How Are Emotions Expressed in Literary Fiction, and Can Language Models Detect Them?

ACL ARR 2025 May Submission619 Authors

Abstract: This paper investigates how emotions are expressed in 19th-century Danish and Norwegian literature and whether contemporary language models can detect them. We introduce a linguistically and culturally grounded annotation scheme distinguishing conceptual, expressive, and non-verbal emotion expressions. Applying this scheme, we construct a multi-label dataset of sentences from the MeMo corpus, annotated with nine emotion categories based on Plutchik’s theory. We evaluate seven Danish and Norwegian pre-trained language models on this task and propose a novel Plutchik-aware Soft-F1 metric that accounts for affective proximity between emotion categories. Our results show that while models like DFM-Large achieve strong performance on standard metrics, they still struggle with overlapping and subtle emotional expressions common in literary texts. The study highlights the challenges of operationalizing emotion theories in NLP and the importance of interdisciplinary approaches to modeling affect in historical and narrative domains.

Paper Type: Long

Research Area: Special Theme (conference specific)

Research Area Keywords: emotion detection and analysis, NLP tools for social analysis, corpus creation, benchmarking, language resources

Contribution Types: NLP engineering experiment, Data resources, Data analysis

Languages Studied: Danish, Norwegian

Submission Number: 619