Automated Dialogue Generation for Behavior Intervention on Mobile Devices

Communication in the form of dialogues between a virtual coach and a human patient (coachee) is one of the pillars in an intervention app for smartphones. The virtual coach is considered as a cooperative partner that supports the individual with various exercises for a behavior intervention therapy. To perform its supportive behavior, the coach follows a certain interaction model and its requirements, such as alignment, mutual commitment and adaptation. In this paper, we propose E-Coach MarkUp Language (ECML), a standard XML specification for scripting discourses that define how the virtual coach maintains a dialogue with a coachee following the interaction model. The format of the language allows messages to be tailored at a fine-grained level. Each sentence is synthesized based on the inferred goals of the coaching process and the current beliefs of the user, incorporating everything that has been said previously in the conversation. The design enables inexpensive implementation on mobile devices for a flexible, seamless coaching dialogue. With expert-based evaluations, we validated the language using scenarios on implemented ECML in the field of insomnia therapy.