The Language of Dreams: Application of Linguistics-Based Approaches for the Automated Analysis of Dream Experiences

New strategies for the cognitive science of dreaming

Dreams have long captivated human curiosity, but empirical research in this area has faced significant methodological challenges. Recent interdisciplinary advances have now opened up new opportunities for studying dreams. This review synthesizes these advances into three methodological frameworks and describes how they overcome historical barriers in dream research. First, with observable dreaming, neural decoding and real-time reporting offer more direct measures of dream content. Second, with dream engineering, targeted stimulation and lucidity provide routes to experimentally manipulate dream content. Third, with computational dream analysis, the generation and exploration of large dream-report databases offer powerful avenues to identify patterns in dream content. By enabling researchers to systematically observe, engineer, and analyze dreams, these innovations herald a new era in dream science.

The influence of intentions on dream content

Study Objectives

The "Zeigarnik effect" refers to the phenomenon where future intentions are remembered effectively only as long as they are not executed. This study investigates whether these intentions, which remain active during sleep, influence dream content.

Methods

After an adaptation night, each of the 19 participants (10 women and 9 men) received three different task plans in the evening before the experimental night, each describing how to perform specific tasks. One of the task plans (completed) was then to be executed before the sleep period, another task (uncompleted) was told to be executed in the next morning, and on the third task (interrupted) participants were interrupted during the enactment before sleep and told to resume it the next morning. Polysomnography and multiple awakenings were conducted, resulting in 86 dream reports, 36 in NREM stage 2, and 50 in rapid eye movement sleep. After a traditional rating-based analysis of dream reports yielded inconsistent results, we analyzed the reports using a transformer-based assessment of dream incorporation, which quantified the semantic similarity between the dreams and pre-sleep tasks.

Results

The number of dreams showing above-criterion similarity to the respective task was significantly lower for the completed than the uncompleted or interrupted tasks (p < .05, χ2 test). This pattern was confirmed through a forced choice approach, where-based on the similarity of single sentences of the dream reports-each dream report was allocated to one of the three task plans (p < 0.01, one-tailed χ2 test).

Conclusions

Active intentions increase the likelihood of dream content being semantically similar to these intentions.

A novel method for objectively classifying sequential emotion within dreams: a proof-of-concept pilot study

Traditionally, emotions in dreams have been assessed using subjective ratings by human raters (e.g., external raters or dreamers themselves). These methods have extensive support and utility in dream science, yet they have certain innate limitations due to the subjective nature of the rating methodologies. Attempting to circumvent several of these limitations, we aimed to develop a novel method for objectively classifying and quantifying sequential (word-for-word) emotion within a dream report. We investigated whether sentiment analysis, a branch of natural language processing, could be used to generate continuous positive and negative valence ratings across a dream. In this pilot, proof-of-concept study, we used 14 dream reports collected upon awakening following overnight polysomnography. We also collected pre- and post-sleep affective data and personality metrics. Our objectives included demonstrating that (1) valence ratings derived from sentiment analysis (Valence Aware Dictionary for sEntiment Reasoning [VADER]) could be used to visualize (plot) positive and negative emotion fluctuations within a dream, (2) how the visual properties of emotion fluctuations within a dream (peaks and troughs, area under the curve) can be used to generate novel "emotion indicators" as proxies for emotion regulation throughout a dream, and (3) these emotion indicators correlate with sleep, affective, and personality variables known to be associated with dreaming and emotion regulation. We describe 6 novel, objective dream emotion indicators: Total number of Peaks, total number of Troughs, Positive, Negative, and Overall Emotion Intensity (composites from an "area under the curve" method using the trapezoid rule applied to the peaks and troughs), and the Emotion Gradient (a polynomial trendline fitted to the emotion fluctuations in the dream chart). The latter signifies the overall direction of sequential emotion changes within a dream. Results also showed that ⅚ emotion indicators correlated significantly with at least one existing sleep, affective, or personality variable known to be associated with dreaming and emotion regulation. We propose that the novel emotion indicators potentially serve as proxies for emotion regulation processes unfolding within a dream. These preliminary findings provide a methodological foundation for future studies to test and refine the method in larger and more diverse samples.

