This repository accompanies the study "Encoding Imagism? Measuring Literary Imageability, Visuality and Concreteness via Multimodal Word Embeddings."

The main directory contains three experiments, each with its own .py file.

• In experiment_2.py, set the target file at the top of the script (Hemingway, Woolf, or the Chicago corpus).
• functions.py provides helpers for generating embeddings and dictionary-based scores.
• The data directory holds the textual input, precomputed embeddings (data/embeddings/), and precomputed dictionary scores (data/measures/). These can be regenerated by deleting the corresponding files and rerunning the scripts.
• resources contains the dictionaries used in the analyses.
• figs and results are populated automatically when the experiments are executed.

---

Computing Requirements (Experiment 2)

• Runs comfortably on CPU only; on an Apple M3 (16GB RAM) it completes in a few minutes.
• No GPU needed — all CLIP embeddings are precomputed and loaded directly.
• Requires Python 3.9+ and standard scientific libraries:
  numpy, pandas, matplotlib, seaborn, scipy, scikit-learn, spacy, torch, transformers
• Install the spaCy model:

  python -m spacy download en_core_web_sm