Excavate

A NaNoGenMo 2019 project by Mike Lynch

A recurrent neural net trained on Dumas' The Three Musketeers extracts 50,000 words from Burton's Anatomy of Melancholy, after the style of Tom Phillips' A Humument

I PROGNOS MEMBERS

The final, 51,115 word version:

I PROGNOS MEMBERS plain | humument-style

Athos & Porthos & Aramis & d’AtGuy

A spin-off: two simulations of Dumas omitting a most common glyph.

Drafts

Earlier versions which didn't quite get to 50k words.

• streaming250k10: plain | humument-style
• streaming250k40: plain | humument-style

Technical notes

A blog post outlining the excavate algorithm, how I modified the base torch-rnn code and some of the technical issues, with a few more samples of its output.

Instructions on how to run the code, along with the pre- and post-processing scripts, are in the Excavate project's README.md file

Source code

• My fork of J C Johnson's torch-rnn extended to constrain the RNN's output to a vocabulary file extracted from another work.
• Pre- and post-processing scripts for extracting vocabularies and formatting the HTML versions.