This paper presents a conceptual framework for the study of social intelligence in a real-life environment. It is focussed on the dialogue organisation in argumentation, in particular how our understanding of dialogue phenomena in mediated communication may help us to support natural interaction in classroom debates. Dialogue organisation is explored in terms of the cohesive structure of dialogue that emerges as the result of information maintenance and change, specified locally by the adjacency pair and turn-taking, and globally by topic threads. We give a descriptive analysis of their interplay in real chat debates, based on lexical chains of repeated words in an attempt to clarify what feedback participants in the debate require in order to reflect on their contributions to the debate. We then show how this analysis can be used as a basis for conceptual design of a feedback adviser tool to automatically search for topics in student chat debates. Much of the data used in our analysis has been generated as part of the SCALE project ¹ focussed on specifying tools for computer supported argumentation in the classroom.