Sleep involves dynamic and state-specific reconfiguration of the computational mechanisms underlying decision-making, rather than a passive decline, with lexical decision-making preserved during N1 and lucid REM sleep but relying on distinct computational strategies.
Key Findings
Results
Lexical decision-making was preserved during N1 sleep and lucid REM sleep, but relied on distinct computational strategies in each stage.
Study used polysomnographically-verified sleep with facial electromyography and hierarchical drift diffusion modeling (HDDM).
Participants included both healthy individuals and participants with narcolepsy.
N1 sleep supported decisions through both enhanced sensory-motor processing and increased evidence accumulation.
Lucid REM sleep supported decisions exclusively through evidence accumulation processes.
Results
During N1 sleep, lexical decisions for words were maintained while those for pseudowords were selectively impaired.
This selective preservation indicates that cognitive resources during sleep are preferentially allocated to meaningful stimuli.
The authors describe this as 'Selective preservation' — one of two fundamental principles identified in cross-state comparisons.
Pseudoword processing, which requires more effortful phonological and lexical analysis, was disproportionately affected in N1 sleep.
Results
During lucid REM sleep, participants increased their decision thresholds, requiring more evidence before responding.
This threshold increase is described as a 'Parallel strategic adaptation,' the second of two fundamental principles identified.
The increased decision threshold helped maintain accuracy even though the efficiency of evidence accumulation was reduced in lucid REM sleep.
This pattern suggests an active compensatory mechanism operating during lucid REM sleep rather than passive cognitive degradation.
Results
In N1 sleep, both enhanced sensory-motor processing and increased evidence accumulation supported decisions about words.
Sensory-motor processing is captured in HDDM by the non-decision time parameter, reflecting perceptual encoding and motor response components.
Evidence accumulation is captured by the drift rate parameter in the drift diffusion model.
Both parameters were altered in a direction supporting word-specific decision-making during N1 sleep.
This contrasts with lucid REM sleep, where only evidence accumulation (drift rate) was implicated.
Methods
The study employed hierarchical drift diffusion modeling (HDDM) to decompose the computational mechanisms of lexical decision-making across sleep states.
HDDM separates decision-making into distinct parameters including drift rate (evidence accumulation efficiency), decision threshold (response caution), and non-decision time (sensory-motor processing).
Facial electromyography was used to detect motor responses during sleep without requiring full awakening.
Polysomnographic verification confirmed sleep staging throughout the experiment.
The paradigm was applied across multiple sleep states: N1 sleep, lucid REM sleep, and presumably wakefulness as a baseline.
Discussion
The findings demonstrate that sleep involves dynamic and state-specific reconfiguration of computational mechanisms underlying decision-making rather than passive cognitive decline.
Traditional conceptualization of sleep as a state of cognitive disconnection is challenged by these results.
The two principles identified — selective preservation and parallel strategic adaptations — together indicate active reorganization of cognitive processing.
The authors note 'important implications for understanding consciousness and cognitive flexibility.'
Narcolepsy participants were included, likely because narcolepsy involves disrupted boundaries between sleep and wake states, particularly with cataplexy and REM intrusions.
What This Means
This research suggests that the sleeping brain is not simply 'switched off' but instead actively reorganizes how it makes decisions depending on which stage of sleep a person is in. Using a task where participants had to distinguish real words from made-up words (pseudowords), the researchers found that people could still make these decisions during light sleep (N1) and during lucid dreaming (a special form of REM sleep where the sleeper is aware they are dreaming). However, the brain used different strategies in each stage to accomplish this — in light sleep, both faster sensory-motor processing and better evidence gathering supported word recognition, while in lucid REM sleep, only the evidence-gathering process was involved, and the brain compensated by becoming more cautious and requiring more evidence before making a response.
One particularly striking finding is that during light sleep, the brain preferentially processed real, meaningful words while struggling more with made-up pseudowords. This suggests that even during sleep, limited cognitive resources are automatically directed toward stimuli that carry meaning — a kind of selective filtering that prioritizes semantically relevant information. This selective preservation, combined with the compensatory threshold-raising seen in lucid REM sleep, points to sleep as a state of active cognitive adaptation rather than uniform shutdown.
This research matters because it deepens our understanding of consciousness and the boundaries between waking and sleeping mental life. The use of a mathematical modeling framework (drift diffusion modeling) allowed the researchers to precisely identify which aspects of the decision-making process were preserved or changed, going beyond simply asking whether people can respond during sleep to explaining *how* and *why* they can. These findings also have potential relevance for understanding conditions like narcolepsy, where the boundaries between sleep and wakefulness are blurred, and for broader questions about what level of cognitive processing remains active during sleep.
Xia T, Hu C, Türker B, Musat E, Naccache L, Arnulf I, et al.. (2026). How sleeping minds decide: State-specific reconfigurations of lexical decision-making.. PLoS computational biology. https://doi.org/10.1371/journal.pcbi.1014007