Researchers at Italy's IMT School for Advanced Studies Lucca have used natural language processing to analyse more than 3,700 accounts of dreams and waking experiences, concluding that dreams are neither random noise nor straightforward replays of the day's events. Instead, the sleeping brain appears to blend memories, emotions, familiar places and imagined scenarios into entirely new sequences, offering fresh evidence of structure behind seemingly chaotic dream content and a new way for AI to accelerate this kind of large-scale psychological research.

The study, published in Communications Psychology, drew on two weeks of diaries from 287 adults, combined with data on personality, sleep quality and cognitive traits, comparing the semantic structure of dream and waking reports via NLP. It found everyday settings such as workplaces and hospitals were often merged with unfamiliar places or shifting viewpoints, with dream style varying by personality: mind-wanderers reported more fragmented dreams, while those who found dreams meaningful described more vivid ones. A comparison with dream reports gathered during COVID-19 lockdowns by Sapienza University of Rome showed these contained markedly more references to confinement and restriction, reflecting the circumstances of the time; the researchers note the method's reliance on self-reported recall and that it does not explain why humans dream at all.