A Diachronic Corpus for Literary Style Analysis

A Diachronic Corpus for Literary Style Analysis Carmen Klaussner, Carl Vogel Computational Linguistics Group, Trinity Centre for Computing and Language School of Computer Science and Statistics Trinity College Dublin klaussnc@tcd.ie, vogel@tcd.ie Abstract This research presents a resource for diachronic style analysis in particular the analysis of literary authors over time. Temporal style analysis has received comparatively little attention over the past years in spite of the possibility of an author’s style frequently changing over time, a property that is not only interesting in its own right, but which also has implications for synchronic analyses of style. The corpus contains 22 American literary authors from the mid 19th to early 20th century that wrote largely in parallel. After describing the resource, we show how the corpus can be used to detect changing features in literary style. Keywords: Diachronic literary corpus, style analysis, linear regression 1. Introduction The analysis of the authorial fingerprint through compu- tational means, referred to as stylometry, is to be distin- guished from manual analyses relying on the verdict of lit- erary scholars particularly acquainted with the author(s) in question. A matter complicating this type of analysis is the fact that an author’s style only takes shape through the comparison with other contemporaneous authors, the ex- act selection of which determines how close one comes to the “actual” fingerprint. For instance, if Mark Twain uses the word ‘but’ more frequently than other conjunctions then this is only interesting or useful in telling his style apart from others if comparable American authors of his time use the same feature at different rates. Additionally, one may consider this type of analysis to hold an intrinsic flaw: despite the fact that most authors com- pose their published writings over a period of 20-40 years, this diachronic property is not widely taken into account. This neglect poses an issue for synchronic style analyses in at least one way. Unless style is found to be invariant for an author and does not change with age and experience, temporality can be a confounding factor in stylometry and authorship attribution (Daelemans, 2013). In the best case, the (synchronic) stylistic analysis might merely select those elements stable over time, as this renders them also stable and consistent over the author’s corpus, heedlessly discard- ing those elements that show temporal variation and inci- dentally possibly style development. In the worst case, all features considered are affected by time and the most sta- ble features over the corpus are too variable to discriminate well. Analysis of diachronic elements of style requires accurately time-stamped data, i.e. either reflecting the time of compo- sition or publication year. This paper describes the devel- opment of a parallel literary corpus that allows for compar- ison among authors as these are both temporally annotated and partially aligned. To the best of our knowledge, this is the first open-source derived literary corpus containing time-stamped texts, which have been collected with the first publication date in mind. Hence, this paper is aimed at pre- senting …

Content truncated to 500 words