Spoken dialogue systems in real applications as well as research have attracted increased attention in recent years. With the limitations of current speech technologies, both for recognition and understanding and for speech generation, this interest in `real' systems has led to an increased awareness in the problems raised by system errors, especially in recognizing user input, and the consequent confusion they may lead to for both users and the system itself over the dialogue. The need to devise better strategies for detecting problems in human-machine dialogues and then dealing with them gracefully has become paramount for spoken dialogue systems.

Papers are invited on innovations in ways that systems can detect their own errors (e.g. through features such as ASR confidence scores); on methods for evaluating spoken dialogue systems that include system errors and error recovery as a major component; on strategies for determining on-line when dialogues are `going wrong'; on mechanisms for recovering once errors are detected; on laboratory and corpus-based studies of human behavior relevant to human-machine problem detection/recovery; on methods for minimizing dialogue problems (e.g. by varying dialogue strategy, system prompts). Position papers are also invited for a special session on aspects of error handling are most in need of additional attention and to propose research approaches in such areas.

Supported by CLIF
Computational Linguistics in Flanders
Endorsed by SIGdial
Special Interest Group on Discourse and Dialogue