Comprehensive Notes on Sleep-Dependent Language Consolidation, Auditory Verbal Cueing, and Slow Oscillation-Spindle Coupling Dynamics

Theoretical Foundations of Sleep and Language Learning

• Active System Consolidation Hypothesis (ASC):

  • System-level memory consolidation during sleep relies on the repeated, spontaneous reactivation of newly encoded memory representations during subsequent non-rapid eye movement (NonREM) sleep.

  • Prolonged wakefulness leads to elevated cortical glutamatergic synaptic strength, which weakens during sleep to facilitate memory stabilization, integration, and long-term storage.

  • System consolidation redistributes transient memory traces initially encoded in the hippocampus to neocortical networks for permanent storage and schema integration.

• Sleep Microarchitecture Dynamics:

  • Slow Oscillations (SOs): Synchronized neocortical membrane potential fluctuations centered at $0.5\text{--}1.0\text{ Hz}$ ($<1\text{ Hz}$), oscillating between hyperpolarized downstates and depolarized upstates. SOs coordinate the global timing of memory replay.

  • Sleep Spindles: Transient bursts of oscillatory brain activity ($11\text{--}16\text{ Hz}$) generated by the thalamic reticular nucleus. Subdivided into slow spindles ($11\text{--}13\text{ Hz}$) and fast spindles ($13\text{--}15\text{ Hz}$ or $13\text{--}16\text{ Hz}$).

  • SO–Spindle Coupling: The phase-amplitude coupling where sleep spindle crests are nested within the depolarized upstate of neocortical slow oscillations. This precise temporal coordination provides an optimal receptive window for influx of $Ca^{2+}$ into excitatory neurons, facilitating synaptic plasticity, memory trace replay, and hippocampal-to-neocortical information transfer.

• Targeted Memory Reactivation (TMR) and Verbal Cueing:

  • External cueing during NonREM sleep using associated sensory stimuli (e.g., odors, auditory sounds, melodies) biases hippocampal replay, selectively strengthening target memory traces.

  • Complex verbal cues (such as foreign language words) can reactivate associated translations in long-term memory during NonREM sleep without causing global micro-arousals or disrupting ongoing, uncued memory consolidation processes.

• Electrophysiological Indices of Wake and Sleep Processing:

  • Theta Oscillations ($3\text{--}7\text{ Hz}$ or $5\text{--}7\text{ Hz}$): Dominant in hippocampal and prefrontal networks during active encoding, episodic retrieval, relational binding, and tracking sequential dependencies. Post-stimulus changes in task-evoked theta power index memory trace strength and cognitive control during rule processing.

  • Aperiodic Activity ($1/f$ Power Law): Non-oscillatory background spectral activity characterized by its spectral slope exponent, reflecting the excitation-inhibition ($E/I$) balance of cortical networks.

Verbal Cueing and Foreign Vocabulary Consolidation

Methodological Parameters & Experimental Design

• Subject Demographics & Screening:

  • Total cohort consisted of $N = 68$ healthy, right-handed native German speakers without prior Dutch language skills. Four participants were excluded due to sleep disturbances.

  • Main Sleep Group ($n = 15$, $8\text{ females}$, mean age $= 25.1 \pm 1.17\text{ years}$).

  • Control Sleep Group ($n = 15$, $8\text{ females}$, mean age $= 23.87 \pm 0.68\text{ years}$).

  • Active Waking Group ($n = 17$, $7\text{ females}$, mean age $= 24.7 \pm 1.11\text{ years}$).

  • Passive Waking Group ($n = 17$, $8\text{ females}$, mean age $= 23.9 \pm 0.97\text{ years}$).

  • All participants possessed normal working memory capacity, defined as an Automated Operation Span (OSPAN) score $\ge 20$.

  • Participants received financial compensation of $120\text{ Swiss Francs (CHF)}$ for sleep protocols or $100\text{ CHF}$ for wake protocols.

