The complexity of Multi-Agent Systems is constantly increasing. With the growth of the number of agents, interactions between them draw complexan d huge conversations, i.e. sequences of messages exchanged inside the system. In this paper, we present a knowledge discovery process, mining those conversations to infer their underlying models, using stochastic grammatical inference techniques. We present some experiments that show the process we design is a good candidate to observe the interactions between the agents and infer the conversation models they build together.