Book Reviews: Semantic Interpretation and the Resolution of Ambiguity

Similar discussions apply to parsing. Fifty thousand words of the corpus have been parsed by hand, and this has been used to make a table of the relative frequencies of different syntactic constructions. Assuming that the correct parse tree is the one ma~de of the most probable constituents (to greatly oversimplify in the interests of saving space), a program was written to parse with about 50% accuracy. Since the preparation of this book, continuing work by Eric Atwell and Geoffrey Sampson at Leeds has greatly improved on this figure, using a simulated annealing technique (see Sampson 1986). Other chapters of the book discuss the history of corpora in linguistic research, a defense of probabilistic methods, a discussion of speech synthesis and an outline of a sophisticated spelling corrector. Not much has been done on speech synthesis, partly because we do not as yet have good data on the relation between syntax and prosody. The spelling corrector is aimed at errors of word selection, i.e., finding words that, although they appear in the dictionary, should not appear in the particular sentence being studied (e.g., "They kingdom come, thy will be done"). All these tools follow the same model: reliance on statistics from the corpus. It is a great relief to read a book like this, which is based on real texts rather than upon the imaginary language, sharing a few word forms with English, that is studied at MIT and some other research institutes (see Postal 1988). It is amazing that computers, which are distinguished for their ability to deal with vast quantities of bytes and their incompetence with even simple patterns and models, have been used in linguistics primarily for the implementation of complex logical models. This book is a start on the exploitation of large database methods for linguistic information. It is remarkable for the performance of its methods combined with their simplicity. Unlike many books on linguistics, it is easy to understand; it makes one think of the Moli~re character who suddenly found out he had been speaking prose all his life. I heartily recommend this book to anyone who wishes to process language for a useful purpose. Other workers such as John Sinclair (1987) and Yaacov Choueka (1988) have also used large text databases for deriving linguistic information. When I was an undergraduate, one of my professors said that "mathematical intuition means having seen the problem before." Similarly, there is no substitute in linguistics for knowing that a particular construction is likely because it has appeared many times. This book is a testimony to the superiority of experience over fantasy.