Enhance Incomplete Utterance Restoration by Joint Learning Token Extraction and Text Generation

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 291 - 309 July 10-15, 2022 ©2022 Association for Computational Linguistics Towards Process-Oriented, Modular, and Versatile Question Generation that Meets Educational Needs Xu Wang Simin Fan Jessica Houghton Computer Science and EngineeringLu Wang University of Michigan Ann Arbor, MI {xwanghci, oliviaaa, houghj, wangluxy}@umich.edu Abstract NLP-powered automatic question generation (QG) techniques carry great pedagogical poten- tial of saving educators’ time and benefiting student learning. Yet, QG systems have not been widely adopted in classrooms to date. In this work, we aim to pinpoint key impediments and investigate how to improve the usability of automatic QG techniques for educational purposes by understanding how instructors con- struct questions and identifying touch points to enhance the underlying NLP models. We perform an in-depth need finding study with 11 instructors across 7 different universities, and summarize their thought processes and needs when creating questions. While instructors show great interests in using NLP systems to support question design, none of them has used such tools in practice. They resort to multiple sources of information, ranging from domain knowledge to students’ misconceptions, all of which missing from today’s QG systems. We argue that building effective human-NLP col- laborative QG systems that emphasize instruc- tor control and explainability is imperative for real-world adoption. We call for QG systems to provide process-oriented support, use modular design, and handle diverse sources of input. 1 Introduction Decades of educational research has shown the ben- efit of active learning in improving students’ learn- ing outcomes where students actively engage with the materials, e.g., through question answering and problem solving (Crouch and Mazur, 2001; Deslau- riers et al., 2019; Koedinger et al., 1997), compared with passively receiving lectures or reading texts (Chi and Wylie, 2014; Freeman et al., 2014). How- ever, most instructors still use traditional teaching methods, e.g., lecturing, in large-enrolment college courses (Henderson and Dancy, 2007; Handelsman et al., 2004; Stains et al., 2018), due to their lim- ited time and resources in creating opportunities for active learning (Dancy and Henderson, 2007; Silverthorn et al., 2006; Fagen et al., 2002). Mean- while, there is a growing interest in leveraging NLP methods to serve educational needs. Specifically, automatic question generation (QG) is promising in enhancing students’ active learning experience by creating problem-solving activities at scale (Al- subait et al., 2016). Still, the adoption of automatic QG systems in classrooms is low (Kurdi et al., 2020; Alsubait et al., 2016), mainly because those models are only suit- able for specific domains, e.g., language learning, and the generated questions are often of low qual- ity and limited in types and difficulty levels (Kurdi et al., 2020; Alsubait et al., 2016). For QG systems to provide more meaningful support to instructors’ work, we need to address at least two fundamen- tal challenges. First, designing questions to assist learning is a highly complex, creative, and knowl- edge intensive process. Typical college instructors have received years of …

Content truncated to 500 words