Browse Books

Reviewer: Mona Singh

Designed as an introduction to the basic concepts of natural language understanding for undergraduate and graduate students, this book has a layout similar to that of the first edition [1]. The book is divided into three parts and 17 chapters, and has three appendices. Part 1 deals with syntactic processing. It provides a gentle introduction to grammars and the parsing techniques that might be employed in natural language systems. The basic formalism of the syntactic part of the book has changed from augmented transition networks to feature-based context-free grammars. This second edition includes a new chapter dealing with the basic principles of the use of statistical methods in parsing. This chapter provides an overview of how statistically based methods are used in estimating probabilities from corpora. Allen explains how probabilities can be used for part-of-speech tagging, for obtaining lexical probabilities, and for building probability-based parsers. The addition of this chapter is an accurate reflection of current trends. Part 2 deals with semantics. Allen describes the logical forms of sentences, which are an important component of meaning. Several different syntactic parses may map onto the same semantic representation. Semantic processing can help eliminate syntactic parses on the basis of meaning. Allen describes how the logical form can be derived from the syntactic structure and how preferences can be used to identify the most likely logical form. Finally, Part 2 discusses the more elusive issues in semantics, including scoping, anaphora, and noun modifiers. Part 3 deals with the centrality of knowledge and reasoning in natural language understanding. Allen first discusses “situational context,” that is, world knowledge and knowledge of the situation. Allen also discusses “global discourse context,” that is, discourse structure. He shows that understanding the causal structure of events is extremely crucial. Allen goes on to define a “conversational agent” as an agent that can communicate via natural language and participate in natural language dialogues. The reader gets a good idea of how beliefs, intentions, and desires affect a conversational agent. Appendix A provides an introduction to logic and model-theoretic semantics. Appendix B provides an introduction to symbolic computation. Appendix C, a new addition to the second edition, gives the reader an introduction to speech recognition and natural language understanding. The book includes software, written in Common Lisp, which can be downloaded from an ftp site. The software includes the implementation of algorithms and examples used in the text. In general, the entire book has undergone some changes reflecting the advances in research since the first edition. All in all, Allen provides a good background in natural language understanding. However, readers will not get a good feel for the applications of natural language understanding systems, the difficulties such systems have in real applications, and possible ways of engineering natural language systems.

Reviewer: Alice Louise Davison

The formal and semantic properties of natural language that are important for various approaches and goals in computer science are covered in this textbook. It gives a clear yet compact introduction to sentence structure, inflectional and derivational morphology, and acoustics of language sounds. It also contains a comprehensive survey of meaning in natural language, from discourse and knowledge representation, to sentential meanings and ambiguities, to subtle distinctions of word meaning. Each topic is somewhat self-contained, allowing a reader to select a range of chapters to consider in depth. The author's intention is to make the book accessible to either graduate or undergraduate students, on the assumption, perhaps, that graduate students would read more chapters, read them more deeply, or do independent projects. For students with little background in linguistics, the first six chapters give an overview of English syntactic structures and morphology. Appendix C gives a brief introduction to the distinctive sounds (phonemes) of English and their acoustic properties, though, unfortunately, not with all the standard phonetic symbols. Each chapter describes some central, related concepts, with copious examples of how they are represented in a given system of formal description. Each chapter ends with exercises, labelled according to level of difficulty, and a usually excellent set of references for further reading. Chapter 1 is a useful overview, which explains succinctly which problems of natural language are of most interest in the various disciplines grouped together under the general umbrella of cognitive sciences (linguistics, psychology, philosophy, and computer science). The self-contained nature of each chapter means that the author does not call attention to the connections among chapters by presupposing knowledge of all the related previous topics. But within the three main sections of the book, there is a core of basic chapters, followed by others that extend the discussion to more complex issues and the computational approaches to them. The first section, “Syntactic Processing,” first gives the basic vocabulary of sentence description, based mainly on work in linguistics, and summarizes the classic literature on the theory of parsers applied to natural language. Two chapters present the most recent work on feature-based and unification grammars, which have been used extensively in computer-based natural language processing. The section concludes with two chapters on computationally efficient approaches to ambiguous features of language, such as syntactic category and constituent structure. Appendix B provides an overview of symbolic computation. The second section, “Semantic Interpretation,” starts with several chapters on the central categories that enter into semantic interpretation of sentences in addition to word meanings: thematic roles; quantifiers and other semantic operators in questions and relative clauses; negation; tense; and modality. There is a brief and somewhat advanced introduction to the logical formalism used for representing semantic relations, and how these relations are linked to the hierarchically organized syntactic structures studied in the first section. (Appendix A provides a more basic introduction to standard logical and model-theoretic semantics.) The following chapters present computational approaches to the recognition and relation of the factors that enter into the interpretation of a sentence. The last section, “Context and World Knowledge,” first gives an overview of the issues in knowledge representation. Separate chapters then deal with local discourse contexts and with reference to entities in them; general knowledge bases; the overall structure of discourse, including the links between individuals, events, and concepts; and finally, purposes and goals in discourse, including actions performed by using words. The book as a whole has a well-thought-out, coherent structure. The discussion of sometimes opposed traditions is even-handed and focused on the strengths of particular ideas. Insofar as I am able to judge by its reference to work I am familiar with, it gives a clear, coherent account of the important research on language and computational problems. It would serve as an excellent reference work for people who are trained in one of the cognitive science disciplines and want to extend their knowledge in other areas. It is probably too intensive for total novices. I expect, however, that it would also be an excellent textbook, if the sections to be covered were chosen with some sensitivity to the level of preparation of the students and to the primary goals of the course. The comprehensiveness of the topics discussed in this book, and the wealth of different approaches represented, give the reader some idea of the magnitude of the problems of interpreting natural language structures.