongoing process. A team from BUET has researched this very topic in Bangla, but the tools required for grading a paper in Bangla are still far from reaching a distinctive platform. In our research, we have collected datasets containing versatile content to build a competent database and have analyzed the requirements teachers used to grade a paper using natural language processing (NLP) tools. After listing the criteria, we fine-tuned a model using deep learning, in accordance with the criteria to grade a paper written in Bangla with enough accuracy to be considered as relevant as having the same paper graded manually by a professor or a faculty. Our goal is to use transformer models, and embedding along with NLP techniques to grade the essays more precisely, to achieve an industry-standard state-of-the-art system for the Bangla Essay Scoring System.