In this article, we discuss how to control nonverbal conversational signals (NCS) including eye gazes, nods and facial expressions displayed by a pair of embodied agents in dialogue. Works on information agent proposes to employ embodied agents for presenting information to the viewers. Since embodied agents have faces and bodies in order to be "embodied", the viewers read various meanings in the NCS on the faces and the bodies of the agents, even if the agents are not actually designed to send NCS to their viewers but only to speak. In order to cope with this problem, it has been proposed in some previous work to make the NCS appropriate for the speech utterances or the task of the agent. However, when two embodied agents are in dialogue, we also need to consider interdependences between NCS displayed by those agents. We discuss how to maintain those interdependencies for producing dialogues between a pair of embodied agents for information presentation.