In order to achieve effective human-machine communication, it is very important that a natural language system produces the same language it understands. Specifically for a dialog system a user will expect that the system will be able to understand the utterances the system has produced. For example, a user should be free to use the same words or phrases that have been introduced by the system itself. But if the system is not be able to understand its own words, it will be very difficult for the user to accept this system as an interlocutor having equal rights because the system ``makes promises it cannot keep''. A reversible system produces utterances only from that subset of language that it is capable of understanding. Therefore, inconsistencies in the language behaviour of the system can be avoided [Jacobs1988].