Learning Content Selection Rules for Generating Object Descriptions in Dialogue