Work progress 2004

Work performed

The partners have successfully completed a range of key tasks in the first year of the project, both in advancing the theory and developing implementations. Many of these tasks have involved close collaboration between the sites.

The active project tasks in 2004 were in the areas of integrating the multilingual Grammatical Framework (GF) with ISU dialogue management, and extending the system to use a unified approach to multilinguality and multimodality; integrating ontology-based dialogue management with the ISU approach, and developing reconfigurable systems; extending Information State modelling for multimodal turn planning and developing modality specific resources; integrating machine learning techniques with the ISU approach; software infrastructure and system integration for the in-car and in-home showcases; and the definition of research questions and evaluation methods, and the design and conduct of data collections.



Significant progress has been made in each of these active tasks, measurable by the fact that each one has produced either a deliverable or a status report. Progress has been made in 3 main ways: in theoretical advances, in corpus collection and annotation, and in system implementation (at the levels of whole dialogue systems, system components, and interfaces.)

 

Results achieved

The project has made good progress on the core research issues which were planned for the first year. The partners have successfully completed a range of key tasks in the first year of the project, both in advancing the theory and developing implementations. In the areas of data collection and annotation, integrating multimodality and multilinguality, and combining learning with ISU dialogue management, progress has been particularly strong.

For example, we have shown in early results that use of learning techniques with ISU representations of dialogue context significantly improves the robustness of speech recognition, natural language understanding, and dialogue strategies in dialogue systems:

 Over 50% reductions in error rates using learning techniques for filtering speech recognition and parsing hypotheses

 Learning of a COMMUNICATOR dialogue strategy (in the domain of flight booking) which outperforms comparable DARPA COMMUNICATOR systems