Book Reviews: Advances in Natural Language Generation: An Interdisciplinary Perspective

expression or syntactic rule in one there is a corresponding expression or syntactic rule in the other with the same meaning. Rosetta differs from the strict Montague model in making no use of intensional logic or of categorial grammar, but it adheres to the compositionality principle (in relating syntactic rules and semantic structures). The paper includes many formal proofs and only trivial sentence examplesnwhich might well not impress those skeptical of the value of a project whose influence on MT theorists has been considerable in recent years. It is followed by a description of another experimental system (albeit less radically innovative in linguistic conception), an account by Margaret King and Sergei Perschke of the Eurotra project of the European Communities. After outlining the historical background (the need for a decentralized multilingual project capable of stimulating MT a n d computational linguistic research in each of the participating countries), the basic features are described: a linguistics-based transfer model, modular and robust, no interactive facilities, multilevel tree-structure interfaces, a controlled production system operating on series of well-defined grammars, etc. Eurotra is admitted to be relatively traditional in its linguistics, but it is claimed to be more advanced in its computational design and to have been successful in promoting the study of languages not previously the subject of MT or detailed linguistic research. This book represents a primary source of information for some of the most significant and influential MT projects in the last decade. It is greatly enhanced by a substantial bibliography, which includes not only all references by the contributors, but also many other items covering the historical development of MT, the current (1984) state of research, and related topics in computational linguistics and artificial intelligence. The publication delay has not diminished the value of the contributions: many papers provide the most detailed and comprehensive accounts of individual MT systems. Only in the cases of Rosetta, GETA, and Eurotra can it be said that there have been substantial changes in the years since the Lugano tutorial. There are some omissions in the historical surveys as already noted, and readers should also be made aware that a number of systems current at the time are not mentioned at all, e.g., the DLT system in Utrecht, the PAHO system in Washington, many of the largely AI-inspired projects in the United States, the systems in the Soviet Union, and the activity in Japan (e.g., the influential Kyoto research). These are minor points. This is a publication that will remain for many years an invaluable source on MT research and it deserves to find a place on the shelves of anyone seriously interested in computational linguistics and in machine translation.