Can Sentiment Analysis Reveal Structure in a "Plotless" Novel?

About This Research

Author: Katherine L. Elkins, Ph.D.
Co-Author: Jon Chun

This repository contains materials related to the early research paper "Can Sentiment Analysis Reveal Structure in a 'Plotless' Novel?" examining Virginia Woolf's To the Lighthouse through computational sentiment analysis.

Research Significance

This work represents pioneering research in computational narratology, published in 2019. It was among the first studies to apply sentiment analysis methodologies to modernist literature, demonstrating that seemingly "plotless" modernist novels contain distinct emotional arcs and underlying narrative structures.

Key Contributions

Methodological Innovation:

• Developed comparative approach using multiple sentiment analysis models (VADER, Syuzhet.R)
• Introduced "middle reading" methodology combining computational analysis with close literary reading
• Validated simple lexical approaches against more sophisticated sentiment detection methods

Literary Discovery:

• Introduced the concept of the "distributed heroine model" - showing how emotional arcs in modernist fiction distribute across multiple characters rather than following a single protagonist
• Demonstrated that Woolf's To the Lighthouse, while lacking traditional plot structure, exhibits strong underlying emotional architecture
• Revealed thematic patterns of connection/separation, coherence/dissolution across the novel

Theoretical Framework:

• Established methods for analyzing "relative" vs "absolute" emotional valence in narrative
• Developed approaches for handling emotionally ambivalent or contradictory moments in modernist texts
• Created framework for selecting appropriate smoothing techniques (DCT, LOESS, Rolling Mean) for different analytical purposes

Foundational Work

This research laid the groundwork for:

• The SentimentArcs methodology published in The Shapes of Stories (Cambridge University Press, 2022)
• Broader applications of computational sentiment analysis across literary texts, medical narratives, social media discourse, and policy documents
• Integration of humanistic close reading with computational distant reading approaches

Publication

Elkins, K. & Chun, J. (2019). "Can Sentiment Analysis Reveal Structure in a 'Plotless' Novel?" arXiv:1910.01441v1.

Read the paper: arXiv:1910.01441

Methodology Overview

Sentiment Analysis Techniques

The study employed multiple approaches:

• Lexical sentiment analysis using Syuzhet.R
• VADER (Valence Aware Dictionary and sEntiment Reasoner) for validation
• Multiple smoothing methods: DCT (Discrete Cosine Transform), LOESS, Rolling Mean

Novel Analysis

Analyzed Virginia Woolf's To the Lighthouse to:

• Test whether modernist fiction exhibits traditional narrative arcs
• Examine how emotional valence patterns reveal underlying structure
• Compare computational findings with traditional literary criticism

Validation

• Compared simple vs. sophisticated sentiment detection approaches
• Validated computational results against human expert interpretation
• Tested against randomized "word salad" versions to verify intentional structure

Key Findings

1. Modernist novels contain strong emotional arcs: Despite lacking traditional plot structure, To the Lighthouse exhibits distinct patterns of emotional valence

2. Distributed narrative structure: Emotional highs and lows distribute across multiple characters, creating a "distributed heroine" rather than following a single protagonist

3. Thematic patterns correlate with emotional arcs:

   • Moments of connection and coherence = emotional highs
   • Moments of separation and dissolution = emotional lows
   • Ambivalent emotions register as "neutral" valence

4. Simple lexical approaches work surprisingly well: Comparison with VADER showed that basic sentiment analysis performs remarkably well on literary texts

Implications for Digital Humanities

This work demonstrates:

• Computational methods can reveal hidden structures in texts thought to break with narrative convention
• Middle reading (between close and distant reading) offers unique insights
• Humanistic expertise + computational tools produce richer analysis than either alone
• Literary language, despite its complexity, is amenable to computational analysis when methods are carefully validated

Related Work

This research contributed to:

• The Shapes of Stories (Cambridge UP, 2022) - comprehensive methodology for sentiment analysis of narrative
• Comparative studies of narrative emotional structure across cultures and media
• Development of human-centered AI approaches to text analysis

Repository Contents

• paper/ - Full research paper (arXiv version)
• docs/ - Additional documentation on methodology
• examples/ - Sample analyses and visualizations

Citation

If you use this work in your research, please cite:

Elkins, K., & Chun, J. (2019). Can Sentiment Analysis Reveal Structure in a 
"Plotless" Novel? arXiv preprint arXiv:1910.01441.

Related Research

By Katherine Elkins:

• The Shapes of Stories: Sentiment Analysis for Narrative (Cambridge University Press, 2022)
• "Beyond Plot" Journal of Cultural Analytics (2025)
• "The Shapes of Cinderella: Emotional Architecture and the Language of Moral Difference" Humanities (2025)

Contact

Katherine L. Elkins, Ph.D.
Professor of Humanities and Comparative Literature
Kenyon College
Email: elkinsk@kenyon.edu

Acknowledgments

This research was conducted at the KDH AI Collaboratory, Kenyon College. Special thanks to the digital humanities community for feedback on early versions of this work.

License

This repository is licensed under the MIT License. See LICENSE file for details.

---

This work represents early computational humanities research (2019) that helped establish sentiment analysis as a valuable tool for analyzing narrative structure in literary texts.