Data-Driven Methods for Adaptive Spoken Dialogue Systems (PDF)

The EC FP7 project "Computational Learning in Adaptive Systems for Spoken Conversation" (CLASSiC) was a European initiative working on a fully data-driven architecture for the development of conversational interfaces, as well as new machine learning approaches for their sub-components. It developed a variety of novel statistical methods for spoken dialogue processing, for extended conversational interaction, which are now collected together in this book. A major focus of the project was in tracking the accumulation of information about user goals over multiple dialogue turns (i.e.\ extended conversational interaction), and in maintaining overall system robustness even when speech recognition results contain errors, by managing uncertainty through the processing chain.

Other advances were made in the areas of adaptive natural language generation (NLG), statistical methods for spoken language understanding (SLU), and machine learning methods for system optimisation, either during online operation, simulation, or from small amounts of data.

This book collects together the main research results and lessons learned in the CLASSiC project. Each chapter provides a summary of the specific methods developed and results obtained in its particular research area. In addition, leading researchers in statistical methods applied to industrial-scale dialogue systems (from SpeechCycle) have contributed a chapter surveying their recent work.

This volume will serve as a valuable introduction to the current state-of-the-art in statistical approaches to developing conversational interfaces, for active researchers in the field in industry and academia, as well as for students who are considering working in this exciting area.