Modeling Textual Emotions in Literary Fiction

This paper explores how finetuned transformer models can be used to examine the vast and underexplored emotional-historical dimensions of literary history. Drawing on a corpus of 859 nineteenth-century Danish and Norwegian novels, we develop an interdisciplinary framework for analyzing the expression of emotions in literary texts. This paper presents three main contributions: (1) we introduce an annotated multilabel dataset of emotion categories for 859 nineteenth-century Danish and Norwegian novels; (2) we evaluate the performance and generalization capabilities of large language models in classifying emotions in long and complex texts; and (3) we propose a theoretical framework that integrates insights from psychology, linguistics, literary theory, and NLP to support future research on cultural and historical formations of emotions. By combining computational methods with explicit humanistic theorization, this approach not only enables large-scale analysis of literature’s emotional dimensions but also demonstrates how methods used in the digital humanities can serve as a gateway to new theorization of emotions and their textual manifestations, offering valuable contributions to humanities scholars engaged in the so-called affective turn.