Briefly Noted

Briefly Noted

Linguistic and Computational Techniques in Machine Translation System Design (second edition)

Peter Whitelock and Kieran Kilby

(Sharp Laboratories of Europe; late of UMIST)

London: UCL Press (Studies in Computational Linguistics, edited by Harold Somers), 1995, xii+208 pp; hardbound, ISBN 1-85728-216-7, £30.00

This is a slightly enhanced second edition of the 1983 report 84/2 of the Centre for Computational Linguistics at UMIST in Man­chester, England. As Harold Somers states in the preface to this edition, "it was the first at­tempt anyone had made to take a critical and objective look at the details of MT systems." This explains to a large extent the interest it may arouse and justifies to some extent its (re)publication within the Studies in Compu­tational Linguistics series a decade after its original completion. The book comprises two parts. Part I is a concise general discussion about ma­chine translation (MT). It covers succinctly the linguistic and computational informa­tion necessary to be able to follow the sec­ond part of the book. Part II surveys, in great detail, and evaluates six MT systems: SYSTRAN, TAUM-Météo, TAUM Aviation, GETA Ariane-78, Yorick Wilks's Preferential Semantics (PS) system, and METAL. Neither of the authors were involved in the develop­ment of any of these software projects. They did, however, have an in-depth understand­ing of the intricacies of the systems, having collected numerous relevant reports and in­ternal memos. Their description is given in a relatively clear, easy-to-follow manner with­out compromising on technical detail. Harold Somers has provided a number of additional footnotes trying to put (part of) the discus­sion into a mid-90's perspective. The book is not, nor does it claim to be, either an introduction to MT or a general reference book on MT. In essence, it is a critical, detailed description of the six afore­mentioned MT systems. Somers claims in the preface that "this effort has not yet been du­plicated." As humble as he may be, the facts are that by the time he wrote this preface (De­cember 1993) Somers himself had already co-authored Hutchins and Somers (1992). The latter is a widely available introductory MT textbook and reference book that contains, in­ter alia, four full chapters dedicated to criti­cal, detailed descriptions of most of these MT systems (all excepting TAUM Aviation and Wilks's PS system). Although Hutchins and Somers may seem at times to exhibit more of a socio-historical interest than a technical one—in contrast to the book under review— nonetheless, they too go into substantial tech­nical detail and give overall comparable de­scriptions. The effort, thus, has actually been duplicated to a significant extent, casting cer­tain doubt on any serious need for this sec­ond (partially redundant) edition of White-lock and Kilby's book. To summarize, the book is interesting, well written, not too thick, and easily read. How­ever, unless one is explicitly interested in detailed critical descriptions of TAUM Avi­ation or Wilks's PS system, then Hutchins and Somers (1992)—which covers most of that discussed by Whitelock and Kilby and much much more—should probably suffice. This renders the book under review some­what dispensable.—Daniel Radzinski, Lexical Technology Inc., Alameda, California.

Hutchins, W. John and Harold L. Somers. 1992. An Introduction to Machine Translation. London: Academic Press.

Connectionist, Statistical, and Symbolic Approaches to Learning for Natural Lan­guage Processing

Stefan Wermter, Ellen Riloff, and Gabriele Scheler (editors)

(University of Hamburg; University of Utah; and Technical University of Munich) This book proved itself, literally, to be a very effective survey of the recent work in a va­riety of approaches to learning natural lan­guage. I began reading this book while on a flight to attend the third International Collo­quium on Grammatical Inference (ICGI-96), where I was invited to give a lecture on grammatical inference of natural language. By the end of the book, and the flight, I had completely rethought the introduction to my talk.

Berlin: Springer (Lecture Notes in Artificial Intelligence 1040), 1996, ix+468 pp; paperbound, ISBN 3-540-60925-3, no price listed

The book is a collection of revised pa­pers from a workshop at the 1995 Interna­tional Joint Conference on Artificial Intelli­gence, with a very worthwhile introduction written by the members of the workshop or­ganizing committee. As the title implies, the book is divided into three sections: connec-tionist networks and hybrid approaches, sta­tistical approaches, and symbolic approaches to language learning.

Most of the selections in the first sec­tion are proof-of-concept papers, in contrast to the more results-oriented papers in the other two sections. The reader is left with the impression that while there is interest­ing work being done in applying neural net­works and genetic algorithms to language learning, the approach is still in its infancy, and lags somewhat behind the statistical and symbolic movements. For example, the final paper in the section, "Learning language us­ing genetic algorithms," by Smith and Wit-ten, established the robustness of its proposed learning algorithm by trying to learn the rule S —> NP VP from a half-dozen five- and six-word sentences.

