Spoken Natural Language Dialog Systems: A Practical Approach

This book gives a detailed yet readable account of the design, implementation, testing, and analysis of a spoken task-oriented dialogue system. Although there is a thorough reporting of related work, the plural in the main title is a bit misleading, as the book focuses on only one system, built by the authors at Duke University. The subtitle is well warranted, as Smith and Hipp put a premium on achieving useful interaction rather than adherence to any particular psychological or linguistic theory of language processing. The writing is generally clear, even when discussing details of the sometimes complex algorithms. The index is also very helpful. In addition, for those who want more speciics, an appendix gives instructions for electronically obtaining the dialogue system code and transcripts of user interactions with the system. The code is also commented with pointers to aspects of the book, including sections, gures, and pages. Smith and Hipp carefully point out not only the successes of their system, but also its shortcomings and some prospects for improvement. Smith and Hipp do a good job of describing and analyzing related work and comparing it to their own system. Chapter 2, in particular, reviews foundational work on dialogue processing. The main focus is on work in the sub-areas of speech act processing, user modeling, use of expectations, and mixed initiative. Also included are several other proposals for integrated dialogue models. The highlight of the chapter is a comparison of 26 other dialogue systems, mostly from the 1980's, listing the application domain and output modality, as well as the types of processing performed and how the systems were evaluated. Other related work is introduced as appropriate in subsequent chapters. The authors are perhaps not quite as successful in recasting the innovations from their own implementation in a more general conceptual framework that could be useful for designers of systems with very diierent architectures. A task-oriented spoken language dialogue system is a large piece of software, requiring successful integration of a number of heterogeneous sub-components to perform at least the following functions: speech recognition, language understanding, dialogue reasoning, task reasoning, language generation, and speech generation. The thrust of the authors' work is on the language understanding and dialogue reasoning components, as well as the overall architecture that ties these subsystems together. These aspects are also presented more brieey by Smith, Hipp, and Biermann (1995), who also include a more detailed discussion …