This paper presents a system for the analysis of discourse threading in chats based on the polyphonic model. The approach is based on the investigation into the usage of cue phrases as a basis to determine rhetorical schemas implying frequent words. The goal is to offer a basis for analyzing discourse threading, inter-animation, cohesion and coherence in "Computer - Supported Collaborative Learning" ( CSCL ) chats, according to the ideas of Mikhail Bakhtin's dialog theory and Stefan Trausan-Matu's polyphonic model. The used chat environment is that used in the Virtual Math Temas ( VMT ) project, which supports and helps the appearance of multiple threads by providing explicit reference facilities. In our system, we first analyze the dialogue for the identification of its topics and discussion threads (repeated words).Then we such find cue phrases and use them to detecting rhetorical schemata. Furthermore we detect the links between threads that are important for CSCL in order to analyze the involvement of students in the chat and their ability to problem solving. Additionally, we shall study links resulting from the model in order to generate be student feedback. The distribution of rhetorical schemata may be an indicate of cohesion and coherence .The paper analyzes chats from the K-Teams laboratory at University Politehnica of Bucharest . The system is using Stanford NLP software in order to process chats : tokenization, stemming, part of speech tagging, remove stop word, parsing, named-entity recognition ( NER ) and the rhetorical relations that are used.

Discourse analysis, chat, dialogism, polyphony, Computer-Supported Collaborative Learning, Rhetorical relation.