• Vocabulary Encoding Protocol:

  • Stimuli comprised $120\text{ Dutch--German}$ word pairs presented across 3 consecutive learning rounds starting at $\sim 22.00\text{ h}$.

  • Auditory Dutch stimuli were delivered at $70\text{ dB SPL}$ with acoustic durations between $400\text{ ms}$ and $650\text{ ms}$.

  • Round 1: Auditory Dutch word $\rightarrow$ Fixation cross ($500\text{ ms}$) $\rightarrow$ Visual German translation ($2000\text{ ms}$). Inter-trial interval (ITI): $2000\text{--}2200\text{ ms}$.

  • Round 2: Auditory Dutch word $\rightarrow$ Visual question mark (up to $7\text{ s}$) $\rightarrow$ Participant vocalizes German translation or says "next" ("weiter") $\rightarrow$ Visual German translation feedback ($2000\text{ ms}$).

  • Round 3 (Preretention Baseline): Cued recall test without visual feedback. Average baseline recall score was $60.88 \pm 1.1\text{ words}$ (mean accuracy $= 50.41\%$), establishing medium task difficulty without floor or ceiling effects.

• Reactivation Protocol ($23.00\text{--}02.00\text{ h}$):

  • During a $3\text{--hour}$ retention interval, $90\text{ Dutch words}$ were presented aurally at $50\text{ dB SPL}$ with randomized ITIs of $2.8\text{--}3.2\text{ s}$ over a $90\text{--minute}$ window, yielding $10\text{--}11\text{ exposures}$ per word.

  • The $90\text{ reactivation stimuli}$ contained: $30\text{ cued hits}$ (remembered before retention), $30\text{ cued misses}$ (unremembered before retention), and $30\text{ new control words}$ (unlearned baseline stimuli).

  • Main Sleep Group: Stimuli replayed during NonREM sleep stages N2 and SWS (N3). Auditory replay was paused upon polysomnographic signs of REM sleep, arousal, or awakenings (mean interruptions $= 5.2 \pm 0.5\text{ times}$)…

  • Control Sleep Group: Words categorized into cued/uncued subsets via identical automated MATLAB algorithms, but no auditory cues were presented during sleep.

  • Active Waking Group: Replay delivered while participants accomplished a $3\text{--back}$ working memory task across three $30\text{--minute}$ blocks intermixed with online computer games (Bubble Shooter).

  • Passive Waking Group: Replay delivered during passive sitting in three $30\text{--minute}$ blocks, where participants listened attentively to the auditory words.

• Retrieval & Testing Phase ($\sim 02.15\text{ h}$):

  • Auditory presentation of $180\text{ words}$ ($120\text{ original}$ $+ 30\text{ reactivation controls} + 30\text{ entirely novel words}$).

  • Participants judged old/new recognition status and provided the vocal German translation for recognized words.

  • Cued recall relative index: $\%\text{ Change} = \left(\frac{\text{Post-retention Score}}{\text{Pre-retention Score}}\right) \times 100$.

  • Recognition sensitivity index: $d' = z(\text{Hits}) - z(\text{False Alarms})$, with extreme proportion corrections adjusted to $\frac{1}{2N}$ or $1 - \frac{1}{2N}$ where $N = 60$.

Behavioral Outcomes & Sleep-Specificity

• Recall Enhancement in Sleep:

  • Re-exposure to Dutch words during NonREM sleep significantly enhanced subsequent cued recall of German translations compared to uncued words ($105.15 \pm 2.64\%$ vs. $95.43 \pm 2.07\%$; $t_{14} = 3.43$, $P = 0.004$).

  • Verbal cueing during sleep produced a $5.14\%$ memory gain above pre-sleep baseline performance ($+5.14 \pm 2.64\%$; $P = 0.072$, two-sided $t\text{--test}$). Uncued words exhibited significant post-sleep forgetting ($-4.75 \pm 2.07\%$; $P = 0.045$).

  • Individual consistency: $12\text{ out of }15\text{ participants}$ demonstrated a net positive memory advantage for cued relative to uncued words (range $+1\text{ to }+11\text{ words}$).

