Spoken Dialogue Systems Kristiina Jokinen and Michael McTear (University of Helsinki, University of Ulster) Princeton, NJ: Morgan & Claypool (Synthesis Lectures on Language Technologies, edited by Graeme Hirst, volume 5), 2009, xiv+151pp; paperback, ISBN 978-1-59829-599-3, $40.00; ebook, ISBN 978-1-59829-600-6, doi 10.2200/S00204ED1V01Y200910HLT005, $30.00 or by subscription
Mary Ellen Foster · Computational Linguistics · 2010
This book gives a short but comprehensive overview of the field of spoken dialogue systems, outlining the issues involved in building and evaluating this type of system and making liberal use of techniques and examples from a wide range of implemented systems.It provides an excellent review of the research, with particularly relevant discussions of error handling and system evaluation, and is suitable both as an introduction to this research area and as a survey of current state-of-the-art techniques.The book is structured into seven chapters.Chapter 1 provides an introduction to the research area and briefly introduces the topics covered in the book.The end of the chapter consists of a list of links to tools and components that can be used for dialogue system development, which-although currently useful-seems likely to go out of date quickly.Chapter 2 addresses the task of dialogue management, beginning by describing simple graph-and frame-based methods for dialogue control, and continuing with a discussion of VoiceXML.The chapter ends with an extended discussion of recent work in statistical approaches to dialogue control and modeling.It is unfortunate that the discussion of other methods such as the information state approach and plan-based models is postponed to Chapter 4, but otherwise this chapter provides a good summary of both classic and recent approaches.Chapter 3 discusses error handling, which is divided into three processes: error detection, error prediction (i.e., the online prediction of errors based on dialogue features), and error recovery.After surveying a range of previous approaches to these three subtasks, the authors go on to discuss several more recent, data-driven approaches.Error handling is both a vital component of any spoken dialogue system designed for realworld use and an active area of current research, so this compact summary of techniques and issues is welcome.Chapter 4 contains case studies illustrating a range of dialogue control strategies and models.It begins with a description of the information state approach, and then moves on to discuss plan-based approaches as exemplified in the TRAINS and TRIPS projects.This is followed by a discussion of two systems that employ software agents for dialogue management: the Queen's Communicator and the AthosMail system.Finally, two systems which make use of statistical models are presented: the Microsoft Bayesian receptionist, which models conversation as decision making under uncertainty, and the DIHANA system, which employs corpus-based dialogue management.The case studies in this chapter provide detailed examples of a range of techniques, along with