A method is developed for the dialogue record management and the response content generation and is implemented on a computer as the dialogue processing subsystem for the speech response system. In order to realize a feasible system for the speech communication between human and machine, several functions are required for the subsystem. Since the current technology for continuous speech recognition is limited, the subsystem is designed to operate only with the information on the content words that may be obtained at a high recognition rate by the word spotting techniques. As for the configuration, the word dictionary and the topic-dependent rules are constructed to suit the topic specified by the knowledge base. Other parts of the subsystem can be constructed independently from the topic. This configuration leads to the high adaptability to the dialogue topic of the subsystem. A prototype for the dialogue processing subsystem is constructed, and smooth dialogue between user and system is performed, indicating the validity of the method.