Book Review

This is a wonderful book-and not just for people actively involved in machine translation.Anyone with an interest in the history of computational linguistics will find much to relish and learn from in this weighty collection of articles past.Lest we forget, MT was one of the first nonnumerical applications proposed for the digital computer following the Second World War, and its often tumultuous 50-year history has had a significant impact on the entire field of computational linguistics.Indeed, this very journal can trace its lineage back to the journal whose original title was Mechanical Translation.Though not a proper history of MT, Readings in Machine Translation is certainly a historical collection.The editors have sought to bring together in one volume "the 'classical' MT papers that researchers and students want, or should be persuaded, to read" (page xi) and that, alas, are often so difficult to find.For this alone, Nirenburg, Somers, and Wilks deserve our gratitude.The volume begins with the famous memorandum that Warren Weaver sent out to some 200 professional acquaintances in 1949, which is generally taken to mark the genesis of machine translation; and the most recent paper included dates back to the fourth MT Summit in 1993.The 36 articles that span the intervening period are said to constitute "MT's communal inheritance."By what criteria were the papers selected?The editors cite three: personal taste; the aforementioned problem of availability, which is certainly a real one, particularly in the case of some of the early classics; and historical significance, which is said to be the main criterion for inclusion in the volume.The articles selected by the editors are supposed to represent "the most important papers from the past 50 years" of MT (p.xi).Well, as criteria go, that certainly sets a high standard!And yet many of these articles seem to meet it with ease.In addition to Weaver's memorandum and a monumental state of the art published by Bar-Hillel in 1960, both of which absolutely must be read, there is a jewel of a piece written by Victor Yngve in 1957 that argues for what later became known as second-generation MT: systems that analyze the input text into an essentially syntactic intermediate representation that serves as the basis for transfer, rather than applying a bilingual dictionary directly to the input string, as first-generation systems did.There is Martin Kay's "The Proper Place of Men and Machines in Language Translation"-an MT classic if ever there was one-in which the author derides the pursuit of fully automatic MT, not as a legitimate goal of basic research, but as a strategy promising a short-term solution to the burgeoning demand for translation; in its place, Kay proposes a modest, incremental program of machine aids for human translation: "little steps for little feet."There is Makoto Nagao's 1984 paper that launched the entire example-based MT paradigm (still very much alive and kicking), as well as a stellar piece by Jan Landsbergen laying out the advantages for

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.