A team of researchers in Italy has used natural language processing to analyze more than 3,700 dream reports and found that what people dream about is shaped in measurable ways by their personality and by what is happening in their lives. The findings, reported on April 28, add to a growing body of work that treats dream content as data that can be studied at scale.
The study drew on dream diaries kept by 287 participants. Rather than relying solely on human coders to read and categorize each report, the researchers applied automated language analysis tools to detect themes, emotional tone and the degree of coherence in each narrative, according to coverage of the research by Euronews and Earth.com.
Personality shows up in dream structure
One of the clearest results concerned how organized or fragmented a dream was. Participants who scored higher on measures of mind wandering, the tendency to drift into spontaneous, unfocused thought during the day, tended to produce dream reports that were more disjointed and less coherent, the researchers found.
By contrast, people who said they paid attention to their dreams and considered them meaningful tended to report dreams that were more vivid and more richly described. The researchers interpret this as evidence that the way a person relates to their own inner life during waking hours carries over into how they experience, or at least how they remember and describe, their dreams.
External events change the emotional tone
The dataset spanned a period that included COVID-19 lockdowns, which gave the team a natural experiment. Dreams recorded during periods of restriction carried stronger emotional content than those recorded at other times, according to the study. The researchers say this supports the idea that major external circumstances feed directly into the emotional character of dreaming, not only into waking mood.
That result is consistent with earlier pandemic-era surveys that found increases in vivid and anxious dreaming, but the new work adds a quantitative, text-based method for detecting the shift.
Machines matched human raters
A methodological finding may prove as significant as the psychological ones. When the researchers compared automated scoring of dream reports with assessments made by trained human raters, the AI-based analysis reached a level of accuracy comparable to the humans, the study reports.
Human coding of dream reports is slow and expensive, which has historically limited dream studies to small samples. If language models can reliably reproduce human judgments about emotion, coherence and theme, researchers could analyze tens of thousands of reports rather than a few hundred, and track changes over time in large populations.
Open questions
The study relies on self-reported dream diaries, which capture only what participants remember and choose to write down. The researchers acknowledge that the method analyzes dream narratives rather than the dreaming brain itself, so it measures how people describe their dreams as much as the dreams themselves.
Still, the work points toward a future in which sleep and dreaming become another stream of behavioral data that can be mined for insight into mental states. That prospect raises familiar questions about privacy and consent, since dream reports can reveal emotional states and personal concerns that people might not otherwise share.
