Mining Gap-fill Questions from Tutorial Dialogues

Gap-fill questions are fill-in-the-blank questions which consist of a sentence with one or more gaps (blanks) and a number of choices for each gap. Such questions play crucial roles in creating test materials and tutorial dialogues. In this paper, we present a system that automatically generates such questions by exploiting previously recorded student-tutor interactions with an Intelligent Tutoring System. Our method is novel because it relies on mining questions’ distractors, i.e. tempting incorrect answers, from tutorial dialogues unlike most of the existing approaches that rely on instructional contents. Experimental results show that the proposed system can generate high quality gap-fill questions.