Dreaming:A Neurocognitive Approach

Dreaming: A Neurocognitive Approach

J. Allan Hobson and Robert Stickgold

Laboratory of Neurophysiology, Harvard Medical School
74 Fenwood Road, Boston MA 02115

Acknowledgments: This project was funded by grants from NIMH (MH-13,923, and MH-48,832) and from the MacArthur Foundation.

Table of Contents

  1. Abstract
  2. Introduction
  3. Background
  4. Dreaming
  5. Conclusions
  6. References


The studies reported in the following articles are aimed at providing a comprehensive, detailed, and quantitative picture of cognition in human dreaming. Our main premises are that waking, REM sleep, and non-REM (NREM) sleep represent physiologically distinct and identifiable brain states, and that the differences between waking, REM, and NREM mentation reflect these physiological differences. We have studied dreams at a formal level of analysis and, in these papers, have studied the specific dream properties of emotions, bizarre transformations, scene shifts, and plot coherence, in adults and 4-10 year old children, as part of a larger effort to map state-dependent mental phenomena back onto the varying neurobiological processes which must underlie them. We believe that such efforts will enhance our understanding not only of dreaming and its neurophysiological substrates, but of the cognitive processes that dreaming shares with other unusual mental states.

Table of Contents


This paper introduces a series of studies, presented in the following papers, that continues our development of a neurocognitive analysis of dreaming. This introductory article places our new findings within the context of the activation-synthesis hypothesis (Hobson & McCarley, 1977) and its neurobiological infrastructure (Hobsonet al.,1975), and then discusses our new findings in terms of a set of general questions about dream cognition and consciousness.

We know that mental activity occurs not only during waking, but during REM sleep and NREM sleep as well (see Arkinet al.,1978 for an overview), and that the formal cognitive characteristics of these three states are distinctly different from one another (Antrobus, 1990; Foulkes, 1985). One of our basic assumptions is that dream consciousness and cognition are both highly state-dependent phenomena. Our goal is to identify aspects of cognition that depend on the wake-sleep state of the brain and to show how changes in these cognitive behaviors result from changes in underlying brain functions.

Our general strategy is to map relationships between the physiological and cognitive domains by examining brain and cognitive functions at a formal and global level, while varying wake-sleep state. An example of this is our investigation of visual hallucinosis. We have explored the hypothesis that pontogeniculooccipital (PGO) waves, which are observed during REM sleep but are either absent or powerfully suppressed in NREM sleep and waking, constitute an internal stimulus source for the visual system in REM sleep (Mamelak & Hobson, 1989a; Kahn & Hobson, 1993). We have further hypothesized that the memory loss in dreaming sleep is a consequence of the aminergic demodulation of the brain that occurs when noradrenergic (locus coeruleus) and serotonergic (Raphe) neurons cease firing in REM sleep. This model of mental activity during sleep has relevance to theories regarding the genesis of hallucinations, amnesia, delusions and confabulation in abnormal mental states, as well as informing our thinking about sleep mentation and normal cognition.
Figure 1

Our research works at the interface of three distinct disciplines. From neurobiology, we obtain information on the anatomical, physiological, and biochemical basis of the wake-sleep cycle. From psychophysiology we determine the relationship between REM sleep and dreaming. Finally, from the cognitive sciences we gain insights into approaches to the analysis of sleep mentation. To put our current work in perspective, it is necessary to briefly review the contributions of these three areas of research.

Table of Contents


Psychophysiology of Waking, Sleeping and Dreaming
The discovery of REM sleep and its correlation with dreaming (Aserinsky & Kleitman, 1953) opened a new era of research in the relation of brain to mind. In the early days of the human sleep-dream laboratory, much attention was paid to the specificity of the REM-dream correlation (Arkinet al.,1978). Normal subjects, usually students, were awakened from either the REM or NREM phase of sleep and asked to report their recollection of any mental experience preceding the awakening.

The main conclusions of these cross-sectional normative studies were that (1) reports from REM sleep awakenings were typically longer, more perceptually vivid, more motorically animated, and more emotionally charged than NREM reports, and (2) NREM reports tended to be more thought-like and contained more representations of current concerns than did REM sleep reports. Based on the measures available, it seemed reasonable to conclude that the activation level of the brain was the major determinant of these observed differences. This hypothesis was supported by the observation that all measures of dream intensity peaked during the eye movement clusters within REM periods.

