Manfred's Column, July/August 2005: Learning during sleep: off-line reprocessing


July/August 2005

Learning during sleep: off-line reprocessing

Readers might have made the following observation: despite learning or practising something assiduously (e.g. juggling), you are still not able to perform it correctly. Disappointed by the results of your efforts, you forget about it and are surprised when you come back to it some time later that it works perfectly.

Evidently, after learning, further steps in the process take place, leading to better learning success. This reprocessing of learned subjects, which leads to their consolidation in memory, is termed consolidation. For decades this process has been linked to sleep, because sleep deprivation after learning impairs memory.

Assumptions based on animal experiments on the involvement of certain hormones in the mechanisms of consolidation during sleep led to my PhD thesis on the possible hormonal effects of different sleep phases 1, 4. Essentially, sleep phases are divided up into deep sleep, which takes predominantly place during the first half of the night, and REM sleep, which predominantly takes place in the early hours of the morning, during the second half of the night. Together with the group around Allan Hobson and Robert Stickgold, I investigated priming on awakenings from different sleep phases. At that time, we found increased indirect priming effects after awakenings from REM sleep [5] and were later able to replicate these results in for German-speaking people [6]. We interpreted these results as a more diffuse activation of the semantic networks during dreams as compared to the awake state. Further studies, as well as results from animal experiments, led Stickgold [8] to the following hypothesis: the nightly succession of different sleep phases makes sense in a neuroinformatic way. In deep sleep, data is stored and then reanalysed during dream sleep. For example, experiments on zebra finches and rats were able to show that the coupled ordered activity of two information storage structures (hippocampus and neocortex) leads to better interchange and improved storage, i.e. consolidation of new memory contents, during sleep [7].

Studies on humans further identified and clarified the role of sleep in learning (2, 9, 10). For the investigation of learning processes, Stickgold and colleagues used visual discrimination tasks. Such tasks consist, for example, of the recognition of briefly presented lines in the peripheral visual field (see Fig. 1).

It has been known for 10 years that training improves the performance of such tasks. This improvement is especially restricted to the area of the visual field in which training takes place (3). Thus, visual training is comparable to muscle training: if, for example, only the right bicep is trained, then only the right bicep becomes stronger and thicker. If the left upper quadrant of the visual field is trained, then the skill of being able to distinguish – the discrimination performance, – between briefly presented stimuli becomes better for this quadrant (and only for this quadrant!).

Interestingly, performance does not improve immediately after training, but a few hours after training. Therefore, the point in time for the maximal improvement lies between a couple of hours and days after training. It became even more exciting when researchers found out that sleep is necessary after training for the improvement of performance – it is not sufficient simply to have time pass after training. In fact a number of experiments pointed out that it is sleep after a training phase that leads to an improvement of performance. In one experiment, Stickgold and his colleagues were able to show that it is not sleep that occurs anytime after learning which matters, but, above all, the sleep in the night following the training. Different groups of subjects were either kept awake (sleep deprivation group) or were allowed to sleep. On various days following the training session, the extent of improvement in the test was assessed. In order to avoid additional training effects through the necessary test repetitions, the assessment of improvement was varied in the different groups. The results showed that the group with sleep deprivation following training was not able to improve performance even after two additional nights with full-night-long sleep. On the contrary, subjects, who had been allowed to sleep the night, which immediately followed the training, showed a significant and continuous improvement of performance. Thus learning or training respectively and sleeping any time afterwards is less effective than sleeping directly after learning. And one might add: Someone, who continuously jumbles his natural day and night rhythm with artificial light, shift work or keeping going all night, impairs not only his immune system, but also his memory.

If sleep represents a differential action of off-line reprocessing of information, which was received during the day, and if especially deep sleep and dream sleep have different functions, then differences of memory performance should be found depending of the sleep architecture. Since deep sleep occurs above all in the first half of the night and dream sleep during the second half of the night, it should make a difference if something is learned in the evening followed by sleep until 2 a.m., or if something is learned between 2. 30 a.m. and 3.30 a.m. followed by sleep from 4 a.m. until 7 a.m. Gais and his colleagues explored exactly this question (2). In their sleep laboratory, they used the simple visual discrimination task from Stickgold and chose a corresponding experimental design, with which they investigated 15 healthy subjects. They could show that the improvement after practicing the visual discrimination task occurred after early sleep and improved even more over a whole night's sleep, while it did not improve after late sleep alone.

My grandmother used to tell us children, when we did not want to go to bed, that the sleep before midnight is the best. I don’t know where she had this information from, but apparently she was right!



1.  Clarenbach P, Ortlieb R, Schopp D, Spitzer M. Does the nocturnal release of antidiuretic hormone correlate to polygraphic sleep events? Sleep 1983: 193-5.
2.  Gais S, Plihal W, Wagner U, Born J. Early sleep triggers memory for early visual discrimination skills. Nature Neuroscience 2000; 3: 1335-9.
3.  Karni A, Tanne D, Rubenstein BS, Askenasy JJ, Sagi D. Dependence on REM sleep of overnight improvement of a perceptual skill. Science. 1994 Jul 29;265(5172):679-82.
4.  Spitzer M. Nächtliche Vasopressin-Freisetzung bei selektivem REM-Schlaf-Entzug. Dissertation M (1984), Freiburg i. Br. 
5.  Spitzer M, Memelak A, Stickgold R, Williams J, Koutstaal W, Rittenhouse C, Maher BA, Hobson JA. Semantic priming in a lexical decision task on awakenings from REM-sleep: Evidence for a disinhibited semantic network. Sleep Research Abstracts 1991: 131.
6. Spitzer M, Walder S, Clarenbach P. Semantische Bahnung im REM-Schlaf. In: Meier-Ewert K, Rüther E (eds.): Schlafmedizin, S. 168-78. Stuttgart: Gustav Fischer Verlag 1993.
7.  Spitzer M. Nicht im Traum: Lernen im Schlaf. Nervenheilkunde 1999; 18:221-2.
8.  Stickgold R. Sleep: Off-line processing. Trends in Cognitive Sciences 1998; 2: 484-92.
9. Stickgold R, Whidbee D, Schirmer B, Patel V, Hobson JA. Visual discrimination task improvement: A multi-step process occurring during sleep. Journal of Cognitive Neuroscience 2000; 12 (2):246-54.

(translation of Spitzer, M. (2002), “Lernen im Schlaf Off-line Reprocessing von Gelerntem” in Schokolade im Gehirn: und weitere Geschichten aus der Nervenheilkunde, Schattauer GmbH, Stuttgart, pp. 58-62)