• Absence of Consolidation Costs for Uncued Items:

  • In the Control Sleep Group (no cueing), recall change for cued-equivalent ($93.55 \pm 2.37\%$) and uncued-equivalent words ($92.80 \pm 3.10\%$) did not differ ($t_{14} = 0.24$, $P = 0.81$).

  • Recall performance for uncued words in the Main Sleep Group ($95.43 \pm 2.07\%$) was equivalent to recall performance in the Control Sleep Group ($93.55 \pm 2.37\%$; $t_{14} = 0.71$, $P = 0.48$).

  • Cued recall in the Main Sleep Group significantly surpassed performance in the Control Sleep Group ($105.14 \pm 2.64\%$ vs. $92.80 \pm 3.10\%$; $t_{14} = 3.26$, $P = 0.003$). Verbal cueing directly improves target memories without compromising uncued consolidation.

• Wakefulness Control Groups:

  • Active Waking: Cued recall did not differ between cued ($85.53 \pm 2.81\%$) and uncued words ($84.21 \pm 2.16\%$; $t_{16} = 0.56$, $P = 0.58$).

  • Passive Waking: Cued recall did not differ between cued ($79.86 \pm 4.58\%$) and uncued words ($81.25 \pm 2.09\%$; $t_{16} = -0.35$, $P = 0.74$).

  • Retention state comparison: Sleep overall supported memory retention significantly better than wakefulness ($F_{3,60} = 13.06$, $P < 0.001$; post-hoc sleep vs. wake $t_{62} = 5.61$, $P < 0.001$).

• Item Recognition Dynamics:

  • Sleep overall enhanced recognition sensitivity $d'$ compared to wakefulness ($F_{2,46} = 15.87$, $P < 0.001$). Main sleep group ($d' = 2.32 \pm 0.15$) and Control sleep group ($d' = 2.01 \pm 0.13$) outperformed Active wake ($d' = 1.44 \pm 0.15$) and Passive wake ($d' = 1.13 \pm 0.17$).

  • In-sleep verbal cueing had no effect on word recognition memory ($F_{3,60} = 1.35$, $P = 0.15$), confirming that cueing specifically strengthens inter-item associations (Dutch–German link) required for recall rather than single-item familiarity.

Electrophysiological Correlates & Spectral Dynamics

• EEG Recording & Trial Partitioning:

  • High-density $128\text{--channel}$ Geodesic Sensor Net sampled at $500\text{ Hz}$, online vertex referenced (Cz), offline re-referenced to averaged mastoids.

  • Trial Categories: "Gains" (unremembered pre-sleep, remembered post-sleep), "Losses" (remembered pre-sleep, forgotten post-sleep), "HitHit" (remembered pre- and post-sleep), "MissMiss" (unremembered pre- and post-sleep), and "Control" (new stimuli).

• Event-Related Potential (ERP) Negativity:

  • Replayed words that yielded memory gains ("Gains") evoked a pronounced late frontal negativity relative to all other trial categories at electrode Fz between $800\text{ ms}$ and $1100\text{ ms}$ post-stimulus ($F_{6,84} = 4.52$, $P = 0.001$; pairwise post-hocs $P < 0.04$).

  • Topographical mapping revealed a fronto-central scalp distribution (false discovery rate corrected $P < 0.05$ across electrodes E4, E5, E6, E11, E12, E13, E16, E19, E20, E23, E24, E28, E29, E35, E112).

• Post-Stimulus Slow Oscillation Evocation:

  • Single-trial analysis identified slow oscillations as negative deflections $>75\mu\text{V}$ lasting $>500\text{ ms}$ starting between $0\text{ and }800\text{ ms}$ post-cue.

  • "Gain" trials triggered post-stimulus slow oscillations significantly more frequently ($31.09 \pm 3.6\%$) than "Loss" trials ($18.48 \pm 3.4\%$; $t_{14} = 5.35$, $P < 0.001$) at frontal sites (Fz, F3, F4).

• Induced Oscillatory Theta Dynamics:

  • Induced theta power ($5\text{--}7\text{ Hz}$, Morlet parameter $c = 7$) exhibited significant differences between cue conditions ($F_{4,56} = 7.38$, $P = 0.002$).

  • "Gains" elicited a robust increase in induced theta power from $700\text{ ms}$ to $900\text{ ms}$ post-stimulus relative to "Losses", localized to right frontal (FC6 / E117: $t_{14} = 3.68$, $P = 0.009$) and left parietal sensor clusters.

• Spindle Band Dynamics:

  • SWS-specific time-frequency analysis revealed a selective increase in slow spindle power ($11\text{--}13\text{ Hz}$) $600\text{--}800\text{ ms}$ following cue presentation for "Gains" versus "Losses" ($P < 0.05$). No overall main effects emerged for fast spindles ($13\text{--}15\text{ Hz}$)…

Sequence-Based Rule Learning and Grammatical Consolidation

Mini Pinyin Architecture & Linguistic Paradigms

• Linguistic Composition:

  • Mini Pinyin is a modified artificial language modeled on Mandarin Chinese containing $32\text{ transitive verbs}$, $25\text{ nouns}$, $2\text{ coverbs}$, and $4\text{ classifiers}$.

  • Nouns encompass $10\text{ human entities}$, $10\text{ animals}$, and $5\text{ small/large objects}$.

  • Classifiers mandatorily precede noun phrases: ge (human), zhi (animal), da (large object), xi (small object).

  • Full corpus contains $576\text{ unique sentences}$ ($288\text{ grammatical}$, $288\text{ ungrammatical}$).

• Word Order Conditions:

  • Fixed Word Order (AbaUV): Requires the presence of coverb ba (actor-undergoer-verb marker). Instantiates a strict, invariant sequence where the first noun phrase is the actor and the verb is positioned at the absolute end of the sentence (e.g., zhi xiong ba xi pingguo chile $\rightarrow$ "[animal] bear ba [small object] apple eats").

  • Flexible Word Order (AVU, UVA): Does not contain coverb ba. Relies on animacy cues and verb placement between noun phrases. Actor-verb-undergoer (AVU; e.g., zhi xiong chile xi pingguo) and undergoer-verb-actor (UVA; e.g., xi pingguo chile zhi xiong) are both grammatical.

  • Ungrammatical Constructions:

  • Fixed violations: UbaAV (incorrect noun ordering, e.g., xi pingguo ba zhi xiong chile) or AbaVU (incorrect verb position, e.g., zhi xiong ba chile xi pingguo).

  • Flexible violations: AUV (verb-final without ba, e.g., zhi xiong xi pingguo chile) or UAV (e.g., xi pingguo zhi xiong chile).

Experimental Workflow & Behavioral Dynamics

• Participants & Setup:

  • $N = 36$ monolingual, right-handed native English speakers ($16\text{ males}$, mean age $= 25.4 \pm 7.10\text{ years}$). One participant excluded due to technical recording errors ($n = 17\text{ sleep group}$, $n = 18\text{ wake group}$)…

  • Pre-screening verified right-handedness (FLANDERS) and normal sleep quality (Pittsburgh Sleep Quality Index, PSQI scores $1\text{--}5$, mean $= 2.9 \pm 1.33$). Self-perceived sleepiness tracked via Stanford Sleepiness Scale (SSS).

  • Pre-study phase: Participants studied noun visual vocabulary booklets for $\ge 15\text{ minutes/day}$ over $\sim 1\text{ week}$, passing a keyboard translation screening at $\ge 90\%$ accuracy.

• Task Protocol:

  • Sentence Learning Task: $96\text{ picture-sentence trials}$. Event image shown for $5000\text{ ms}$, followed by word-by-word sentence presentation ($700\text{ ms}$ word duration, $200\text{ ms}$ inter-stimulus interval, ISI).

