Experimental design.

A: Design overview: There were three experimental sessions per participant. Participants were played with pairs of foreign words and translation words during sleep starting around 11 pm. They took a first retrieval test at 12 hours and a second retrieval test at 36 hours after the acoustic stimulations during sleep. The EEG was recorded during sleep and during the first retrieval at 12 hours. B: Experimental conditions: We varied between subjects whether the words were played during peaks or troughs of slow-waves. Within subjects we played pairs of foreign words and translation words in the experimental condition and foreign words alone in the control condition. Both word pairs (experimental condition) and foreign words (control condition) were presented four times in succession to facilitate sleep-learning. In the experimental condition, translation words were played into the left ear and foreign words into the right ear. C: Retrieval tasks: Retrieval tasks were the same at the 12-hour and the 36-hour retrieval. Previously sleep-played and new foreign words were presented at test in both the visual and auditory modality simultaneously (a word appeared on screen and was simultaneously spoken). During each presentation of a foreign word, participants needed to answer three questions. First, they were asked to indicate whether they had a feeling of having heard (FoHH) the presented word during their sleep in the laboratory. Next, they were asked to assign the presented foreign word to a superordinate category (animal, tool, place; categorization task). Lastly, they were asked to rate their confidence on a four-point scale regarding their category assignment.

Memory performance.

A: Bar plots illustrate how well participants had assigned the foreign words presented at test to a superordinate category in the experimental condition; the data are averaged (SEM) over both retrieval time points. Categorization accuracy was above chance (chance = 1/3 correct assignments, red dotted line) in the Trough condition (blue) and exceeded accuracy in the Peak condition (yellow). B, C: Categorization accuracy split by the Peak/Trough condition and by the 12-hour/36-hour retrieval condition (x-axis). Displayed are average box plots with 95% confidence intervals, average density plots and dot plots of the participant average. The red dotted line indicates chance performance (chance = 1⁄3 correct assignments).

Word-related EEG potentials recorded during sleep.

A: Word pair-related voltage response plotted for the Trough condition (blue) and the Peak condition (yellow). B: Comparison of subsequently correctly (dark blue) and incorrectly (light blue) assigned foreign words to categories at the 36-hour retrieval in the Trough condition. C: Topographical voltage distributions of the averaged significant cluster for the contrast of correct versus incorrect category assignments at the 36-hour retrieval. D: Time course of the topographical voltage distribution for correctly and incorrectly assigned foreign words at the 36-hour retrieval (* indicates significant time points; cluster p< 0.05).

Trough features compared between correct and incorrect category assignments at the 36-hour retrieval.

We display the comparison between correctly (blue) and incorrectly (brown) assigned foreign words in the experimental condition at the 36-hour retrieval using the amplitude of the correlation between the ERP and the template (Prototypicality) at word onset (A), using global field power (GFP) of the ERP at word onset (B), using the inter-trial phase coherence (ITC) between the four presentations of a word pair (C), and using the time difference (Time Difference) between the actual acoustic stimulation and the measured trough maximum (D). Participant averages (dots), group averages (line) and 95% confidence intervals (boxes) are displayed as boxplots.

Theta and fast spindle power following word onset.

Time course of the topographical distribution of theta (A) and fast spindle (B) power, shown for the experimental condition (EC), where word pairs were played, and for the control condition (CC), where foreign words were played (left, average of representative electrodes). The blue box demarcates clusters that reflect a significant difference between EC and CC (*: cluster p<0.05). Topographical distribution of the theta and the fast spindle power averaged over the significant time period and electrode displayed for the experimental condition (EC, top) and the control condition (CC, bottom) in the Trough condition. Correlation of the theta enhancement and the fast spindle enhancement measured from the cluster of significant electrodes with the individual accuracy of category assignments in the experimental condition at the 36-hour retrieval (right).

Neural complexity.

EEG-derived neural complexity measured as Higuchi Fractal Dimension (HFD) displayed for the Trough condition and the Peak condition before the presentation of word pairs (Baseline) and following the presentation of word pairs (Evoked) in the experimental condition. The right panel of A displays the significant difference between the Trough and the Peak condition in the neural complexity gain from before to following word onset. (cluster-level Monte Carlo: p = 0.002, * indicates a significant comparison; cluster p< 0.05).

Accuracy of the targeting of slow-wave phase.

Data showing the precision of phase targeting for slow-wave Troughs (A, blue) and Peaks (B, yellow). On the left side in A and B, we present rose plots that indicate the phase of slow-wave filtered voltage (08.-1.2 Hz). The arrows demarcate the average phase of acoustic stimulation; 360° represents a peak and 180° represents for a trough. On the right side in A and B, we present the histogram illustrating the time delay between a real peak/trough and the acoustic stimulation. The highlighted line indicates the average of all trials. On the top right we explain the translation of slow-wave phases into angles. We targeted the acoustic stimulation ahead of the local maxima (peak or trough) to fit the entire word utterance into a peak/trough. The onset of word presentation was on average 82 ms or 66° before the trough maximum and 115 ms or 54° before the peak maximum.