Mapping open-domain ASR output to dialogue systems grammars