  • Baseline & Delayed Sentence Judgment Tasks: $288\text{ sentences}$ ($144\text{ grammatical}$, $144\text{ ungrammatical}$; $156\text{ core target subset}$ presented word-by-word at $600\text{ ms}$ duration, $200\text{ ms}$ ISI). Sentence judgment prompt ("?") shown for $4000\text{ ms}$. Baseline task provided trial-by-trial accuracy feedback; delayed task provided no feedback.

  • Sleep Group: Learning ($21.00\text{ h}$) $\rightarrow$ Baseline test ($22.00\text{ h}$) $\rightarrow 8\text{--hour}$ nocturnal sleep opportunity ($23.00\text{--}07.00\text{ h}$) with continuous PSG $\rightarrow$ Delayed test ($08.00\text{ h}$).

  • Wake Group: Learning ($08.00\text{ h}$) $\rightarrow$ Baseline test ($09.00\text{ h}$) $\rightarrow 12\text{--hour}$ wakefulness period $\rightarrow$ Delayed test ($21.00\text{ h}$).

• Behavioral Results:

• Logistic mixed-effects regression of response accuracy revealed a significant interaction of $\text{Grammaticality} \times \text{Sentence Type} \times \text{Condition}$ ($\beta = 0.13$, $\text{SE} = 0.03$, $p < 0.001$).

• Holm–Bonferroni post-hoc comparisons confirmed that sleep significantly increased post-delay response accuracy relative to wake for fixed grammatical word orders (Odds Ratio $\text{OR} = 0.55$, $\text{SE} = 0.12$, $z = -2.60$, $p_{\text{adj}} = 0.03$).

• Within the sleep group, post-delay response accuracy was higher for grammatical fixed versus grammatical flexible constructions ($\text{OR} = 0.58$, $\text{SE} = 0.06$, $z = -4.63$, $p_{\text{adj}} < 0.001$).

Task-Evoked Theta Oscillations & Spectral Exponents

• Time-Frequency Signal Processing:

  • EEG recorded from $32\text{ channels}$, re-referenced to linked mastoids, filtered $0.1\text{--}40\text{ Hz}$. Epochs extracted $-0.5\text{ to }6.5\text{ s}$ relative to sentence onset.

  • Individual Alpha Frequency (IAF) quantified from resting-state parietal-occipital channels (P3, P4, O1, O2, P7, P8, Pz, Oz) to define individualized harmonic theta frequency bands.

  • Task-evoked theta power calculated using complex Morlet wavelets with baseline correction spanning $-0.5\text{ to }0\text{ s}$ pre-sentence onset. Background scale-free spectral decay ($1/f$ exponent) estimated via Irregular-Resampling Auto-Spectral Analysis (IRASA) and included as a model covariate.

• Cluster-Based Permutation & Sensitivity Models:

  • A non-parametric cluster-based permutation test contrasting $\text{Condition (Sleep, Wake)} \times \text{Sentence Type (Fixed, Flexible)}$ isolated a significant theta cluster during the delayed session ($p = 0.008$).

  • Linear mixed-effects modeling on judgment sensitivity ($d'$) demonstrated a significant $\text{Theta} \times \text{Condition} \times \text{Sentence Type}$ interaction ($\beta = -1.09$, $\text{SE} = 0.34$, $p = 0.001$).

  • For fixed word order sentences, reduced theta desynchronization (higher relative theta power) was significantly associated with improved $d'$ sensitivity in the sleep group compared to wake ($\beta = -4.70$, $\text{SE} = 1.10$, $p_{\text{adj}} < 0.001$). For flexible word order sentences, elevated theta synchronization correlated with poorer judgment sensitivity following sleep.

NREM Slow Oscillation–Spindle Coupling Mechanics

• Algorithmic Detection of NREM Oscillations:

  • Sleep Spindles: Automated detection across individual peak spindle frequencies ($12\text{--}16\text{ Hz}$) on derivative power spectral density time series. Morlet wavelet filtering flagged spindle events exceeding $6 \times \text{median amplitude}$ for a duration $\ge 400\text{ ms}$.

