Clarification questions to improve dialogue flow and speech recognition in spoken dialogue systems

Within human-machine conversation, clarification is vital and may consist of various forms, as it is may by due to many different effects on different levels of communication. In this paper, we present a strategy for detecting situations where a need for clarification exists in a natural spoken dialogue system. We define rule sets which enable us, via an anomaly analysis, to detect these critical situations. Through the use of such rule sets, we show that ist is possible to enhance the strategy in such a manner that more different situations are detected. In a user test, we evaluate the success of the strategy and show that strategies with explicit clarification improve the naturalness of human-machine interaction.