Applying Generalized Linear Mixed Models to Word Counts to Analyze the Literary Style of Pre-1920 Detective Fiction

• Activity Number: 571
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Roger Bilisoly*+ and Krishna K. Saha
• Companies: Central Connecticut State University and Central Connecticut State University
• Keywords: GLMM ; Stylometry ; Categorical Data ; Text Mining

Abstract

We investigate differences in literary style in pre-1920 detective short stories by A. C. Doyle, G. K. Chesterton, and E. W. Hornung, where generalized linear mixed models (GLMMs) are applied to word counts. To simplify the analysis, only a few restricted sets of words are considered, which form word classes that share both (1) a related meaning and (2) common grammatical features. For instance, one class is the "break verbs," which includes "break," "crack," and "crash," and these use similar verb alternations (types of grammatical structures) in sentences. That word classes exist satisfying both (1) and (2) exist has been shown by the linguists Beth Levin and R. M. W. Dixon. Literary style is measured by fitting GLMMs of word counts, where some of the factors used are author, time period, word choice, and type of grammatical structure. By analyzing both words and grammatical structures, it is hoped that both conscious and habitual patterns of an author can be detected. Finally, the GLMM fitting is done with both SAS and R, and the respective results are compared.