Intel l igent User Interfaces

Given the range of systems, tools, and techniques discussed in this collection, it appears that the editors have adopted quite a broad definition of the notion of 'intelligence.' Although the book contains descriptions of computer systems that would be conventionally classified as intelligent--those that aim to develop models of their users--it also includes descriptions of sophisticated tools for assisting with building interfaces, for defining human-computer interaction styles, and for presenting information on computer displays. Consequently, the collection appears to be intended not only for practitioners in AI but also for those in Human-Computer Interaction (HCI). There are two principal reasons why this book may be of interest to computational linguists. Firstly, some of the systems described use computational methods for analyzing and generating language; for example, parsers, semantic networks, and text generators. Secondly, as implementations of techniques from artificial intelligence, the theoretical foundations of many of the papers are similar to those used by many computational linguists; for example, both can rely on notions of users' plans and goals and the assumption that it is possible to infer the users' intentions, plans, and goals from their utterances (cf. Perrault and Allen 1980). Therefore, intelligent interfaces could be viewed as a testbed for both developed and more novel techniques of natural language understanding and generation. The introduction to the collection, by Miller, Sullivan, and Tyler, suggests four requirements of intelligent interfaces: human--computer interactions need to be made clearer; interfaces need to support users' plans and goals; information needs to be presented clearly; and interfaces need to be designed and implemented more easily. These requirements are then addressed in the four sections of the book: multi-modal communication; models, plans, and goals; dynamic presentation design; and knowledgebased tools for interface design. Although Miller et al. state that the requirements of intelligent interfaces are essentially those of human-computer interaction research in general, it is unclear how these requirements have been derived. Rather than defining a framework in which to consider each contribution, the editors use the questions to divide up the contents of the book. This lack of a framework has important consequences for the coherence of the collection as a whole; there are obvious overlaps between what is described in individual sections, and it is also unclear how papers in one section relate to those in another. In this review, only the first two sections of the book will be examined in any detail. These sections contain research and developments that are likely to be of direct interest