A sleeper hears a maths problem, works out the answer inside a dream, and moves their eyes in a rehearsed pattern. Electrodes beside the eyes record the reply while brain and muscle signals still meet the standard criteria for rapid-eye-movement sleep.
That is the core of a remarkable 2021 proof-of-concept study conducted by four independent teams. It showed that a small number of lucid dreamers could receive new information and send a limited answer back without waking for the exchange. The finding turned the usual retrospective study of dreams into something closer to a live experiment.
It was also much more constrained than the word “conversation” might imply. The exchanges were brief, answers used codes learned before sleep, and most questions received no detectable response. The significance lies not in fluent dream chat, but in showing that a two-way channel can sometimes remain open while a person is verifiably in REM sleep.
Sleep and lucidity had to be demonstrated separately
The Current Biology study brought together work from Northwestern University in the United States, Osnabrück University in Germany, Sorbonne University in France, and Radboud University Medical Center in the Netherlands. The researchers studied 36 people. Some regularly experienced lucid dreams, some had little previous experience, and one French participant had narcolepsy and frequent lucid dreams.
A lucid dream is one in which the dreamer recognises that the ongoing experience is a dream. It is not the same as being awake, and it does not necessarily give the dreamer complete control. A meta-analysis of 50 years of survey research estimated that about 55 percent of people have experienced at least one lucid dream, but only about 23 percent report one at least monthly. Producing one on demand in a laboratory is considerably harder.
To establish sleep stage, the teams used polysomnography: electroencephalography to measure brain activity, electro-oculography to record eye movements, and electromyography to monitor muscle activity. Participants who realised they were dreaming sent a rehearsed sequence of large left-right eye movements. That lucidity signal and the physiological classification served different purposes. One reported the dreamer’s awareness; the other documented REM sleep.
Four laboratories built the channel in different ways
The protocols differed enough to make the convergence more persuasive. In the United States and the Netherlands, experimenters spoke simple arithmetic questions. The German team encoded arithmetic through flashes of coloured light or patterns of sound, sometimes using Morse code. The French team asked spoken yes-or-no questions and tested whether a participant could distinguish words, lights, sounds, or a series of taps.
Replies had to pass through movements that were deliberate but still measurable during REM sleep. Most participants answered numbers by making the corresponding count of left-right eye sweeps. The French participant used two contractions of either the zygomatic muscle, associated with smiling, or the corrugator muscle, associated with frowning, to mean yes or no.
Crucially, the correct answer was not known to the sleeper in advance. A participant might hear “8 minus 6” only after signalling lucidity, calculate two, and then make two eye signals. That sequence required perception of a new stimulus, short-term retention, a simple computation, recall of the response rule, and a voluntary reply.
The result was convincing but uncommon
The numbers set the boundary of the claim. The teams attempted two-way communication during REM sleep in 57 sessions. Fifteen sessions, or 26 percent, produced a signal-verified lucid dream. In 47 percent of those lucid-dreaming episodes, the researchers obtained at least one correct answer.
They asked 158 questions during the signal-verified lucid dreams. The paper’s full trial table records 29 correct responses, five incorrect responses, 28 ambiguous responses, and 96 occasions with no response. Correct answers therefore accounted for 18.4 percent of all attempts, while 60.8 percent drew no detectable reply. The 29 correct answers came from six of the 36 participants.
Correct answers nevertheless greatly outnumbered incorrect ones. In control attempts made during REM periods without a lucidity signal or a later report of lucidity, only one correct response appeared across 379 trials. The pattern is difficult to explain as random eye movements that experimenters merely interpreted hopefully.
“Without waking” means during the exchange
The strongest evidence was the physiological record at the time each question and answer occurred. All 158 trials were scored as REM sleep by at least two of three expert sleep scorers. For 26 of the 29 correct answers, all three experts agreed that the recording showed REM. Separate blinded raters decoded the eye or facial signals, and a response was called ambiguous when the required consensus was missing.
The participants did not wake up to hear a question or deliver the coded answer. They were often awakened after successful communication so the researchers could immediately collect a dream report. The careful formulation is therefore that the exchange occurred without waking, not that the person was never awakened at any later point.
Those reports revealed how porous and unreliable dream memory can be. Some sleepers accurately remembered the incoming question. Some did not remember it, and others recalled a changed question or answer. External events sometimes entered the dream story: speech seemed to come from a car radio or from a narrator at a party, while flashing lights became an object glowing within the scene. The live signals were valuable precisely because a later recollection could diverge from what the instruments recorded.
This was not unrestricted access to a dream
The study’s “real-time dialogue” was a prepared, low-bandwidth exchange. The questions had objectively checkable answers and the response vocabulary was agreed before sleep. Researchers did not hear spontaneous dream speech, reconstruct the dream’s imagery, or gain access to private thoughts. They could ask a small question and sometimes receive a small coded answer.
The participants were also not representative of all sleepers. Several were selected for dream recall or lucid-dreaming experience, and the especially responsive French participant had narcolepsy. Most experimental prompts failed to produce a clear answer. The work established that communication is possible under particular conditions, not that anyone can routinely hold a conversation in sleep.
REM classification is evidence under accepted sleep-scoring rules, not proof that every part of the brain occupies a perfectly uniform state. The authors noted that REM physiology shares features with both waking and light N1 sleep, and that local mixtures of state remain scientifically possible. Sensory stimulation can also trigger brief arousal or waking, which is why the physiological review around each exchange mattered.
A new instrument for studying a private world
ScienceBlog’s original report on the interactive-dreaming experiments focused on the methodological breakthrough: researchers no longer had to wait until morning to ask everything. A live prompt can test what a dreamer perceives or computes at a known moment, reducing the distortions introduced by awakening and memory.
Later research suggests the opening is not limited to expert lucid dreamers. In 2023, a Nature Neuroscience study found brief periods in which sleepers across several sleep stages could respond to spoken words with prearranged facial movements. The finding did not turn sleep into continuous wake-like awareness; it suggested that connection to the environment fluctuates.
The method has also moved beyond arithmetic. A 2026 real-time reporting study asked frequent lucid dreamers to signal dream-eye position and the presence or absence of visual content using distinctive sniff patterns. Researchers recorded 150 signals over 19 sessions, showing how voluntary markers can timestamp subjective events for comparison with brain activity.
That is the lasting importance of the original conversation experiment. It did not abolish the border between waking and sleep. It found a narrow, intermittent gate through it, allowing a question from the laboratory to enter an ongoing dream and a deliberate answer to return.