Starting from dialogism in which every act is perceived as a dialogue, we shift the perspective towards multi-participant chat conversations from Computer Supported Collaborative Learning in which ideas, points of view or more generally put voices interact, inter-animate and generate the context of a conversation. Within this perspective of discourse analysis, we introduce an implemented framework, ReaderBench, for modeling and automatically evaluating polyphony that emerges as an overlap or synergy of voices. Moreover, multiple evaluation factors were analyzed for quantifying the importance of a voice and various functions were experimented to best reflect the synergic effect of co- occurring voices for modeling the underlying discourse structure.