Whale Communication Decoded by AI: A Phonetic Alphabet of Clicks

For decades, marine biologists have listened to the rhythmic clicking of sperm whales with a mix of wonder and confusion. We knew they were communicating, but the complexity of their exchanges remained locked behind a barrier of biology and acoustics. Now, a massive breakthrough using artificial intelligence has cracked the door open. Researchers have identified what functions as a “phonetic alphabet” in whale songs, suggesting these animals share a communication system with a structure surprisingly similar to human language.

Project CETI and the Dominica Recordings

The discovery comes from Project CETI (Cetacean Translation Initiative), a nonprofit organization dedicated to decoding animal communication. Their research focused on a specific group of about 400 sperm whales known as the Eastern Caribbean Clan, which lives off the coast of Dominica.

This was not a small sample size. The team utilized a dataset collected by the Dominica Sperm Whale Project, which has been recording these specific families for nearly two decades. The study, published in Nature Communications in May 2024, analyzed nearly 9,000 distinct “codas.” A coda is a short burst of clicks that sounds somewhat like Morse code.

Before this AI analysis, scientists believed these whales had a limited repertoire of about 21 distinct coda types. It was assumed these were simple signals, perhaps identifying the whale or the family group. However, the machine learning algorithms revealed something far more intricate. The AI detected subtle variations that human ears—and previous computer models—had missed entirely.

The Components of Whale Speech

The machine learning models identified that sperm whale vocalizations are combinatorial. This means they can combine small, meaningless distinct sounds to create a massive variety of meaningful signals. This is a feature previously thought to be unique to human language.

The AI broke the vocalizations down into distinct auditory features, effectively creating an alphabet of movements and sounds. The researchers identified four primary elements that whales manipulate to change the meaning of a click sequence:

  • Rhythm: The classic pattern of the clicks.
  • Tempo: The speed at which the clicks are delivered.
  • Rubato: A term borrowed from music theory, referring to the smooth change of tempo within a single sequence. The whales speed up or slow down their clicking during a coda.
  • Ornamentation: The addition of an extra “grace note” or click at the end of a coda.

By mixing and matching these four variables, the whales can generate a vast array of unique calls. The researchers compare this to how humans use phonetic sounds (like the sound of the letter ‘b’ or ‘a’) to build words, and then words to build sentences. The 21 distinct codas scientists originally identified are actually part of a much larger, flexible system capable of expressing complex information.

How the AI "Heard" the Difference

The human ear is not designed to hear underwater nuances perfectly, and traditional analysis was limited by human bias. We tend to categorize sounds into boxes that make sense to us. To get around this, Project CETI used machine learning visualization tools.

The AI mapped the acoustic properties of thousands of clicks into a visual space. Instead of seeing 21 tight clusters (which would indicate 21 standard calls), the AI revealed a fluid spectrum. It showed that whales were actively modifying the duration and timing of their calls in a predictable, structured way.

For example, two whales might use the same “rhythm” (pattern of clicks), but one might perform it with a distinct rubato (slowing down at the end) to convey a different context or intent. The AI demonstrated that these weren’t accidental variations or errors. They were consistent, intentional choices made by the whales during conversations.

Why Sperm Whales Need Complex Language

Sperm whales possess the largest brains in the animal kingdom, weighing up to 20 pounds. They are highly social creatures that live in matrilineal pods (led by females). They dive thousands of feet deep to hunt giant squid, often coordinating their movements in pitch-black water where visual signals are useless.

Sound is their primary way of seeing and connecting. Their nose contains a massive organ filled with spermaceti oil, which acts as an amplifier. This allows them to produce clicks up to 230 decibels—loud enough to vibrate a human diver’s body from the inside out.

While they use echolocation clicks to “see” their environment and hunt, the “codas” analyzed by Project CETI are social. They happen when the whales are at the surface, socializing, babysitting calves, or making collective decisions about where to travel. The complexity of their social structure requires a complex way to share information. If they need to decide to move the pod 10 miles north to find better hunting grounds, a simple grunt or single signal likely isn’t enough. They need a way to negotiate and agree. This “alphabet” of clicks provides the bandwidth necessary for that level of negotiation.

The Next Step: Meaning and Interaction

Identifying the alphabet is step one. Step two is figuring out the dictionary. Currently, we know the structure exists, but we do not know what specific combinations mean.

Lead biologist Shane Gero and computer scientist GaĹĄper BeguĹĄ are now working on the next phase. They intend to use this new map of vocalizations to predict what a whale might say next in a conversation. If an AI can predict the response, it proves the model understands the grammar of the language.

Eventually, the goal is to attempt two-way communication. This involves playing back synthesized whale sounds based on this new “alphabet” to see if the whales respond appropriately. If successful, we may one day ask the whales directly about their lives, their history, and their view of the ocean.

Frequently Asked Questions

What is a coda? A coda is a stereotyped series of clicks produced by sperm whales. It usually lasts less than two seconds. Think of it as a rhythmic pattern, like “shave and a haircut,” but used for social communication rather than music.

Did the AI translate what the whales are saying? No, not yet. The AI identified the structure of the language. It found the phonetic alphabet (the sounds they use to build words). It has not yet translated the meaning of those words. That is the next phase of research.

Who is funding this research? This work is largely driven by Project CETI (Cetacean Translation Initiative), a nonprofit organization. It involves a multidisciplinary team of marine biologists, cryptographers, linguists, and robotics experts.

How is this different from whale songs? Humpback whales sing long, melodic songs that evolve over time, mostly for mating. Sperm whales do not “sing” in that sense. They click. Their communication is more like a rapid-fire telegraph or a conversation than a musical performance.

Where can I find the study? The findings were published in the scientific journal Nature Communications in May 2024 under the title “Contextual and combinatorial structure in sperm whale vocalisations.”