Using dependency parsing and machine learning for factoid question answering on spoken documents