Identifying salient utterances of online spoken documents using descriptive hypertext.