The authors emphasize the need to evaluate separately three things: "the trained expression," "behavior under the questioning format," and "preservation of non-target behavior."