tankanizer

a Markov chain poetry generator with syllabic constraints

Using Allison Parrish's Pronouncing to count syllables, I generate poetry in the tanka form with a corpus of Walt Whitman poems and a Markov chain. A tanka is a poetic form originating in Japan, comprised of five lines in a distinct 5-7-5-7-7 syllabic structure.

NOTE: The corpus was created in this notebook within my poetrydb_poem_generator project.

To use the app locally, run the following in Terminal (after cloning the repo):

streamlit run app.py

To use the app on the World Wide Web, click here!

Future steps

I'd like to allow users to upload a corpus within the app, which would then process the text and create a usable Markov dictionary to act as the source for the tankanizer.

In a larger context, I'd like to use Markov chains across several corpora to generate poetry that I can then use to train a more sophisticated RNN (LSTM) generator. I envision some sort of text generation loop, the visual/written equivalent of audio feedback.

List of files

• data folder - text files to use as a source for the generator.
• .gitignore - list of files to ignore.
• Procfile - requirement for Heroku deploy.
• README.md - this very file!
• app.py - file with app layout and tankanizer function call.
• functions.py - text file with functions, from processing text to the final poetry generator.
• requirements.txt - requirement for Heroku deploy.
• setup.sh - requirement for Heroku deploy.
• tankanizer.ipynb - the main workbook, from loading and processing the corpus to generating some tankas.