Dreams

Abstract

Dreams are a form of imagination: the imagery, narrative, and emotion generated by the brain during sleep, experienced without voluntary control and, while they unfold, mistaken for reality. This article defines dreaming and distinguishes it from the sleep stage it was once identified with, traces the discovery of rapid-eye-movement (REM) sleep and the neuroanatomy of the dreaming brain, sets out the major theories of why we dream, examines what dream content reveals through the continuity hypothesis, surveys lucid dreaming and the recent demonstration of two-way communication with dreamers, and derives the arithmetic of REM sleep across the night that explains why we mostly remember the dreams we wake from.

Keywords: REM sleep, activation-synthesis, threat simulation, lucid dreaming

Everyone dreams, yet dreaming is among the hardest mental states to study, because its only direct evidence is a report given after the experience is over, by a person who was asleep while it happened. For most of history the dream was a message to be interpreted; the science of dreaming began when it became a physiological event that could be caught in the act. The discovery that vivid dreaming coincides with a distinctive stage of sleep, with darting eyes and a brain nearly as active as waking, turned the dream from a subject of divination into a subject of measurement. This article treats dreaming as a cognitive and neural process — a mode of internally generated experience — and follows the evidence from the discovery of REM sleep to the laboratory in which a sleeping person answered a question by moving their eyes.

What Dreams Are

A dream is the succession of imagery, narrative, thought, and emotion that the brain generates during sleep. MeSH classifies Dreams under imagination, and the placement is apt: dreaming is imaginative experience — the construction of scenes and events not present to the senses — with two features that set it apart from waking imagination. It is involuntary, arising without an act of will, and it is immersive, experienced from within as a real world rather than as an image summoned before the mind. The dreamer does not usually know the experience is self-generated; the hallmark of the ordinary dream is precisely the loss of that insight.

Two boundaries fix the concept. Dreaming is distinct from the physiological state of sleep in which it most often occurs: for decades dreaming was effectively equated with REM sleep, but the two can be pulled apart, and treating them as identical was a mistake the field had to correct. Dreaming is also distinct from the report of a dream, which is all the researcher ever receives — a recollection, given on awakening, of an experience that cannot be observed directly and may be edited by the very act of recall. These two gaps, between dreaming and its physiology and between dreaming and its report, are the methodological problems around which the science of dreams is organised, and they structure the sections that follow.

The Sleeping Brain

The modern science of dreaming begins with an instrument rather than a theory. Recording eye movements and brain activity through the night, Aserinsky and Kleitman found recurring periods in which the eyes darted rapidly beneath closed lids while the electroencephalogram resembled waking; woken from these rapid-eye-movement periods, sleepers reported vivid dreams far more often than at other times (Aserinsky & Kleitman, 1953). REM sleep gave dreaming a physiological signature and a clock: the periods recur through the night and lengthen toward morning, a regularity the worked example below makes quantitative.

Neuroimaging later revealed why dreams feel as they do. Positron-emission tomography during REM showed a brain that is not uniformly active but selectively so: limbic and paralimbic regions handling emotion are strongly engaged, while the dorsolateral prefrontal cortex that supports logical control and self-monitoring is relatively deactivated (Maquet et al., 1996). This pattern — hot emotion, cool executive oversight — matches the phenomenology of dreams as emotionally intense yet uncritical, accepting bizarre transitions without surprise. The link from this physiology to the felt experience of dreaming became the organising problem of the field's neuroscience (Nir & Tononi, 2010). More recent high-density electroencephalography has refined the picture, decoupling dreaming from REM as such: whether in REM or non-REM sleep, the presence of a dream is predicted by reduced low-frequency activity in a posterior cortical hot zone, suggesting that this region's activation, not the sleep stage, is the proximate correlate of dreaming (Siclari et al., 2017). The demonstration below reconstructs a night's sleep architecture and shows how the REM periods that carry most vivid dreaming lengthen across the night.

