The dataset behind the study is unusually large for this field. The researchers analysed 1,434 dream reports and 1,619 mind-wandering reports collected from 176 individuals. Rather than reading each narrative and coding it by hand with subjective categories, the team converted the language itself into networks. In this approach, each word in a report becomes a node, and the relationships between words become edges. Once a report is transformed into a graph, the entire toolkit of network science becomes available: measures of how tightly words cluster together, how easily information can travel across the network, how often words repeat, and how central particular words are to the overall structure of the text.

The first major finding concerns the global architecture of the two kinds of reports. Dream reports turned out to have a more complex structure than mind-wandering reports, displaying more word repetitions, more local cliques, tightly interconnected clusters of words, and greater global integration. The authors describe this pattern as resembling a small-world structure, a configuration familiar from studies of brain networks and social networks, in which dense local clusters are linked by shorter paths across the whole system. Mind-wandering reports, by contrast, contained fewer word repetitions and were more verbose, with distinct transition points marking shifts in the flow of the narrative.

Why would dreams and daydreams leave such different structural traces in language? The researchers point to a plausible cognitive explanation rooted in memory. Dream reports depend on recall over longer periods of time, because a dream must be held in mind and then reconstructed in words after waking. That reconstruction process appears to encourage repetition of key words and a more integrated, clustered organization of the narrative. Mind wandering, sampled closer to the moment of experience, produces longer, more lexically varied accounts in which the speaker moves through topics with fewer loops back to earlier material. The graph metrics capture this difference objectively, without any researcher needing to judge what a dream is about.

Perhaps the most striking result is that the network structure of these reports is not generic. By capitalising on a repeated measures design, in which the same individuals contributed multiple reports across both states, the team found that the structure of dream and mind-wandering narratives contains individual-specific information. In other words, the way a particular person connects words when describing their inner life is stable enough to distinguish them from other people. This suggests that narrative structure reflects enduring characteristics of how an individual’s mind organizes experience, rather than being merely a by-product of the topic that happened to drift through consciousness on a given day.

The study also uncovered a link between language structure and mental health. Word centrality, which corresponds to lower lexical diversity, meaning that a smaller set of words dominates the network of a report, was predictive of symptoms of depression and anxiety. Notably, the association was state-specific: centrality in dream reports predicted depressive symptoms, while centrality in mind-wandering reports predicted anxiety symptoms. This dissociation hints that different forms of inner experience may carry different kinds of clinical information, and that the architecture of spontaneous thought could serve as a window onto distinct dimensions of psychological distress.

Technically, the approach has significant advantages over traditional content analysis. Classic dream research often relies on trained judges rating reports on scales such as bizarreness, emotion, or thematic content, which is labour-intensive and vulnerable to inter-rater variability. Graph-theoretical analysis, once implemented, can process large volumes of naturalistic text cheaply and consistently. The authors emphasise that the method seems promising for cost-effective analyses of large, naturally occurring qualitative datasets, a category that includes not only dream and mind-wandering reports but potentially other forms of unstructured self-narration collected in clinics, sleep laboratories, or experience-sampling studies.

The work also speaks to a long-standing debate in consciousness science. Dreams and mind wandering are both examples of spontaneous, stimulus-independent thought, yet they arise in radically different states: one during sleep, the other during wakefulness. By showing that the two can be quantitatively differentiated using nothing more than the structure of the language used to report them, the study provides a new empirical handle on how the state of consciousness shapes the organization of thought. The small-world-like structure of dream narratives, with its repetitions and integrated clusters, may reflect how sleep-dependent memory processes and the absence of external input sculpt the flow of experience, while the more verbose, transition-rich structure of waking mind wandering may mirror the restless, forward-moving character of unguided daytime thought.

There are, of course, caveats. The measures capture the structure of reports, not the experiences themselves, and reporting is always filtered through memory, language ability, and the demands of the moment. The clinical findings are predictive associations rather than diagnostic tools, and the authors note that the article was shared early as a citable, peer-reviewed accepted version subject to further editorial refinement. Even so, the implications are considerable. If the architecture of a person’s dream narratives is stable, measurable, and sensitive to their mental state, then something as ordinary as telling a researcher about last night’s dream could, with the right mathematics, reveal enduring features of the mind that produced it. The study, supported by funders including the Alfred Kordelin Foundation, the Emil Aaltonen Foundation, the Finnish Cultural Foundation, the Bial Foundation, the Cambridge ESRC Impact Acceleration, and the Australian Research Council, was jointly supervised by Natália Bezerra Mota and Tristan Bekinschtein, and brings together teams from Cambridge, Stanford, Turku, Skövde, Monash, Queen Mary University of London, and the Federal University of Rio de Janeiro.

What makes the research resonate beyond the laboratory is its simplicity and reach. The raw material is free: people describing their inner lives in their own words. The analysis requires no brain scanners, no electrodes, and no invasive procedures, only computational tools that turn stories into networks and networks into numbers. As large language datasets become ever easier to collect through smartphones and experience-sampling apps, methods like this one could transform how psychologists study spontaneous thought, how clinicians track symptoms over time, and how consciousness researchers compare states of mind that have always been, until now, notoriously difficult to measure. The mind’s private narratives, it turns out, have a shape, and that shape can be charted.

Subject of Research: Graph-theoretical analysis of the linguistic structure of dream and mind-wandering reports and its links to individual differences and mental health symptoms

Article Title: A graph theory approach to chart the architecture of dream and mind-wandering reports

Article References: Coppola, P., Sikka, P., Valli, K., Tuominen, J., Revonsuo, A., Loukola, V., Bernstein, R., Strid, N., Kirberg, M., Ezquerro-Nassar, A., Windt, J., Noreika, V., Mota, N. B., & Bekinschtein, T. A. (2026). A graph theory approach to chart the architecture of dream and mind-wandering reports. Communications Psychology, 4(1), Article 135. https://doi.org/10.1038/s44271-026-00509-5

Image Credits: AI Generated

DOI: 10.1038/s44271-026-00509-5

Keywords: graph theory, dreams, mind wandering, consciousness, network analysis, language, depression, anxiety, small-world networks, lexical diversity, mental health, spontaneous thought