Deciphering Storytelling Events: A Study of Neural and Prompt-Driven Event Detection in Short Stories

Detecting events in literary narratives is relatively challenging due to the presence of both realis and non-realis events. Realis events are those that actually take place. Unlike other domains, such as biomedical and news articles, events in literary narratives are not always factual. This paper introduces “Gatha-200”, a manually annotated dataset of 200 short stories for realis events. We employed several neural models for realis event detection. The best-performing neural model reached a macro Fl score of 92.2 %. Furthermore, we also evaluated the capability of the prompt-based learning approach to detect the realis events in zero-shot and few-shot settings and got promising results.