There may be more than 1 semantic tag for an utterance. The method could be like duplicating utterance and using tags as label respectively.