The Quality of the Silence

In the middle of October, on Rick Rubin's podcast, a man who has spent his life publishing secret histories offered a test anyone could run.

Mark Booth was a London editor for decades, and under the name Jonathan Black he writes the esoteric history of the world: the material universe as a small, recent island in an ocean of mind. He had come to talk about his new book, which argues that pioneering scientists, from Curie to Einstein, drew on mystical traditions to make sense of what they were seeing. For most of the hour the talk ran where that thesis runs: entanglement, the fine-structure constant, angels, Pauli's broken equipment. Then, near the end, he said something small.

When you ask someone a question on the telephone, and the answer matters to you, he said, listen to the quality of that silence. Do you still love me. What did the test results say. You will know the answer before it arrives. Rubin, who had already praised the book in print, agreed at once: that sounds right. Booth called it mind-reading, and said we do it all the time and have been trained not to notice.

Almost everything else said that hour lives where no instrument can follow. This claim does not. It is precise, it is testable, and it has been tested — on telephone calls, among other places — by people who had never heard of Booth and would not have accepted his explanation. What they found does not need the explanation. It is stranger without it.

Begin with the gap itself: the silence between one person stopping and the next person starting.

In 1974 three sociologists, Harvey Sacks, Emanuel Schegloff and Gail Jefferson, published a paper in Language proposing that ordinary conversation runs on something like machinery — a small set of rules, applied afresh at every possible handover, that lets any number of people talk while keeping both overlap and silence to a minimum. It was an odd thing to claim. Conversation feels like the least mechanical thing we do. But the paper became one of the most cited in the study of language, because the evidence was everywhere once you were told to look: one speaker at a time, handovers arriving at the points where a sentence, a clause, a phrase or even a single word could be complete, gaps short enough that nobody notices them.

How short, nobody measured carefully across cultures until 2009. Tanya Stivers and ten colleagues took video of natural conversation in ten languages — Danish, Dutch, English, Italian, Japanese, Korean, Lao; Tzeltal in Mexico; Yélî Dnye in Papua New Guinea; ‡Ākhoe Haiǁom in Namibia — and timed the gap between a yes-or-no question and its answer. Folk wisdom promised large differences — the slow Scandinavian, the impatient New Yorker. The recordings found one pattern. In every language the most common gap was between zero and two hundred milliseconds. The averages differed — Japanese speakers came in fastest, at seven milliseconds, Danes slowest, at under half a second — but every language's mean fell within roughly a quarter of a second of the cross-language mean. The authors' conclusion was careful and large: a shared infrastructure for human talk, with cultures tuning a single dial.

The precision has its sceptics, and they belong in the story. Mattias Heldner and Jens Edlund, measuring every change of speaker in three corpora rather than answers to questions alone, found turn-taking messier than the textbook picture: clean handovers with neither gap nor overlap were rare, and a large share of gaps were long enough for the listener simply to have heard the other stop and then reacted. Both things are true at once. The handover is not a metronome. But the most common gap, in language after language, is shorter than any reply can be built from scratch.

A fifth of a second is about the length of a blink. And here is the problem, which took the field a few more years to state plainly: it is not enough time to answer.

Psycholinguists have timed how long it takes a person to produce speech from a standing start. Naming a single pictured object — one practised word — takes about six hundred milliseconds before the speaker even starts to say it. Describing a simple scene, a boy kicking a ball, takes around a second and a half. If we waited for the other person to finish before beginning our reply, the gaps in every language on earth would be several times longer than they are. They are not. So we are not waiting. Stephen Levinson and Francisco Torreira laid out what that implies: the listener is doing two things at once. As soon as the gist of the question is clear — often halfway through it — the reply begins to be assembled, and it is held ready, like a breath, until the cues of an ending arrive: a falling pitch, a sentence closing. Then it is launched, aimed at a moment that has only just come. Not everyone accepts the full model. Some experiments find speakers putting off the hardest planning until near the end of the other's turn, and one recent paper argues that corpus gaps overstate the precision. The disagreement is about how early the work begins. Nobody argues that it waits for silence.

That is what happens, all day, in every kitchen and market and phone call on earth. Two nervous systems, locked to a shared clock with a tolerance of a fraction of a second, each modelling the other well enough to finish its thoughts.

The manifesto this series extends opens on that fact, though when the line was written it was not reaching for a laboratory:

Humanity was born with a superpower.

