The First Real Translation of Whale Language Is Here — And It’s Bad News

追加: 2026-05-13

[00:00:02]Are whales talking to each other?

[00:00:04]Definitely. Scientists just ran the first real translation of whale language, and what came back wasn't supposed to be possible. After feeding 9,000 recordings into the most advanced AI on Earth, researchers didn't find simple animal signals. They found vowels, phonemes, a combinatorial grammar, >> [music] >> the same architectural blueprint underlying every human language ever spoken. And buried inside that discovery is something nobody was prepared for.

[00:00:32]It's possible that they're communicating in ways that we've never thought of before. A finding so unsettling that it reframes everything humans have been doing to the ocean for the last 100 years. That's the bad news, and it's worse than you think.

[00:00:47]The assumption that collapsed. For decades, the scientific consensus on sperm whale communication was considered settled. Researchers believed sperm whales used roughly 21 distinct click patterns called codas, each one carrying a fixed, simple meaning. A warning, a direction, a contact call. Scientists treated whale communication like a traffic light, a handful of states, a fixed set of meanings, and nothing approaching real complexity. That assumption wasn't just wrong, it was one of the most spectacular miscalculations in the history of modern biology. Goldwasser had spent her [music] career cracking actual encrypted communication systems, the kind governments use to hide state secrets.

[00:01:31]When she stopped outside Gruber's door at Harvard's Radcliffe Institute and heard those clicks, >> [music] >> her instinct wasn't casual curiosity. It was professional pattern recognition. To an expert in hidden structure, those whale clicks were not noise. They were a code that had never been run through a decoder. The four words she said to Gruber were, "Have you tried machine learning?" He hadn't. Nobody had. That conversation led to a third researcher, AI specialist Michael Bronstein, who explained that the same deep learning algorithms powering large language models like ChatGPT could in theory be applied to animal communication, as long as researchers could gather enough training data. The question was whether anyone had the will, the resources, and the time to try. What followed would take 5 years, millions of dollars, and a level of scientific ambition the field had never seen. And it would end with a finding that nobody in that room was fully prepared for when it came back.

[00:02:31]Building the machine. In 2020, that question became Project CETI, the Cetacean Translation Initiative, funded by the TED Audacious Project with partnerships from MIT, Harvard, UC Berkeley, and 15 other institutions. The team built a permanent research station in Dominica, where multi-generational sperm whale families have been observed for decades. Individual animals are known by name. Social hierarchies have been mapped through years of patient surface observation. The island sits in a stretch of Caribbean water that sperm whale clans return to reliably, year after year. The team then went further than any previous whale study in history. Hydrophone arrays were permanently installed along migration routes. Bio-logger tags were attached directly to whale skin via suction cup, each carrying three synchronized microphones capable of distinguishing individual voices inside a group conversation. GPS logging, depth sensors, accelerometers tracking body movement to within fractions of a degree. The system was engineered to capture not just what the whales said, but who they were saying it to, what they were physically doing at that exact moment, and what was happening in the acoustic environment around them. Every variable that could give meaning to sound was being logged simultaneously.

[00:03:53]Not just to hear the language, to build the contextual record that might eventually allow it to be understood.

[00:03:59]9,000 recordings went into the AI. The same class of deep learning that powers the most advanced language models on Earth was turned on a species that had been clicking in the dark for 30 million years. If you've made it this far in, you're not subscribed, do it now.

[00:04:16]Because what the machine returned is the thing this entire video has been building toward, and you'll want to be here when it surfaces.

[00:04:23]What the AI found. The AI didn't find 21 codas, it found 156, but the number was almost beside the point. What mattered was what was inside them. Each coda has internal structure, tempo variations, rhythm shifts, ornamentation clicks layered in recurring patterns. Those elements aren't decorative, they're functional. They work the way phonemes work in human language. Modular units combined according to rules to generate meaning that no single unit carries alone. The whales were not transmitting 156 preset messages like a fixed menu.

[00:04:59]They were constructing communication dynamically from a modular phonetic system. [music] The same fundamental architecture that underlies every human language ever documented on Earth. One analogy makes this concrete. English builds millions of words from 44 basic sounds. [music] Remove one of those sounds and thousands of words vanish. Add a rule about how two sounds combine and new words become possible. Sperm whales appear to operate on exactly the same principle. A finite acoustic inventory producing a generative system with effectively unlimited expressive [music] range. The building blocks are clicks.

[00:05:37]The grammar is real. Dr. Daniela Rus from MIT described the findings plainly.

[00:05:44]The results show far more complexity here than anyone previously believed.

[00:05:48]Let me talk about the extra click because I still remember the meeting when we talked about that and we were all so excited and started wondering, >> [music] >> is it like a full stop at the end of a sentence? Challenging the current state of the art on what we thought was possible in the animal world. That is the most understated sentence in modern science because nothing about what the AI found was a small adjustment to existing knowledge. It was a structural demolition of the assumption that nothing in the animal kingdom had independently evolved anything resembling the architecture of human language. Except, something had. And it had been doing it in the dark a mile underwater for 30 million years. But that wasn't the finding that stopped the room. For that, a linguist at UC Berkeley had to sit down with the recordings. And what he was about to see would shake the foundations of his entire field.

[00:06:42]The finding that stopped the room.

[00:06:44]Gaspar Bakus studies the acoustic architecture of human language. He knows the phonetic building blocks of speech the way a musician knows intervals. Not as abstraction, but as physical fact, as shapes of frequency he has spent years learning to read. He has mapped the vowel systems of dozens of human languages. He knows what a vowel looks like when it appears on a spectrogram.

[00:07:06]He knows the precise form and frequencies that distinguish an A from an E, an E from a diphthong. When Project CETI's recordings landed on his desk at UC Berkeley, he was not expecting a discovery. He was expecting to map coda sequences, run his analysis, and confirm what the AI had already shown: structure, complexity, combinatorial grammar. Nothing that would require him to rewrite his assumptions about what language was. He pulled up the acoustic spectrogram. He ran his standard phonetic analysis. And then, he saw it.

[00:07:40]The whales were producing vowel sounds, specifically the A vowel, the A in father, the E vowel, the E in C, and diphthongs, those gliding double vowel combinations like the OI in boy. Stop.

[00:07:54]Read that again. The defining acoustic features of human spoken language had independently evolved in a creature living in complete darkness under thousands of pounds of water pressure.

[00:08:05]Not similar sounds, not analogous sounds. Vowels, precise controlled modulations of frequency that require specialized vocal anatomy, specific musculature, and the neural capacity to produce them with intention. Backus described the finding directly, "In the past, researchers thought of whale communication as a kind of Morse code, but this paper shows their calls are more like very, very slow vowels, suggesting a complexity that approaches human language." Azzizz complexity that approaches human language from a species with no shared ancestor with humans for hundreds of millions of years, developing its communication system in a completely different sensory world, arriving at vowels anyway. That is convergent evolution. Two entirely separate lineages, separated by hundreds of millions of years of independent evolutionary history, arriving at the same complex solution through entirely different routes. The human vocal tract evolved toward vowels through a descended larynx, specific tongue musculature, and precise control of air flow through a resonance chamber shaped for speech.

[00:09:15]Sperm whales arrived at them through echolocation.

[00:09:19]In an acoustic environment where high frequency precision determines whether you eat or starve, we evolved toward vowels because it helped us survive in open air. They evolved toward vowels because it helped them survive in open ocean. Two completely different worlds, two completely different bodies, the same architecture. And when convergent evolution produces something as specific as vowels, the only scientific conclusion that holds is this.

[00:09:46]Vowel-based communication is not an accident of human anatomy. It is a feature of intelligence itself. We thought language was ours, not borrowed from nature, not shared with cousins on the evolutionary tree, ours alone, the crown achievement of our specific lineage. The sperm whale vowel data does not challenge that belief. It removes the ground it was standing on. And somewhere in the Caribbean, in water too deep for sunlight to reach, a 70-year-old whale who was born before sonar existed has been using that system every day of her life. We are only now learning it exists.

[00:10:22]A conversation a mile down. She doesn't have a name in this script, but she is real, the matriarch of one of the Dominica whale families that Project CETI has been observing for years. Her lineage documented, her family relationships mapped, her voice logged.

[00:10:38]Across decades of recordings, she is the oldest living member of her group. She has watched calves born and raised and grown into [music] adults who now have calves of their own. She communicates every day. Shane Gero has spent 13 years alongside sperm whale families in Dominica. He describes watching these interactions unfold from the surface. It is hard not to see cousins playing while chatting, not to see a mother hand her calf to a babysitter, and exchange what looks for all the world like a few parting words before diving deep to hunt.

[00:11:10]>> [music] >> The same individual whale produces different codas depending on who it is addressing, what the other whale just said, and what is happening in the immediate environment. That is not a signal system with fixed inputs and fixed outputs. That is pragmatic language use, the ability to deploy communication responsively in real time, adjusted to a shifting social situation.

[00:11:32]The same flexibility [music] humans show when they speak differently to a child if then to a colleague or soften a message depending on who [music] is listening. The exchanges last up to an hour. Multiple whales overlap, respond, build on each other's contributions.

[00:11:48]There is measurable conversational timing, rhythm, turn-taking that occasionally breaks down when two whales click simultaneously, which Jero notes appears to be socially acceptable in sperm whale culture. The way interrupting can be among close family.

[00:12:04]Consider what that requires from a brain. To produce the right coda for the right whale in the right moment demands not just memory and acoustic control. It requires theory of mind, an awareness that the other individual holds a perspective different from your own and that what you say should be calibrated to reach [music] them specifically. That capacity was once considered uniquely human. Researchers studying sperm whales are no longer certain that framing holds. The brains sustaining those conversations are the largest of any [music] animal on Earth, six times the mass of a human brain. [music] Sperm whales live in matrilineal families with multi-generational social structures, cultural traditions transmitted from grandmother to mother to calf across centuries and coordinated deep water hunting strategies executed in pitch darkness across hundreds of meters of open ocean, behaviors that require real-time multi-party information exchange to work at all. The matriarch coordinates it. She has been coordinating it, communicating it, transmitting it for 70 years >> [music] >> in a language we couldn't read until now.

[00:13:13]Structure is not meaning yet. Here is where the script has to be precise because the gap between what is known and what is not known is exactly where the bad news lives. The AI has mapped the grammar. It has not translated a single word, not one. The phonetic inventory has been identified. The combinatorial rules have been modeled.

[00:13:33]The vowels have been cataloged.

[00:13:35]Individual whales have been profiled by their acoustic signatures.

[00:13:38]Conversational patterns between family members have been documented across years of recording. All of that is real, verified, and peer-reviewed. And none of it tells researchers what any specific coda sequence means. What the difference in content is between a rising tempo pattern and a falling one. What a matriarch says to her whole group versus what a mother says only to her calf.

[00:14:01]Nobody knows yet. Meaning requires correlation. A data set deep enough that patterns emerge between specific vocalizations and specific observable behaviors. Next-generation biologger tags are now logging audio from three synchronized microphones alongside depth, GPS coordinates, body orientation, and the acoustic output of every nearby whale simultaneously.

[00:14:28]Feeding into a behavioral archive cross-referenced against decades of observational records. Project CETI has committed to making this data openly available to the global scientific community. Every discipline is invited in. But here is the weight of what that gap actually means. The matriarch was born before sonar existed. She has spent her entire life communicating with her family in a system that science only just admitted might be language.

[00:14:56]Whatever she has said across those seven decades, whatever knowledge she transmitted, whatever warnings she issued, whatever she told the young ones about how the world works and what threatens them, all of it happened in a language we could not read. We still can't read it. We have only just found the alphabet, and the world she has been describing in that language has been getting louder, and louder, and louder.

[00:15:19]She can hear it. She has always been able to hear it, and now comes the bad news.

[00:15:25]What changes? If they're right, in the 1960s, researchers Roger Payne and Scott McVay proved humpback whales sing complex evolving songs. Before that, most people thought of whales as large commercially valuable marine animals, ecologically present, but not cognitively significant. That single finding changed the public's entire mental model of what a whale was. It sparked the save the whales movement, led to the Marine Mammal Protection Act, and pulled multiple species back from the edge of extinction. One discovery about whale intelligence rewrote international conservation law. Proving sperm whales have language, not songs, not signals, but a compositional linguistic system with vowels, phonemes, and grammar, doesn't update that precedent. It detonates it. Gero explains why the framing matters. When people can talk about how important whale grandmothers are part of their families, or the importance of being a good neighbor in a whale community, that resonates in ways abstract conservation arguments never can. People protect what they relate to.

[00:16:31]Language creates relation, and if the public comes to understand that [music] sperm whales are not just intelligent animals, but communicating ones, You're dealing with an incredibly complex Morse code click type with infinite amount of possibilities. So, it's possible they're having elaborate, dramatic even, conversations. It's possible that they're communicating in ways that we've never thought of before. Beings that tell each other things, remember things, pass knowledge down through generations the way human cultures do, then every whale becomes not a marine statistic, but a speaking member of a community.

[00:17:04]Here is the bad news, not the theoretical bad news, the operational bad news happening right now in every ocean [music] basin on Earth, every single day. Commercial whaling killed hundreds of thousands of sperm whales in the 20th century. Ship strikes kill dozens every year and underwater noise pollution from industrial shipping and military sonar is broadcasting interference continuously, permanently into the acoustic space where an entire civilization holds every conversation it has ever had. Every family exchange, every transmission of cultural knowledge, every word a matriarch has ever spoken to the family she has spent 70 years building. The noise doesn't stop. It doesn't sleep. Cargo shipping runs 24 hours a day, 365 days a year through the exact migration corridors and feeding grounds where sperm whale families have communicated for millennia. Military sonar exercises are conducted in the same waters. Seismic surveys pulse acoustic energy through the deep ocean at volumes that temporarily deafen the animals that live there. [music] The interference in the acoustic environment that sperm whales rely on for communication, for navigation, for cultural transmission has been growing continuously for over a century and shows no sign of reversing.

[00:18:25]If that community speaks a language and the evidence now says it does, then the noise is not environmental damage. It is the ongoing systematic erasure of a speaking civilization sentence by sentence, year by year in every ocean basin on Earth simultaneously.

[00:18:42]That is the bad news the title promised.

[00:18:44]That is what makes this finding matter beyond the science. Legal scholars are already asking what follows. If sperm whales use language to maintain relationships, transmit cultural knowledge, and discuss the world they inhabit, do they qualify for legal personhood? Do they have rights the law has never been forced to confront?

[00:19:04]Several legal theorists in the United States and Europe [music] have begun drafting frameworks extending certain protections to non-human animals demonstrably capable of language use.

[00:19:14]Those frameworks were largely theoretical before this research. They are becoming less theoretical every month. The science is moving faster than the law is built to handle. The question nobody is ready for. The AI is still running. The data keeps arriving. The day is coming when a researcher sits at a screen and sees not just structure, but words. Not just grammar, but meaning. Not just a phonetic system, but a sentence. The first sentence any human being has ever read in a language 30 million years older than our own. And somewhere in the Caribbean, in water too deep for sunlight, the matriarch is still talking. She has watched the ocean change across her entire lifetime in ways no human has documented from the inside. She has watched the acoustic world she was born into, the world her mother taught her to navigate by sound, fill with industrial interference that had no ceiling and nowhere left to go.

[00:20:11]She has lost family members to ships she could hear coming from miles away and could not escape. She has spoken to her family about all of it in a language we are only now learning exists. The sperm whales alive in the Caribbean today have been talking about the warming water, the disappearing prey populations, the noise that now saturates every ocean basin on Earth. They have been passing that knowledge from grandmother to mother to calf, the way every human culture has ever passed knowledge forward in language across generations against forgetting.

[00:20:44]If they have language, and the evidence now says they do, then they have been talking about all of this. They have been talking about us. That is the weight of what project CETI is reaching toward. Not the translation of an animal communication system, the decoding of a 30-million-year perspective on this planet that watched humans arrive, watched us industrialize, and has had something to say about what we've done to the ocean in a language we are only now learning to read at the exact moment we may have already done the damage that cannot be undone. The alphabet has been found, the vowels have been identified, the matriarch is still speaking. We just don't know yet what she's saying.

[00:21:24]Subscribe because the day project CETI reads the first full sentence, you'll want to be here when it surfaces.