Resumen de Digui: a flexible dialogue system for guilding the user interaction to access web services
Current dialogue systems can handle friendly and collaborative communication that supports diverse types of interactions, such as menus in which the user is asked to choose an option, form filling in which the user is asked for specific information, commands for expressing users' orders, and complex questions that can even reference previously evoked entities. Consequently, dialogue systems seem to be useful for accessing different types of applications. In particular, spoken dialogue systems are appropriate for devices that do not allow web browsing, such as telephones, or the use of hands, such as vehicle GPS.
Practical dialogue systems are mostly adapted to the functionality of the specific application they access, specially those incorporating speech, since voice technology still presents recognition problems in open domains. Because the adaptation of those DS to new applications is expensive and has to be done by experts, many studies have focused on the problem of developing dialogue systems that can be used for different applications and languages. Most relevant of those systems use domain and dialogue models, and have reusable domain-independent components. However the cost of adapting those systems to other types of applications is still high, specially in mixed-initiative systems, in which the dialogue initiative can be taken either by the user or the system.
The work in this thesis is particularly concerned with dialogue systems for guiding the user to access web services. The huge amount of web information increase the need of communication systems adaptable to different types of users, languages, services and channels. For this reason, dialogue systems can improve the usability and accessibility of web contents.
We have developed a multilingual mixed-initiative dialogue system for guiding the user when accessing web contents. To facilitate the adaptation of the dialogue system to new services and languages, the main functions of the system (language processing, dialogue control and task management) is performed by independent modules). The dialogue system developed consist of four independent general modules (Language Understanding, Language Generator, Dialogue Manager and Task Manager) as well as two knowledge structure accessible by all these modules (dialogue context and domain description).
Application and language-dependent resources are incorporated in separated data structures. In order to minimize the need for application and language-restricted data we propose the use of separated general bases for representing linguistic, application and domain knowledge.
In order to achieve a friendly communication, the dialogue management module follows a formal dialogue model, based on the information state model, that uses of a rich representation of the dialogue context.
The dialogue manager modules also incorporates an independent submodule to adapt the dialogue strategies, according to how well the communication is progressing. This component uses a conversation model (obtained from a previously analysed corpus of dialogues) to determine the degree of the system initiative in the next intervention considering several features on how well the conversation is doing.
