Neurocomputing: Foundations of Research

ion for Knowledge Acquisition” by T. Bylander and B. Chadrasekaran. Chandrasekaran’s papers are usually illuminating, and this one does not fail: He and Bylander re-examine such traditional beliefs as knowledge should be uniformly represented and controlled and the knowledge base should be separated from the inference engine. The final 10 papers in volume 1 discuss generalized learning and ruleinduction techniques. They are interesting and informative, particularly “Generalization and Noise” by Y. Kodratoff and M. Manango, which discusses symbolic and numeric rule induction. Most rule-induction techniques focus on the use of examples and numeric analysis such as repertory grids. Kodratoff’s and Manango’s exploration of how the two complement each other is refreshing. Because of their technical nature and the amount of work it would take to put their content to use, most of the papers in this section of the volume are more appropriate for a specialized or research-oriented group. For those just getting involved in knowledge-based–systems development, Knowledge Acquisition Tools for Expert Systems is the more useful volume. In addition to discussing the tools themselves, most of the papers contain details of the knowledgeacquisition techniques that are automated, thus providing much of the same information which is available in the first volume. As an added benefit, they also often discuss the underlying architectures for solving domain-specific problems. For instance, the details of the medical diagnostic architecture laid out in “Design for Acquisition: Principles of Knowledge System Design to Facilitate Knowledge Acquisition” by T. R. Gruber and P. R. Cohen are almost as useful as the discussion of how to build a knowledge-acquisition system. Volume 2 is particularly germane given the rise in commercial interest about automated knowledge acquisition following this year’s introduction of Neuron Data’s NEXTRATM product and last year’s introduction of Test Bench by Texas Instruments. Test Bench is actually discussed in “A Mixed-Initiative Workbench for Knowledge Acquisition” by G. S. Kahn, E. H. Breaux, P. De Klerk, and R. L. Joseph. This volume provides the background necessary to evaluate knowledge-acquisition tools such as NEXTRA, Test Bench, and AutoIntelligence (IntelligenceWare). The vendors of knowledge-based–systems development tools, for example, Inference, IntelliCorp, Aion, AI Corp., and IBM, would do well to pay heed to these books because they point the way to removing the knowledge bottleneck from knowledge-based–systems development. Overall, the papers in both volumes are comprehensive and well integrated, a sometimes difficult state to achieve when compiling a collection of papers resulting from a small conference. The collection is comparable to Anna Hart’s Knowledge Acquisition for Expert Systems (McGraw-Hill, 1986), but it is broader in scope and not as structured. The arrangement of the papers is marred only by an overly brief index. Few readers can be expected to read a collection from beginning to end, and a better index would facilitate more enlightened use. Less important—but nevertheless distracting—is the large number of typographical errors in both volumes. In conclusion, the set is recommended for both the commercial and research knowledge-based–systems practitioner. Reading the volumes in reverse order might be more useful to the commercial developer given the extra information available in volume 2. Neurocomputing: Foundations of Research