The study involves the consideration of a theory of dialogue with meta-control features that previous research have indicated as desirable for improved efficiency in human-machine voice-interactions. In particular these features include: (1) process meta-statements which overview and report progress in the interaction; (2) an automatically set level of initiative in the system; and (3) perform dialogue corrections and automatic adjustment of other parameters. The first feature serves the purpose of informing the user of the system's intentions to encourage cooperation, educating the user about the process of debugging, and maintaining morale to maximize efficiency. The second feature enables the system to take charge and lead the interaction when the user lacks the knowledge required and to release control to the user when he or she has sufficient expertise. The third is used to optimize aspects of system performance. The effects of these meta-behaviors in human-human dialogues on the efficiency and educational value of the exchange are also considered with views toward their implementation in automatic dialogue systems.
