In recent years, the most widely discussed function of sleep and dreaming concerns the consolidation of declarative memory, including semantic, episodic, and autobiographical information; and procedural memory including perceptual and motor skills (Rasch & Born 2013). In particular the role of REM sleep in memory consolidation has been studied for several decades. While many studies from the 1970s have been criticized for being heavily confounded by too stressful REM sleep deprivation procedures (Horne & McGrath 1984), research in the 1990s raised interest in the role of REM sleep for memory consolidation: Karni (1994) demonstrated that a basic visual discrimination task improved after a normal night’s sleep, but not after selective REM sleep deprivation. Following this, a leading research aim in the field has been to identify which memory systems benefit from which sleep stages: it was demonstrated that early deep sleep benefits declarative memories, while late REM-rich sleep supports procedural skills (Plihal & Born 1997). Further support for the role of REM sleep in procedural memory consolidation came from studies showing that REM sleep intensity (total number of REMs and REM densities) increased following procedural-task acquisition (Smith et al. 2004) and improvements in procedural memory performance after a night of sleep were proportional to time spent in REM sleep (Fischer et al. 2002). Moreover, brain areas activated during a procedural learning task were more active during REM sleep in subjects who were trained at the task (Maquet et al. 2000; Peigneux et al. 2003).
More recent studies, however, speak against a prominent role of REM sleep in the consolidation of procedural motor skills or other forms of non-emotional memories, and instead emphasize non-REM sleep processes (Genzel et al. 2014). On the neurophysiological level, it has been suggested that dreaming represents the phenomenological reflection of a neural replay of activation patterns associated with recent learning experiences (Wilson & McNaughton 1994; Wamsley & Stickgold 2011; Wamsley 2014). Although memory reactivations have been observed in REM sleep as well (Louie & Wilson 2001), the most advanced models of sleep-related memory consolidation propose that neural replay is orchestrated by an interaction of non-REM sleep microprocesses, including slows oscillations and sleep spindles (Genzel et al. 2014).
Events and episodes from waking life are sometimes incorporated into dreams, either as classical day-residues the following night or after a “dream lag” of about 5–7 days (Nielsen & Powell 1989; Nielsen et al. 2004). Supporting the idea that such dream incorporations reflect processes of memory consolidation, items that were incorporated into dreams have been observed to lead to better memory retention (de Koninck et al. 1990; Cipolli et al. 2004). While an actual episodic replay of waking events was found in no more than 1–2% of the dream reports (Fosse et al. 2003), with NREM-sleep dreams appearing to include more identifiable episodic memory sources than REM-sleep dreams (Baylor & Cavallero 2001), it has been suggested that particularly engaging learning experiences have a more robust influence on dream content relative to more passive experiences (Wamsley 2014).
In contrast to recent episodes, incorporations of autobiographical memory features could be identified in the majority of dreams (Malinowski & Horton 2014). This suggests that dreaming might serve to assimilate recent memory fragments into autobiographical memory schemas and thus supports autobiographical self-model maintenance (Metzinger 2013). For semantic memories, evidence of a relationship between dreaming and neural memory reactivations stems from studies of declarative memory that present memory cues during sleep: these cues, when associated with the pre-sleep learning session, induce associated dream imagery (Schredl et al. 2014) and enhance post-sleep memory retrieval (Rasch et al. 2007). For procedural memories, learning of an engaging visuomotor task led to integration of task-related imagery into dream-like activity during non-REM sleep (Wamsley et al. 2010a), and such dream-incorporations of recent learning experiences were associated with later memory performance (Wamsley et al. 2010b). This memory-enhancing re-experience reminds us of motor imagery training during wakefulness, which has been repeatedly demonstrated to improve motor skills (Driskell et al. 1994; Schuster et al. 2011).
Recently it has been suggested that instead of consolidating memories, REM sleep serves as a state of elaborative (re-)encoding, during which the hippocampus integrates recent episodic memory fragments into remote episodic memories (Llewellyn 2013). It has been proposed that this process relies upon principles that also underlie the mnemonic encoding strategies of ancient orators, such as vivid, complex and often bizarre associative imagery, narratives with embodiment of oneself, and associations with known locations, later serving as retrieval cues. Subjectively, this process would be experienced as the typical dream mentation with its hyper-associative and bizarre imagery. However, despite being; intuitively appealing, several theoretical considerations and empirical findings are inconsistent with the idea of mnemonic encoding strategies acting during dreaming (Dresler & Konrad 2013).
To sum up, a first important function of sleep and dreaming is memory consolidation and integration, including the rehearsal of procedural motor skills, replay of episodic and semantic memories, and integration of memory episodes into autobiographical memory schemas.