Electrophysiologically-inspired evaluation of dialogue act complexity

Abstract

This paper proposes a computationally appropriate approach to evaluation of the dialogue act complexity in human-machine interaction inspired by a functional interpretation of the N400 and P600 components of electroencephalographic signals. Two main parameters of the dialogue act complexity are: the retrieval cost and the integration cost. At the conceptual level, the retrieval cost of a dialogue act reflects the cognitive load related to the retrieval of lexical information from long-term memory necessary to process the dialogue act. The integration cost of a dialogue act reflects the cognitive load related to integration difficulty caused by an incomplete, ambiguous or semantically irregular dialogue act. The proposed approach is illustrated for a dedicated therapeutic scenario including speech-based three-party interaction between the child, the therapist, and the robotic system with an integrated conversational agent.