In a scene familiar to all, a fractious child is sent to bed in the belief that the sleep that will surely follow will allow the child to get some much needed `rest'. But the idea that sleep is simply a rest period has been repeatedly challenged over the years. Initially, the challenge was provided by the electrophysiological descriptions of the sleep cycle. More recent findings have linked sleep and the hippocampus with the consolidation of neuronal temporal codes for spatial memory. However, we still know little about sleep. Two recent papers in Science provide thought-provoking additions to this literature.

In the first paper, Robert Stickgold and colleagues focused on the effect of extended mental activity on hypnagogic mentation — the replay of mental imagery at the onset of sleep. The experiment was designed to evaluate the source and the function of hypnagogic images. The participants played the computer game Tetris for extended periods over three days and were asked to report mentation during the first hour of attempted sleep each evening. Three groups participated in the experiment: novices, experts and amnesic patients with bilateral medial temporal lobe damage. All subjects reported intrusive stereotypical images of the blocks used in the game at sleep onset. This was true even for the amnesic patients who could not recall playing the game. One interpretation of this result is that hypnagogic images do not require intact declarative memory systems. Another intriguing aspect of this study is that although playing Tetris might be expected to involve procedural learning (which is intact in amnesic subjects), the amnesic patients in this study did not show the same level of improvement as the novice players. Further work will be required to clarify this area, but it does seem that the source of hypnagogic images and the improved performance may rely on different areas of the brain.

The second study provides evidence for the role of sleep in birdsong learning. Dave and Margoliash recorded the activity of single neurons in the forebrain nucleus robustus archistriatalis (RA) of zebra finches. This area is analogous to the motor cortex in mammals and its descending projections represent the output of the forebrain song system. Recordings from RA in individual birds were made while each bird was singing and again during playback of the bird's song while it was asleep. The timing and structure of the neural activity elicited by the playback of song during sleep matched the activity recorded during singing. Activity during singing leads the sounds that are produced, thus the matching auditory responses represent a prediction of the following sound based on the sequence of the preceding sounds. Such predictions may lie at the heart of the computations involved in vocal learning. Moreover, the spontaneous activity of the RA neurons during sleep matched their sensorimotor activity, and thereby may provide a form of song replay for motor learning.

In conclusion, these two studies indicate that the replay of sensory experience during sleep may have complex effects on different forms of learning and memory. So although it may be some time before the secrets of the Sandman are finally revealed, these two studies demonstrate that we are now able to tackle this problem with a wide range of techniques.