July 15, 2020 – ,
Have you ever received a grade without any explanation – just a fat 0 next to your answer? How helpful was that for being able to learn from your mistakes?
Probably not at all. The lack of individual feedback is not only a problem humans with a lot of pressure to grade answers from many students produce but also the kind of feedback you can expect from automatic short answer grading systems. However, recent advances in the explainable AI field may be able to alleviate this problem. Using gradient information or attention scores to mark the student answer's parts and how they contributed to the final predicted grade could give valuable clues as to which parts of a free-text answer were correct/incorrect. Alternatively, we could attempt to directly train a model, such as T5, to explain its decision and thus, provide the student with in-depth, individual feedback.
Keywords: Explainable AI, Automatic Short Answer Grading
Research Area(s): Knowledge & Educational Technologies
Student: Siddharth Singh Parihar