Attempts to further detail the psychophysiology of dreaming (Roffwarget al.,1962) and to understand the functional benefits of REM sleep for cognition (Dement, 1960) were impeded by severe methodological difficulties that led to replication failures (Moscowitz & Berger, 1969; Kaleset al.,1970). These in turn cast doubt upon the whole sleep psychology enterprise. It is our view that many of the ensuing controversies (Herman, 1989; Antrobus, 1990) regarding these fundamental issues can be resolved only by the introduction of new concepts and methods such as those of the cognitive sciences.

The Neurocognitive Study of Waking, Sleeping and Dreaming
Many of the sleep and dream scientists who have continued their research despite the decline of the psychophysiological paradigm now seek to establish a cognitive neuroscience of brain-mind states. For example, John Antrobus (1990) has modeled mentation across states, using a non-quantitative parallel processing approach that is informed by neurobiology as well as by sleep laboratory data. Antrobus has also introduced important new definitions of the wake state to serve as controls for his studies of sleep mentation (see Reinsel et al., 1992).

Inspired by the tonic-phasic model of Molinari and Foulkes (1969) as well as by our activation-synthesis model, Seligman and Yellen (1987) added consideration of emotions to the concepts of primary visual activation and secondary cognitive elaboration. Their result was a more comprehensive and descriptive approach to some of the very same dream features that interested Freud, but which obviated the ad hoc psychology of psychoanalytic dream theory. David Foulkes, another leading dream psychologist, has also emphasized the importance of the cognitive approach (Foulkes, 1985).Figure 2

Since 1977, our own work on sleep mentation has gradually assumed a more deliberate, quantitative, and analytic approach; recently we have been examining the several mental faculties that are altered across states from the perspective of orientational stability (see Figure 2). We began with an attempt to identify and formalize types of dream "bizarreness" (McCarley & Hoffman, 1981; Hobson, et al., 1987), initially defining dream bizarreness as "impossibility or improbability in the domains of dream plot, cognition and affect" (Hobson et al., 1987). The bizarre features of dreams were divided into discontinuities, incongruities and uncertainties. Discontinuities were further characterized as "interruptions in orientational stability" (Hobson et al., 1987), including instabilities in plot, scene location, characters and objects. We call this distinctive cognitive feature of dreaming "orientational instability." We have attempted to quantify discontinuity in the orientation domain using several techniques, and several of the papers that follow are concerned with this issue.

Neurobiology of Waking, Sleeping and Dreaming
The discovery of the ubiquity of REM sleep in mammals (Dement, 1958; Jouvet, 1962) provided an animal model for the study of the brain side of the brain-mind question of wake-sleep mentation. While animal studies showed that potent and widespread activation of the brain did occur in REM sleep, it soon became clear that Moruzzi and Magoun's (1949) concept of a brainstem reticular activating system required extension and modification to account for the differences between the kinds of EEG activation seen during waking and that seen in REM sleep.
A possible source of the differentiation between the two activated states, waking and REM sleep, was provided by the discovery of the chemically specific neuromodulatory subsystems of the brainstem and of their differential activity in waking (noradrenergic and serotonergic systems on, cholinergic system inhibited) and REM sleep (noradrenergic and serotonergic systems off, cholinergic system disinhibited). The resulting model of reciprocal interaction (Hobson et al., 1975; McCarley & Hobson, 1975) provided a theoretical framework for experimental interventions at the cellular and molecular level that have vindicated the notion that waking and dreaming are at opposite ends of an aminergic-cholinergic neuromodulatory continuum with NREM sleep holding an intermediate position (Figure 3). This implies that the simple activation model of many dream psychologists may be inadequate.

The articulation of the reciprocal interaction model and the emergence of a wealth of detail regarding the functional reorganization of the visual system in sleep (Hubel, 1959, Evarts, 1961, Callaway et al., 1987), suggested a new conceptual approach to the science of brain-mind state determination (McCarley & Hobson, 1977). First expressed as the activation-synthesis hypothesis of dreaming (Hobson & McCarley, 1977), this new global brain-state to mind-state mapping effort has dealt with the contributions of three factors to mental state differentiation (Hobson, 1990b): brain neuromodulation (aminergic vs. cholinergic), stimulus origin (external vs. internal) and cortical activation level (low voltage vs. high voltage). The model is shown in Figure 4. We specifically proposed that the changes in thinking and memory that occur in REM sleep could be accounted for by the decrease in the ratios of aminergic to cholinergic neurotransmitters (Flicker et al., 1981), but were unlikely to reflect simply different activation levels in REM sleep compared to waking.

