Character Identification in Short Stories

Abstract

Character identification in short stories is crucial for story segmentation and literature understanding but remains difficult due to the intricacies of language and the variety of narrative styles. This paper presents a system that identifies both speaking and non-characters and resolves their references. Speaking characters are identified by speaking references using regular expressions and noun phrase (NP) analysis; non-speaking characters are identified through NP extraction, filtered by animacy detection and Named Entity Recogniton (NER). Pronouns are resolved with a hybrid coreference resolver (CoreNLP + Hobbs). The system was evaluated on five manually annotated stories from Oscar Wilde’s The Happy Prince and Other Tales . The hybrid coreference system achieved an F1 of 0.82, and the character identification pipeline achieves an F1 of 0.95. This work contributes a reproducible pipeline for character identification in literary texts, and its output serves as a foundation for character-based scene segmentation and narrative analysis.