Abstract. We have developed an artificial agent based on a computational model of peer learning we developed. That model shows that shifts in initiative are conducive to learning. The peer learning agent can collaborate with a human student via dialog and actions within a graphical workspace. This paper describes the architecture and implementation of the agent and the user study we conducted to evaluate the agent. Results show that the agent is able to encourage shifts in initiative in order to promote learning and that students learn using the agent. Key words: Peer Agent, Knowledge Co-construction, Initiative 1

For tutorial dialogue systems, selecting an appropriate dialogue move to support learners can significantly influence cognitive and affective outcomes. The strategies implemented in tutorial dialogue systems have historically been based on handcrafted rules derived from observing human tutors, but a data-driven model of strategy selection may increase the effectiveness of tutorial dialogue systems. Tutorial dialogue projects including CIRCSIM-TUTOR [1], ITSPOKE [2], and KSC-PAL [3] have utilized corpora to inform the behavior of a system. Our work builds on this line of research by directly learning a hierarchical hidden Markov model (HHMM) for predicting tutor dialogue acts within a corpus. The corpus was collected during a human-human tutoring study in the domain of introductory computer science [4]. We annotated the dialogue moves with dialogue acts (Table 1). The subtask structure and student problem-solving action correctness were also annotated manually.

A central challenge for tutorial dialogue systems is selecting an appropriate move given the dialogue context. Corpus-based approaches to creating tutorial dialogue management models may facilitate more flexible and rapid development of tutorial dialogue systems and may increase the effectiveness of these systems by allowing data-driven adaptation to learning contexts and to individual learners. This paper presents a family of models, including first-order Markov, hidden Markov, and hierarchical hidden Markov models, for predicting tutor dialogue acts within a corpus. This work takes a step toward fully data-driven tutorial dialogue management models, and the results highlight important directions for future work in unsupervised dialogue modeling.