Verbmobil: Foundations of Speech-To-Speech Translation

Author: Wolfgang Wahlster List Price: $84.95 Our Price: Click to see the latest and low price ISBN: 3540677836 Publisher: Springer Verlag (September, 2000) Edition: Hardcover Sales Rank: 740,516 Average Customer Rating: 5 out of 5

Buy now directly from Amazon.com -

Customer Reviews

This exceptionally readable (in English) account of the German Verbmobil project (1992-2000) is comprised of 47 contributions from the principal participants in the project. These contributions cover all aspects of this major effort in natural language processing under the headings:

- Introduction (Overview by editor) - From Speech Input to Augmented Word Lattices - Lexical and Syntactic Processing - Semantic Processing - Dialogue Translation - Dialogue Processing and Context Evaluation - Language Generation and Speech Synthesis - Data Collection and Evaluation - System Architecture and Software Integration.

The project culminated in a working prototype which allows bi-directional telephone dialogues in three discourse domains: Appointment Scheduling, Travel Planning and Remote PC Maintenance,- between German, Japanese and American-English speakers.

The work reported,

- represents the state of the art in Speaker-Independent Spontaneous Dialogue Recognition, Translation and Speech Synthesis, - provides quantitative data which illustrate the strength and weaknesses of symbolic and statistical approaches to translation in this context. - illuminates the tradeoffs in various approaches to deep linguistic analysis - illustrates the important role of acoustical and lexical data collection and processing. - gives insight into the interaction between the many modules as reflected in the detailed End-to-End evaluation of the system and its components - demonstrates the role of the underlying system and software architecture in providing both, a test-bed development environment and the basis for a near-real-time demonstration system. - reflects upon some of the management issues encountered in a project of this scope

The reported work includes a number of advances, including

- a speech controlled telephone dialogue system including speaker-language identification - integration of deep and shallow processing: results from concurrent translation threads. - systematic use of (multilingual) prosodic information at all processing stages, - Parsing, Dialog Understanding, Translation, Generation and Speech Synthesis - understanding of spontaneous speech repair - generation of dialogue summaries

The people responsible for editing this volume have raised the bar for technical writing. The result is an unusually lucid, concise and consistent exposition of work from several disciplines, generally only accessible to the specialists. Limiting the length of the individual contributions, has confined the reporting to what was accomplished in the course of the project., without compromising scientific rigor. If I have any criticism at all, it is the lack of a subject and author index and the occasional use of abbreviations not spelled out, at first use, within an article.

To those interested in speech recognition and machine translation, this is essential reading; - they'll be surprised to find it satisfying as well.

Those concerned with large-scale, distributed software development will find several new benchmarks in this project.

See also for a project overview. .