a recurrent neural net trained on a dictionary of Regency slang

Francis Grose was an antiquarian and lexicographer who compiled one of the most famous and entertaining compilations of English slang, the Classical Dictionary of the Vulgar Tongue. (The Grose River in New South Wales was named in honour of his son, who was lieutenant-governor.)

I've trained the same kind of recurrent neural network that powers GLOSSATORY on the Project Gutenberg text of the expanded 1811 edition. As with GLOSSATORY, the formatting of the output into paragraphs defining terms in all caps is learned automatically by the network. The only postprocessing is filtering to ensure the definitions aren't too long.

Toots by @vulgar

Part of the appeal of Glossatory is the puerile sport of hoping that the recombination of mundane definitions will make something rude. VulgaRNN works in the opposite direction, turning the bawdy original into something that's often less obscene but weirder, while still retaining the Georgian style of the original, and most of its recurring cast of bawds, rogues, harlots, sailors, dolts and country fellows. There's a much longer sample of the output in my blog post.

I'll be tuning the settings over the next few weeks to try to maximise the entertainment value.

It's also available in bigger slices on Mastodon:

Mike Lynch 2018

Technical notes and credits

VulgaRNN runs on the same Python bot and torch-rnn stack as @GLOSSATORY.