# Study Maps How Traits and Lockdown Shaped Dream Reports > Source: > Published: 2026-07-29 09:15:00+00:00 # Study Maps How Traits and Lockdown Shaped Dream Reports A Communications Psychology study published April 28 analyzed 3,366 dream and waking-experience reports from 207 adults, plus 351 dream reports from 80 adults during Italy's first COVID-19 lockdown. Using LLM-assisted ratings and lexical analysis, the researchers found associations between dream content, stable personal traits, and shared external events, while emphasizing that the observational results do not establish causation. A study published in **Communications Psychology on April 28, 2026** used language models and lexical analysis to quantify how dream reports differ from waking accounts and how those patterns vary with personal traits and shared events. The research separates two datasets that should not be treated as one uniform cohort. The main dataset contained **3,366 analyzed reports from 207 Italian adults** collected between 2020 and 2024: 1,687 dream reports and a matched set of 1,679 waking-experience reports. A second dataset contributed **351 dream reports from 80 adults** during Italy's first COVID-19 lockdown. Together, the analyzed data covered 287 adults, but only the main group supplied the paired dream-and-waking corpus. ### What the models measured The researchers scored each report across 16 semantic dimensions, including visual detail, emotion, social interaction, bizarreness, changes of setting, and perceived constraints. They combined ratings from Llama 3, GPT-4, and GPT-4 Turbo with a separate lexical-domain method built from word embeddings. Validation exercises compared the automated ratings with trained human raters and with participants who scored their own dreams. Compared with waking reports, dreams shifted away from thought-centered, self-referential accounts toward more perceptual narratives involving spatial detail, multiple characters, and unusual events. The study also found that greater interest in dreams and a stronger tendency to mind-wander were associated with differences in vividness, bizarreness, and scene changes. These are population-level associations in reported language, not a tool for interpreting an individual's dream. ### Lockdown patterns and limits The independent lockdown dataset contained more emotionally intense dreams and more references to limitations than reports collected after restrictions eased. Those differences diminished over subsequent years, which the authors interpret as evidence that shared external conditions can be reflected in dream semantics. Important limits remain. The study was observational, its analysis plan was not preregistered, and the cohorts were recruited in Italy. Dream reports depend on recall and narrative reconstruction after waking, so the methods cannot cleanly separate the original experience from how a participant later described it. The paper therefore supports a scalable way to study subjective text, not causal claims about why a specific person dreamed something. For data and ML practitioners, the useful pattern is methodological: combine structured participant measures with transparent text representations, validate model ratings against people, and keep inference at the level the study design can support. ## Key Points - 1The main analysis used 3,366 paired dream and waking-experience reports from 207 adults; a separate lockdown dataset added 351 dream reports from 80 adults. - 2LLM-assisted ratings and lexical-domain analysis linked reported dream features with mind-wandering, attitudes toward dreams, sleep measures, and lockdown-era experience. - 3The findings are observational and vulnerable to recall and narrative-reconstruction effects, so they do not support individual dream interpretation or causal conclusions. ## Scoring Rationale The peer-reviewed study shows a reproducible use of LLM-assisted semantic ratings and lexical analysis on subjective text, with human validation and explicit limitations. It is relevant to computational social science and cognitive research, but it does not introduce a deployable general-purpose model or support individual-level interpretation. ## Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)