Sleep and dream disturbances associated with dissociative experiences

Some dissociative experiences may be related, in part, to REM intrusion into waking consciousness. If so, some aspects of dream content may be associated with daytime dissociative experiences. We tested the hypothesis that some types of dream content would predict daytime dissociative symptomology. As part of a longitudinal study of the impact of dreams on everyday behavior we administered a battery of survey instruments to 219 volunteers. Assessments included the Dissociative Experiences Scale (DES), along with other measures known to be related to either REM intrusion effects or dissociative experiences. We also collected dream reports and sleep measures across a two-week period from a subgroup of the individuals in the baseline group. Of this subgroup we analyzed two different subsamples; 24 individuals with dream recall for at least half the nights in the two-week period; and 30 individuals who wore the DREEM Headband which captured measures of sleep architecture. In addition to using multiple regression analyses to quantify associations between DES and REM intrusion and dream content variables we used a split half procedure to create high vs low DES groups and then compared groups across all measures. Participants in the high DES group evidenced significantly greater nightmare distress scores, REM Behavior Disorder scores, paranormal beliefs, lucid dreams, and sleep onset times. Validated measures of dreamed first person perspective and overall dream coherence in a time series significantly predicted overall DES score accounting for 26% of the variance in dissociation. Dream phenomenology and coherence of the dreamed self significantly predicts dissociative symptomology as an individual trait. REM intrusion may be one source of dissociative experiences. Attempts to ameliorate dissociative symptoms or to treat nightmare distress should consider the stability of dream content as a viable indicator of dissociative tendencies.

Influencing dreams through sensory stimulation: A systematic review

Sleep is typically considered a state of disconnection from the environment, yet instances of external sensory stimuli influencing dreams have been reported for centuries. Explaining this phenomenon could provide valuable insight into dreams' generative and functional mechanisms, the factors that promote sleep continuity, and the processes that underlie conscious awareness. Moreover, harnessing sensory stimuli for dream engineering could benefit individuals suffering from dream-related alterations. This PRISMA-compliant systematic review assessed the current evidence concerning the influence of sensory stimulation on sleep mentation. We included 51 publications, of which 21 focused on auditory stimulation, ten on somatosensory stimulation, eight on olfactory stimulation, four on visual stimulation, two on vestibular stimulation, and one on multimodal stimulation. Furthermore, nine references explored conditioned associative stimulation: six focused on targeted memory reactivation protocols and three on targeted lucid reactivation protocols. The reported frequency of stimulus-dependent dream changes across studies ranged from 0 to ∼80%, likely reflecting a considerable heterogeneity of definitions and methodological approaches. Our findings highlight a lack of comprehensive understanding of the mechanisms, functions, and neurophysiological correlates of stimulus-dependent dream changes. We suggest that a paradigm shift is required for meaningful progress in this field.

Structural differences between non-lucid dreams, lucid dreams and out-of-body experience reports assessed by graph analysis

Dreaming is a complex phenomenon that occurs during sleep, involving various conscious dream experiences. Lucid dreams (LDs) involve heightened awareness within the dream environment, while out-of-body experiences (OBEs) involve the sensation of being outside one's physical body. OBEs occur during sleep paralysis (SP), where voluntary movements are inhibited during sleep/wake transitions while remaining aware of the surroundings. The relationship between LDs and OBEs is debated, with some viewing them as distinct phenomena and others considering them different manifestations of the same underlying experience. This study aimed to characterize non-lucid dreams, LDs, and OBEs by analyzing dream reports' structural properties. OBE reports displayed a condensed and interconnected network structure compared to non-lucid dreams and LDs. Additionally, OBE reports exhibited a specialized network structure, with specific nodes playing a more central role. These findings suggest that OBE dreams may have a more coherent and unified narrative, with certain nodes being pivotal in the network structure.

Mapping dreams in a computational space: A phrase-level model for analyzing Fight/Flight and other typical situations in dream reports

This article demonstrates that an automated system of linguistic analysis can be developed - the Oneirograph - to analyze large collections of dreams and computationally map their contents in terms of typical situations involving an interplay of characters, activities, and settings. Focusing the analysis first on the twin situations of fighting and fleeing, the results provide densely detailed empirical evidence of the underlying semantic structures of typical dreams. The results also indicate that the Oneirograph analytic system can be applied to other typical dream situations as well (e.g., flying, falling), each of which can be computationally mapped in terms of a distinctive constellation of characters, activities, and settings.