Extracting information in spontaneous speech