Educational Question Generation of Children Storybooks via Question Type Distribution Learning and Event-centric Summarization

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 5073 - 5085 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Educational Question Generation of Children Storybooks via Question Type Distribution Learning and Event-Centric Summarization Zhenjie Zhao♠ ♡ Yufang Hou♣ Dakuo Wang♢ Mo Yu⋆∗Chengzhong Liu♦ Xiaojuan Ma♦ ♠ Nanjing University of Information Science and Technology ♡ Nankai University ♣ IBM Research Europe ♢ IBM Research ⋆ WeChat AI ♦ The Hong Kong University of Science and Technology zzhaoao@nuist.edu.cn, yhou@ie.ibm.com, dakuo.wang@ibm.com moyumyu@tencent.com, chengzhong.liu@connect.ust.hk, mxj@cse.ust.hk Abstract Generating educational questions of fairytales or storybooks is vital for improving children’s literacy ability. However, it is challenging to generate questions that capture the inter- esting aspects of a fairytale story with edu- cational meaningfulness. In this paper, we propose a novel question generation method that first learns the question type distribution of an input story paragraph, and then sum- marizes salient events which can be used to generate high-cognitive-demand questions. To train the event-centric summarizer, we fine- tune a pre-trained transformer-based sequence- to-sequence model using silver samples com- posed by educational question-answer pairs. On a newly proposed educational question- answering dataset FairytaleQA, we show good performance of our method on both automatic and human evaluation metrics. Our work in- dicates the necessity of decomposing ques- tion type distribution learning and event-centric summary generation for educational question generation. 1 Introduction Listening to and understanding fairy tales or story- books are very crucial for children’s early intellec- tual and literacy development (Sim and Berthelsen, 2014). During the storybook reading process, prompting suitable questions with educational pur- poses can help children understand the content and inspire their interests (Zevenbergen and Whitehurst, 2003; Ganotice et al., 2017). There is evidence that high-cognitive-demand (HCD) questions usually relate to good learning achievement (Winne, 1979). HCD questions usu- ally correspond to application, analysis, synthe- sis, and evaluation questions in Bloom’s taxon- omy of cognitive process (Winne, 1979; Ander- son et al., 2000), which are salient events merged from different elements across a session (Greatorex ∗This work was done while Mo was at IBM Research. and Dhawan, 2016). However, it is challenging even for humans to ask educationally meaningful questions to engage children in storybook read- ing, which could be due to adults lacking the skills or time to integrate such interactive opportunities (Golinkoff et al., 2019). Recent research shows that AI-powered conversational agents can play the role of language partners to read fairy tales to children and ask them educational questions (Xu et al., 2021). This motivates us to investigate tech- niques to generate HCD educational questions for children’s storybooks automatically. Automating the generation of such questions can have great value in supporting children’s language develop- ment through guided conversation. During storybook reading, HCD questions re- quire children to make inferences and predictions. In contrast to low-cognitive-demand (LCD) ques- tions describing facts in stories (e.g., Who is Snow White’s mother?), HCD questions are often related to events and their relations (e.g., Why did the queen …

Content truncated to 500 words