Model-Based Personalization within an Adaptable Human-Machine Interface Environment that is Capable of Learning from User Interactions

The paper describes a multimodal interface architecture that is capable of automatic adaptation to the user by learning from the user's interaction patterns. The introduced architecture consists of a general way to specify multimodal HMI systems based on models with modes, transitions and guards as well as a mechanism to apply adaptations during the runtime process. Adaptations are defined as modifications of models which are specified during the specification process and used during the runtime process to control the program flow. To provide the developer with a flexible but secure way to define personalization uses cases in form of adaption rules we introduce modification boundaries that are defined as an additional model based on the same formalism as the models used for the program flow. The framework will be discussed by means of an example: Analyzing the interaction of the user after a recognition of a yet unknown speech command to infer and apply adequate modifications of the model to connect the unknown speech command with a typical user interaction.