Color Me Intrigued: Quantifying Usage of Colors in Fiction

Siyan Li

Abstract

We present preliminary results in quantitative analyses of color usage in selected authors' works from LitBank. Using Glasgow Norms, human ratings on 5000+ words, we measure attributes of nouns dependent on color terms. Early results demonstrate a significant increase in noun concreteness over time. We also propose future research directions for computational literary color analytics.

Introduction

While literary scholars have long analyzed color symbolism—from the green light in The Great Gatsby to the "red death" in Poe—most analyses remain qualitative and limited to specific authors or eras. This research seeks to bridge the gap between traditional literary criticism and computational linguistics by quantifying the "vividness" and "emotional value" of colors in text. By analyzing how color usage shifts across centuries, we can better understand the evolution of descriptive language in English fiction.

Methodology

The researcher, Siyan Li, employed a multi-step computational pipeline to analyze color usage:

  1. Dependency Parsing: Using the Stanza library, the study identified "color-modified nouns" through dependency links (e.g., "eyes of sapphire"), allowing for the capture of complex relationships beyond simple adjective-noun pairs.
  2. Psycholinguistic Mapping: Words were mapped to the Glasgow Norms, a dataset of human ratings for dimensions like Imageability (how easily a word conjures a mental image), Concreteness (sensory experience), and Valence (emotional positivity/negativity).
  3. Predictive Modeling: To handle words not found in the Glasgow Norms, a Multi-Layer Perceptron (MLP) was trained on FastText word embeddings to predict psycholinguistic scores.
  4. Diachronic Analysis: The team calculated Pearson’s correlations between publication years and average psycholinguistic scores to track historical shifts in usage.

Key Insights: The Concrete Evolution and Authorial Fingerprints

The most striking finding is the "Concrete Evolution": color usage has become significantly more concrete and imageable over time. In pre-1800 literature, colors were frequently tethered to abstract emotions or moral states (e.g., "pink" associated with "shame" or "guilt"). In contrast, post-1900 literature primarily uses colors for physical, sensory descriptions (e.g., "pink" associated with "cheeks" or "flowers").

Beyond broad historical trends, the methodology successfully captured unique authorial "fingerprints." For example, the color red in Edgar Allan Poe’s The Masque of the Red Death showed an abnormally low valence because it was linked to "death," "stain," and "horror," whereas other authors might use red for "roses" or "lips."

Comparative usage of 'yellow' in works by Fitzgerald (red) and Joyce (blue)

As visualized above, different writers use the same color in distinct ways. In F. Scott Fitzgerald’s work, "yellow" often modifies facial hair (beards, pompadours), while in James Joyce’s work, it frequently modifies food items like soup, cheese, or pancakes.

Quantitative Results

Statistical analysis revealed significant upward trends in Imageability (IMAG) and Concreteness (CNC) over time for several colors, most notably black, white, yellow, and pink. For the color pink, the correlation between publication year and Imageability was particularly robust (r = 0.534, p < 0.001).

Historical trends in Imageability (IMAG) for various colors over time

The predictive models also performed exceptionally well, achieving high Pearson correlation scores on held-out test sets (0.79 for IMAG and 0.83 for CNC), suggesting that modern word embeddings carry significant psycholinguistic information that aligns with human perception.

Limitations and Future Directions

While these results are promising, the study acknowledges several limitations. The current dataset (LitBank) is heavily skewed toward 19th-century realistic novels and lacks a deep well of pre-1800 data. Future research will explore how color usage differs between prose and poetry—where colors may remain more symbolic—and how color associations reflect socio-economic class or cultural biases. Additionally, the challenge of "domain shift" (how word meanings change over time) remains a key area for improving NLP models in historical contexts.

Summary updated

Abstract

We present preliminary results in quantitative analyses of color usage in selected authors' works from LitBank. Using Glasgow Norms, human ratings on 5000+ words, we measure attributes of nouns dependent on color terms. Early results demonstrate a significant increase in noun concreteness over time. We also propose future research directions for computational literary color analytics.
BibTeX
@inproceedings{li2022color,
  author = {Siyan Li},
  title = {{Color Me Intrigued: Quantifying Usage of Colors in Fiction}},
  booktitle = {creativeAI @ AAAI},
  year = {2022},
  url = {https://openreview.net/forum?id=tqCf3xklDG},
}
Figure