Spoken Language Dialog Modeling

Conclusion

In this chapter we have discussed aspects for modeling human-machine spoken language dialog.

Several systems have been adapted the Geneva model for operational purposes. Nevertheless, there always exists a separation between the structural model component in terms of exchange, intervention and language acts, and the functional component in terms of illocutionary and interactive function. In fact, the grammar rules of the different models are not sufficient to direct and to control the dialog.

The presented dialog model combines the structural and functional components. The conversational (interactional, sequential and structural) constraints of the Geneva model need to be taken into account during the dialog act identification. This model includes different sub-dialogs aligned on three axes, the task, the dialog and the meta-dialog. With respect to the hierarchy (exchange, intervention and language act), these sub-dialogs separate the different dialog phases, therefore enabling a precise control of the dialog flow. The identification of the sub-dialogs and the dialog acts has been carried out on the basis of corpus analyses, with the aim to offer a functional model, that may directly be used for a specific application.