The line after it names the power: not flight, not fire — listening, and answering. It is easy to read that as poetry. The turn-taking literature reads it as a timing measurement.

It is also old — older than the words carried in it.

Mothers and infants take turns long before infants have anything to say. Mary Catherine Bateson, studying film of a mother and her two-month-old at MIT around 1970, named what she saw proto-conversation: the two of them trading sounds in alternation, each leaving room for the other, months before the first word. The exchange is slow by adult standards — a 2022 meta-analysis pooling twenty-six studies put infants' replies at about a second — and how the timing changes over the first years is still argued; one longitudinal study found babies quick at three to five months, slower at nine, and picking up speed again by eighteen. What nobody disputes is the order of events. The alternation comes first. The words are poured into it later.

And it is not only ours. Common marmosets — small South American monkeys, on a branch of the primate tree that split from ours some forty million years ago — take vocal turns, rarely calling over one another, on a slower clock of several seconds; a team at Princeton showed in 2013 that the timing behaves like two coupled oscillators settling into a shared rhythm, and infant marmosets learn it from their parents' feedback, much as ours do. The finding is contested: one linguist argued it shows call-and-response rather than turn-taking in any rich sense, and the marmoset is a case of convergence, not ancestry. Alternation of the same kind has been reported in songbirds, in gibbons' duets, in primates on every major branch — though a 2018 review of the animal evidence called its empirical basis surprisingly weak. Levinson has argued that what makes human language possible is not grammar first but an older interaction engine: timing, contingent response, the recognition of intentions — machinery that language came along to use. That is a hypothesis, argued with care, and it is contested not at the edges but at the centre. Whether the machinery was inherited or evolved more than once is unknown, and the leading rival account, Michael Tomasello's, places language inside a broader and uniquely human adaptation for cooperation instead. But on the narrow fact that matters here the evidence points one way: the timing of the gap is not a late cultural refinement. It looks like something we were born with.

Now the silence that Booth was talking about.

Not every answer is equal, and conversation knows it. Conversation analysts call some replies preferred — the yes to an invitation, the acceptance of an offer — and others dispreferred: the no, the refusal, the turned-down request. The terms are not about what anyone wants. They describe the shape the reply takes, and which shape counts as which depends on the action: answer someone who has just run themselves down, and agreeing with them is the dispreferred move. Anita Pomerantz showed it in 1984 for the plainest case, agreeing and disagreeing with an opinion. Preferred replies arrive quickly and plainly. Dispreferred ones come late, and they come wrapped — an uh, a well, a token of agreement before the disagreement, the edge taken off. The wider literature added the rest of the wrapping: the account of why, the apology, the softened no. The timing tendency has turned up in every language compared so far, from Danish to Tzeltal to Yélî Dnye. It is, among other things, a kindness — the bad news is slowed so the hearer can see it coming.

Which means the hearer can see it coming. In 2006 Felicia Roberts and two colleagues played listeners staged telephone calls between friends, built from a real one, with the silence before each reply edited to nothing, to six hundred milliseconds, or to twelve hundred. With every step of added silence, listeners heard less willingness behind a sure and weaker agreement behind a yeah. A second study, with American, Italian and Japanese listeners hearing the same reply tokens in their own languages, found the same direction in all three — the longer the silence, the weaker the yes — though the groups started from different places and fell at different rates, which the authors read, cautiously, as cultures expecting agreement to come at different speeds.

The brain is quicker than the rating. In 2015 Sara Bögels, Kobin Kendrick and Levinson recorded the brain activity of people overhearing real questions lifted from Dutch telephone calls — proposals, requests, offers, invitations — with the reply arriving after either three hundred milliseconds or a full second. A quick no set off a measurable signal of surprise that a quick yes did not. After a second of silence, the surprise was gone. Nothing about the word had changed; only the wait before it. Somewhere inside that second, the listeners had already started to expect the refusal.

So Booth is right about the experience. You do hear the answer in the silence. Listeners in three languages hear it in the ratings, and in the laboratory it registers in the brain.

Here is where the evidence turns, and the turn is the reason to tell the story at all.

In 2015 Kendrick and Torreira went to the recordings to measure the thing directly. If silence carries the answer, the gaps before refusals should be reliably longer than the gaps before acceptances. They took a hundred and ninety-five replies to requests, offers, invitations and proposals from classic collections of real telephone calls — the very instrument Booth chose for his test — and timed them.

