Poster
Mancini, Maurizio
(LINC - Université Paris 8, Paris)
Hartmann, Björn
(Stanford University, Palo Alto)
Pelachaud, Catherine
(LINC - Université Paris 8, Paris)
Gesture Expressivity in Embodied
Conversational Agent
Embodied Conversational Agents (ECAs) are a powerful user interface
paradigm, aiming at transferring the inherent richness of
human-human interaction to human-computer interaction. ECAs
are virtual embodied representations of humans that communicate
with the user (or other agents) through different communicative
channels (called modalities) : voice, facial expression, gaze, gesture,
and body movement. Effectiveness of an agent is dependent
on her ability to suspend the user’s disbelief during an interaction.
To increase believability and life-likeness of an agent, we seek to
move away from a generic acting agent model and, instead, to
simulate individualized agents that portray idiosyncratic behaviors.
Human individuals differ not only in their reasoning, their set of
beliefs, goals, and their emotive states, but also in their way of
expressing such information through the execution of specifi c behaviors.
Based on what we should call “behavioral infl uences”, the same
kind of information should be conveyed by the agent using one or
more modalities, on which different kinds of signals will be transmitted
using the proper degree of intensity. We will use the work
“expressivity” to refer to all of these kinds of behavioral differences.
In this paper, we present our model of capturing expressivity in
human gesturing and propose a set of parameters to characterize
this individual variability in a conversational agent generation system.
We then suggest a mapping of the identifi ed dimensions of
expressivity onto particular sets of animation parameters for gesture
animation. We demonstrate synthesized behaviors with different
expressivity settings in our existing ECA system and present an out-look of how to integrate our work with higher-level agent functions
such as simulations of personality or emotion. We will also present
some evaluation studies we conducted to validate our model and how
these results will direct our work on expressivity in the near future.
