Feedback source modality effects on training outcomes in a serious game: Pedagogical agents make a difference

The aim of this research is to enhance game-based training applications to support educational events in the absence of live instruction. The overarching purpose of the presented study was to explore available tools for integrating intelligent tutoring communications in game-based learning platforms and to examine theory-based techniques for delivering explicit feedback in such environments. The primary tool influencing the design of this research was the open-source Generalized Intelligent Framework for Tutoring (GIFT), a modular domain-independent architecture that provides the tools and methods to author, deliver, and evaluate intelligent tutoring technologies within any instructional domain. Influenced by research surrounding social cognitive theory and cognitive load theory, the resulting experiment tested varying approaches for utilizing an Embodied Pedagogical Agent (EPA) to function as a tutor during interaction in a game-based training environment. Conditions were authored to assess the tradeoffs between embedding an EPA directly in a game, embedding an EPA in GIFT's browser-based Tutor-User Interface (TUI), or using audio prompts alone with no social grounding. The resulting data supported the application of using an EPA embedded in GIFT's TUI to provide explicit feedback during a game-based learning event. Analyses revealed conditions with an EPA situated in the TUI to be as effective as embedding the agent directly in the game environment.