by Tao Xia, Chuan-Peng Hu, Basak Türker, Esteban Munoz Musat, Lionel Naccache, Isabelle Arnulf, Delphine Oudiette, Xiaoqing HuSleep has traditionally been conceptualized as a state of cognitive disconnection, yet emerging evidence indicates that decision-making capacities persist across sleep stages. Here, we elucidate the computational mechanisms underlying real-time lexical decision-making during polysomnographically-verified sleep, using facial electromyography and hierarchical drift diffusion modeling in both healthy individuals and participants with narcolepsy. We found that lexical decision-making was preserved during N1 and lucid REM sleep, but relied on distinct computational strategies: in N1 sleep, both enhanced sensory-motor processing and increased evidence accumulation supported decisions about words, whereas in lucid REM sleep, lexical decisions were driven exclusively by evidence accumulation processes. Cross-state comparisons revealed two fundamental principles: (1) Selective preservation—during N1 sleep, lexical decisions for words were maintained while those for pseudowords were selectively impaired, indicating that cognitive resources during sleep are preferentially allocated to meaningful stimuli; (2) Parallel strategic adaptations—during lucid REM sleep, participants increased their decision thresholds, requiring more evidence before responding, which helped maintain accuracy even though the efficiency of evidence accumulation was reduced. Our findings demonstrate that, rather than a passive decline, sleep involves dynamic and state-specific reconfiguration of the computational mechanisms underlying decision-making, with important implications for understanding consciousness and cognitive flexibility.