In this paper, we propose a new computational framework for human/agent communication. The main objective is to allow software agents to participate in flexible communications with human and to be efficient in these communications. To be flexible and efficient, artificial agents in our model are able to: 1) deal with the implicit aspects of conversations by considering the non literal level of speech acts; 2) reason on their internal states and on the conversation state using argumentation abilities; and 3) manage the social influence. Our framework is based on the philosophical foundations provided by speech act theory, argumentation theory, and social commitments.