With this book, Wermter, Riloff, and Scheler have done a commendable job of providing the community with a good start­ing point for exploring these very differ­ent approaches to the problem of language learning.—David M. Magerman, Renaissance Technologies Corp.

Instrumentation: An Introduction for Students in the Speech and Hearing Sci­ences (second edition)

Lawrence Erlbaum Publishers, 1996, xx+163 pp; paperbound, ISBN 0-8058-2186-4, $24.50

Although written primarily for students in audiology, speech pathology, and related ar­eas, Decker's book will also be helpful to any linguist who wants to learn the practical as­pects of electronics for collecting and ana­lyzing speech data. The book assumes that the reader has no scientific or technical back­ground. The topics covered include the fol­lowing: the basics of setting up and connect­ing equipment; microphones and other trans­ducers; amplifiers, attenuators, mixers, and filters; digital signal processing; spectrum analysis; digital and analogue tape recording; and instrumentation for studying the physi­ology of speech.—G.H.

Amsterdam: IOS Press, 1997, v+440 pp; hardbound, ISBN 90-5199-252-1, Dfl 180.00, $100.00, £68.00, DM 160.00; distributed in Japan by Ohmsha Ltd, ISBN 4-274-90080-0-C3000

This Eurocentric guide to the language in­dustries and some of the basic concepts of language builds upon the 1994 edition, but the organization is now encyclopedic: one simple alphabetical list. The listings include major companies, organizations, technolo­gies, and media. A typical entry is one or two paragraphs, with addresses and phone num­bers where appropriate. The coverage may be inferred from the first few M entries:

Manx

Unusually for a reference book, the Lan­guage Industries Atlas is set in a large font with much white space. (For example, the list of standard abbreviations for languages takes six pages for a table that would have required little more than a single page in a conventional layout.) The physical size of the book could have easily been halved, signif­icantly reducing the price, with no loss of clarity.—G.H,

Dordrecht: Kluwer Academic Publishers (Text, Speech and Language Technology series, edited by Nancy Ide and Jean Véronis, number 1), 1996, xv+415 pp; hardbound, ISBN 0-7923-4152-X, Dfl 195.00, $128.00, £86.00

The book contains revised versions of papers presented at two workshops on parsing tech­nologies in 1993 and 1994. The contents of the volume are as follows: "Parsing Technologies, and Why We Need

Them" by Harry Bunt "Fully Incremental Parsing" by Mats Wirén

and Ralph Ronnquist "Increasing the Applicability of LR Parsing"

by Mark-Jan Nederhof and Janos Sarbo "Towards a Formal Understanding of the Deterrninism Hypothesis in D-Theory" by James Rogers and K. Vijay-Shanker "Varieties of Heuristics in Sentence Parsing"

by Makoto Nagao "Parsing as Dynamic Interpretation of Fea­ture Structures" by Harry Bunt and Ko van der Sloot

"Efficient Parsing of Compiled Typed Attrib­ute-Value Logic Grammars" by Bob Car­penter and Gerald Perm

"Predictive Head-Corner Chart Parsing" by Klaas Sikkel and Rieks op den Akker

"GLR* — An Efficient Noise-Skipping Pars­ing Algorithm for Context-Free Gram­mars" by Alon Lavie and Masaru Tomita

"Evaluation of the Tagged Text Parser, A Pre-Uminary Report" by Tomek Strzalkowski and Peter Scheyen

"Estimation of Verb Subcategorization Frame Frequencies based on Syntactic and Multi­dimensional Statistical Analysis" by Akira Ushioda, David Evans, Ted Gibson, and Alex Waibel

"Stochastic Lexicalized Tree-Insertion Gram­mar" by Yves Schabes and Richard Waters

"The Interplay of Syntactic and Semantic Node Labels in Parsing" by David McDon­ald

"Integration of Morphological and Syntac­tic Analysis based on GLR Parsing" by Hozumi Tanaka, Takenobu Tokunaga, and Michio Aizawa

"Structural Disambiguation in Japanese by Case Structure Evaluation based on Exam­ples in a Case Frame Dictionary" by Sadao Kurohashi and Makoto Nagao

"Predictive Parsing for Unordered Relational Languages" by Kent Wittenburg