Co-speech Gesture Generation for Embodied Agents and its Effects on User Evaluation - Coverbal Synchrony in Human-Machine Interaction - page 238

making to account for both inter-individual differences in gesture
use, as well as common patterns of how meaning is mapped onto
gesture form. In addition, it accounts for the fact that the physical
appearance of generated gestures is influenced by multiple factors
such as characteristic features of the referent accounting for iconicity,
contextual linguistic/discourse factors, or a speaker's previous gesture
use. Evaluations show that the gestures producible with this model
can reasonably approximate a certain range of human iconic gesturing
and are acceptable to human users.
The results demonstrate that automatically generated gestural
behavior is actually beneficial with regard to the impact of embodied
agents on human addressees. Notably, different models were found to
result in different behavior, with consistently differing perception and
evaluation by human recipients. As a consequence, it seems reasonable
to detect particularly appropriate speakers and to individualize an
embodied agent's communicative behavior accordingly. This does, of
course, raise the question concerning the characteristics of 'successful'
gesture style. At this point, the potential of embodied agents comes
to the fore as they provide the flexibility to turn on and off aspects
of the behavior model to observe how human addressees respond.
Individualization of communicative behavior, however, bears the
danger of narrowing acceptance down to a certain population of
users, since gesture perception, like production, may be subject to
inter-individual differences. For instance, Martin et al. (2007) found
the rating of gestural expressivity parameters to be influenced by
a human addressee's personality traits. Accordingly, an important
lesson to be learned, therefore, concerns the role of evaluation studies
as an integral part of the communicative behavior modeling process.
While prediction accuracy is highly prized in many evaluations of
behavior simulation, the impact of how humans perceive an embodied
agent's expressive behavior should always be a major criterion to help
producing adequate behavior and increase the acceptance of the agent.
Acknowledgements
This research is partially supported by the Deutsche
Forschungsgemeinschaft (DFG) in the Collaborative Research Center
673 “Alignment in Communication” and the Center of Excellence 277
in “Cognitive Interaction Technology” (CITEC).