A modelled night of five ~90-minute cycles. Deep slow-wave sleep (N3) dominates early; the REM periods that carry vivid dreaming lengthen toward morning. Move the wake time to see which stage you surface from and how much dreaming you would recall.

You wake mid-REM from a 30-minute period — the kind of awakening that yields a remembered dream about 80% of the time. A natural end-of-night waking (7.5 h) lands in the fifth and longest REM period — three times the first — which is why the morning dream is the one most often recalled, not because it is special but because the REM that carries it is longest just before waking.

Theories of Dream Function

Why we dream is the oldest and least settled question in the field, and the major answers form a genuine debate rather than a consensus. The psychoanalytic tradition begins with Freud, for whom the dream is the disguised fulfilment of a repressed wish: a latent content, censored and transformed by the dream-work into the remembered manifest content, whose interpretation reveals the unconscious (Freud, 1900). Freud's specific mechanism found little empirical support, but he established the dream as a psychological product worth decoding rather than noise.

The first thoroughgoing neurophysiological alternative was the activation-synthesis hypothesis: dreaming reflects the forebrain's attempt to make sense of essentially random signals generated by the brainstem during REM, so the dream's bizarreness is the byproduct of synthesising a narrative from noisy activation rather than a disguised message (Hobson & McCarley, 1977). Hobson later broadened this into the protoconsciousness hypothesis, casting REM dreaming as a virtual-reality generator that builds and rehearses a model of the world and self, a substrate on which waking consciousness develops (Hobson, 2009). The identification of dreaming with brainstem-driven REM was itself challenged by lesion evidence: Solms found that dreaming can be abolished by damage to forebrain regions while REM persists, and can persist when REM-generating brainstem structures are damaged, implying that dreaming and REM are controlled by different mechanisms and that dreaming depends on forebrain motivational circuitry (Solms, 2000).

A separate line asks not how dreams are made but what they are for. Revonsuo's threat-simulation theory proposes that dreaming is an evolved function: by repeatedly simulating threatening events and rehearsing avoidance in a safe offline arena, dreaming improved ancestral survival, which is why negative emotions and threatening scenarios are over-represented in dream content (Revonsuo, 2000). A contemporary memory-based account, the NEXTUP model (network exploration to understand possibilities), holds that dreaming explores weak, unexpected associations among memories to discover useful but non-obvious connections, extracting value from experience in a way waking thought does not (Zadra & Stickgold, 2021). Table 1 sets the major accounts side by side. The demonstration below lets the reader take a salient feature of a dream and see how each theory would explain it.

Pick a salient feature of a dream. Each major theory of dreaming explains the same feature differently — the contrast is the debate itself, not a settled answer.

The theories are not all rivals on the same question: some explain how a dream is generated, others what it is for, and others what its content means. A feature that is a meaningless byproduct on one account is the whole point on another.

Dream Content and Continuity

Whatever dreams are for, their content is not arbitrary, and the most productive empirical program treats dream reports as data to be coded and counted. The continuity hypothesis holds that dreams largely reflect the dreamer's waking life — the people, activities, and emotional concerns that occupy them — rather than a separate symbolic realm. Domhoff built this into a neurocognitive theory of dreaming, arguing that dreaming is a form of intensified mind-wandering produced by a mature default network, and that its content is continuous with waking thought (Domhoff & Fox, 2015); large samples of home dream reports show stable, replicable frequencies of characters, social interactions, and emotions that differ systematically by age, gender, and culture. Systematic content analysis is the method: Schredl's questionnaire studies repeatedly find that the frequency of a waking activity predicts how often it appears in dreams, the core continuity prediction (Scarpelli et al., 2019).

