Towards Automatic Analysis of Science Classroom Talk: Focus on Teacher Questions

Introduction

This chapter seeks to answer the call to reduce the complexity and laboriousness of classroom interaction analysis by modernising and complementing human analysis with novel computational methods. The role instruction and classroom interactions play in the learning process has been addressed in quantitative studies (Fischer et al., 2014; Howe et al., 2019; Müller et al., 2016), but the nature of this connection has not yet been sufficiently addressed due to the complexity and temporal nature of classroom interactions and the laborious methods involved in their analysis. As Hennessy et al. (2020) noted, the time is right for researchers, practitioners, and professional development leaders to explore modern and complementary computer-supported approaches for analysing and developing dialogue in classrooms. In this light, approaches that transcend human boundaries when studying classroom interactions in science will be explored. The specific focus is the automatic detection of teacher questions in the context of science. The focus on teacher questioning moves is rationalised by the essential role questions play in knowledge building and more engaging discursive activities developing reasoning and argumentation skills (O’Connor & Michaels, 2019; Sohmer et al., 2009).

Approaches to harness the power of automatic speech recognition (ASR) in analysis of content and semantic meaning (i.e., what is said) have been developed (Song et al., 2021); the next step is the automatic detection of pragmatic features (Nikula, 2005) of teacher talk (i.e., how it is said). In this study, the latter is addressed by targeting the focus on teacher questioning within the context of science classrooms. The theoretical background is foregrounded by sociocultural (Vygotsky, 1978) and dialogic (Bakhtin, 1986) perspectives highlighting the essentiality of social interactions. There is some evidence stressing that how scientific concepts are introduced in relation to one another makes a difference in terms of student learning (Viiri & Helaakoski, 2014). As shown later through more automated approaches, these results align with the results of automatic network analysis (Caballero et al., 2017). Similarly, an automatic keyword centrality approach—where three centrality measures were used to describe teachers’ discourse—conformed to these findings (Schlotterbeck et al., 2020). These automated approaches made it possible to automatically identify differences between teachers by studying the way the physics concepts are presented during the lesson. By applying these measures, a new way of describing the content of teachers’ talk was then introduced. In general, these findings underline the cruciality of the way the concepts are introduced and discussed in subject disciplines, such as science, as part of knowledge building (Barreto et al., 2021; Viiri & Helaakoski, 2014). In other words, it makes a difference how facts and concepts are interconnected and connected in relation to one another in condition and time (Badreddine & Buty, 2011; Barreto et al., 2021) and what kind of communication is used in knowledge building (Bereiter & Scardamalia, 2014).