Pedagogical agents that support learning by explaining: Effects of affective feedback

Pedagogical agents that support learning by explaining: Effects of affective feedback Yugo Hayashi (yhayashi@fc.ritsumei.ac.jp) Mariko Matsumoto (is039081@ed.ritsumei.ac.jp) Hitoshi Ogawa (ogawa@airlab.ics.ritsumei.ac.jp) College of Information Science and Engineering, Ritsumeikan University, 1-1-1 Nojihigashi, Kusatsu, Shiga, 525--8577, Japan Abstract The present study investigates how a conversational agent can facilitate explanation activity. An experiment was conducted where pairs of participants, who were enrolled in a psychology course, engaged in a task of explaining to their partners the meanings of concepts of technical terms taught in the course. During the task, they interacted with a conversational agent, which was programmed to provide back-channel feedbacks and metacognitive suggestions to encourage and facilitate conversational interaction between the participants. Results of an experiment suggested that affective positive feedbacks from conversational agent facilitate explanation and learning performance. It is discussed that a conversational agent can play a role for pedagogical tutoring and triggers a deeper understanding of a concept during an explanation. Keywords: pedagogical affective learning. agents; explanation activities; Introduction The ever-evolving information and communication technology has made it possible to support human cognition by using systems which aids human interaction. Many researchers in computer science are tackling on the theme of developing embodied conversational agents to support education. Recent studies on cognitive science and learning science show that collaborative learning facilitates understanding or acquisition of new concepts depends greatly on how explanations are provided. In this study a collaborative activity of making explanation is experimentally investigated by using a conversational agent that serves as a teaching assistant. The goal of the experiment is to find out what kind of feedback from the agents is most conducive to successful learning performance. Related work Explanations during collaborative activities Number of studies on collaborative problem solving in cognitive science revealed how concepts are understood or learned. Researchers have shown that asking reflective questions for clarification to conversational partners is an effective interactional strategy to gain a deeper understanding of a problem or a concept (e.g. Chi, Bassok, Lewis, Reimann, & Glaser, 1989; Miyake, 1986; Salomon, 2001; Okada & Simon, 1997). It has also been demonstrated that the use of strategic utterances such as asking for explanation or providing suggestions can stimulate reflective thinking and meta cognition involved in understanding a concept. Playing different roles during explanation is also said to help problem solvers reconstruct external representation and concepts (Shirouzu, Miyake, & Masukawa; 2002). Studies that are discussed above suggest that how well one can explain is the key to understanding and learning of a concept. However, explanation becomes successful if people have difficulties in retrieving and associating relevant knowledge required for explanation activity. Researches on collaborative learning have reported that these difficulties rise among novice problem solvers (Coleman, 1998; King, 1994). Also, it may not help learn a concept if people cannot communicate with each other as in when, for example, they use technical terms or phrases unknown to others (Hayashi & Miwa, 2009). It is assumed that one of the ways to help collaborative problem solvers is to introduce a third-person or a mentor who can facilitate the task by using prompts such as suggestions and back-channels. However, it is often difficult for one teacher to monitor several groups of collaborators and to supervise their interaction during explanation in actual pedagogical situations. Recently there are studies which demonstrate that the use of conversational agents that act as educational companions or tutors can facilitate learning process (Holmes, 2007; Baylor & Kim, 2005). Unfortunately, it has not been fully understood if and what kinds of support by such agents would be more helpful for collaborative learners. In this paper, the author will further investigate this question through the use affective expressions. Pedagogical conversational agents as learning advisers Researchers in the field of human computer interaction have conducted a number of experimental studies which involve the use of pedagogical agents (e.g. Kim, Baylor & Shen, 2007; Reeves & Nass, 1996; Graesser & McNamara, 2010). One point to be taken into consideration in studies of human performance is the affective factor. This factor influences people's performance in either negative or positive ways and