Continuity is not literal replay. The memory sources of dreams follow a characteristic dream-lag: fragments of a day's experience are incorporated most heavily the following night and again about a week later, and dreams recombine memory elements rather than reproducing whole episodes (Nielsen & Stenstrom, 2005). Emotion is central to what is selected and recombined: reviews of dream content link dreaming to the overnight processing of emotional memories, with the affective tone of dreams tracking waking mood and stress (Scarpelli et al., 2019). This convergence — continuity of content, selective incorporation of recent emotional experience, recombination rather than replay — is what memory-based theories such as NEXTUP aim to explain (Zadra & Stickgold, 2021). The demonstration below models content analysis in the tradition of the Hall and Van de Castle coding system, whose normative frequencies for characters, interactions, and emotions made dream content quantitatively comparable across samples (Hall & Van de Castle, 1966), letting the reader compare a dream series against those norms.

Systematic content analysis codes dream reports into categories and compares their frequencies against large normative samples. Choose a category and set how prominent the matching concern is in waking life; the continuity hypothesis predicts the dreamed frequency should track it.

The continuity hypothesis is not that dreams replay waking life literally, but that the frequency of a concern predicts how often it surfaces in dreams. Because these regularities hold across thousands of coded reports, dream content becomes quantitative data rather than private anecdote.

Lucid Dreaming

Ordinarily the dreamer lacks insight into the dream, but not always. In a lucid dream the sleeper becomes aware that they are dreaming while the dream continues, and may gain some control over it. Lucidity is not merely a vivid dream: high-density electroencephalography shows that lucid REM carries a distinct signature, with increased activity in the gamma frequency band and in frontal regions normally quiet in REM, placing lucid dreaming as a hybrid state with features of both waking and ordinary dreaming (Voss et al., 2009). The reduced deactivation of prefrontal cortex fits the return of self-reflective awareness that defines lucidity.

Lucidity turned out to solve the field's oldest methodological problem — that dreams can only be reported after the fact. Because a lucid dreamer retains volitional control over eye movements, which are not paralysed in REM, they can signal to the outside world from within the dream by a prearranged pattern of eye movements. Building on this, four laboratories independently established two-way communication: experimenters posed questions — spoken words, tones, tactile taps, even simple arithmetic — to which lucid dreamers responded accurately in real time with coded eye or facial-muscle signals, and later confirmed the exchange in their dream reports (Konkoly et al., 2021). Interactive dreaming demonstrates that a sleeping, dreaming person can perceive external input, hold a task in mind, compute, and answer — collapsing the gap between the dream and its report.

Worked Example

REM sleep is not distributed evenly through the night, and its distribution explains a familiar fact: we mostly remember the dreams we wake from in the morning. Model a night as five sleep cycles of about 90 minutes each, 450 minutes of sleep in total, and let the REM period within each cycle lengthen across the night. A simple linear rule captures the real pattern well enough to reason with:

REM(n) = 5 + 5n minutes, for cycle n = 1 to 5

This gives REM periods of 10, 15, 20, 25, and 30 minutes across the five cycles. Two consequences follow. First, the total REM time is:

10 + 15 + 20 + 25 + 30 = 100 minutes

which is 100 / 450 ≈ 22% of the night, matching the roughly one-fifth to one-quarter of adult sleep that is normally spent in REM. Second, and more instructive, the final REM period is three times the length of the first — 30 minutes against 10 — so REM-dominated, vivid dreaming is concentrated in the last cycles before waking. Because dream recall on awakening directly from REM is high (on the order of 80%), the expected vivid-dream time available to be recalled is about:

0.8 × 100 = 80 minutes

heavily weighted toward the morning. A sleeper who wakes naturally at the end of the night wakes from the longest REM period of all, which is why the last dream is the one most often remembered — not because morning dreams are special, but because the REM that carries them is longest just before waking. Figure 1 plots the lengthening REM periods across the night.

Discussion

Dreaming has resisted a single settled account because it sits at the intersection of three questions that need not have the same answer: how dreams are generated, what they are for, and what their content means. The activation-synthesis tradition answered the first and was too quick to treat the second as vacuous; the psychoanalytic and adaptationist traditions pressed the second and third but strained the evidence on mechanism. The most durable finding is a negative one that reorganised the whole field — that dreaming and REM sleep are dissociable (Solms, 2000); (Siclari et al., 2017). Once dreaming is not identified with a brainstem-driven sleep stage but with the activation of specific forebrain and posterior cortical systems, the neurophysiology and the psychology can be reconciled: the same posterior hot zone whose activity predicts dreaming is where perceptual experience is constructed in waking, so a dream is experience built by the same machinery, running on internally generated rather than sensory input.

