Pedagogical Agents in Constructivist Multimedia

Objectives and Theoretical Framework What are the cognitive consequences of introducing the image of a pedagogical agent in a discovery-based multimedia learning environment? How should pedagogical agents communicate with students to promote constructivist learning? In order to answer these questions, we conducted two sets of studies. First, we varied whether the agent's words were presented as speech or on-screen text and whether or not the agent's image appeared on the screen, both with an animated fictional agent (Experiment 1), and a video of a human face (Experiment 2). Next, we varied whether or not the agent's words were presented in a self-referenced or neutral style (i. e. as dialogue or monologue) both using speech (Experiment 3), and on-screen text (Experiment 4). The dissemination of software agents in instructional design brings up the need to investigate the effects of agents' presence in human-computer interaction. Software pedagogical agents may carry the modeling, scaffolding, and coaching attributes of successful mentors in apprenticeship learning (Collins, Brown, & Newman, 1989). However, in addition to the considerations about what type of advise and feedback the agent should give to the student, it is necessary to investigate how the agent should look and talk to the student to convey effectively his message. For example, as mentors are usually physically present during the apprenticeship interaction, software agents may carry the attribute of visual presence (Towns et al., 1998). In addition, as mentors do not merely send out messages into the void to be picked up by students but rather interact with students in the process of knowledge construction, software agents may carry the attribute of human communication (Tharp & Gallimore, 1991). Regarding the image and voice of the personal agent, we tested the predictions arising from two opposing hypothesis: the social-cue hypothesis, and the interference hypothesis. According to the former, students learn better and rate more favorably computer lessons that include social cues-such as facial expressions or human voices than those which do not include social cues. On the other hand, according to the interference hypothesis, social cues may be entertaining but interfere with the student's ability to make sense out of the presented materials (Moreno & Mayer, in revision-a; Harp & Mayer 1998). Given the constraints on the learner's working memories, instructional programs which include the additional information contained in the agent's image or voice-called seductive details-hinder learning (Moreno & Mayer, in press). Regarding the language style …