What Would a Teacher Do? Predicting Future Talk Moves

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4739–4751 August 1–6, 2021. ©2021 Association for Computational Linguistics 4739 What Would a Teacher Do? Predicting Future Talk Moves Ananya Ganesh, Martha Palmer and Katharina Kann University of Colorado Boulder {ananya.ganesh, martha.palmer, katharina.kann}@colorado.edu Abstract Recent advances in natural language process- ing (NLP) have the ability to transform how classroom learning takes place. Combined with the increasing integration of technology in today’s classrooms, NLP systems leverag- ing question answering and dialog processing techniques can serve as private tutors or par- ticipants in classroom discussions to increase student engagement and learning. To progress towards this goal, we use the classroom dis- course framework of academically productive talk (APT) to learn strategies that make for the best learning experience. In this paper, we in- troduce a new task, called future talk move pre- diction (FTMP): it consists of predicting the next talk move – an utterance strategy from APT – given a conversation history with its cor- responding talk moves. We further introduce a neural network model for this task, which outperforms multiple baselines by a large mar- gin. Finally, we compare our model’s perfor- mance on FTMP to human performance and show several similarities between the two. 1 Introduction The field of natural language processing (NLP) has made rapid progress over the last few years (Wang et al., 2019). Success on natural language under- standing, dialogue generation, and question an- swering tasks has spurred advances in NLP-based systems for educational applications. (McNamara et al., 2013; Litman, 2016; Burstein et al., 2020). Systems that can simulate human teachers in spe- cific situations such as small-group discussions have the potential to aid learning by promoting student engagement. Research has shown that deep conceptual learn- ing is heightened when students are active partici- pants in the classroom and contribute to discussions with their questions and ideas (McNamara, 2011; Bransford et al., 1999). However, large class sizes Scenario: Learning about proportional relationships in a classroom. The teacher gives an example of toasting two slices in a toaster, for 2 minutes. Teacher: So we’ve just seen that 2 slices of toast gets done in 2 minutes. (None) Teacher: What if I had 3 slices of toast? (Press for Accu- racy) Student: 4 minutes! (Wait) Teacher: Why would it take 4 minutes? (Press for Reason- ing) Student: Because you’d have to use the toaster twice. (Wait) FTMP: Getting Students to Relate (e.g., who else agrees it would be 4?) Table 1: Our proposed FTMP task; the teacher talk move corresponding to each utterance is shown in parentheses. often make it difficult for all students to actively participate. Discussions in sub-groups increase each student’s speaking time, but, in turn, make it impossible for a single teacher to guide all in- dividual conversations. In this paper, we present a first step towards a system that can solve this problem by taking the teacher’s role in facilitating sub-group discussions. For this, we turn to a classroom discourse frame- work by Michaels …

Content truncated to 500 words