Robustness in Language and Speech Technology (review)

This volume is primarily a collection of lectures presented at the 6th European Summer School on Language and Speech Communication held in Barcelona, Spain in the summer of 1998, but also includes several invited external contributions. The focus of this volume is robustness for the speech-recognition component and the natural-language parsing components. In particular, feature extraction, robust recognition of noise, adaptive systems, language modeling, the interface between speech recognition and natural language understanding, and general language parsing are considered at length. The underlying assumption of this volume is that language and speech technology systems are modular, allowing a range of components to be assembled on the basis of specific functional requirements.

In Ch. 1, Jean-Claude Junqua and Gertjan van Noord introduce the component approach and provide an overview of the volume, relating how various aspects of the subject are addressed by the contributors. Next, Johan de Veth, Bert Cranen, and Louis Boves (Ch. 2) discuss acoustic features and distance measurement, presenting several techniques used to alleviate the effects of unknown transmission channels and providing acoustic background on the acoustic features used in automatic speech recognition. In Ch. 3, Daniel Tapias Merino discusses speaker compensation in the context of automatic speech recognition, positing that adaptive automatic speech recognition provides a useful path to improve robustness in such systems. In Ch. 4, Jerome Bellegarda considers robustness in statistical language modeling through a review of a number of recent approaches to robust language modeling.

In Ch. 5, Peter Heeman and James Allen discuss how to enhance robustness in modeling spontaneous speech events and advocate that such events should be explicitly modeled in order to include and account for them in the overall language model. Mehryar Mohri and Mark-Jan Nederhof present in Ch. 6 an algorithmically based approach for approximating context-free languages with regular languages. This is complemented by Ch. 7, where Mohri describes a general grammar library that provides tools and algorithms for building dynamic grammars for large vocabulary recognition. In Ch. 8, Jean-Pierre Chanod introduces the concept of robust parsing, discussing linguistic phenomena that motivate this approach, and summarizes foundational and recent work in this area. Following this, Gertjan van Noord presents in Ch. 9 a general model for robust parsing of word graphs and discusses a number of search techniques aimed at optimization of concept extraction while maintaining accuracy. Finally, Carolyn Penstein Rosé and Alon Lavie (Ch. 10) provide an alternative approach to the preceding chapters in endeavoring to find the analysis that represents the closest approximation to grammatical input.

Overall this volume provides an accessible and well-rounded overview of the subject and is complemented by an extensive bibliography.