Many kinds of expert systems have been developed in various fields to date. To realize interaction through spoken Japanese between a user and such an expert system, we are currently developing a dialog management system called MASCOTS in addition to the speech understanding system and the speech synthesizer. MASCOTS manages the complex flow of dialog using two stacks and plan information. It sends useful information tailored in appropriate forms to the expert system and helps the language processing system by predicting the next user utterance. This paper describes the architecture of MASCOTS focusing on exchanges of the information with the language processing system.