An attempt to experimentally test Edward Said's claims about late style using computational text analysis and principal component analysis.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
bibliography
data
paper
results
.gitignore
LICENSE.md
README.md
get-all-authors.ipynb
get-corpora.ipynb
late-style-PCA.ipynb
late-style-rewrite-everyone-semantic.ipynb
late-style-rewrite-everyone.ipynb
late-style-rewrite.ipynb

README.md

Does Late Style Exist?

This repository contains code and prose related to a computational literary analysis which attempt to test literary critical claims of authorial periodicity known as "late style," and popularized most recently by Edward Said's book On Late Style.

Slides

A presentation from the Digital Humanities 2018 conference, showcasing some early results is available here on my website.

Contents

  • Single-author corpora used are in /bibliography.
    • everyone.csv contains the full corpus used for the larger "corpus B" analysis, created using Corpus-DB
    • late-style.bib is a BibTeX file containing references to academic books and articles discussing late style. If you know of any others I'm missing, please let me know! And/or submit a pull request with the changes.
  • /paper contains a very early draft of a paper, which discusses very preliminary results. Warning: this doesn't represent the current state of the project!
  • late-style-PCA.ipynb is the first attempt at this experiment, using a small, curated corpus.
  • late-style-rewrite.ipynb is the first attempt at the more probabilistic, Bayesian approach to this problem.
  • late-style-rewrite-everyone.ipynb tests "corpus B," about 51 authors, and about 1K texts from Project Gutenberg
  • late-style-rewrite-everyone.ipynb tests "corpus B" again, but uses semantic document embeddings instead of word frequencies

License

GPLv3. See LICENSE.md.

Contributing

I could use some help with this! Get in touch, or open a new issue. Pull requests are also welcome.