Extended Finite State Models of Language András Kornai (editor) (BBN Technologies) Cambridge University Press (Studies in natural language processing), 1999, xii+278 pp and CD-ROM; hardbound, ISBN 0-521-63198-X, $59.95

In this volume, AndrOs Kornai has put together a collection of articles that strongly argue his contention that finite-state approaches to natural language processing (NLP) are now part of the mainstream, both theoretically and computationally. The papers included were first presented at a 1996 workshop, held in Budapest as part of the European Conference on Artificial Intelligence (ECAI '96), and have been chosen for inclusion in this volume to complement previous works on finite-state approaches. Articles referring to the Xerox regular expression calculus and the AT&T Bell Labs system of weighted finite-state transducers that appeared in Roche and Schabes's 1997 Finite State Language Processing are followed up by new articles in this book. In addition this volume includes a CD-ROM, which, although it does not contain the Xerox and AT&T Bell Labs toolkits, does have source code and executables for several of the systems described in the book, including an older version of Bruce Watson's FIRE Lite toolkit for constructing and minimizing finite automata. The attraction of finite-state approaches to NLP is their speed and efficiency. The question is their adequacy--just how powerful a formal system is needed to describe natural language? Consider the following sentence: