Posted by Somewhere off the west coast of Dominica a sperm whale is producing one of the loudest sounds any animal makes, and most of the time it is using that sound for something entirely unromantic: locating squid in water where light has stopped being useful. Sperm whale clicks have been measured underwater at levels around 230 decibels. A fraction of that output, though, is not aimed at squid at all. It is aimed at other whales, in short rhythmic bursts, and that is the point where whale communication stops being a question about acoustics and turns into a question about minds.
Nobody in the field disputes that whales exchange information. The arguments, and there are plenty, are about what kind of information, how much of it, and whether any of the vocabulary we use for human language should be allowed within a hundred metres of the subject.
The loudest nose in the ocean
Roughly a third of a sperm whale is nose. Inside that enormous forehead sits the spermaceti organ, a case of waxy oil that gave the species its name and nearly finished it off during the whaling centuries. The sound starts at the phonic lips just under the blowhole: air driven through them makes a sharp pulse that travels backwards through the oil, reflects off an air sac braced against the skull, then comes forward again and leaves the head as a beam. Bertel Mohl and his colleagues described this arrangement as a bent horn, and the mechanics have a useful side effect: a sperm whale click is never a single pulse but a train of them, and the gap between pulses depends on how long the organ is. From a recording alone you can estimate the body length of a whale you have never seen. It is one of the very few easy things about this animal.
The clicks come in flavours. Slow, steady usual clicks while an animal descends and searches. A rapid buzzing creak in the last seconds before a squid is taken. Long, ponderous slow clicks from big males, audible for many kilometres. And then codas, which are the ones that keep people awake.
Codas, clans, and a large bet on machine learning
A coda is a short patterned burst, usually three to a dozen clicks packed into under two seconds. Researchers name them by rhythm rather than by meaning, so the catalogue reads like a drummer’s shorthand: five regular clicks, or a pattern written 1+1+3. Codas turn up when whales are socialising at the surface, not when they are hunting, and they get traded back and forth in overlapping exchanges that look, to a human eye, uncomfortably like conversation.
Shane Gero has been following the same sperm whale families off Dominica since 2005, long enough to know individuals by name and to watch calves grow up. That kind of longitudinal work produced one of the strangest findings in cetacean biology: vocal clans. Whales that share a coda repertoire associate with each other and largely ignore whales that use a different one, even when both groups are working the same stretch of ocean. The repertoire is learned, not inherited. Calves babble before they get it right.
Project CETI, the Cetacean Translation Initiative, launched in 2020 to attack that system with the tools of modern machine learning. David Gruber leads it, with Daniela Rus from MIT’s robotics lab and Michael Bronstein on the deep learning side, and Gero supplying the whales and the decades of context. The plan runs on moored hydrophone arrays, suction-cup tags, aerial and underwater robots, and a target of collecting clicks by the billion rather than the thousand.
Their 2024 paper in Nature Communications is the result people cite. Working through roughly nine thousand codas from the eastern Caribbean clan, the team argued that variation in rhythm and tempo combines with two features they called rubato, a smooth stretching of tempo across an exchange, and ornamentation, an extra click tacked onto the end. Combine those dimensions and the repertoire is far larger than the couple of dozen coda types the field had used.
Which is genuinely interesting and also much narrower than the headlines suggested. A wider channel is not a message. Showing that a system can carry more distinctions than anyone assumed says nothing yet about whether those distinctions are used to mean anything.
Ten hertz and a very long road
At the other end of the spectrum sit the blue whales, and everything about their signal is inverted. Where sperm whales produce brief, brutally loud broadband clicks, blue whales produce moans near 10 to 20 hertz, at or below the floor of human hearing, lasting twenty seconds or so and repeated in patterns for hours. In the northeast Pacific the two main call types are labelled, with no poetry at all, A and B. On a boat you do not hear them. You feel them, if you feel anything.
Low frequency is the whole trick. Sound at ten hertz loses very little energy to absorption, and the ocean happens to provide a waveguide for it. Maurice Ewing and Lamar Worzel worked out in the 1940s that sound speed reaches a minimum around a kilometre down, so energy entering that layer is refracted back into it instead of escaping. The deep sound channel, or SOFAR channel, was a military discovery before it was a biological one.
In 1971 Roger Payne and Douglas Webb ran the numbers for fin whales and concluded that in a pre-industrial ocean, calls could in principle have carried across entire basins. That calculation has been quoted ever since, usually without its conditions. The ocean is louder now, low frequency shipping noise having climbed steeply through the second half of the twentieth century, and a whale that has to surface to breathe does not spend its life parked on the sound channel axis. Measured detection ranges in real conditions run to tens or a few hundred kilometres. Remarkable, and not the same claim.
There is also a puzzle nobody has solved. Since the 1960s the pitch of blue whale song has been drifting downward across every population that has been monitored, by a few percent. Recovering numbers, changing ocean chemistry, shifting body sizes and social display have all been proposed. Nobody knows.
The word nobody wants to use
So is any of this language? Linguists have a checklist for that, most of it descended from Charles Hockett’s design features, and the demanding items are displacement, the ability to refer to things absent in space or time, and duality of patterning, where meaningless units combine into meaningful ones which then combine again.
Whales clear some bars easily. Vocal learning is established. Dialects are established. Individually distinctive signals, context-linked calls, cultural transmission across generations, all established. What has not been demonstrated in any cetacean is a single acoustic unit with a specific, testable, agreed meaning. That is the gap, and it is not a small one.
Statistical structure by itself will not close it. Sequences produced by many non-linguistic processes fall into the same skewed distributions that human words do, so finding Zipf-like statistics in a coda catalogue is a weak argument. The deeper problem is that decipherment has always needed a crib. Champollion had Greek on the same stone. There is no whale Greek, no parallel text, no bilingual informant, and the behaviour we would need to align the signals against mostly happens several hundred metres below anyone’s ability to watch it.
The obvious experiment is playback: broadcast a coda, record what happens. It is also the one most likely to go wrong, scientifically and ethically, because you are shouting words of unknown meaning at an endangered animal in its own social space.
Why the SETI people keep turning up
The overlap with the search for extraterrestrial intelligence is not decorative. Both fields face the same ugly first step, which is deciding that a pattern is a signal at all, without knowing anything about who made it.
SETI’s history is mostly a history of that step failing. The Wow! signal ran for 72 seconds through the Big Ear telescope in August 1977, looked exactly like what people were hoping for, and was never seen again. BLC1, picked out of Breakthrough Listen data from Parkes near 982 megahertz and pointing in the direction of Proxima Centauri, held up for months until the team traced it back to human equipment in 2021. Green Bank and the Allen Telescope Array keep grinding through candidates that dissolve the same way.
Set that against the whale problem and the asymmetry is sobering. A coda arrives with a body attached. Researchers know the depth, the group, who was present, what the animals did in the next ten minutes. A radio detection arrives with a frequency, a sky position and nothing else, no biology, no chance to run the source again tomorrow. If we cannot yet read whale communication with every contextual advantage this planet offers, the confidence some people bring to decoding an interstellar transmission deserves a second look.
That is not an argument for abandoning either search, only for precision about what has been shown. Whale communication is real, structured and socially learned, and studying it is teaching us to measure a signal system before telling a story about it.
SETIworld follows both ends of that problem, the ocean and the sky, because they sit closer together than they look. Read on through our work on cetacean intelligence and signal detection, and bring your objections with you. The best conversations here tend to start with someone disagreeing.
