Modeling and Tracking Surface Instructional Discourse Dynamics

In this work we focus on classroom discourse. Specifically, we describe an approach to modeling and tracking the classroom verbal interaction between teacher and students. We assume that the evolution of the classroom discourse takes place in a space whose dimensions correspond to specific discourse categories and thus the rate of occurrence of these events can be monitored, categorized and tracked. In other words, we assume that the teacher relies on a collection of verbal mechanisms in guiding the classroom discussion. The speed and effectiveness with which the discourse progresses in these dimensions (or reacts to the teacher stimuli) is thus controlled by the teacher's actions. We propose a dynamical system representation (linear Discrete Wiener Process Acceleration model under system and observation noise) to model the classroom interaction. This model can be used to track discourse, and to predict and smooth discourse trajectories with applications to problem solving, conversational analytics, dialog understanding, etc. 1 Classroom Discourse Classroom interaction constitutes arguably one of the most rich and interesting environments in which to study human communication dynamics. This is so, because the objectives and roles in a typical classroom are well defined and it's common that the participants are clearly motivated to achieve their goals (i.e., learning). Teachers are responsible for leading the learning process and they typically adjust their style to match the response of the classroom. To guide, motivate, prompt and teach, teachers rely mostly on verbal interaction. There has been substantial previous work on education-theoretical based analysis of instructional discourse (e.g.,[2,3,4,5]). From those studies, it is clear that there not only there are multiple frameworks and styles for teaching, but also many frameworks available for the analysis and interpretation of the role of discourse in the process of learning. To constrain the scope of the paper we focus on the constructivism perspective to education and future work can focus on other perspectives and include other interaction modalities. Detailing the history and foundations of constructivism is beyond the scope of this paper, so we refer the reader to [4] for overview and [2,5] as examples of discourse analysis in education. From the educational-oriented research we borrow two ideas: (a) that the teacher uses a well defined set of dialog mechanisms (i.e., the classroom discourse is based on a taxonomy of events [3]) and (b) that the teacher carries out the task of leading and directing the learning process by using mechanisms in this taxonomy. In future work we can extend these ideas to analyze the extent in which learning is internalized and reflected by the students in their Workshop on Modeling Human Communication Dynamics at NIPS 2010 Page 18