  • Slow Oscillations: Continuous NREM data bandpass filtered $0.5\text{--}4.0\text{ Hz}$. Positive-to-negative zero-crossings separated by $0.5\text{--}2.0\text{ s}$ isolated, with events in the top amplitude quartile classified as SOs.

  • Coupling Metric: Quantified as the percentage of detected spindles coinciding with SO events and the mean vector length of SO phase at peak spindle amplitude (SO–spindle coupling strength).

• Coupling Strength & Word Order Consolidation:

  • Circular Rayleigh testing confirmed significant phase preference regularity across participants ($p \le 0.001$).

  • Linear mixed-effects regression showed a significant interaction of $\text{Coupling Strength} \times \text{Sentence Type} \times \text{Sagittality}$ ($\beta = 3.05$, $\text{SE} = 0.97$, $p = 0.002$).

  • Pairwise contrasts established that stronger SO–spindle coupling predicted superior delayed judgment sensitivity ($d'$) selectively over anterior channels for fixed word order rules ($\beta = 6.85$, $\text{SE} = 2.01$, $p_{\text{adj}} < 0.001$). Coupling strength did not predict sensitivity in central ($\beta = -0.75$, $p_{\text{adj}} = 0.77$) or posterior regions ($\beta = -3.90$, $p_{\text{adj}} = 0.26$).

Cross-Frequency Coupling & Task-Evoked Interplay

• Predictive Interaction on Behavioral Sensitivity:

  • A linear mixed-effects model regressing anterior SO–spindle coupling strength, task-evoked anterior theta power, sagittality, and sentence type onto delayed $d'$ sensitivity yielded a significant three-way interaction of $\text{SO-Spindle Coupling Strength} \times \text{Task Theta Power} \times \text{Sentence Type}$ ($\beta = -41.60$, $\text{SE} = 16.70$, $p = 0.01$).

  • The combination of robust anterior NREM SO–spindle coupling and elevated task-evoked anterior theta power during retrieval strongly predicted optimal delayed judgment sensitivity ($d'$) for sequence-based fixed word order rules.

  • Demonstrates a direct link between offline NREM neurophysiological coupling dynamics during sleep and online task-evoked oscillatory retrieval dynamics during waking language processing.

Methodological Specifications & Quantitative Summaries

Sleep Macroarchitecture Parameters

• Vocabulary Reactivation Experiment Sleep Macroarchitecture:

  • Total Sleep Time (TST) in retention interval: Main Sleep $= 185.31 \pm 6.80\text{ min}$; Control Sleep $= 185.47 \pm 5.12\text{ min}$.

  • Stage N1 Duration: Main Sleep $= 7.76 \pm 1.66\text{ min}$ ($4.02 \pm 0.84\%$); Control Sleep $= 5.20 \pm 1.46\text{ min}$ ($2.72 \pm 0.70\%$); $P = 0.16$.

  • Stage N2 Duration: Main Sleep $= 93.16 \pm 5.93\text{ min}$ ($48.70 \pm 2.64\%$); Control Sleep $= 100.27 \pm 4.71\text{ min}$ ($53.73 \pm 2.95\%$); $P = 0.71$.

  • Slow Wave Sleep (SWS / N3) Duration: Main Sleep $= 62.26 \pm 5.80\text{ min}$ ($33.11 \pm 3.26\%$); Control Sleep $= 57.93 \pm 5.37\text{ min}$ ($31.13 \pm 2.95\%$); $P = 0.94$.

  • Stage REM Duration: Main Sleep $= 22.13 \pm 3.18\text{ min}$ ($11.38 \pm 1.59\%$); Control Sleep $= 22.07 \pm 2.73\text{ min}$ ($11.65 \pm 1.36\%$); $P = 0.37$.

  • Wake After Sleep Onset (WASO): Main Sleep $= 4.66 \pm 1.71\text{ min}$ ($2.35 \pm 0.82\%$); Control Sleep $= 0.37 \pm 0.14\text{ min}$ ($0.002 \pm 0.00\%$); $P = 0.03$.

