Variation in English: Multi-Dimensional Studies. Edited by Susan Conrad and Douglas Biber. Harlow: Longman, 2001, 255 pp.
Multi-Dimensional analysis is Biber's term for a method for describing language variation based on the statistical procedure called factor analysis, which is used to group grammatical features that co-occur into what are termed Dimensions. For example, Dimension 1 contrasts personal involvement against information density, and texts with a high score on Dimension 1 (corresponding to a high level of personal involvement) tend to have a large number of contractions, present tense verbs, second-person pronouns, and hedges (among a host of other features), but relatively few nouns, prepositions, attributive adjectives, or long words. As one might expect, face-to-face conversations and romantic fiction tend to have a high score on Dimension 1, while official documents and academic prose tend to have a low score.
The basic idea of Multi-Dimensional analysis is that, while certain grammatical features are found more regularly in some linguistic registers than others, the true picture only emerges if clusters of features are considered together rather than each feature in isolation. This book seeks to illustrate the use of this analysis technique in a wide range of studies, including diachronic consideration of language change in scientific papers, comparison of British and American usage, differences in the language used by various writers commenting on American nuclear policy, variation in the oral proficiency interviews of second language learners of English, and gender-specific contrasts in the dialogues found in episodes of Star Trek.
The application of Multi-Dimensional analysis depends on the use of statistical software to conduct a factor analysis on frequency counts of linguistic features in corpora of texts, and an in-depth understanding of the technique would require substantial knowledge of statistics. However, this book aims to make the findings of the technique easily accessible for non-specialists, and it is remarkably successful in presenting the findings of the various studies without getting bogged down in heavy-duty statistics. Moreover, in each chapter, in addition to the charts that show how the scores along the various Dimensions capture register variation, there are a number of example texts to illustrate the points, and this does bring the exposition down to earth very effectively.
Multi-Dimensional analysis originated with the work of Douglas Biber in the 1980's, and Biber is both one of the editors of this book and the sole or joint author of seven of the fourteen papers included in the book. Furthermore, in the References, there are no less than twenty-seven papers or books with him as the sole or first author, and his influence is so pervasive that it seems to create an orthodoxy which the other writers are hesitant to challenge. Even when some of the results are unexpected, such as the scores for Dimension 4 (which reflects overt persuasive effort) for texts on American nuclear arms policy being "dramatically higher than all registers analyzed by Biber" (p.91), the author of the study seems reluctant to question the model itself.
It also seems likely that, while the use of general-purpose Dimensions may facilitate comparisons between a wide range of different texts, many studies might benefit from consideration of alternative combinations of features. For example, in an analysis of historical shifts in gender-specific language, it is reported (p.163) that both hedges and downtoners are important as markers of tentativeness in reflecting personal involvement for Dimension 1, but the introductory chapters indicate (p.22) that downtoners actually have a very slight negative loading for Dimension 1 (though the value is too small to be considered important). This seems to suggest that analysis of gender-specific language would benefit from an alternative combination of linguistic features, and when the general-purpose Dimensions are used, there is a danger that the joint significance of grammatical features such as hedges and downtowners as markers of tentativeness might get drowned out by the other components of Dimension 1.
In fact, there are some substantial reservations about Biber's model of Multi-Dimensional analysis. David Lee (forthcoming) tries out the model on texts from the British National Corpus, and he suggests that Biber's analysis of English is not the only possible one, particularly since the quantitative definition of 'general English' in terms of proportions among the constitutive genres has not yet been agreed upon. Since the results of the statistical procedure used to derive Biber's Dimensions are crucially dependent on and reflect the actual make-up of the corpus Biber used, this poses the question of how stable and valid Biber's Dimensions for 'general English' really are. In support of this, Lee reports difficulties in replicating some of Biber's original results, and overall he concludes that caution is needed in interpreting the outcome of Multi-Dimensional analysis techniques on mixed-data corpora.
Despite these reservations, this book does provide a very clear overview of an important field. Even if it turns out that the model needs some refinement or even substantial revisions, it has certainly provided valuable insights into the analysis of a variety of texts, and this highly accessible survey both of the basis of the technique and of some of its applications is most welcome.
Lee, David (forthcoming) Modelling Variation in Spoken and Written English: the Multi-Dimensional Approach Revisited. (To be published by Routledge.) [Abstract available at: http://clix.to/davidlee ]
Acknowledgement: I am grateful to my colleague, Paul Doyle, for his valuable comments on Multi-Dimensional analysis.
From: SAAL Quarterly Vol 59, August 2002