Review: Fitzpatrick (ed., 2007) Corpus Linguistics Beyond the Word: Corpus Research from Phrase to Discourse Amsterdam: Rodopi

Abstract

This book, edited by Eileen Fitzpatrick, is a collection of fifteen papers that were first presented at the North American Symposium on Corpus Linguistics in 2004. Given the breadth of corpus studies included in the book, it is logically separated into two sections: the first section contains papers addressing corpus tools and strategies, and the second section contains papers relating to the application of corpus analysis. In the first paper Barrett, Greenberg and Schwartz explore syntacticlevel grammatical structure uses within texts of specific domains or genres. The development of statistical methods for automatically classifying texts into domains for the purposes of creating training and testing corpora for machine translation systems is the end goal of the authors’ research. Results show that syntactic-level differences exist within specific domains, providing evidence that text classification for testing and training corpora is possible beyond the lexical level. In the second paper Grieve-Smith examines register and genre variability using multi-dimensional analysis. Drawing from sociolinguistics, the concept of ‘envelope of variation’ is integrated into multi-dimensional analysis and is used to analyse the co-occurrence of two grammatical features (third-person pronouns and demonstrative adjectives) in twelve texts. While the inclusion of envelope of variation into the multi-dimensional analysis method did not result in statistically significant outcomes, the conclusion drawn from the study demonstrates that grammatical (and situational) variation is complex and requires detailed analysis of the feature(s) in advance of applying multi-dimensional analysis. An original method for determining word similarity scores is presented by Deane and Higgins in the third paper. In contrast to other methods (e.g., Latent Semantic Analysis, Random Index) that rely on topicbased texts, Deane and Higgins’ method uses the local contexts in which a