  • Reactivation counts in Main Sleep: Stage N2 $= 442.86 \pm 40.68\text{ exposures}$; SWS $= 508.80 \pm 54.42\text{ exposures}$.

• Artificial Language Experiment Nocturnal Sleep Parameters:

  • Total Sleep Time (TST): $400.00 \pm 67.02\text{ min}$.

  • Sleep Onset Latency (SOL): $15.23 \pm 12.23\text{ min}$.

  • Wake After Sleep Onset (WASO): $52.64 \pm 55.60\text{ min}$.

  • Stage N1: $38.05 \pm 29.47\text{ min}$ ($10.05 \pm 8.21\%$ of TST).

  • Stage N2: $196.30 \pm 46.29\text{ min}$ ($49.52 \pm 10.36\%$ of TST).

  • Slow Wave Sleep (SWS / N3): $104.23 \pm 42.27\text{ min}$ ($25.84 \pm 9.60\%$ of TST).

  • Stage REM: $61.30 \pm 39.39\text{ min}$ ($14.57 \pm 8.56\%$ of TST).

Signal Processing, ERP & Oscillatory Metrics

• Vocabulary Cued Recall Performance Statistics:

  • Main Sleep Group: Cued baseline $= 29.87 \pm 0.09\text{ words}$, retrieval $= 31.40 \pm 0.16\text{ words}$, absolute change $= +1.53 \pm 0.79\text{ words}$, relative change $= 105.15 \pm 2.64\%$. Uncued baseline $= 33.20 \pm 2.54\text{ words}$, retrieval $= 31.33 \pm 2.17\text{ words}$, absolute change $= -1.87 \pm 0.70\text{ words}$, relative change $= 95.43 \pm 2.07\%$. Change difference $t_{14} = 3.52$, $P = 0.003$.

  • Control Sleep Group: Cued baseline $= 30.00\text{ words}$, retrieval $= 28.07 \pm 0.71\text{ words}$, relative change $= 93.55 \pm 2.37\%$. Uncued baseline $= 31.93 \pm 1.84\text{ words}$, retrieval $= 29.27 \pm 1.66\text{ words}$, relative change $= 92.80 \pm 3.10\%$.

  • Active Waking Group: Cued relative change $= 85.53 \pm 2.81\%$; Uncued relative change $= 84.21 \pm 2.16\%$.

  • Passive Waking Group: Cued relative change $= 79.86 \pm 4.58\%$; Uncued relative change $= 81.25 \pm 2.09\%$.

• NREM SO–Spindle Coupling Regional Metrics:

  • Channel C3: Coupling strength $= 0.53 \pm 0.11$, mean SO phase $= -0.27 \pm 0.29\text{ rad}$, percent coupled $= 16.95 \pm 4.15\%$, coupled density $= 1.33 \pm 0.38\text{ events/30s}$, uncoupled density $= 6.55 \pm 1.50\text{ events/30s}$.

  • Channel C4: Coupling strength $= 0.55 \pm 0.11$, mean SO phase $= -0.24 \pm 0.32\text{ rad}$, percent coupled $= 17.36 \pm 4.58\%$, coupled density $= 1.32 \pm 0.39\text{ events/30s}$, uncoupled density $= 6.33 \pm 1.56\text{ events/30s}$.

  • Channel F7: Coupling strength $= 0.44 \pm 0.11$, mean SO phase $= 0.08 \pm 0.73\text{ rad}$, percent coupled $= 18.39 \pm 4.88\%$, coupled density $= 1.09 \pm 0.38\text{ events/30s}$, uncoupled density $= 4.81 \pm 1.38\text{ events/30s}$.

  • Channel F8: Coupling strength $= 0.52 \pm 0.11$, mean SO phase $= -0.14 \pm 0.29\text{ rad}$, percent coupled $= 18.78 \pm 5.17\%$, coupled density $= 1.13 \pm 0.42\text{ events/30s}$, uncoupled density $= 4.81 \pm 1.22\text{ events/30s}$.

