Despite observed good performances, the use of vocal systems in realistic conditions appears to be not completely satisfactory. This is due to the fact that researchers have mainly focused their efforts on speech recognition difficulties, minimizing the importance of both user-system interaction and recognition-understanding interaction. This paper attempts to show that, in task-oriented dialogues, the knowledge concerning the task, the dialogue and the user, can be used in a both predictive and corrective way, in order to increase the performance of a vocal dialogue system. Predictions allow the system to reduce the recognition search space and therefore to prune the bad solutions. Correction strategies are used after the recognition phase to correct erroneous messages.
Cite as: Matrouf, K., Néel, F. (1991) Use of upper level knowledge to improve human-machine interaction. Proc. 2nd VENACO Workshop - The Structure of Multimodal Dialogue, 75-88
@inproceedings{matrouf91_smmd,
author={K. Matrouf and Francoise Néel},
title={{Use of upper level knowledge to improve human-machine interaction}},
year=1991,
booktitle={Proc. 2nd VENACO Workshop - The Structure of Multimodal Dialogue},
pages={75--88}
}