a simulaton of Robert Burton's Anatomy of Melancholy
By this art you may contemplate the variation of the twenty-three letters, which may be so infinitely varied, that the words complicated and deduced thence will not be contained within the compass of the firmament; ten words may be varied 40,320 several ways...
Robert Burton's Anatomy of Melancholy is one of my favourite books, a vast and sprawling compilation of seventeenth-century wisdom about mental illness. A few years ago I trained a neural net on it, but I didn't get around to doing anything with it until I used it as one of the sources for my NaNoGenMo entry for 2019, I PROGNOS MEMBERS.
The book is full of Latin verse maxims followed by English translations, and the RNN has picked up this habit. I like the verses, so I adapted the Python bot framework that runs @GLOSSATORY to look for them and prefer them over ordinary sentences when it samples the RNN's output.
There is also an oulipian version, which takes advantage of the modifications I made to the RNN library to allow suppression of certain characters. It has its own page on this site which has a little more information about Burton and his literary descendants.
Making obscure and strange bots seems like an appropriate activity for the lockdown, and I can't help adding another quote (Burton is the most quoting of authors, and, for me, one of the most quotable) from the Preface of the Anatomy, which is longer than many books, in which Burton, under his alias of Democritus Junior, defends his cloistered life:
I have followed all, saving that which I should, and may justly complain, and truly, qui ubique est, nusquam est, which Gesner did in modesty, that I have read many books, but to little purpose, for want of good method; I have confusedly tumbled over divers authors in our libraries, with small profit, for want of art, order, memory, judgment. I never travelled but in map or card, in which mine unconfined thoughts have freely expatiated, as having ever been especially delighted with the study of Cosmography.
It was strange that on the day I launched these bots ("toys", Burton might have called them, rightly, although the meaning of that word has shifted since his day), I read, in this moving essay on the pandemic by another of my favourite writers, the philosopher Justin E H Smith:
None of that discourse is any more germane to thinking about the present situation than, say, Robert Burton, or Galen, or St. Theresa of Ávila. Read whatever you want to read now, and don’t be distracted by those writers who are so set in their ways that they know no other strategy than to recover formulae devised back in the old world, and to retry them in the new one, like stubborn Norsemen struggling to graze cattle in Greenland, when the world they find themselves in demands they learn to hunt seals.
If nothing else, I hope these toys inspire you to look into their original.
So that I could write a self-contained parser for the features of the Anatomy of Melancholy RNN (which had quirks like generating fake inset page numbers -  - copied from the Library of Gutenberg text I trained it on) I tidied up and modularised the Python bot code I use to run my RNN bots.
Here is a download link for the Anatomy of Melancholy RNN itself (103M t7 file) - you'll need Torch-RNN to use it.