How artificial intelligence is cracking the sperm whale alphabet

Imagine trying to decode an alien language where every word sounds like a heavy wooden door creaking shut. That is essentially what marine biologists face when studying sperm whales.

Sperm whales possess the largest brains on Earth, live in complex matriarchal societies, and coordinate deep-sea hunts in pitch-black water. To communicate across miles of ocean, they fire off rapid, rhythmic bursts of clicks known as codas.

For a long time, researchers assumed these codas were fairly basic. Because human ears and early audio software could only parse so much detail, the working theory was that these clicks functioned like rudimentary Morse code—simple acoustic name tags used to broadcast identity or clan affiliation. It turns out the whales were saying a lot more; we just lacked the processing power to hear it.

By feeding decades of underwater recordings into machine learning models, scientists are discovering that sperm whale clicks contain a highly structured, phonetic alphabet.

The limits of the human ear

Historically, analyzing marine communication meant a grad student sitting in a lab with a pair of headphones, staring at visual sound graphs called spectrograms, and manually counting the number of clicks in a sequence.

Under that manual system, biologists categorized about 150 distinct coda types. A whale might emit a “5R” coda (five evenly spaced clicks) or a “1+3” coda (one click, a pause, then three quick clicks). It looked like a closed, finite catalog of sounds.

The problem is that biological communication is messy. When a whale sped up a click sequence or added a slight delay, human researchers generally wrote it off as a biological error or meaningless background noise. But algorithms do not get bored, and they do not ignore microscopic variations in timing.

Feeding the ocean to an algorithm

Researchers with Project CETI (Cetacean Translation Initiative) decided to treat whale codas the exact same way tech companies treat human speech. They took a dataset of nearly 9,000 recordings collected from a clan of sperm whales in the Caribbean and processed the audio using the same foundational machine learning techniques used to build modern predictive text and voice recognition software.

As reported by Smithsonian Magazine, the algorithms found that the subtle variations in the clicks were not mistakes at all. They were deliberate, systematic features of a complex language.

The computer models identified that whales actively manipulate multiple acoustic properties simultaneously. They change the base rhythm. They use rubato—a musical term for slightly stretching or compressing the tempo of a phrase depending on the conversational context. They also use ornamentation, tossing in a rapid, extra click at the very end of a sequence, similar to how a human might raise their pitch at the end of a sentence to indicate a question.

A combinatorial language

The real breakthrough is how these whales use those acoustic features. They do not just have a static list of 150 calls. They mix and match them.

In human language, we take a small set of meaningless building blocks—vowels and consonants—and combine them to form tens of thousands of unique words. The machine learning analysis revealed that sperm whales do the exact same thing. By combining different rhythms, tempos, rubatos, and ornaments, the whales can generate a massive, open-ended acoustic vocabulary.

Scientists have not yet translated what a specific combination of clicks actually means. We do not have a Rosetta Stone for the deep ocean. But proving that sperm whales use a combinatorial phonetic alphabet completely upends how we view animal intelligence. It means they are not just broadcasting simple emotional states; they have the mechanical hardware and the acoustic structure to share complex, novel ideas in the dark.

This article is made and published by Asger Risom, who may have used AI in the preparation