One of the most interesting applications of natural language processing is the automatic generation of natural language text by computer. Many areas of language-related computer use would become much more interesting and approachable if the text generation problem were tractable and within the current state of the art, including the design of human-computer conversation systems, the implementation of more refined mechanical translation procedures, and the design of text summarization and abstracting systems. Unfortunately, text generation is not well understood, and except for special-purpose applications in narrow contexts--for example, the preparation of weather forecasts or stock market reports--text generation is not currently within the state of the art.

It would have been nice if a book had come along that actually explained the text generation problem, the approaches that have been pursued in the past (for example, the design of schemas for special-purpose texts), and the rudiments of rhetorical structure theory (the theory purporting to explain the overall structure of ordinary discourse). Unfortunately, as the title indicates, this book is not at all suitable for this purpose. Instead, we have the somewhat expanded proceedings of the Second Workshop on Natural Language Generation, held in 1989 in Edinburgh.

This book contains 12 individual reports by mostly British and Dutch authors, and while some well-known contributors to the text generation field are included, the nonspecialist has no good reason to look at this material. The book has no nice overview of the field, no proper summary of where we stand, no discussion of future prospects in the field, and no explanation of the main approaches currently used. Why the proceedings of a workshop designed for specialists should be published as a hardcover book that is disseminated to the general public is not clear. The material is not likely to be useful even to specialists because more than two years have passed since the original workshop was held, and the contents will be well known to insiders by now. So far as the wider public is concerned, this somewhat random collection of disconnected systems and approaches does not provide the well-structured overview that is needed to render the field reasonably transparent.

If it was not obvious before, this book makes it clear that general-purpose text generation is not around the corner, and that much more needs to be learned about language understanding and discourse structure before flexible text generation components can be incorporated into general-purpose natural language processing systems.