  • Channel P3: Coupling strength $= 0.58 \pm 0.11$, mean SO phase $= -0.34 \pm 0.26\text{ rad}$, percent coupled $= 17.36 \pm 3.65\%$, coupled density $= 1.45 \pm 0.35\text{ events/30s}$, uncoupled density $= 6.93 \pm 1.29\text{ events/30s}$.

  • Channel P4: Coupling strength $= 0.57 \pm 0.12$, mean SO phase $= -0.39 \pm 0.23\text{ rad}$, percent coupled $= 17.11 \pm 4.53\%$, coupled density $= 1.37 \pm 0.39\text{ events/30s}$, uncoupled density $= 6.69 \pm 1.40\text{ events/30s}$.

Statistical Models & Regression Estimates

• Sentence Accuracy Generalized Linear Mixed Model (GLMM):

  • Formula: $\text{Logit(Response)} = \beta_0 + \beta_1(\text{Grammaticality}) \times \beta_2(\text{Sentence Type}) \times \beta_3(\text{Condition}) + \beta_4(\text{Baseline Accuracy}) + \beta_5(\text{SSS}) + (1|\text{Subject}) + (1|\text{Item})$.

  • Three-way interaction: $\beta = 0.13$, $\text{SE} = 0.03$, $p < 0.001$.

  • Fixed Grammatical (Sleep vs. Wake): $\text{OR} = 0.55$, $\text{SE} = 0.12$, $z = -2.60$, $p_{\text{adj}} = 0.03$.

  • Fixed Ungrammatical (Sleep vs. Wake): $\text{OR} = 0.89$, $\text{SE} = 0.19$, $z = -0.52$, $p_{\text{adj}} = 1.00$.

  • Grammatical Fixed vs. Grammatical Flexible (Sleep): $\text{OR} = 0.58$, $\text{SE} = 0.06$, $z = -4.63$, $p_{\text{adj}} < 0.001$.

• Task-Evoked Theta Sensitivity Linear Mixed Model (LMM):

  • Formula: $d' = \beta_0 + \beta_1(\text{Task Theta Power}) \times \beta_2(\text{Condition}) \times \beta_3(\text{Sentence Type}) + \beta_4(\text{Baseline } d') + \beta_5(\text{Aperiodic } 1/f) + \beta_6(\text{Channel}) + (1|\text{Subject})$.

  • Three-way interaction: $\beta = -1.09$, $\text{SE} = 0.34$, $p = 0.001$.

  • Fixed Word Order Theta Slope (Sleep): $\beta = -4.70$, $\text{SE} = 1.10$, $p_{\text{adj}} < 0.001$.

• SO–Spindle Coupling Strength Linear Mixed Model (LMM):

  • Formula: $\text{Accuracy} = \beta_0 + \beta_1(\text{Coupling Strength}) \times \beta_2(\text{Sentence Type}) \times \beta_3(\text{Sagittality}) \times \beta_4(\text{Laterality}) + \beta_5(\text{Baseline Accuracy}) + \beta_6(\text{Sleep Stage}) + (1|\text{Subject})$.

  • Three-way interaction: $\beta = 3.05$, $\text{SE} = 0.97$, $p = 0.002$.

  • Anterior Fixed Word Order Slope: $\beta = 6.85$, $\text{SE} = 2.01$, $p_{\text{adj}} < 0.001$.

  • Central Fixed Word Order Slope: $\beta = -0.75$, $\text{SE} = 2.62$, $p_{\text{adj}} = 0.77$.

  • Posterior Fixed Word Order Slope: $\beta = -3.90$, $\text{SE} = 3.47$, $p_{\text{adj}} = 0.26$.

• Coupling $\times$ Theta $\times$ Rule Interaction Model:

  • LMM predicting delayed $d'$ from anterior coupling strength, task theta power, sagittality, and sentence type yielded a significant interaction: $\beta = -41.60$, $\text{SE} = 16.70$, $p = 0.01$.