How prosody in chatbots influences communication effects
Chatbots allow communication with machines to be more human-like by their general concept: the human user does not have to interact with a database using commands, but talks to chatbots as if they were human. Speech is by now a key addition to chatbots requiring broader research. More specifically, we argue that current approaches to chatbot speech ‘ignore’ the role of a key characteristic of human speech, i.e. prosody. This project answers the question whether proper use of prosody makes speaking chatbots more human-like. WP1 involves constructing a chatbot generating answers that can be enhanced with prosody. In WP2, we apply deep learning (RNN encoder/decoders) to extract prosodic patterns from speech elicited in psycholinguistic experiments. These patterns can be applied to speech generated by WP1 chatbots, allowing them to represent hypotheses on appropriate prosody. Finally, WP3 entails an experiment to compare the chatbots in terms of human responses (e.g. anthropomorphism, concerns, trust, evaluations).
Dr. Hilde Voorveld is Associate Professor of Persuasion & New Media Technologies in the Amsterdam School of Communication Research ASCoR, Communication Science department, University of Amsterdam.
Dr. Tom Lentz is Assistant Professor of Formal Modelling of Language and Cognition at the Institute for Logic, Language and Computation, University of Amsterdam.
Prof. dr. Evangelos Kanoulas is Associate Professor at Informatics Institute, and Professor of Information Retrieval and Text Analytics at Amsterdam Business School, University of Amsterdam.