To illustrate this concept, let us consider the mechanism and functional consequences of the shift in the source of visual system input from the retina (in waking) to the brainstem (in REM sleep). On entry into REM sleep, the noradrenergic and serotonergic neurons of the locus coeruleus and Raphe nucleus stop firing. This in turn leads to the cessation of both aminergic modulation and inhibition in brainstem oculomotor networks, the lateral geniculate bodies, and the visual cortex. As a result of disinhibition, the peribrachial cholinergic neurons of the pons become hyperexcitable and fire in synchronous bursts which initiate the phasic activation of the geniculate bodies and visual cortex. These pontogeniculooccipital (PGO) waves are recordable in REM sleep and normally precede the rapid eye movements that characterize REM sleep. In both cats and humans, this cholinergically mediated activation of the visual system produces a signal which contains clear information about the direction of the eye movements which, in REM sleep, becomes completely uncoupled from external stimulus control (see Callaway et al., 1987 and Hobson, 1990b for details and references).Figure 4

The net result of this shift is an electrically activated visual system which is: (a) aminergically demodulated and (b) cholinergically stimulated by signals arising in the brainstem which (c) convey information about eye movement direction. These changes could determine such cognitive features of dreaming as (1) the formed visual imagery, (2) the frequent reorientation, (especially as evidenced by complete scene shifts), (3) the loss of volition (an admittedly controversial point), and (4) the loss of voluntary control of internal attention (possibly through the loss of control of eye movement, and with the notable exception of lucid dreaming; see LaBerge, 1985). The loss of these last two faculties prevent the guided thought and action that are crucial to waking mentation. In addition, a shift for aminergic to cholinergic dominance has been reported to increase both the noise level of cortical neurotransmission (Foote et al., 1983) and the ease of attentional shifts in a perceptual cueing task (Clark et al., 1989, Doricchiet al.,1991), which together could be responsible for many of the bizarre features of dreaming (Mamelak & Hobson, 1989a).

Obstacles to Progress
The study of sleep mentation suffers from several major limitations (Arkin et al., 1978) which must be overcome or minimized by any scientific strategy. First, no behavioral tests of cognition are possible during sleep, except insofar as the behavior involves automatic functions, such as heart rate, PGO activity (McCarley et al., 1983), or rapid eye movements (Roffwarg et al., 1962). Second, since sleep mentation is not usually under conscious control, it is impossible to manipulate the content of the cognition, eliminating many types of exploration (except see LaBerge, 1990). Third, reports of sleep cognition are suspect, because they are subjective and because recall of sleep cognition is so short lived and uncertain (Badia, 1990). Fourth, while the sleep laboratory setting allows accurate assessment of brain states, reports gathered in this setting are more constricted than those which are collected in more familiar home settings (Monroe, 1967). Fifth and finally, long-term longitudinal studies are infeasible in the sleep laboratory both because of the time commitment required of the researcher and the demands on the subjects. This last problem precludes longitudinal studies which could establish large databases from individual subjects that would be sensitive to the normal vicissitudes of the life cycle and to experimental interventions.

A further confounding factor in the study of mentation is the methodology often used to analyze reports. Much effort has been spent attempting to analyze their mental content in order to elucidate the experiential history and current emotional status of the dreamer. With a few important exceptions, both Freudian psychology and modern empirical dream research have tended to be content-analytic and to ignore both the global and formal levels of analysis. We believe that the study of such global and formal properties provides a more productive paradigm for the neurocognitive study of dreaming and for sleep mentation in general.

Table of Contents

Dreaming: definition, physiology and cognitive analysis

A dream is popularly defined as a sequence of sensations, images, emotions, and thoughts passing through a sleeping person's mind. The two most critical features of a dream are the presence of hallucinatory imagery, most commonly visual and auditory, and of a sequence of perceptions which takes on a scenario or story-like quality. This does not describe all sleep mentation. Awakenings from NREM sleep often elicit from subjects the comment that they were not asleep, or that they were thinking about one thing or another. While polysomnographic evidence clearly indicates that these subjects were asleep, they do not describe their prior mentation as dreaming, and it is important, we believe, to emphasize this distinction, and to characterize such thoughts as non-dreaming sleep mentation. We have more precisely characterized dream mentation as having the following six characteristics (Table 1): hallucinoid imagery, narrative structure, cognitive bizarreness, hyperemotionality, delusional acceptance, and deficient memory of its mental content.

The papers that follow address three major questions. The first question is methodological - how can we obtain subjective reports that most accurately reflect the actual content of normal sleep mentation? The second question is cognitive - what are the characteristics of dreaming and non-dreaming sleep mentation; how do they differ from each other and from waking mentation? The third question is physiological - what underlies the different types of sleep mentation; do dreaming and non-dreaming sleep mentation occur in physiologically distinct sleep stages? The remainder of this paper summarizes the answers to these questions provided in the series of papers that follows.