That framing brings dreaming close to the account of imagination under which MeSH files it. A dream is imaginative experience with the volitional control removed and the immersive realism turned up — the constructive systems of perception and memory run offline, without the executive oversight that in waking tags an image as unreal. The continuity of dream content with waking concerns, the selective incorporation of recent emotional memory, and the exploratory recombination that memory-based theories describe all follow naturally from treating the dream as the brain's constructive apparatus operating in a state of reduced constraint rather than as a separate faculty with a message to decode.

Current Directions

The most striking recent advance is methodological: the demonstration of real-time, two-way communication with lucid dreamers across four independent laboratories has opened interactive dreaming as an experimental paradigm, in which questions can be posed and answered during the dream itself rather than reconstructed afterward (Konkoly et al., 2021). If the approach scales, it promises to attack the field's founding problem — the reliance on retrospective report — by sampling experience while it is happening.

A second front is the neural decoding of dream content. If dreaming and perception share posterior cortical machinery, the activity that predicts the presence of a dream should also carry information about what is being dreamed, and high-density recording has begun to relate specific dream contents to localized activation (Siclari et al., 2017). A third concerns function: the emotion-regulation and memory-consolidation roles of dreaming are being tested directly, connecting dream content to overnight changes in emotional memory and mood (Scarpelli et al., 2019). Whether dreaming performs these functions or merely accompanies the sleep processes that do remains the central open question.

Common Misconceptions

Dreaming and REM sleep are the same thing.

Vivid dreaming is strongly associated with REM, but the two are dissociable: dreams occur in non-REM sleep, dreaming can be lost while REM is preserved, and the presence of a dream is best predicted by posterior cortical activity rather than by sleep stage (Solms, 2000); (Siclari et al., 2017).

Dreams are meaningless random noise.

Even if dreaming begins with spontaneous activation, dream content is systematic and continuous with waking concerns, showing stable, replicable regularities across large samples rather than randomness (Revonsuo, 2000); (Scarpelli et al., 2019).

A sleeping person is cut off from the outside world.

Lucid dreamers can perceive external questions and answer them in real time with coded eye or facial-muscle signals, demonstrating that a dreaming brain can register and respond to input from outside (Konkoly et al., 2021).

Glossary

Activation-synthesis hypothesis.

The theory that dreaming is the forebrain's synthesis of a narrative from spontaneous brainstem activation during REM sleep.

Continuity hypothesis.

The proposal that dream content largely reflects the dreamer's waking activities, relationships, and emotional concerns.

Dream-lag effect.

The tendency for daytime experiences to be incorporated into dreams most heavily the following night and again about a week later.

Interactive dreaming.

An experimental paradigm in which experimenters exchange questions and answers with a lucid dreamer in real time during REM sleep.

Lucid dream.

A dream in which the sleeper is aware that they are dreaming while the dream continues, sometimes with volitional control over its course.

Manifest and latent content.

In psychoanalytic theory, the remembered surface of a dream (manifest) versus the hidden wish it is held to disguise (latent).

Neurocognitive theory of dreaming.

Domhoff's account of dreaming as intensified mind-wandering produced by a mature default network, with content continuous with waking thought.

NEXTUP model.

Network exploration to understand possibilities: the account of dreaming as exploration of weak associations among memories to discover useful connections.

Posterior hot zone.

A posterior cortical region whose reduced low-frequency activity predicts the presence of dreaming independently of sleep stage.

Protoconsciousness hypothesis.

Hobson's proposal that REM dreaming is a virtual-reality generator building and maintaining a model of the world and self.

Rapid-eye-movement (REM) sleep.

The sleep stage marked by darting eye movements and wake-like brain activity, most strongly associated with vivid dreaming.

