The cartography of dreams: application of computational linguistics to the study of sleep conscious experiences

The study of dreams represents a crucial intersection between philosophical, psychological, neuroscientific, and clinical interests. Since dreams are subjective experiences spontaneously generated by the brain when it is partially disconnected from the external environment and thus is let free to operate in an unconstrained manner, their study could reveal specific mental processes that are different from those occurring during wakefulness and might provide crucial insights into brain functioning, both in physiological and pathological conditions. Given the high cost of sleep and dream research in terms of human effort and funding, open science and the building of large- scale datasets and repositories will constitute a key for significant advances in the field. At the same time, the analysis of large datasets will require a methodological shift, from human-based assessments to more automated approaches. For instance, methods based on natural language processing (NLP) could replace manual scales and rating approaches for the assessment of dream content. Such a methodological shift could also have positive consequences concerning the reproducibility and reliability of scientific results. Based on the above premises, we created Somnieve, a multimodal, open-source database collecting dream reports along with demographic information and psychometric, cognitive, and electroencephalographic measures obtained from a representative sample of the healthy Italian adult population. In particular, participants were asked to wear an actigraph and to record a report of their last dream experience each morning upon awakening for 14 days. Moreover, they completed a battery of questionnaires and cognitive tests. The database currently includes 1324 dream reports obtained from 161 healthy adult individuals (66M, 18-65y). Beside presenting and describing the Somnieve database, this Thesis work exploited the database to investigate the individual determinants of physiological dream content and recall frequency. We relied on computational linguistics to test whether it might be possible to implement computational linguistics based tools to automatically and objectively code dream content and verify the existence of generalizable semantic patterns in dream narratives. Moreover, we evaluated the inter- and intra-individual factors affecting dream recall frequency. Present results highlight the potential benefits that large multimodal databases like Somnieve could bring for the field of dream research. It is our hope that this, and similar independent efforts by other laboratories, will contribute to improve reproducibility in dream research and identify the individual determinants of dream content and recall frequency in physiological conditions, as well as quantify their possible pathological alterations.