Sofian Audry’s for the sleepers in that quiet earth. is a generative text produced by a deep recurrent neural network trained on Emily Brontë’s Victorian novel Wuthering Heights (1847). The machine learning algorithm analyzes the frequency of character sequences in the novel and generates successive outputs according to the statistical model it derives from them. The algorithm trains itself by analyzing the frequency and probability of increasingly longer character sequences that eventually become syllables, words, and sentences, and periodically produces generative outputs based on the current state of these data. These outputs are collected in for the sleepers in that quiet earth.

Accordingly, the book starts with single characters scattered seemingly arbitrarily across the page, condensing more and more into legible sequences of letters, and finally revealing a legible new text on the final pages that most closely reflects Brontë’s novelistic style. By this, Audry’s generative book not only documents and visualizes the neural network’s learning curve, but also gives examples of the aesthetic quality of different stages of probabilistic style, which depends on the model’s current level of complexity.

The text in for the sleepers in that quiet earth. is divided into chapters based on the occurrence of the character sequence “chapter” followed by some letters as its own line, in which case this line is typeset as a chapter heading on a new page.

The book was produced with an Espresso Book Machine in an edition of thirty-one unique and signed copies, each copy capturing a different reading session of the algorithm in 642,746 characters, the same length as the version of Wuthering Heights. The page count varies between 272 and 282 pages. The text for the copy archived in our library was generated from 09:19:50 on July 6, 2017 through 18:07:23 on the same day.