For most of the range, the timing did not tell the two apart. The most common moment for a yes and for a no was the same, about a quarter of a second after the question ended, and most replies of both kinds arrived within seven hundred milliseconds. Only beyond that did refusals clearly take a larger share. Where the news was carried was elsewhere. A reply that began with well turned out to be a refusal, or a grudging, hedged yes, about four times in five; one that began with um, or an audible in-breath, leaned the same way; one that began clean, about one time in five. And when the authors separated plain answers from hedged ones — the yes that comes with a maybe, a condition, a flattened tone — the timing sharpened: past about three hundred milliseconds, the plain, unqualified yes became steadily less likely. Their conclusion was that the timing of a reply is best understood not as a gap before the answer but as part of the answer — in their words, the first virtual component of a turn that is already being built.

The silence, in other words, is not silent. What we hear in it is a mind already answering — the intake of breath, the fractional delay of someone assembling well, the throat beginning to soften the no before it is spoken. We read this in a fraction of a second, from sounds too small to name, and we arrive at the meaning before the words do. Booth called it mind-reading. It is reading. The text is the other person's body, beginning to speak.

That is a smaller claim than telepathy, and a far stranger fact. Nothing has to travel between two minds by unknown means. Two people on a telephone line are simply so finely tuned to each other that a few hundred milliseconds of hesitation, and the faint sound of an in-breath, is enough to deliver the news. Booth's worldview needs the universe to be more connected than it looks. The measurement says something else that is just as good: people are more connected than they look.

Which raises the question of what happens when the clock is broken.

The speed of the gap is not only information; it is felt. In 2022 Emma Templeton, Thalia Wheatley and their colleagues at Dartmouth recorded strangers, and then friends, in ten-minute conversations and found that the faster two people answered each other, the more connected they felt. Then they played short clips of the strangers' talk to outside listeners, some with the gaps artificially shortened, and the listeners judged those speakers more connected to each other. The difference that told them lived in fractions of a second.

The machinery assumes that both speakers share a moment — that a silence of six hundred milliseconds means six hundred milliseconds to both of them. Put a network between them and that stops being true. In 2014 Katrin Schoenenberg and her colleagues at Deutsche Telekom's laboratories at TU Berlin ran telephone conversations into which they inserted delay, up to a little over a second. People's ratings of the line's quality did not budge. Their ratings of each other did. With the delay at its longest, partners were judged less attentive, less friendly and less self-disciplined than partners heard on a clean line. The machinery that reads hesitation as a soft refusal was reading a cable's latency as a cooling heart — and blaming the person, not the wire. Schoenenberg's own explanation, writing afterwards for a general audience, was that the awkwardness bites hardest between people unfamiliar with each other, who have no sense of how the other normally sounds.

Video calls carry the same fault in a quieter form. In 2022 Julie Boland and her undergraduate students at Michigan had pairs of friends talk face to face and then over Zoom. Across the table, the average handover took about a seventh of a second. On the call it took nearly half a second — more than three times as long — and the speakers held the floor longer, handing it over less often. The strangest part is the arithmetic. The line itself added only a few tens of milliseconds. The authors' explanation is that the line's small and shifting delay unsettles the shared rhythm the handover depends on, and the rest of the wait is the two people, unable to find the beat, waiting for each other. Anyone who has spent a working year on calls knows the texture of it without the measurement: the double starts, the no, you go, the faint fatigue of talking to someone whose silences you cannot read.

None of this is a scandal. Lag is physics, and the people who build networks fight it hard. But it is worth knowing what a few hundred milliseconds of delay costs, because the cost lands in exactly the faculty Booth was describing. We have never been able to reach so many people, so quickly, so far away. And a growing share of the talk we do with them now passes through a clock that slightly mistunes the oldest instrument we have — the one that hears, in a fraction of a second, whether we are loved.

Booth's book makes claims about physics that physicists will contest. But in the middle of an hour about entanglement he handed his listeners a practice that needs none of it, and the practice is sound. When the answer matters, listen to the silence before it. Not because something crosses the line between two minds, but because the other person has already begun to answer, in a language older than words, and the most attentive thing you can do is hear them start.

It asks almost nothing. Slow down enough to notice six hundred milliseconds. Pay attention to the breath before the well. Give the person you are asking the shared clock the machinery was built for — the same room, if you can; a clean line, if you cannot.

The silence before a yes is very short. You already know that. You were keeping time with another voice before you had a word to put in it.