AI Decodes Whale Clicks: How Machine Learning Revealed a Sperm Whale "Alphabet"

For decades, scientists have listened to the rhythmic clicking sounds of sperm whales with a mixture of awe and confusion. We knew they were communicating, but the complexity of that communication remained a mystery. Now, a breakthrough study utilizing advanced artificial intelligence has cracked part of the code. Researchers have discovered that these marine giants possess a “phonetic alphabet” much closer to human language than previously imagined.

The Breakthrough by Project CETI

The discovery comes from Project CETI (Cetacean Translation Initiative), a non-profit organization dedicated to applying advanced machine learning to animal communication. In a study published in Nature Communications in May 2024, researchers analyzed nearly 9,000 recordings of sperm whales collected off the coast of Dominica in the Caribbean.

These recordings were not new. Many came from the Dominica Sperm Whale Project, which biologist Shane Gero has led for nearly 20 years. However, the sheer volume of data made it impossible for human listeners to identify subtle patterns. This is where AI stepped in.

By feeding these audio files into machine learning algorithms, the team discovered that sperm whale clicks are not static. The whales vary the tempo and rhythm of their click patterns, known as “codas,” in a structured way. This suggests a combinatorial communication system where basic sounds are combined to create distinct meanings, similar to how humans combine phonemes to form words.

Understanding the Sperm Whale "Alphabet"

To understand what the AI found, you first have to understand how sperm whales sound. They do not sing haunting melodies like humpback whales. Instead, they produce rapid bursts of clicks that sound like a creaking door or a hammer hitting a metal pipe. These patterns are called codas.

Before this study, scientists believed there were about 150 standard codas that functioned like simple signals. The AI analysis revealed a much deeper layer of complexity involving three specific variables:

• Rhythm: The timing between the clicks.
• Tempo (Rubato): The speed at which the clicks are delivered.
• Ornamentation: Extra “grace notes” or clicks added at the end of a coda.

The Concept of Rubato

The most surprising finding was the use of “rubato.” In music, rubato refers to the expressive speeding up or slowing down of a tempo. The AI detected that whales actively change the duration of their calls. They might stretch a coda out or compress it.

Crucially, the whales mimic each other’s tempo during conversations. If one whale slows down its clicking pattern, the responding whale often matches that speed. This synchronization suggests that the timing carries specific information or emotional context, rather than just being a random variation.

Ornamentation and Structure

The researchers also identified “ornamentation.” This occurs when a whale adds a distinct extra click at the end of a standard coda. The machine learning models showed that these ornaments function similarly to suffixes in human language. They modify the base coda, potentially changing its intent or directing the message to a specific individual within the pod.

How Machine Learning Solved the Puzzle

The human ear is excellent at processing speech but struggles to track millisecond differences in underwater clicking sounds. This limitation is why the complexity of whale language remained hidden for so long.

Project CETI researchers, including Gašper Beguš from UC Berkeley and David Gruber from CUNY, used computer models to visualize the sounds. They mapped the clicks into a high-dimensional geometric space. In this mathematical view, clusters of similar sounds became visible.

The computer did not just look for identical matches. It looked for relationships. It recognized that Coda A played at high speed was mathematically related to Coda A played at low speed. By identifying these variables, the AI determined that the whales have a “phonetic alphabet” capable of generating roughly 7,000 to 10,000 distinct combinations. This is a massive leap from the 150 calls scientists previously assumed existed.

What Are They Talking About?

While the AI has identified the alphabet, it has not yet written the dictionary. We know how they structure their language, but we do not know what they are saying.

However, the context of the recordings offers clues. Sperm whales have the largest brains in the animal kingdom and live in complex matriarchal societies. They babysit each other’s calves, coordinate deep-sea hunts, and maintain lifelong family bonds.

Biologists speculate that this complex language is used to:

• Identify Individuals: Unique click patterns may serve as names.
• Coordinate Hunting: Organizing dives into the deep ocean to hunt giant squid requires precise timing.
• Social Bonding: Much like human gossip or storytelling, these conversations happens during “social hours” near the surface.

The Future of Interspecies Communication

This research marks a turning point in bioacoustics. Project CETI plans to deploy underwater microphones and swimming robots to capture more data. The ultimate goal is to attempt two-way communication.

This involves playing back synthesized whale sounds to the animals to see if they respond. If a whale mimics a computer-generated coda or replies with a predictable variation, it would confirm that we have cracked the syntax of their language.

This possibility raises significant ethical questions. If we can speak to whales, we must decide what to say. Furthermore, proving that whales possess a complex, language-like culture could drastically change legal frameworks regarding ocean conservation. It becomes much harder to justify the destruction of a habitat when you know the inhabitants are discussing it.

Frequently Asked Questions

Did the AI translate what the whales said? No. The AI identified the structure of the communication, revealing an alphabet and grammar system. It did not translate the meaning of the clicks into English.

What is a “coda” in whale communication? A coda is a stereotyped pattern of clicks. It is the basic unit of sperm whale communication, similar to a word or a syllable in human language.

How does this compare to human language? The study suggests sperm whale communication is combinatorial. This means they combine small, meaningless units (clicks) to form larger, meaningful structures. This features is a hallmark of human language known as “duality of patterning.”

Who is funding this research? This research is primarily conducted by Project CETI, a non-profit organization comprised of biologists, cryptographers, and roboticists from institutions like MIT, Harvard, and National Geographic.