Student project about machine learning. An AI algorithm analyzes the Romand de la Rose by Guillaume de Lorris and Jean de Meung and then in turn proposes to produce a short text based on it. The project presents a selection of outputs, made with different levels of training, which can result in accidents. I intentionally interrupted the training prematurely for certain models so that their future generated output would be incomplete. These different text outputs are then used to illustrate the machine’s learning process step by step. These text snippets are then taken and laser engraved into concrete slabs. The project aims to highlight the relationship between the transformation of the generated text and that of language.
The corpus chosen for training the algorithm is an ancient manuscript dating back from
the 13th century, written in Old French, the “ langue d’oïl ”.
The augmentation would take place on a blank concrete “ page ” onto which text would be projected. Through a discreet webcam, the projected text would change according to ambient motion.The text transforms, from barely comprehensible to coherent. The text excerpts are displayed on top of each other, creating a form of digital palimpsest.
