מרק לסט

The long-term goal of this research is to evaluate the accuracy of known literature patterns in narrated text.Following Lima et al. who suggested literary patterns for each of the following five narrative genres: Comedy, Mystery, Romance, Satire and Tragedy, we explore the prevalence of those patterns (or their elements) are indeed prevalent in Award-Winning Novellas. We have manually annotated 35 award-winning novellas for their narrative genres and compared the results to automated annotation by a Large Language Model (LLM), which analyzes the five narrative genre patterns and their structural elements. Statistical tests (such as binomial test, Herfindahl-Hirschman index (HHI)) indicate that there is a high recall and low precision of the LLM with respect to human annotation. Our feature selection analysis reveals that Closure, Final Confrontation, and Return are the most structurally central pattern elements across all genres. We also perform an element by element analysis to identify the structural elements, which are most informative for genre identification. This study provides the first step towards our long-term goal by proposing a methodology evaluation of literature patterns and demonstrating the proposed methodology on five common patterns in a small dataset of 35 award-winning novellas.