Book Reviews: Participating in Explanatory Dialogues: Interpreting and Responding to Questions in Context

Moore's book, based on her doctoral thesis, presents her work on the automatic generation of natural language explanations and also gives an excellent summary of previous work in the field. The book is well written, and should be accessible with limited prior knowledge of the field. I would recommend it both to those wanting a detailed overview of Moore's work (partly summarized by Moore and Paris [1993]) and to those wanting a solid introduction to recent work on explanation generation with an emphasis on dialog issues. The book should be of interest to researchers both in computational linguistics (particularly in text generation and discourse structure) and in expert systems. Moore focuses on textual explanations given by expert and advisory systems, and she describes in detail the explanation component of an expert system that advises the user on how to improve their Lisp program. However, it is clear how the basic ideas and techniques presented apply to a wide range of applications in which explanations need to be provided, such as help systems, online documentation systems, tutorial systems, and so on. Moore starts from the point of view that explanation should be an essentially interactive process, requiring a dialog between the person or system giving the explanation (the advisor) and the person receiving the explanation (the advisee). Without such a dialog the chances of the advisee obtaining the information required, and in a form they understand, are much reduced. In order to participate effectively in such a dialog the advisor must be able to respond appropriately to follow-up questions after the initial explanation is given. Moore argues that this requires that the advisor understand the context of these questions, and in particular the context created by the advisor's previous responses. To do this, the advisor must be able to reason about its own previous responses. This theme is developed throughout the book, through the detailed discussion of a particular implementation of the ideas. Moore starts off with two very useful and clearly written review chapters. Some people might choose to study the book primarily for these reviews, which might indeed be suitable as a basis for postgraduate-level introductory courses on explanation in expert systems. The first chapter argues that if an expert system is to provide effective explanations, then the system itself must be designed with that in mind, to ensure that the knowledge required for possible explanations is explicitly represented. Prior