Natural speech dialogue systems

The application of speech recognition and understanding to real-life situations generates new scientific problems. To illustrate the latter, the example of an automatic switchboard system is given that can be used in a very natural communication style. It is based on a high-end speech recognizer that outputs a word lattice which is subject to further evaluation by a speech understanding and a dialogue component. Although word error rates are above 20% for this task, the integration of additional knowledge sources like the telephone directory and the dialogue history can add to both the acoustic and the language model in order to achieve highly improved performance. As recognition is not perfect, the user has to confirm what the system understood. In order to avoid lengthy confirmations, it was requested to base the level of confirmation (repeated question, explicit or implicit confirmation) on the probability that the request was understood correctly. This was formulated within the statistical framework of speech understanding and successfully implemented into our prototype system. The degree of reliability controls the further dialogue flow. Other useful applications of this reliability measure will be described.