"ARC" is a large scale audiovisual installation consisting in 60 states of consciousness representations. This project explores a comprehensive collection of human-labelled texts depicting various human states of consciousness. The body of text is classified according to a double individual/collective classification system (see map), and serves as input for machine learning algorithms. The conditionally generated text, which draws on seemingly contradictory emotional areas, emerges as manifestations of latent voices. Part of the text dataset is based on the DRA employed in the project "L".