is listening to a language while sleeping effective?

what the careful studies found

the hope is old, and so is the doubt. in the mid-1950s charles simon and william emmons criticised the methods of the studies that had reported sleep learning, and showed that when sleep was checked properly with brain recordings, recall and recognition of the material played during it did not occur.

a 1992 study by james wood, richard bootzin, john kihlstrom and daniel schacter looked for something subtler than recall. they played pairs of words to sleeping people during rem sleep or light sleep, then tested them straight away with tasks designed to pick up memory that people cannot report. a group who heard the same pairs awake showed the effect. the sleepers showed none. the few cases of recall they did find came only when the words were followed by the sleeper starting to wake.

that last detail matters. if you have ever woken up feeling that a recording got through to you, the likeliest explanation the research offers is that you were partly awake when it did.

what a sleeping brain can do

the newer findings are real, and each one is narrower than the idea it gets attached to.

in 2012 anat arzi and colleagues played tones to sleeping people, each tone paired with a pleasant or an unpleasant smell. later in the night, the tones alone made sleepers sniff more or less deeply, according to the smell each tone had come with, and the response carried on after they woke, though they had no awareness of having learned anything. in 2017 thomas andrillon and colleagues played repeated patterns of noise to sleepers. heard during rem or light sleep, the patterns were recognised better on waking. heard during deep sleep, they were recognised worse, as if that stage of sleep had pushed them out.

closest to language is a 2019 study by marc züst, katharina henke and colleagues in bern. during a daytime nap, sleepers in deep sleep heard pairs made of an invented word and a real german word. when the second word happened to fall on a particular peak of the brain's slow waves, a link between the pair was more likely to form. after waking, that link nudged people's choices on a test they did not know was about memory. the authors themselves describe earlier evidence for learning in sleep as inconclusive, and what they found was implicit: the test measured people's choices, not what they could recall.

the night as a second pass

a second line of research does not try to teach anything new at night. it tries to help with what was learned during the day.

in a study published in 2015, thomas schreiner and björn rasch had german speakers learn dutch words before they slept. while they slept, some of the dutch words were played again during non-rem sleep. the next morning, people remembered the german translations of the replayed words better than those of the words left unplayed. playing the same words to people who were awake did not help.

a follow-up from the same lab changed one thing. when each dutch word was followed straight away by more sound, whether the correct translation, a wrong one or a simple tone, the benefit disappeared entirely. the brain seemed to need a quiet moment after the cue to do its work.

this matters more than any other finding here, because it describes almost exactly what bilingual audio does. a line, then its translation, then the next line. in the lab, that pattern cancelled the benefit. the cues that worked were few, timed to a measured stage of sleep, and followed by silence.

why a night of audio is a poor bet

put the studies together and the conditions that help are narrow: words learned awake first, a handful of quiet cues, played during the right stage of sleep as measured by an electrode cap, and nothing straight after them. a playlist running through the night meets none of those conditions. some of it will land in deep sleep, where the andrillon study found sound pushed memories down rather than up. some of it may simply wake you a little, which is the one condition under which the older studies found anything.

what sleep reliably offers is different, and more useful. a long line of research summarised by susanne diekelmann and jan born describes sleep as a state that helps fix newly acquired information in memory, depending on how it was learned and when the sleep came, and it does that with no soundtrack at all.

the hour before bed

so the practical version reverses the original idea. give the language some of your evening, while you are awake enough to follow it, and then stop. a short session of listening you can understand, with meaning attached, is the part you can control. the night that follows is the part sleep handles on its own.

keep it short enough to finish before you are drowsy, and keep it to material you can follow, because sleep can only work on what you took in. then turn it off. silence is not a wasted opportunity. in the one study that tested sound after a cue, silence was the condition that worked.

made for the hours you are awake

driftspeak is not sleep audio. it is subtitles for your ears, made to be listened to, not slept through. words i kept is free in full in the app: thirteen short tracks of lines kept from records and films on the shelf, about nineteen minutes in all. the mariya takeuchi track opens on "また、昼と夜が逆なの。" ("my day and night are reversed again."), the kind of line that belongs to an evening.

while you are still awake, listen to a sample of words i kept, then switch it off, and see in the morning whether that line came through the night with you.