How can we obtain reports that most accurately reflect the actual content of normal sleep mentation?
We believe that two methodological improvements can lead to much more accurate reports of normal sleep mentation. The first is to move the site of report collection out of the unfamiliar and disquieting environment of the sleep laboratory and into the home. The second is to use affirmative phenomenological probes in the collection of the reports themselves.

Toward a naturalistic database of sleep cognition:
We have developed a new home-based method of data collection that combines the advantages of subjective mentation reports obtained in a naturalistic environment with objective verification of brain state in a natural field setting. By monitoring postural shifts (which demarcate human sleep at NREM - REM - wake state transitions) and eye movements, we have developed and sleep-laboratory tested the reliability of a simple, portable, two channel sleep-stage detector which we call the "Nightcap" (Mamelak & Hobson, 1989b). This approach has already demonstrated sensitivity to subjective estimates of goodness of sleep (Hobson et al., 1978) which might, in turn, be correlated with measurable aspects of cognitive capability.

Nightcap: A home-based sleep laboratory methodology:
The results of studies of the Nightcap reported in the second paper in this series (Stickgold et al., 1993b) support the following methodological conclusions: 1) subjective reports of dreaming following spontaneous arousal from sleep in the home are more plentiful, more expressive, longer, and richer than their laboratory counterparts. 2) Home-based dream reports collected in conjunction with sleep stage measurements made with the Nightcap corroborate the laboratory-based hypothesis that long, bizarre reports are reliably related to REM sleep physiology. 3) The veracity of home dream reports is supported by the consistency of their formal structure both within and across subjects; the validity of these measures, derived from home-based subjective reports, is strengthened by their correlation with objective physiological measures of REM sleep obtained at home, in the human sleep laboratory, and from basic neurobiological studies in animals.

Affirmative Phenomenological Probes:
One lesson that we have learned is that the open ended inquiry into dream mentation that has been typical of most past work (including our own) is grossly inadequate. If we simply ask subjects to describe the conscious experience they can recall upon awaking (especially in the sleep laboratory where they are often groggy and eager to return to sleep) their accounts are generally sketchy and poor in detail. With affirmative probes, and in the home setting, we have been able to show that children's dreams have bizarre features that are remarkably similar to those in adult dreams (see our third paper, by Resnick et al., 1993) and to detect ten times the usual levels of emotion (see our fourth paper, by Merritt et al., 1993). The analysis of reports collected using affirmative probes for bizarre uncertainties in dreams is currently in progress. We intend to design affirmative probes for assessing dream memory and dream attention in the near future.

What are the characteristics of dreaming sleep mentation?
We have characterized dream mentation in terms of six distinctive features which distinguish it from waking mentation and other forms of sleep mentation. Table 1 summarizes these features and indicates which of our papers address each of them. While most of the papers in this series are concerned with the cognitive discontinuities, incongruities, and uncertainties that we have called dream bizarreness (Rittenhouse et al., 1993, Sutton et al., 1993a,b, Stickgold et al. 1993b, and Resnick et al., 1993), it is notable that two of our papers deal with dream emotion (Merritt et al., 1993, Sutton et al., 1993b) and three with narrative structure (Stickgold et al. 1993b, Sutton et al., 1993a,b), (See Table 2).

Why have we focused so sharply on this one dream feature, paid only modest attention to two others (emotion and narrative structure) and totally ignored the other three (hallucinoid imagery, delusion and memory loss)? Our strong interest in dream bizarreness has both historical and strategic roots.

Our historical rationale is that we initially considered bizarreness to be the most distinctive and intriguing feature. Other theorists, especially psychoanalysts (Freud, 1900), had focused on the apparent meaninglessness of bizarre elements as evidence of their determined disguise and/or censorship. The concept of dream bizarreness thus seemed to lie at the very heart of the theory we hoped to replace with our activation-synthesis hypothesis (Hobson & McCarley, 1977).

Our strategic rationale has been two-fold. First, since we have been able to reliably define this robust property (Hobson et al., 1987, Hobson, 1988, Williamset al.,1991), we were encouraged to attempt its more precise quantitative measurement (Suttonet al.,1993a). Second, we believe that dream bizarreness is the most accessible measure of the shifts in cognitive processes that accompany physiological state changes over the wake-sleep cycle. In dream bizarreness we see a mental readout of the chaotic brainstem activity of REM sleep, and thus have focused on this bizarreness as an example of the "activation" portion of the activation-synthesis hypothesis.

