Using Dialog Corrections to Improve Speech Recognition

We propose a preliminary method for automatically correcting errors in spoken dialogue systems1. Current spoken dialogue systems usually show a rather static and rigid behavior regarding recognition errors, therefore a feasible method of correcting system errors might be helpful to successfully support user requests. Moreover, a correction differs from non-correction prosodically [1]. Generally a user correction exhibits a greater prosodic difference the more distant it is from the initial error. In this case it is recognized more poorly, and it involves a longer human-machine interaction because this often leds to the same recognition errors. This paper will focus on methods to adapt the system using the error feedback provided by the user, and basically requires the adaptation of the language model.