SIRIDUS will contribute to Community economic development through scientific
and technological advances in spoken language dialogue systems. The central objective is to expand our understanding of and develop computational tools to support the development of more robust, functional
and user-friendly dialogue systems. First we will extend the range of dialogue types to which the Information State Update view of dialogue applies. We will exploit it in improvements in speech recognition
and synthesis. We also aim to develop architectures highly suitable for robust processing capable also of integrating and upgrading different and linguistically sophisticated components. We will increase
dialogue sensitivity to prosodic information. Project results will include pure research and two pieces of demonstrator software: an interface to a telephone using a Natural Command Language and an
integrated toolset for dialogue researchers embodying our distinctive Information State Update view.
SIRIDUS is divided into six main Work Packages. In "Dialogue Moves:
Extensions and Specifications", we extend the range of dialogue types to which the Information State Update view of dialogue applies. This view of dialogue has been promoted by the EU TRINDI project. We
consider in addition to Information Seeking Dialogues, natural command language dialogues and negotiative dialogues. In "Relating Information States to Speech Input and Output", we associate the
generic TRINDI dialogue move engine with both speech recognition and speech synthesis. We also investigate possibilities for enhancing speech understanding and semantically marked speech synthesis. In
"Implementing a Natural Command Language Dialogue System", we design and implement a particular spoken language dialogue system in Spanish, namely an interface to a telephone using a Natural
Command Language. We also evaluate it from the perspective of our Information State Change View of dialogue. In "Robustness and Scalability for Dialogue Systems", we develop methods for achieving
robustness in spoken dialogue systems. We attack the problem on two fronts: adding goal-centred dialogue management strategies to linguistically sophisticated processors; and adding increasingly
sophisticated linguistic knowledge to goal-centred dialogue management. Our objective is to identify the most habitable middle-ground. In "Prosodic Information in a Dialogue Move Framework" we
undertake fundamental research aimed at relating information structure in information states to features of prosodic information. The final work package is an integration package for producing an integrated
toolset for dialogue researchers embodying our Information State Update view of dialogue based on the research results of the project.