Machine learning helps decode a sperm whale 'alphabet'

For decades, humans thought complex language was our exclusive club. A pod of Caribbean sperm whales and some clever machine learning just begged to differ.

Researchers at MIT CSAIL and Project CETI published a study revealing they have uncovered a sperm whale phonetic alphabet. Using machine learning to analyze the vocal exchanges of ocean giants, the team discovered that sperm whale clicks follow a structured, combinatorial coding system rather than random noises.

Sperm whales communicate using short click bursts known as codas. Biologists previously documented around 150 coda types, knowing only that a fraction identified specific whales or clans. The broader structure and information capacity remained a complete black box.

Borrowing music theory to decode clicks

Inspired by the late pioneering marine biologist Roger Payne, the team tapped into an extensive archive of 8,719 codas recorded off the coast of Dominica by researcher Shane Gero.

MIT CSAIL director Daniela Rus noted that their pipeline involved feeding the audio into machine learning models, tweaking the system for better visualization, and then analyzing the outputs shoulder-to-shoulder with field biologists.

Instead of analyzing individual codas in isolation like older studies did, the researchers looked at how sounds shifted during interactive exchanges. They categorized variations using musical terminology, including tempo, rhythm, ornamentation, and rubato.

That framing led to the identification of a combinatorial vocal system. In linguistics, this mirrors duality of patterning: taking meaningless sound units and combining them into larger meaningful structures, much like humans assemble syllables into words.

Why it matters

This is a massive proof of concept for non-human language research. For years, decoding animal communication sounded closer to sci-fi than computer science. By showing that machine learning can parse subtle contextual shifts across thousands of recordings, the team gave researchers a repeatable blueprint for bioacoustics.

Sperm whales made an ideal starting point because their clicks form discrete signals, making them significantly easier for models to dissect than the continuous, sweeping songs of humpback whales.

That said, we should keep our expectations grounded. Finding an alphabet is not the same as having a translator. Knowing how the sounds are structured does not mean we understand what the whales are actually saying, let alone allow us to reply.

Still, we think this is one of the most compelling applications of pattern recognition to date. It proves AI is capable of uncovering deep structural rules hidden right under our noses in the natural world.