English grammar multiple-choice question generation using Text-to-Text Transfer Transformer

Abstract

English grammar multiple-choice questions (MCQs) can be automatically generated to reduce preparation time. Previous studies have focused on semiautomated methods based on the transformation of human-made sentences/articles into MCQs, owing to which the number of generated questions is dependent on the size of a given text corpus. This study proposes an artificial intelligence-assisted MCQ generation system that increases the number of generable questions using controllable text generation techniques. In this system, the questions for MCQs are generated using a text generation model trained using the Text-to-Text Transfer Transformer (T5) architecture, a powerful deep learning model for performing text generation tasks, with a keyword and a part-of-speech (POS) template as the input for content and grammar topic control. For the text-to-MCQ transformation process, answer-to-MCQ and distractor-selection strategies are proposed for 10 grammar topics using rule-based algorithms. The quality of the generated MCQs is evaluated by human experts. The acceptance rate of the questions generated using the proposed system is 86%. The controllability of content and grammar topics are 96.86% and 98.57%, respectively. The findings of this study show that the T5 model achieves good performance in terms of controlling the POS structure in a keyword-to-text generation task. Moreover, the good acceptance rate indicates that artificial intelligence has the potential to help teachers speed up the process of selecting examination questions. We also discuss extending the proposed system to other grammar topics and the limitations of the proposed system.