On the Linguistic and Pedagogical Quality of Automatic Question Generation via Neural Machine Translation

Tim Steuer, Leonard Bongard, Jan Uhlig, Gianluca Zimmer

Abstract: Allowing learners to self-assess their knowledge through questions is a well-established method to improve learning. However, many educational texts lack a sufficient amount of questions for self-studying. Hence, learners read texts passively, and learning becomes inefficient. To alleviate the lack of questions, educational technologists investigate the use of automatic question generators. However, the vast majority of automatic question generation systems consider English input texts only. Therefore, we propose a simple yet effective multilingual automatic question generator based on machine-translation techniques. We investigate the linguistic and pedagogical quality of the generated questions in a human evaluation study.