Briefly Noted

Examples play a crucial role in instruction manuals, textbooks, and documentation. If they are carefully chosen they can show clearly how a new facility or language can be used, while also supporting conceptual understanding. To be effective, examples should be linked with any textual explanation of a topic, complementing that description. Language generation techniques have been used in several prototype documentation systems--e.g., Reiter, Mellish, and Levine (1995). Documentation can be generated on-the-fly, tailored to characteristics of the particular user, using a knowledge base that can (sometimes) be automatically updated as a system specification is revised. This provides the possibility of improved system maintainability and effectiveness. If such systems are to begin to challenge conventional methods for documentation, then it is essential that they provide integrated and well-chosen examples. This book looks at how examples can be selected within a system to generate natural language descriptions. This builds on the work of the Explainable Expert Systems project (Swartout, Paris, and Moore 1992) and many of the examples illustrate a plan language used in that project. The book is based on the author 's Ph.D. thesis and includes an interesting review of work on the use of examples, taken from several disciplines, some conclusions and illustrations from a corpus analysis of instructional texts, and a detailed description of an implemented system. The book highlights many of the issues that need to be addressed when generating examples that are based on some underlying specification. A particular contribution is highlighting that the use (and categorization) of examples should depend on any surrounding textual explanation. Given the range of interesting issues discussed, the actual system description is a little disappointing. For example, the rather general issue of how the presentation of examples depends on text type is reduced to a simple expert/novice reader distinction, and the dependence of examples on the type of knowledge being conveyed is not addressed. The system, however, suggests a general direction and approach for integrated example generation, and highlights the issues and problems. One problem that is highlighted by the book is the difficulty in generating effective examples given limitations in the underlying knowledge representation. The system described uses representations of concepts that are purely syntactic (simply the BNF definition of allowed constructs). As a result, some of the examples are a little odd. For example, a generated example of the use of the LISP function let is: