Dialogic learning is a theoretical framework for knowledge building in collaborative learning, based on Mikhail Bakhtin’s dialogism and polyphony. This chapter focuses on the analysis of collaborative learning in small groups of students, starting from a polyphonic model, in which are identified several interanimating voices that have inherently different, conflicting ‘personalities’ but which build knowledge, achieving a coherent whole. It presents the ideas behind the introduction of the polyphonic model of discourse generated in dialogical collaborative learning sessions and its operationalization using Natural Language Processing for analysing collaboration in Computer Supported Collaborative Learning. Starting from the polyphonic model, computer tools were developed. The analysis may be done either on logs of online chats or on recorded dialogs in face-to-face settings, for example, in classrooms, in the latter case starting from both the transcripts of the discussions and the video recordings, considering also the interanimation of non-verbal acts, individual or collective.