I put this corpus together some time back to explore recurrent neural nets using torch-rnn. It is now the foundation of a larger literary project I'm developing called Billy as an extension of the folktale podcast I coproduce, Old Wives.

The work of training and sampling models is left to you, but the minimum groundwork of preprocessing is represented in this repo.

This material is pulled from Project Gutenberg, and has only been scrubbed of extraneous white space and extratextual matter.

It's worth noting that these folktale collections were collated around the turn of the Twentieth Century, and as a result may consist of elements insensitive to cultural progress. As with all machine learning projects, care should be used in generating automatic content.

If you find text missing (it's possible a volume or two were missed, though I don't think so) or that does not belong (less likely), feel free to open a PR. Likewise, I'd like to see any work produced via this repo, and would welcome an index of projects and recipes should any arise.

Please note: the GPT-2 corpus is a work in progress, and it's included here for ease of access. I'm not certain of it's correctness is my point, and it is likely to change.