Rethinking Scene Segmentation. Advancing Automated Detection of Scene Changes in Literary Texts

SummaryAutomated scene segmentation remains a central challenge in computational literary studies (CLS), especially for narrative texts with fluid boundaries. This study presents a new approach to detecting scene changes in 20th-century US-English romance fiction. Building on manually annotated data from the Harlequin “Men Made in America” series and other popular fiction texts, we fine-tuned the pretrained transformer-based model Universal Sentence Encoder (USE) to identify scene changes in six-sentence segments. AcknowledgmentsMany thanks to our student annotators / other project members Mallen Clifton, Agnes Hilger, Jessica Monaco, Alexander J. Sherman, and Ellen Yang, who supported the project with their close readings, manual scene annotations, and discoveries of unconventional scene structures. The conference travel was funded by the priority program SPP 2207 Computational Literary Studies (CLS) of the German Research Foundation (DFG) as part of the Young Researcher Visiting Programme 2025.