Many recent works on agent architectures have focused on polite and optimal turn-transitions. However, real turn-taking is more complex due to several contextual variables, linked to each agent’s own goals (cooperative or non cooperative for example). For mixed-initiative interactions, we need to go beyond the polite agent context to make more complex patterns of turn-taking emerge. We present here an architecture based on a dynamical and continuous model of turn-taking, able to control the turn-taking behaviors of the agent depending on its willingness to speak or not. We show how we implemented our model based on human data and how complex patterns of turn-taking emerge from agent-agent simulations. Finally, we present the results of a perceptual experiment where we questioned participants about the intentions of two agents interacting.
CITATION STYLE
Jégou, M., Lefebvre, L., & Chevaillier, P. (2015). A continuous model for the management of turn-taking in user-agent spoken interactions based on the variations of prosodic signals. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9238, pp. 389–398). Springer Verlag. https://doi.org/10.1007/978-3-319-21996-7_42
Mendeley helps you to discover research relevant for your work.