The Role of Why Questions in Effective Human Tutoring C. P. Ros´ e, D. Bhembe, S. Siler, R. Srivastava,K. VanLehn Learning Research and Development Center, University of Pittsburgh, Pittsburgh PA, 15260 Abstract. It is undoubtedly true that one prominent component of effective human tutoring is collaborative dialogue between student and tutor [7, 12]. Nevertheless, many impor- tant questions remain to be answered about which features of human tutorial dialogue make it effective and how the most effective human tutoring strategies can be imple- mented in a tutorial dialogue system. In this paper we present an analysis of a corpus of human tutoring dialogues where we examine which features correlate significantly with learning gains. In particular we explore the role of Why questions and other open ended questions in creating opportunities for student learning as well as the role of explicit negative feedback for wrong answers. 1 Introduction It is undoubtedly true that one prominent component of effective human tutoring is collab- orative dialogue between student and tutor [7, 12]. Because of this, in recent years a great deal of interest in building tutorial dialogue systems has developed, and many of the resulting systems have been evaluated successfully with students [15, 6, 5]. Nevertheless, it is still true that important questions remain to be answered about what it is that makes human tutorial dialogue effective and how these effective human tutoring strategies can be implemented in a tutorial dialogue system. Previous studies have argued the effectiveness of Socratic and other similar tutoring ap- proaches that encourage students to be actively involved in the conversation and say as much as possible. A previous study [16] revealed a trend for Socratic style tutoring dialogues to be more effective for learning than didactic style ones. The rationale behind this result is that students learn more effectively when they are given the opportunity to discover knowledge for themselves [1, 11, 13]. Collins and Stevens (1982) report that the best teachers tend to use a Socratic tutoring style. Recent research on student self-explanations supports the view that when students explain their thinking out loud it enhances their learning [3, 4, 14]. In support of this, in this paper we present an analysis that shows a significant correlation between aver- age student turn length and learning. This leaves open the question of how best to encourage self-explanation from students. In a recent study [8], a computer system was able to elicit ef- fective self-explanations from students as well as human tutors did without understanding the explanations given by students. However, the system used in that study was used solely for the purpose of eliciting self-explanations from students and not offering any feedback. How- ever, in a typical dialogue situation with a human tutor, the tutor offers the student feedback on what the student says. Our hypothesis is that in a dialogue situation where the tutor offers