Affiliation:
1. University of Jyväskylä, Finland
2. University of Chile, Chile
3. McMaster University, Canada
Abstract
In this study, classroom interaction is addressed by using automatic speech recognition (ASR) for automatic analysis of teacher talk. Educational research on classroom interactions and talk is mainly based on manual analysis, which, although quite accurate, is time consuming and expensive. As there are already many ways to harness the power of automatic approaches to analyse the content of talk, the next step is the automatic detection of different types of talk. In this study, the latter includes the automatic detection of teacher questions and their differentiation into closed and open types. The data consists of ASR text outputs of 25 physics lessons on the same topic, each 90 minutes long, from 25 different science teachers. The interplay between human- and machine-led approaches resulted in promising steps taken towards automatization. Future possibilities for automatic analysis of classroom talk and the implications for teaching and learning are discussed.