This does not mean, however, that we were either unaware of or not interested in the obvious coherence that dream reports normally exhibit. We have heretofore avoided attempts at analyzing this characteristic of dream reports because we were dissatisfied with the methods available for its study. Four of the papers in this series (Rittenhouseet al.,1993; Stickgoldet al.,1993b; Suttonet al.,1993a,b) specifically address issues of dream coherence using new methodological tools which we have recently developed. These studies have allowed us to develop a preliminary description of structural constraints of dream construction, and provide a first view of the "synthetic" component of the activation-synthesis hypothesis.

Do dreaming and non-dreaming sleep mentation occur in physiologically distinct sleep stages?
Two reservations have heretofore restrained enthusiasm for the study of home based dreams. One is the absence of the experimenter - observer and the other is the absence of objective documentation of the state of the sleeper. We believe that the combination of affirmative phenomenological probes and our portable REM detector, the Nightcap, help to overcome or at least diminish these reservations. In doing so, they may actually tip the balance of dream research in favor of the home based paradigm.

Our reasoning runs as follows. Laboratory studies already indicate that most long, and detailed reports, come from REM sleep (Arkinet al.,1978). If we add the measurement of dream bizarreness, and compute its density (Porte & Hobson, manuscript in preparation) we further increase our capacity to distinguish between REM and NREM based reports.

But the clincher comes from our own Nightcap study (Stickgoldet al.,1993a) which shows that these same correlations and distinctions can be quantified in the home setting. Subjects in our study produced long reports (> 250 words) following spontaneous arousal from REM sleep in 43% of all cases, compared to only 8% of reports following arousal from NREM sleep. While it must be admitted that instrumental awakenings of home-based subjects might have yielded a different set of figures and one more similar to the sleep laboratory breakdown, this caveat in no way undermines our confidence that the vast majority of recalled dreams were actually experienced in REM sleep. We note that only one of 72 REM dream reports and none of 72 NREM reports recorded by Antrobus (1983) was over 250 words (personal communication). It is therefore valid to conclude that special dream cognition, as we have defined it in Table 1, is strongly, if not exclusively, correlated with REM physiology. While we continue to explore the details of this correlation, we feel confident that we have a sufficient basis to construct inferences about the possible relationship of the dream phenomenology to REM sleep neurophysiology as well as relationships between non-dreaming sleep mentation and NREM physiology.

Table of Contents


Over the past forty years the strong correlation between REM sleep and dreaming has held up as a robust empirical datum despite evidence that the relationship is non-exclusive and despite the failures of a laboratory-based effort to establish any detailed causal connection between brain state physiology and cognitive state changes. The best chance of obtaining long and detailed accounts of dream mentation remains spontaneous or instrumental awakening from eye movement-rich epochs of REM sleep.

During the same forty year epoch, the neurobiology of REM sleep has advanced apace so that studies conducted in animal models have progressed to the cellular and molecular levels. It is now possible to specify how the brain is activated during REM sleep and even to mimic this causation experimentally. Furthermore, important and informative chemical differences between the brain activation of REM sleep and that of waking have been established which encourage a new paradigm and a new approach to the study of the cognition and consciousness of REM sleep/dreaming.

This new paradigm consists of a cross-disciplinary, state-to-state mapping. It maps the neurophysiological state model developed from animal studies up to the formal and global cognitive properties of state-dependent mentation. It also maps down from the developing model of state-dependent human cognition to the results of animal studies of wake-sleep neurophysiology. This new global state-to-state mapping effort replaces the attempts at moment-to-moment cross-correlations of earlier REM/dream psychophysiological models. Complementing the paradigm shift are the methodological innovation of home-based sleep stage detection, which validates and experimentally empowers new studies of home-based dreaming, and the increased reliance on affirmative phenomenological probes.

The emerging picture is of an increasingly clear isomorphism between distinctive features of dream mentation and REM sleep neurophysiology. Chaotic phasic signals arising in the pontine brainstem constantly impinge upon, and sometimes disrupt ongoing processing in sensory and motor centers of the cortex and subcortex of the forebrain and in emotion centers of the limbic forebrain. These signals are thought to lead directly to the visual and motor hallucinations, emotional intensification, and distinctively bizarre cognition that characterize dream mentation. The change in cortical chemical modulation from aminergic dominance (in waking) to cholinergic dominance (in REM sleep) is a possible substrate for the loss of recent memory (dream amnesia), the orientational instability (dream bizarreness), and the failure of internal attention (loss of self-reflective awareness), as well as the breathtaking leaps of association seen in the narrative plot, and all of which characterize dreaming.

Table of Contents


Table of Contents