Predicting the Constructiveness of Student-Student Classroom Conversations: An Exploratory Study on Uptake

• In Event: Data Analysis and Statistics

Abstract

Conversational uptake is an important phenomenon emphasizing the co-construction of knowledge in classrooms. Historically, uptake has been measured by qualitative methods. Recently, however, computation and language modeling have been applied to analyze audio-recorded classroom conversations. We built on a study seeking to predict and evaluate uptake in teacher-student conversations using both word-overlap and word vector-based similarity measures. We applied similar methods to student-student conversations and found some disparities: (1) the embedding-based similarity model—deriving from BERT-pretrained contextualized representations—outperformed the word-overlap baseline, and (2) measures performed better with text including stop words, highlighting the role of function words in linguistic alignment. Our work suggests that students’ uptake may take on different linguistic patterns than teachers’ and require more contextualized modeling.

Authors

• Hsiaolin Hsieh, Stanford University
  Presenting Author

• Victoria L. Delaney, Stanford University
  Presenting Author

• Simon Wiles
  Non-Presenting Author