The Dream Atlas pursues two objectives that are, by design, inseparable. The first concerns oneiric intelligence: to identify, map, and measure recurrent dream formations — organised, emergent, meaning-bearing structures — across twelve philosophical and religious traditions and across millennia, thereby illuminating the cognitive architecture of human dreaming. The second concerns machine intelligence: to examine the model that reads the corpus as an object of inquiry in its own right — how it structures dream material during unsupervised learning, what patterns it detects that elude human observation, and what epistemological and ethical questions its deployment raises.

If intelligence is understood as organised, emergent, meaning-bearing structure rather than as the deliberation of the waking ego, the dreaming mind emerges as a different kind of knower — one the world’s traditions recognised long before cognitive science named it.

The project addresses four questions concerning the cross-cultural variability, temporal dynamics, epistemological conditions, and empirical foundations of large-scale dream analysis.

1. i. Cross-Cultural Structural Invariance Which dream formations recur across cultures and centuries — and can the distinction between invariant and variant elements within any single dream be operationalised, and empirically validated, at scale?
2. ii. Historical Dynamics and Temporal Change How do structural features of dream content shift across historical periods, and what patterns, thresholds, or discontinuities are detectable through longitudinal analysis?
3. iii. The Epistemic Limits of Machine Intelligence By what principles does machine intelligence structure large dream corpora within black-box latent spaces, what governs this hidden processing — and what does its operation disclose about the nature and limits of machine knowing?
4. iv. Computational Mapping and Novel Discovery How can computationally mapped spatio-temporal patterns in dream content transform the empirical study of dreams in everyday life, and what novel insights can machine intelligence generate by revealing latent structures and dynamics that remain inaccessible to conventional methods?

Dream patterns exhibit a stable structural core that transcends temporal and geographic boundaries.

Stable Structural Core

Recurrent semantic structures persist across cultural and historical contexts, constituting an invariant foundation of dream phenomenology.

Adaptive Variability

Surface-level content remains context-sensitive, reflecting individual, cultural, and historical conditions.

Predicted Layering

Large-scale analysis will reveal distinguishable layers—stable structural architecture common across populations and time periods, and variable features responsive to cultural contexts.

Critical Proposition

Peripheral variability does not alter the underlying structural core.

Empirical Validation

This requires computational analysis of dream reports spanning multiple linguistic, geographic, and temporal domains.

DREAMATLAS brings together four established fields — empirical dream content research, anthropology, digital humanities, and computational text analysis — applying contemporary NLP and large language models to questions that none of these disciplines could answer alone.

The empirical study of dream content has a quantitative tradition extending from the Hall and Van de Castle content analysis system through the corpus-based work developed around the Sleep and Dream Database (SDDb) and DreamBank. These frameworks have produced robust findings on the consistency of dream content within individuals and populations, but their evidentiary base remains predominantly Western, contemporary, and English-language. The interpretive traditions of anthropology, religious studies, and the history of culture document the centrality of dreams across civilizations and historical periods, but until recently lacked the analytical infrastructure to operate at corresponding scale.

DREAMATLAS addresses this gap. The project compiles and analyses dream accounts across continents and historical periods, drawing on primary sources, ethnographic records, manuscripts, religious texts, and oral traditions. Collaboration with monastic, national, and private libraries is integral to the design, with explicit priority given to non-digitized materials that fall outside standard digital humanities infrastructures.

Recent advances in NLP and large language models permit the detection of semantic structures and emergent patterns at a scale and resolution earlier methods could not achieve. The project uses these capacities to test whether the recurrent structures identified in dream content — understood here as cross-culturally observable motif clusters — are temporally stable and culturally widespread, context-dependent, or some combination of both.

This analytical ambition carries epistemological responsibilities. AI methods produce particular kinds of knowledge: pattern-based, probabilistic, and sensitive to corpus composition. These properties are treated as methodological objects in their own right rather than as background assumptions. The interpretive complexity of cross-cultural material — including the risk of imposing contemporary categorical frameworks onto historically distant sources — is addressed in corpus design and analytical protocols.

Dreams are among the most universal and least systematically studied dimensions of human experience. While recurrent motifs in dream content have been examined across anthropology, psychology, religious studies, and cultural analysis, no investigation has yet undertaken a systematic empirical study of how these structures shift — or persist — across historical epochs and cultural environments. The absence reflects the structural difficulty of working across disciplines, languages, and corpora at sufficient scale and diversity. DREAMATLAS addresses that gap.

The project develops a computational framework for identifying, tracking, comparing, and visualising long-term patterns in dream content across cultures and time, opening an empirical space in which the foundational question of whether structures of human dreaming are universal, historically contingent, or both can be pursued with the necessary evidentiary base.

This project contributes on three levels:

1. Humanities & Social Sciences

   A replicable model for large-scale, cross-cultural textual analysis, demonstrating how computational methods can be applied rigorously to culturally diverse material.

2. Psychology & Cognitive Science

   The first systematic effort to integrate dream datasets from as many culturally and temporally diverse traditions as possible, providing the field with an empirical foundation that is global, historical, and truly comparative in scope.

3. AI Research

   A direct engagement with the epistemological limits of machine analysis when applied to culturally complex material, contributing to active debates on what computational methods can — and cannot — reliably produce in humanistic domains.

The open-access framework — designed to represent dream reports from as many cultural and historical contexts as possible — is intended as a lasting shared research resource, extensible as new corpora are digitized and integrated.

DREAMATLAS is, to the project's current knowledge, the first to combine three commitments: large-scale computational analysis applied to a cross-cultural and diachronic dream corpus; the treatment of non-Western interpretive traditions as primary analytical sources rather than supplementary context; and explicit inquiry into the epistemological behaviour of AI when applied to culturally complex humanistic material.

The difficulty is proportionate to the ambition. A representative multilingual corpus across this temporal and cultural range requires access to dispersed, partially digitized, and institutionally sensitive materials. NLP pipelines for pre-modern languages must be developed in collaboration with domain specialists. The assumption that a "dream" in one tradition maps meaningfully onto a "dream" in another cannot be taken for granted: cross-cultural equivalence must be interrogated at every analytical stage.

These risks are recognised and built into the research design. Corpus imbalances may skew findings toward better-digitized traditions; AI methods may surface patterns that reflect data structure rather than dreaming structure; certain cultural categories may resist computational formalisation. The project treats these as productive constraints — each generating a research question in its own right, and each addressed through the human-in-the-loop protocols and domain specialist collaboration central to the methodology.

The project began in 2023 with primary dream reports from historical periods and from traditions in which dreams functioned as a substantive domain of knowledge. It is in active development.

Institutional partners comprise the Dream Library Foundation (United States), Victoria University of Wellington (New Zealand), and the Science and Research Centre Koper (Slovenia, EU). Two papers are under peer review.