printer friendly version of this page

FrB.SS - Machine Learning for Spoken Dialogue Systems

Friday, August 31, 2007, Astrid Plaza hotel, Room Scala 1

Session chairs: Oliver Lemon, Edinburgh University and Olivier Pietquin, Supélec - IMS Research Group

During the last decade, research in the field of Spoken Dialogue Systems (SDS) has experienced increasing growth. Yet, the design and optimization of SDS does not only consist of combining speech and language processing systems such as Automatic Speech Recognition (ASR), parsers, Natural Language Generation (NLG), and Text-to-Speech (TTS) synthesis systems. It also requires the development of dialogue strategies taking at least into account the performances of these subsystems (and others), the nature of the task (e.g. form filling, tutoring, robot control, or database querying), and the user’s behaviour (e.g. cooperativeness, expertise). In addition, new statistical learning techniques are emerging for training and optimizing speech recognition, parsing, and generation and synthesis in spoken dialogue systems, depending on various definitions of context. Automatic learning of optimal dialogue strategies is currently a leading domain of research.

Among machine learning techniques for spoken dialogue strategy optimization, reinforcement learning using Markov Decision Processes (MDPs) and Partially Observable MDP (POMDPs) has become a particular focus. One concern for such approaches is the development of appropriate dialogue corpora (e.g. TALK project, DiSCoH project) for training and testing.

However, the small amount of data generally available for learning and testing dialogue strategies does not contain enough information to explore the whole space of dialogue states (and of strategies). Therefore dialogue simulation is most often required to expand the existing dataset and man-machine spoken dialogue stochastic modelling and simulation has become a research field in its own right. Other areas of interest at the intersection of machine learning and dialogue are statistical approaches in context-sensitive speech recognition, trainable NLG, and statistical parsing for dialogue.

The purpose of this special session is to offer the opportunity to the international community concerned with this topic to share ideas and have constructive discussions in a single, focussed, special conference session. To do so, the session could begin with a short twenty minute overview and introduction followed by contributed papers.

Program – Oral presentations

10:00 - Machine Learning for Spoken Dialogue Systems, Oliver Lemon and Olivier Pietquin, Edinburgh University (UK) and École Superieure d'Electricité (France)

10:20 - Learning dialogue strategies for interactive database search, Verena Rieser and Oliver Lemon, Saarland University (Germany) and University of Edinburgh (UK)

10:40 - Hierarchical Dialogue Optimization Using Semi-Markov Decision Processes, Heriberto Cuayáhuitl, Steve Renals, Oliver Lemon and Hiroshi Shimodaira, University of Edinburgh (UK)

11:00 - Knowledge Consistent User Simulations for Dialog Systems, Hua Ai and Diane Litman, University of Pittsburgh (USA)

11:20 - Reducing Recognition Error Rate based on Context Relationships among Dialogue Turns, Hsu-Chih Wu and Stephanie Seneff, Industrial Technology Research Institute (Taiwan) and MIT CSAIL Laboratory (USA)

11:40 - Bayes Risk-based Optimization of Dialogue Management for Document Retrieval System with Speech Interface, Teruhisa Misu and Tatsuya Kawahara, Kyoto University (Japan)

Contact