Издательство Springer, 1997. — 229 p.EC AY 96 Workshop, Budapest, Hungary, August 13, 1996, Revised Papers
This volume contains a selection of extended and revised versions of papers presented to the European Conference on Artificial Intelligence ECAI-96 Workshop on Dialogue Processing in Spoken Language Systems. The workshop took place on August 13, 1996 in Budapest, Hungary. This workshop received considerable interest from researchers of internationally acclaimed research institutions and companies, indicating that spoken dialogue system development is a burning issue in the fields of artificial intelligence and speech processing.
This volume collects papers from researchers that belong to the leading groups in the community. It covers research being carried out in the United States (7 papers), in Europe (6 papers), and in Japan (1 paper). It is interesting to observe that international cooperation already exists in this field, as testified by one paper in the book which is authored jointly by researchers from Europe, the United States, and Japan. The book reports on work being pursued both in academia and in industry. The papers, though, are mostly biased towards application-oriented research, since most of the projects described in the book are either pursued in industry or heavily (co-)sponsored by companies interested in applications of speech processing technology.
This fact indicates that speech technology, and dialogue processing in particular, is a field that is in the process of transition from the research labs and is entering the marketplace: automatic speech-operated reservation and booking systems are already accessible to the public, as are information systems answering questions concerning timetables, weather conditions, private bank accounts, etc. Also, speech technology is more and more integrated into hands-free applications where a user is unable to employ interaction modalities like text or gestural interaction with graphic entities: typical applications are the operation of speech-controlled devices such as a communication radio or telephone in a car.
Along with the rapid advance of research in the area of spoken dialogue systems, progress is being made in many related fields: areas are identified where results from research in typed dialogues can be transferred to spoken language systems. With growing experience in system development, tasks like the collection and annotation of speech corpora are being made more efficient and less expensive. Wizard of Oz techniques are widely used both to involve the user in the system design from the earliest stages of development and to gather maximally realistic and usable data. Standards currently being developed for the transcription and annotation of speech (e.g., ToBi for phonetic labeling and the Discourse Resource Initiative (dri) for segmentation and dialogue act labeling) lead to the reusability of speech corpora. According to the slogan "No data is better than more data" such massive data collections are crucial for system development, especially for training and testing of statistical models. The treatment and recognition of prosodic features is of growing importance for spoken dialogue systems: since the segmentation of dialogue contributions into meaningful (processing) units is an important issue in spoken dialogue processing, the determination of such units, possibly by means of prosodic information like pauses and/or rising and falling intonation, is a topic which at the moment is receiving considerable attention.
As far as rapid prototyping is concerned, new design principles and architectures have been proposed that allow easy adaptation of dialogue systems to new domains and applications. For the most part, these requirements are met through the introduction of object-oriented design and by a clear-cut separation of application-dependent and application-independent knowledge.
Finally, the evaluation of spoken dialogue systems is a current topic that has been neglected in the past: while many studies have been carried out to assess the quality of speech recognition components, few efforts have been reported to estimate the overall quality of integrated spoken dialogue systems.
The book contains papers that address all of the problems mentioned above and propose methods for their solution.
The most challenging problem in speech processing is the transition from cooperative continuous speech input to truly spontaneous conversation. On the dialogue level, the transition from human-machine to human-human interaction, and eventually to multiparty discourse, is a central issue of current research. This book contributes substantially to the progress towards the next generation of spoken dialogue systems.Foundations of Spoken Language Dialogue System Design
User Errors in Spoken Human-Machine Dialogue
Towards a Dialogue Taxonomy
Using an Interpretation System - Some Observations in Hidden Operator Simulations of 'VERBMOBIL'
Classification of Public Transport Information Dialogues Using an Information-Based Coding Scheme
Dialogue Units and Prosodic Aspects of Spoken Dialogue Processing
Speech Production in Human-Machine Dialogue: A Natural Language Generation Perspective
Input Segmentation of Spontaneous Speech in JANUS: A Speech-to-speech Translation System
"Pause Units" and Analysis of Spontaneous Japanese Dialogues: Preliminary Studies
Syntactic Procedures for the Detection of Self-Repairs in German Dialogues
Utterance Units in Spoken Dialogue
Spoken Dialogue Systems - Design and Implementation
Development Principles for Dialog-Based Interfaces
Designing a Portable Spoken Dialogue System
Minimizing Cumulative Error in Discourse Context
Evaluation of Systems
Automatic Evaluation Environment for Spoken Dialogue Systems
End-to-End Evaluation in JANUS: A Speech-to-speech Translation System
A Task-Based Evaluation of the TRAINS-95 Dialogue System
