Cluster and Time-Series Analyses of Computer-Assisted Pronunciation Training Users: Looking Beyond Scoring Systems to Measure Learning and Engagement

摘要

The present study utilized hierarchical agglomerative cluster (HAC) analysis to categorize users of a popular, web-based computer-assisted pronunciation training (CAPT) program into user types using activity log data. Results indicate an optimal grouping of four types: Reluctant, Point-focused, Optimal, and Engaged. Clustering was determined by aggregate data on seven indicator variables of mixed types (e.g., ratio, continuous, and categorical). It was found that measurements of effort: lines recorded and episodic effort served best to distinguish the user types. Subsequent time-series analysis of cluster members showed that groupings exhibited distinct trends in learning behavior which explain performance outcomes. Four waves of data were collected during one semester of EFL instruction wherein CAPT usage partially fulfilled course requirements. This study follows an exploratory, data-driven approach. In addition to the findings above, suggestions for future research into interactions between individual differences variables and CALL platforms are made.