Machine translation systems Ed. by Jonathan Slocum (review)

BOOK NOTICES 209 hand, he concludes cautiously that the hypothesis that 'the alien [non-Polynesian] numerals from Easter Island' came from South America might be 'correct but impossible to demonstrate because of the lack of linguistic data' (62); on the other hand, he suggests that similarities in the (not very elaborate) consonant phoneme inventories of Karok and Proto-Polynesian may point to early contact between the two languages (50-51). His position with respect to linguistic science is also equivocal. Although his bibliography contains a number of entirely respectable references, for instance, his main methodology for considering possible linguistic influences from South and/or North America in Polynesian languages is confined to drawing nonsystematic typological comparisons and proposing unidentified substrata to account for changes in Polynesian structure (e.g. 16, 29). Since the book lacks both a clear organization and a systematic consideration of any sizable quantity of data, it is not a satisfying evaluation oflinguistic aspects of Heyerdahl's claims. Casual mentions of a few random similarities among relevant languages do not constitute even weak evidence of historical links of any kind. The real puzzle is why Carl Winter, a publisher with a list of illustrious historical linguistic studies, especially in Indo-European, has published two books in the past two years that can most charitably be described as fringe historical works (see Niko Besnier's BN, Lg. 64:817, 1988, on Nors S. Josephson, Greek linguistic elements in the Polynesian languages). [Sarah G. Thomason , University of Pittsburgh.] Machine translation systems. Ed. by Jonathan Slocum . (Studies in natural language processing.) Cambridge : Cambridge University Press, 1988. Pp. ix, 341. Cloth $49.50, paper $16.95. This book is a collection of all but one of the papers originally published in issues 1-3 of the journal Computational Linguistics in 1985. Requests for reprints ofthe articles were numerous enough to warrant republication as a book. Although the articles and bibliography have been slightly updated, they necessarily represent a recent but not current state of the field of machine translation. The book contains seven articles and an extensive bibliography. The first article, ? survey of machine translation: Its history, current status and future prospects', by Jonathan Slocum , surveys the somewhat rocky past of machine translation (MT) and makes a case for the practicality and the importance of the field. The article distinguishes the commonly accepted system design classifications (direct, transfer, interlingua) by which systems in the book are subsequently characterized. The rest of the articles in the book are detailed descriptions of a variety of MT systems. In his survey article Slocum introduces the reader to the historical pillars of MT, all of which were created before the early 1970s: (1) GAT, the Georgetown Automatic Translation project; (2) CETA, the equivalent project in Grenoble, France; (3) METAL, the product of the Linguistics Research Center at the University of Texas at Austin; (4) TAUM, the project in automatic translation at the University of Montreal; and (5) ALP, the project on Automated Language Processing at Brigham Young University. Slocum then discusses current production systems and current research and development systems. His conclusion is a somewhat pointed appeal to the computational linguistics community to reconsider the position that MT cannot be done, and to apply current knowledge to new MT designs. The second article is 'ASCOF: A modular multilevel system for French-German translation ', by Axel Biewer et al. ASCOF is based on the transfer design; it translates French into German. This system has been a research project since 1981 at the University of Saarland at Saarbrücken, West Germany. Biewer et al. examine the project history, the application environment , the general translation approach, the linguistic techniques and computational realization , and future prospects. This same format is followed, with some variation in detail, in all of the articles on specific projects. The next article, 'Automated translation at Grenoble University', by Bernard Vauquois & Christian Boitet. focuses on the research group at Grenoble, which is known as GETA (Groupe d'Etudes pour la Traduction Automatique ). GETA follows the tradition of MT work originally performed at CETA (Centre d'Etudes pour la Traduction Automatique) until 197 1 . The system, which was developed originally for translating Russian to French, has...