Automatically Measuring Question Authenticity in Real-World Classrooms

• DOI: 10.3102/0013189x18785613
• Publication Date: 2018-06-29T16:34:05Z
• Authors: Sean Kelly; Andrew M. Olney; Patrick Donnelly; Martin Nystrand; Sidney K. D’Mello
• Keywords: Complement; Audio equipment

Analyzing the quality of classroom talk is central to educational research and improvement efforts. In particular, presence authentic teacher questions, where answers are not predetermined by teacher, helps constitute serves as a marker productive discourse. Further, questions can be cultivated improve teaching effectiveness consequently student achievement. Unfortunately, current methods measure question authenticity do scale because they rely on human observations or coding To address this challenge, we set out use automatic speech recognition, natural language processing, machine learning train computers detect in real-world classrooms automatically. Our were iteratively refined using audio human-coded observational data from two sources: (a) large archival database text transcripts 451 112 classrooms; (b) newly collected sample 132 high-quality recordings 27 classrooms, obtained under technical constraints that anticipate large-scale automated collection analysis. Correlations between computer-coded at level sufficiently high ( r = .602 for .687 recordings) provide valuable complement