Conversational Agents and Their Longitudinal Affordances on Communication and Interaction

Abstract

In this study, we investigate the effects of conversational agents on communication and interaction when used to assist participants in developing an online portfolio. Data from 52 participants were gathered and analyzed through questionnaires, written reflections, transcripts of student-agent interactions, and focus groups. Data revealed that participants communicated with the agents on issues ranging from portfolio development to popular culture. Although participants did not view the agents as particularly helpful in completing class activities, they did use them as social companions throughout the four-week study. Implications of the findings for future design and research include: (a) learner-developed conversational agents, (b) improved " intelligence " with which agents deliver content-based knowledge, and (c) further developed virtual characters that can meet users' humanistic and utilitarian expectations. Virtual characters are becoming more common within industry and academia where they have been used in a variety of domains for multiple purposes, such as assisting learners on how to complete a task (e.g., Baylor and Ryu, 2003) or to report sports news online conducted on the use of virtual characters in educational environments although these have come to no consensus as to whether virtual characters improve learning and teaching. argued that the use of agents does not generally contribute to improved performance. Specifically, Baylor (2002) observed that student performance on the development of an instructional plan did not differ among agent conditions. Craig, Gholson and Driscoll (2002) discovered that agent properties (agent only, agent with gesture, no agent) were not able to explain differences in students' performances for retention, matching, transfer, and multiple choice questions; and Mayer, Dow and Mayer (2003) found no significant difference on problem-solving transfer performance regarding the presence of the agent's image on the screen. found that students who interacted with an agent and received text in audio format attained higher scores than those working without an agent and receiving instruction via written text. Dehn and van Mulken (2000) and Gulz (2004) examined the proposed benefits of agent-enhanced learning environments and found that the evidence for integrating agents in educational settings is at best mixed. This lack of unanimity is further complicated by differences in the design of agents and experiments (Clark and Choi, 2005). For example, Cole et al. (2003) had used an agent that was described as a disembodied head lacking natural facial expressions, while Louwerse et al. (2005) used a Conversational Agents 4 This DRAFT copy is provided …