Using semantic descriptions in tags for screen readers