This project builds 50,000 word novels with a top-down mechanical process described below. My example, "request, anything, and kindness.txt", was built using novels from Project Gutenberg.
ParseText.py turns blocks of text into word to word edge lists (both forward and backwards), and labels the resulting output with tags to use in MakeStory.py.
MakeStory.py generates a story by:
- Generating a set of main and secondary characters and a narrator, each with a tag to load sets of word to word edge weights (from ParseText.py) for generating sentences
- Generating a set of scenes by choosing a pair of characters to interact, with each main character having a scnece with most other main characters (up to twice), and each secondary character having a scene with a third of all main characters
- Each scene alternates a paragrah from the narrator, then one character then another until a minimum number of words has been reached (50,000 words / number of scenes).
- Each paragraph is a set of sentences starting either with a random capitalized word, or a topic chosen from the nouns in the previous paragraph