Although the Semantic Web is still in its infancy, the theories and practices that support its development are founded in the pursuit of a symbolic form of knowledge representation that can be traced back to Socrates and Plato. This pursuit, and the science developed from it, is the subject of Knowledge Representation, which John Sowa describes as a "general textbook of knowledge-base analysis and design." It is an in-depth investigation into the practical and theoretical foundation underlying modern-day research into machine-accessible knowledge—the Semantic Web's modus operandi.

Unlike many textbooks, Knowledge Representation is both accessible and concise. It isn't packed with formulas or obtuse language like many other "serious" works on logic and ontologies, nor is it filled with props and rhetoric that insult the reader's intelligence. Still, the author doesn't spare the gory details when he feels they're necessary for understanding the topic. At times, the reader will have to put the book down and think about what he or she has just read, or risk not understanding what follows. Sowa doesn't explain anything twice, but simply assumes the reader has "done his homework." This is a challenging assumption, and a dangerous one when applied to a general audience, yet Sowa's style is refreshing in its dependence on the reader's ability to grasp the ideas presented. What emerges is a text that forces the reader to become involved in the same questions the book (and the science it describes) is trying to answer: What is knowledge? How is it organized? How can machines manipulate it? To further challenge the reader, Sowa presents scenario-based exercises at each chapter's end. These are designed to ensure that readers firmly grasp the material covered in the chapter by making them think about what they've just learned.

The book begins with an in-depth survey of the logical, mathematical, and philosophical foundations underlying modern knowledge representation. The author defines knowledge representation as the application of theories and techniques from

The first two chapters are devoted to extensive surveys of logic and ontology, respectively. They also cover knowledge representation's historical evolution, from Aristotle's syllogisms to Douglas Lenat and Ramanathan Guha's Cyc. Sowa particularly emphasizes the logical system of Charles S. Peirce, the originator of two basic forms of logic notation used today: existential graphs and the algebraic notation of predicate logic. Sowa employs both notations throughout the text, although he often prefers conceptual graphs (which evolved from Peirce's existential graphs), because Sowa was extensively involved in developing and standardizing this latter notation in the early 1990s. This graph notation of logic and ontologies is also the basis for the Resource Description Framework model.

For readers who haven't studied logic before, Sowa includes an impressively thorough introduction to both graph and algebraic notations in Appendix A. An even more fundamental introduction to the mathematics and logics involved in knowledge formalization is available on the author's Web site (www.jwsowa.com), along with illustrative excerpts from the book.

Although Knowledge Representation focuses on the theories supporting knowledge formalization, its approach to the subject is still pragmatic. Sowa is careful to stress the limitations of logic as a tool for describing human knowledge, although he also tries to communicate logic's extraordinary power in building knowledge systems from the ground up. Throughout the text, Sowa describes practical scenarios to illustrate how knowledge representation theory applies to real problems. He uses examples in Prolog, the C Language Integrated Production System, the Structured Query Language, and the Unified Modeling Language to relate knowledge formalization concepts to the reader's experience with these tools.

In chapter 4 ("Processes"), Sowa uses state machine and Petri net notations to describe discrete procedures and demonstrates the reducibility of these notations to conceptual graphs and predicate logic. Chapter 5 deals with theories of agency and purpose—an especially enlightening read for those who might be interested in software agents or robotics. Chapters 6 and 7 describe the real challenges knowledge engineers face in translating human knowledge into a machine-understandable form. Sowa suggests alternative systems such as fuzzy and nonmonotonic logic as possible approaches to problems of vagueness, perspective, and exception. Chapter 7 is particularly relevant to readers interested in the Semantic Web; it explores existing ontologies such as WordNet and Cyc as well as computer-science-originated models such as the entity-relationship model (for databases) and parse trees (for grammars). A fascinating study of language patterns follows, and a summary of the tools involved in knowledge acquisition concludes the text.

Although Knowledge Representation isn't easy reading by any standard, anyone with a serious interest in knowledge formalization—from beginners to experts—should not pass it by.

Minor Gordon is a graduate student at the Technische Universität Berlin. Contact him at minorg@cs.okstate.edu.