Computational Stylistics: A Study of Enjambment : Automatic Detection of Enjambment

Technologie des languesUnlike rhyme or metrics, enjambment has not been widely investigated using computational means, and research on the subject has not reaped the benefits of new technologies. Most definitions only highlight the breakup of related units over two lines, notwithstanding the evolution of versification with free verse. Stylistic studies are based on these restricted definitions and often limited to small samples of poems by a specific author. Existing studies of enjambment thus show several shortcomings: the definition of the phenomenon is blurred, especially regarding free verse poetry, and quantitative analyses do not abound. In this context, this study aims to provide a tool to automatically detect enjambment in English poetry. Specifically, a detailed syntax-based typology of enjambment was adopted, which was originally created for Spanish and was adapted in this work to better fit English syntax. A tool was then developed that detects and classifies instances of enjambment according to syntactic criteria. To develop the tool, a corpus was collected and annotated iteratively: first with the Spanish specifications, which were then adapted to better fit English poetry and syntax. This corpus served as the gold standard to evaluate the tool quantitively. Rules for enjambment detection were defined, based on the typology of enjambment, and implemented. The tool detects 85% of enjambments and obtains mixed results for classification. The relevance of the tool regarding research in literature was evaluated by conducting a case study on 632 poems. The results showed strong correlations between the poetic conventions of three periods (18th, 19th, and 20th centuries) and the number of enjambments found per period. Furthermore, it also reflects known facts on Milton’s use of enjambment compared to his contemporaries. These results show that such a tool is capable of drawing valid insights on the phenomenon according to various criteria.Thèses et écrits académique