Using Descriptive Video Services to Create a Large Data Source for Video Annotation Research