Image Change Captioning by Learning from an Auxiliary Task