A combined method for discovering short-term affect-based response rules for spoken tutorial dialog.