Automatic Summarization (Mani) Book Review

Researchers in automatic document summarization have already adopted many techniques from existing machine translation literature. Likewise, there is much that the machine translation community can learn from current research in summarization. Automatic Summarization, by Inderjeet Mani, provides a firm grounding in the primary techniques that have been applied to the summarization task, so that researchers or students unfamiliar with the topic will be sufficiently versed in its primary techniques to begin their own work. The book provides a logically structured, comprehensive survey of what the field of summarization was, up through the time of the book’s publication. Overall, Automatic Summarization provides a strong and effective comparison between techniques that professional human summarizers employ and those used by automatic summarization systems. This is one of the aspects that differentiates the book from Advances in Automatic Text Summarization, edited by Mani and Maybury. For a beginner in the summarization field, the introductory coverage of the various dimensions along which summarization systems can lie, and the coverage of basic definitions, is essential. The format is effective for readers of varying levels of sophistication: for students, the chapter conclusions help solidify the main points; for researchers, they provide a good reference into summarization literature. The book covers much ground: the content flows from simple extraction-based systems, through revision and discourse, to abstraction, multi-document summarization and multi-media summarization. This content is bookended by an introductory chapter on professional summarization and a concluding chapter on summary evaluation. The introductory chapter on professional summarization sets the framework for the remainder of the book, in which each task is considered, as it relates to the methods humans use for producing summaries. The automatic text summarization discussed ranges from simple single-document sentence extraction systems to systems that produce abstracts of document collections. The research presented in the chapters on extraction, revision and discourse has reached some reasonable level of stability, and accordingly, the chapter organization is easy to follow. The quality of the organization decreases in the chapters on abstraction, multi-document summarization and multi-media summarization; this reflects both that the later topics cover research that is, itself, inconclusive and that the topics are, themselves, less interrelated that those discussed in the first few chapters. The concluding chapter covers the problem of summary evaluation. Evaluation is largely an unsolved problem in summarization, due greatly to the fact that different humans typically create very different summaries of the same document. This pre-