Research / Sentiment Analysis and Narrative Intelligence
Sentiment Analysis and Narrative Intelligence
Elkins’ sentiment-analysis work develops methods for tracing emotional structure across narrative. Beginning with hybrid computational-close-reading work on modernist fiction and extending through The Shapes of Stories and later work on narrative intelligence, this research treats emotion not as ornament but as one of the organizing structures of story.
The method has traveled well beyond literary studies. Diachronic sentiment analysis has been used to study novels, translations, film and television scripts, political speeches, end-of-life narratives, social media crises, collective emotion, sustainability narratives, and AI-generated stories. Across these settings, the central claim remains consistent: emotional arcs can reveal structures that plot, character, topic, and event analysis often miss.
Selected work and projects
The Shapes of Stories: Sentiment Analysis for Narrative
Cambridge Elements in Digital Literary Studies, Cambridge University Press, 2022
The Shapes of Stories develops ensemble methods for diachronic sentiment analysis applied to narrative texts. It introduces a vocabulary for narrative shape — storyteller curves, curves-on-a-hill, curves-in-a-hole, tragic curves, person-on-the-plain, and distributed hero(ine) — and proposes middle reading as a methodological bridge between distant reading and close reading.
The book has been taken up across computational literary studies, NLP, visualization, cognitive science, and sustainability research. It is discussed alongside foundational work by Reagan et al., Jockers, and Nalisnick and Baird in computational approaches to literary sentiment. Bilstrup et al.’s Litteraturmaskinen (ACL 2026) names The Shapes of Stories for aesthetic positioning alongside work on character analysis, narrative structure, literary quality, and canonicity. He, Breithaupt, Kübler, and Hills’s Scientific Reports study of 25,728 story retellings uses the book as a methodological foundation, while later uptake appears in Hamilton, Wilkens, and Piper’s NarraBench (EACL 2026), Yeh et al.’s Story Ribbons (IEEE TVCG 2026), and Rebora’s Digital Humanities Quarterly field survey. Nielsen, Christensen, and Bolwig name sentiment analysis, including Elkins 2022, as one of two quantifiable methods that could scale comparative sustainability-narrative analysis.
Can Sentiment Analysis Reveal Structure in a “Plotless” Novel?
Elkins and Chun, arXiv 1910.01441, 2019
This early paper applies sentiment analysis to Virginia Woolf’s To the Lighthouse, a modernist novel often described as plotless. It combines computational analysis with close reading to show that even a novel with minimal conventional plot can exhibit a strong emotional structure.
The paper introduces the distributed hero(ine) model, in which emotional structure is not organized around a single protagonist but distributed across characters, perspectives, and scenes. This finding became one of the conceptual anchors for the later claim that sentiment analysis can surface narrative structures not reducible to plot or character arc.
What the Rise of AI Means for Narrative Studies
Chun and Elkins, Narrative 30.1, 2022
This response essay argues that AI’s role in narrative studies is no longer a question of whether computational systems can matter for literary interpretation, but how scholars should prepare for their use. Written in response to Angus Fletcher’s “Why Computers Will Never Read (or Write) Literature,” the article situates narrative studies within the longer history of AI, explaining why advances in machine learning and language modeling require new forms of collaboration between computational and humanistic methods.
The essay marks a field-level transition: from asking whether computers can read or write literature to asking how narrative scholars can evaluate, guide, and critique AI systems as they enter literary and cultural analysis. Published before the public release of ChatGPT, it connects computational narrative analysis, AI, and literary study at a moment when the field’s assumptions were beginning to shift.
Beyond Plot: How Sentiment Analysis Reshapes Our Understanding of Narrative Structure
Journal of Cultural Analytics, 2025
This article synthesizes the broader methodological argument: sentiment analysis can identify latent emotional structures that challenge traditional assumptions about plot, character, and narrative form. It argues that emotional arcs in fictional and non-fictional narratives often surface structures that are otherwise difficult to see, and that feature extraction frequently correlates with passages critics select for close reading.
A central claim is that close reading itself has often been undertheorized. Critics regularly choose passages because they feel formally or emotionally charged, but the field has fewer ways to describe how affect, compression, intensity, and interpretive salience guide that selection. The article reframes sentiment analysis not as a replacement for close reading, but as a way to make visible why certain passages invite interpretation in the first place.
The article extends sentiment analysis across media and discourse types, including novels, translations, film, end-of-life memoirs, political speeches, television transcripts, financial-crisis social media, and anti-Asian sentiment during COVID. It was selected as an Editors’ Choice by Digital Humanities Now.
eXplainable AI with GPT-4 for Story Analysis and Generation
Chun and Elkins, International Journal of Digital Humanities 5, 1–26, 2023
This paper develops Ensemble Cross-Correlation and Ensemble Polarity Confidence metrics, along with a human-in-the-loop XAI framework for story analysis and generation. It connects sentiment analysis to explainable AI by asking how narrative arcs can be evaluated, compared, and interpreted across models.
The paper has traveled beyond literary analysis into political theory and urban-governance scholarship. Federico Cugurullo and Ying Xu engage it in Policy and Society as theoretical authority on LLMs’ obscure epistemological processes.
The Shapes of Cinderella: Emotional Architecture and the Language of Moral Difference
Humanities 14(10), 198, 2025
This article applies emotional-arc modeling and close reading to Cinderella variants across cultures, including Ye Xian, Perrault’s Cendrillon, and two Grimm versions. It argues that stories with similar recognition scaffolds can encode distinct emotional architectures and moral worlds.
The piece extends sentiment analysis into cross-cultural narrative comparison, showing how emotional architecture can reveal differences in moral language, social world, and narrative expectation even when plots appear structurally similar.