Briefly Noted

With the explosion in the quantity of online information in recent years, automatic abstracting and indexing has received renewed interest and a number of promising approaches have emerged. The goal of this book is to present a complete description of current indexing and abstracting techniques in the context of the underlying linguistic and statistical knowledge. The book has three parts: the indexing and abstracting environment, methods of automatic indexing and abstracting, and applications. The first part covers theories about text, such as rhetorical and thematic structure, and discusses the use of different text representations for information retrieval and abstraction. The second part of the book deals with lexical analysis and weighting schemes, abstraction, and evaluation techniques. The third part describes two summarization systems developed by the author: a summarizer for court cases and a generic magazine article summarizer. I found the application part particularly interesting because it describes all the stages of the development of applied summarization systems, from the initial corpus analysis to the final evaluation of the method. The system developed by the author combines symbolic techniques, such as a text grammar, with statistical methods, such as classification and clustering. The system demonstrates tradeoffs between symbolic techniques based on manual knowledge encoding and corpusbased shallow methods. However, the book's breadth comes at the expense of its depth. This problem is especially acute in the first two parts of the book, which are supposed to give an overview of the subject. The presentation of text theories and abstraction techniques is somewhat superficial and incomplete; the book is not up to date and does not contain the latest developments in the area. For example, it omits computational methods for discourse; cohesion-based approaches; segmentation algorithms; and recent work in evaluation, corpus-based summarization, and regeneration (e.g, Marcu 1997; Mani et al. 1998; Jing and McKeown 1999). These limitations can be a major hurdle for the reader who wants to learn about state-of-the-art developments.--Regina Barzilay, Columbia University