In this paper, we present a statistical approach for dialogue act processing in the dialogue component of the speech-to-speech translation system \vm. Statistics in dialogue processing is used to predict follow-up dialogue acts. As an application example we show how it supports repair when unexpected dialogue states occur.