Investigating Self-Regulated Learning in Teachable Agent Environments

Abstract

We have developed a computer-based learning environment that helps students learn science by constructing causal concept map models. The system builds upon research in learning-by-teaching (LBT) and has students take on the role and responsibilities of being the teacher to a virtual student named Betty. The environment is structured so that successfully instructing their teachable agents requires the students to learn and understand the science topic for themselves. This learning process is supported through the use of adaptive scaffolding provided by feedback from the two agents in the system: the teachable agent, Betty, and a mentor agent, Mr. Davis. For example, if Betty performs poorly on a quiz, she may tell the student that she needs to learn more about the topics on which she is performing poorly. In addition, Mr. Davis may suggest that students ask Betty questions and get her to explain her answers to help them trace the causal reasoning chains in their map and find out where she may be making mistakes. Thus the system is designed to help students develop and refine their own knowledge construction and monitoring strategies as they teach their agent.This chapter provides an overview of two studies that were conducted in fifth-grade science classrooms. A description of the analysis techniques that we have developed for interpreting students’ activities in this learning environment is also provided. More specifically, we discuss the generation of hidden Markov models (HMMs) that capture students’ aggregated behavior patterns, which form the basis for analyzing students’ metacognitive strategies in the system. Our study results show that students who utilized LBT versions of our system performed better than students who used a non-teaching version of the system. Further, students’ performances were strongest when the system explicitly provided support to help them develop self-regulated learning strategies. To gain further insight into the students’ reactions to feedback from the two agents, we present results from a second study that employed a think-aloud protocol. Overall, the results from this study illustrated that students were more receptive to the explicit strategy-oriented feedback from the mentor agent. Interestingly, this study also suggested that students had difficulty in correctly applying Betty’s feedback related to metacognitive monitoring activities.