Agents are situated autonomous entities that perceive and act in their environment, and communicate with other agents. An agent usually starts a conversation by querying another agent because it needs to satisfy a specific goal. This process allocates a new goal to the agent receiving the initial query, starting new dialogs with other agents, generating a recursive interaction. The generation of this kind of dialog is interesting when the system has the possibility of generating conditional answers with imprecise and uncertain values. We consider simple deliberative rule-based agents that proactively try to satisfy their goals. The mechanism to achieve this dialogs is based in the specialization of the mental state of agents, by means of the partial deduction of rule bases.