Composable Multimodal Dialogues Based on Communicative Acts

摘要

In Social Robotics, being able to interact with users in a natural way is a key feature. To achieve this, we need to model dialogues that allow the robot to complete its tasks and to adapt to unforeseen changes in the conversation. We present an approach where these dialogues are modelled as a combination of basic interaction units, called Communicative Acts or CAs. With this, our system aims to provide all the necessary tools so each of the robot's applications can tailor their own dialogues in a simplier way. These applications make the decisions that need task-related information and request the activation of the CAs in order to create complex dialogues. The CAs handle decisions that require communication-related information (e.g. giving the user some information, or asking a question). They also manage some of the problems that can appear in any interaction, like not being able to understand the other peer, or not getting an answer to a question. A case study associated to a cognitive stimulation exercise is presented in this paper to validate our system.