An introduction to text processing by Peter D. Smith, The MIT Press, Cambridge, MA, 1990, pp 300

power of the meta-level in handling the object knowledge a discussion of this would have made for a good book. MDX is discussed in detail in Chapter 4, and is projected onto the "distributed architecture". A much clearer statement of the architecture and its components is needed by this point, together with a discussion of how it differs in substance from other expert system architectures. Chapter 5 looks at the PATREC component of MDX, and finds it lacking in many respects, in particular that its knowledge is encoded as LISP functions' ("an arcane representation of knowledge"). To a certain extent tjie discussion here of data abstraction is a distraction. As long as the meta-level can manipulate the knowledge as required, it is unimportant how it is represented. Chapters 6 to 8 cover the authors' own research in the area of competent expert systems. The principal reason for adopting the "distributed architecture" was to enable competent expert systems to be built, and so competence should have played a much greater part in these chapters. Chapter 6 is a detailed case study of FIBERS, which is a shell for the management of "findings base" (one of the components of the "distributed architecture"). The following chapter illustrates how to integrate a FIBERS findings base with a task-structured hypothesis engine (Appendix B gives an example of the system in action). The transition from Chapter 7 to Chapter 8 is a little disorientating, as we are suddenly thrown into the world of knowledge engineering environments with a "sales-pitch" for KEE. The final chapter concludes that "To corroborate the arguments presented here, one (or more) full distributed expert system needs to be developed in real-world domains of expertise". The book would have benefited here from giving a present-day view has a "full distributed expert system" been built, does it satisfy the requirement of competence, and how does this work now fit into a broader picture of expert systems development? The untimeliness of this book is exacerbated by the realization that there are only about 120 pages of real text, to which is added about 30 pages of code listings. At £25 that does not seem like very good value for money.