In this paper, we propose a new robust discourse processing compensating for misrecognition in a spoken dialogue system. As the current speech recognition is not able to obtain adequate accuracy, a robust discourse processing which compensates for misrecognition is required in a spoken dialogue system. Here, we developed some functions for the robust discourse processing to reduce extra interactions caused by misrecognition. We incorporated the robust discourse processing functions into a spoken dialogue system and evaluated the effectiveness. The experimental result showed 27% reduction of extra interactions.