Briefly Noted

Typological discourse analysis is the study of those factors that affect the choice of constructions in a language, taking the surrounding discourse context into account. Typically, texts from a large number of languages are analyzed quantitatively in order to discover statistical regularities between functional parameters affecting the choice of alternations. Examples of alternations are active and passive in English or subject-verb and verb-subject word order in Spanish, and parameters encode information such as "If the NP representing the agent of the action is non-human, the passive construction is relatively more likely" (p. 3). Typological Discourse Analysis is structured as follows. The first chapter introduces typological discourse analysis. This is an extremely readable overview of what is involved in typological discourse analysis. Typical discourse parameters, such as humanness, recency of mention of referent, specificity, and affectedness of patient, are clearly illustrated, and the methodological questions of objectivity, generality, and functional classification are discussed. The remainder of the book illustrates these issues with detailed analyses of noun-phrase types, tense and aspect, voice alternations, word-order variation, and a series of lessstudied phenomena such as Spanish clitic climbing and subject marking in Korean. The final chapter concludes with a personal view of the difficulties involved in determining differences of meaning between alternations and in relating these differences to mental processes. At first sight, this book would not appear to have much relevance for computational linguists. The approach is primarily descriptive, and more abstract explanatory principles underlying the phenomena described are generally avoided. However, there are several reasons why typological discourse analysis in general, and its presentation in this book in particular, should be of interest. Much linguistic theory, as well as work in computational linguistics, is based on a narrow range of languages-often English and perhaps a few other European languages. Analysis is usually sentence based and syntactically oriented, so that the functional characteristics of constructions in their discourse contexts tend to receive scant consideration. However, it could be argued that any sophisticated languageunderstanding system needs access to such functional parameters in order to determine the meaning of a construction. This book provides sound arguments for a set of universal functional parameters and a rich set of examples of how they are applied in a range of languages. The main usefulness of this work, however, will be for those working in language generation, as a detailed account of the functional properties of constructions is a prerequisite for the most appropriate choice of one alternation over another in a particular discourse context. As most work to date in language generation has tended to be restricted to English and a few other languages, the cross-linguistic data presented here should provide both a guide and a challenge for theories of language generation. In sum, this book provides a detailed account of typological discourse analysis with a clear and balanced account of the methodological problems encountered in the empirical analysis of cross-linguistic data.--Michael McTear, University of Ulster