|Clip art: Clker|
2011 research by van der Helm, Yao, Dutt, Rao, Saletin, and Walker of University of California-Berkeley, summarized by Science Daily, which studied the function of dreams in diffusing the emotional loading of traumatic events, suggests something of why that may be the case. One of the findings was that "normal" REM sleep greatly facilitates that integration and diffusion. (In the case of certain PTSD victims, a blood pressure medication actually restored some REM enabling function.) Without it, things go very differently.
Not that learners be placed in dream states to change pronunciation (although I have tried a bit of that), but that integrating pronunciation effectively requires being able to create optimal REM-like, stress-resolving attention. That is the (achievable) dream of HICP work as well.