The work presented is concerned with the development of multilingual and multidomain Dialogue Systems, using a domain and language independent Dialogue Manager component, and the (semi)automatic generation of the system's responses.
The Dialogue Manager may support different types of dialogue initiatives.
On one hand, the Dialogue Manager may drive the interaction with inexperienced users, asking them all the information the service needs.
On the other hand, experienced users may take the initiative giving information not only about what the system asked for, but also about other aspects.
The task of the Dialogue Manager become more complex when dealing with user-initiative dialogues. The Dialogue Manager should cope with the user's interventions that give more, less and/or different information than requested as well as with the user's interventions that ask for information not related to the last system's intervention.
The Natural Language Generator module may provide natural and efficient communication messages to the Dialogue Manager.
Feedback strategies improve the communication. For this reason, the Dialogue Manager may deal with feedback strategies, and consequently the Natural Language Generator module may generate messages expressing this feedback.
Feedback is related to implicit and explicit confirmation of values, and also with different types of mistakes, either caused by the user or by the system. For example, different messages should be generated for expressing misunderstanding of the user's sentence, an error accessing to the database, or a bad parameter value.