Helbig's Knowledge Representation and the Semantics of Natural Language presents the semantic representations derived from natural language needed for knowledge-rich applications, such as question answering and reasoning systems. The discussion is extremely detailed, providing extensive natural language examples as well as lists of phenomena with analyses, divided by semantic class. This level of detail is the main strength of the book: natural language processing researchers and engineers wanting to create a system involving deep semantic representations derived from natural language texts can look through all the phenomena, decide which are most important for their application, and then implement them. A general, broad-coverage system would presumably aim to incorporate all of the phenomena discussed.
The theory used throughout the book is Multilayered Extended Semantic Networks (MultiNets). The author has proposed and developed this theory for its coverage of the linguistic phenomena needed to interface with knowledge-based systems and in hope that the theory will be adopted more widely among both theoretical and applied researchers and system developers. Having one theory used by the majority of the community would allow systems to build upon one another and to more rapidly advance the field. However, even those who are already working with other theories for practical or principled theoretical reasons could find this book useful because it presents a broad range of natural language-based phenomena that must be covered for use with a knowledge-rich system. For those interested in how MultiNets fit into the larger field of semantics and knowledge representation, Chapter 15 provides a comparison of MultiNet with other semantic formalisms, including other network representations, logic-oriented formalisms, and frame-based representations.
Part I of the book outlines the problems of semantic knowledge representation, while Part II focuses systematically on the details of MultiNet. The book does not concentrate on how to do reasoning over the proposed semantic representations, explicitly setting that aside for another work, although there is an appendix of axioms and occasional reference is made to how the representations are used in systems, including a chapter on question answering and inferencing. There is more focus on how the mapping (semantic interpretation) from natural language to semantic representations might work. There are extensive examples throughout the book, usually on English with some German examples, giving the semantic representations for the example sentences and elaborating on what aspect of the sentences and of the meaning is at issue. In a few cases, the English examples are stilted or not quite grammatical, especially when translating the German examples (e.g. “Peter buys no motorcycle but a bicycle.”); however, this is never of importance for the points being made and would only be of concern if the example sentences were used as a development test suite for a system. In addition, there are diagrams of how the concepts used by MultiNet are connected to each other and of broader concepts such as how language and semantic knowledge representation are linked to knowledge in the real world. For example, Figure 3.1 posits levels for (1) natural language, (2) formal knowledge representation which is divided into intensional and pre-extensional levels, (3) the cognitive level in the brain, and (4) the real world; the figure analyzes the examples “Napoleon lost the battle of Waterloo in 1815.” and “Nobody has ever seen a unicorn.” at each of these levels.
The table of contents lists chapters, sections, and subsections, giving a good feel for the coverage of the book. After an introduction (Ch. 1) and discussion of historical roots (Ch. 2), chapters include: Part I: basic concepts (Ch. 3), semantic characterization of objects (Ch. 4), semantic characterization of situations (Ch. 5), the comparison of entities (Ch. 6), the spatio-temporal characterization of entities (Ch. 7), modality and negation (Ch. 8), quantification and pluralities (Ch. 9), the role of layer information in semantic representations (Ch. 10), relations between situations (Ch. 11), lexicon and knowledge representation (Ch. 12), question answering and inferences (Ch. 13), software tools for the knowledge engineer/sample applications (Ch. 14), and comparison between MultiNet and other semantic formalisms (Ch. 15); Part II: overview and representational principles (Ch. 16), means for expressing classification and stratification (Ch. 17), and relations and functions (Ch. 18).
Depending on your background and interest, the ideal reading order for the book will vary. In fact, the author suggests that some readers might want to look through Part II to get an overview of the coverage of MultiNet before reading Part I in detail. For readers with a strong background in knowledge representation, this approach might work. However, for the majority of the Natural Language Engineering audience, this may be inappropriate since a little background in the earlier chapters is necessary to understand the second part effectively. Those who work on implemented systems, which is probably the majority of the NLE audience, may want to focus on Chapter 14 first since it describes the workbench for knowledge engineers, the system for translating natural language expressions into MultiNet, the lexicographers workbench, and a virtual tutor for students using MultiNet. This chapter gives an application-oriented view of what the system is intended to do, thereby providing a concrete background for the other, more theoretically oriented chapters. Those with a background in formal semantics may want to start with Chapter 15 which compares MultiNet to other semantic formalisms, thereby situating it relative to these formalisms and some of their implementations, and then move to the chapters focusing on classic semantic phenomena such as quantifier scope and modality. Another way to approach the book is to take the virtual tour provided in both English and German on the CD-ROM. The tour has sections on semantic knowledge representation in MultiNet, natural language processing (i.e. mapping from natural language text to MultiNet semantic representations), knowledge management and reasoning, resources (i.e. the lexicon used by the system), and two applications (i.e. the virtual electronic laboratory and a natural language interface to the internet). These sections do not have to be read in order, allowing the reader to first examine the sections most closely related to their expertise, and are cross-linked.
The intense level of detail which makes this book so valuable, also makes it dense to read. If what you want is a quick read to understand the state-of-the-art in semantic representation for knowledge-based systems, this book is not for you. Even those intensely interested in the field or those looking to build or augment such systems will want to first skim over the main sections, or at least take the CD-ROM tour, and then return to sections as the need arises. In this respect, the book is a combination of an overview text, including some philosophical and theoretical background, with sophisticated documentation. The CD-ROM includes a pdf version of the book with cross-linking between chapters, figures, and definitions, which aids in using the book in a documentation-like fashion. The language used in the discussions can be heavy going at times, but the plethora of examples allows the reader to first focus on the phenomenon to be described and then to return to the text for detailed explanation.
Overall, Helbig's Knowledge Representation and the Semantics of Natural Language provides an extensive and detailed overview of the semantic phenomena needed for knowledge-rich applications. Although not an easy read, it is a valuable resource for those developing such systems or using the MultiNet-based systems for applications.
