Experiences with an icon-like expression evaluation system