Sleep cycle.

A recurring ~90-minute progression through non-REM and REM sleep; the REM period lengthens across successive cycles through the night.

Threat-simulation theory.

Revonsuo's evolutionary account that dreaming rehearses the perception of and response to threats in a safe offline simulation.

Wish-fulfilment.

Freud's thesis that a dream is the disguised satisfaction of a repressed, usually unconscious, wish.

Key Researchers

G. William Domhoff. Built the neurocognitive theory of dreaming and the Hall-Van de Castle system of dream-content analysis, grounding the continuity hypothesis in large content databases. Google Scholar - Wikipedia - Wikidata

Sigmund Freud (1856-1939). Founded the psychoanalytic theory of dreams as disguised wish-fulfilment with the 1900 The Interpretation of Dreams and the manifest/latent-content distinction. Wikipedia - Wikidata

J. Allan Hobson (1933-2021). Proposed the activation-synthesis hypothesis and later the protoconsciousness theory, recasting dreams as the brain's synthesis of REM activation. Wikipedia - Wikidata

Antti Revonsuo. Proposed the threat-simulation theory of dreaming, the leading evolutionary account of dream function. ORCID - Google Scholar - Wikipedia

Michael Schredl. Leading empirical dream researcher whose large questionnaire studies underpin the continuity hypothesis relating dream content to waking concerns. ORCID - Wikidata

Mark Solms. Showed by lesion evidence that dreaming and REM sleep are controlled by different brain mechanisms, reviving the role of forebrain motivational circuits in dreaming. ORCID - Google Scholar - Wikipedia - Wikidata

Robert Stickgold. Links dreaming to sleep-dependent memory consolidation and, with Zadra, proposed the NEXTUP model of dreaming as weak-associative memory exploration. Google Scholar - Wikipedia - Wikidata

Frequently Asked Questions

What is a dream in cognitive psychology? A dream is the imagery, narrative, thought, and emotion the brain generates during sleep, experienced involuntarily and, while it unfolds, typically as if it were real. Cognitive psychology treats it as a mode of internally generated experience closely related to imagination.

Do dreams only happen in REM sleep? No. Vivid dreaming is strongly associated with REM sleep, but dreams also occur in non-REM sleep, and studies show that dreaming and REM are dissociable and controlled by partly different brain mechanisms (Solms, 2000); (Siclari et al., 2017).

What is the activation-synthesis hypothesis? It is the theory that dreaming is the forebrain's attempt to synthesise a coherent narrative from spontaneous, largely random signals generated in the brainstem during REM sleep, so that the bizarreness of dreams is a byproduct of that synthesis rather than a hidden message (Hobson & McCarley, 1977).

Why do we dream? There is no settled answer. Leading theories include Freud's wish-fulfilment, the threat-simulation theory that dreaming rehearses responses to danger, and memory-based accounts such as the NEXTUP model that dreaming explores useful associations among memories (Revonsuo, 2000); (Zadra & Stickgold, 2021).

Does dream content mean anything? Dream content is not random: the continuity hypothesis holds that it largely reflects the dreamer's waking activities and emotional concerns, and systematic content analysis reveals stable, quantifiable regularities across large samples (Scarpelli et al., 2019).

What is lucid dreaming? Lucid dreaming is a state in which the sleeper becomes aware that they are dreaming while the dream continues; electroencephalography shows it is a hybrid state with features of both waking and ordinary REM dreaming (Voss et al., 2009).

Is real-time communication with a dreaming person possible? Yes. Across four laboratories, experimenters posed questions to lucid dreamers who answered accurately in real time with coded eye or facial-muscle signals and confirmed the exchange in their dream reports (Konkoly et al., 2021).

Why do we usually remember only the last dream of the night? REM periods lengthen across the night, so the longest, most vivid REM dreaming occurs just before natural waking; because recall from REM awakenings is high, the last dream is the one most often remembered (Aserinsky & Kleitman, 1953).

References