The vocabulary and rhetorical structures in literary studies articles about Molly Bloom: A description, and applications for information science.

摘要

The purpose of this research was to intensively study 11 literary studies articles about the fictional character from James Joyce's Ulysses , Molly Bloom. A typology was qualitatively developed for coding word types, and Rhetorical Structure Theory (RST) (Maim & Thompson, 1987) was used to code sentence roles, or functions, such as background, evidence, thesis, etc. A few sentence roles in addition to those found in RST were used. A goal of the study was to develop rules for selecting sentences from the articles to represent them. A number of analyses were done to further this goal: the distributions of word types and sentence role types within the various articles were counted; the occurrences of word types within sentences of particular roles were found; the bigrams (two-word type pairs) within sentences of different sentence roles were counted. In addition, the number of times sentence role types followed other sentence roles was counted, and the agreement (percentage of word types shared) between sentence roles that followed other sentence roles was calculated. Z-scores (Smajda, 1993) were used to determine when bigrams and followed-by pairs were comparatively frequent. A study of the occurrence of actual “phrases” (real combinations of word-types, as opposed to simple co-occurrence pairs, or bigrams) was done. Chi-square goodness of fit tests revealed that in some cases the distributions of actually occurring word combinations across sentence roles differed from the overall distributions of sentence roles. The quantitative results and qualitative observations were used to generate sentence extraction rules. The qualitatively derived rule for macroparadigmatic articles (those which apply a theory to a literary work or discuss it in comparison with another work) (Fahnestock and Secor, 1991) worked better than the quantitatively derived rules, in many cases yielding coherent, not overly concrete extracts that captured main points of the articles. The macroparadigmatic rule and another less satisfactory one for concrete articles were applied to 21 additional articles.