In 2024, an MIT research team analyzed 9,000 recordings of sperm whale vocalizations using an artificial intelligence system originally designed for code-breaking. The analysis identified 156 distinct click patterns, known as codas, that the whales use to communicate. This was a significant increase from the 21 codas that researchers had previously cataloged since the 1970s.
The findings emerged from Project CETI, the Cetacean Translation Initiative, a research collaboration involving MIT, Harvard, UC Berkeley, and more than 15 institutions. The project was launched in 2020 with funding from the TED Audacious Project. Its goal is to determine whether sperm whale communication meets the criteria for language.

The project traces its origins to 2017, when marine biologist David Gruber was reviewing recordings of whale clicks collected during a jellyfish study in the Caribbean. Gruber’s colleague, MIT cryptographer Shai Goldwasser, heard the clicks and suggested applying machine learning to the data. Goldwasser recognized patterns in the vocalizations that resembled the encrypted codes she worked with professionally.
Researchers built a permanent field station in Dominica, an island in the eastern Caribbean where several generations of sperm whale families have returned to the same waters for decades. The study animals are known individuals, tracked through photo identification and documented family relationships.
The analysis revealed that sperm whale codas have internal structure, including tempo variations and additional clicks that appear to modify meaning. Researchers described this as a phonetic alphabet, a finite set of building blocks that can combine in varied ways. The structure resembles the phoneme system used in human language, in which a limited set of sound units produces a large number of words and sentences.
Linguist Gaspar Becerra at UC Berkeley analyzed individual clicks and identified vowel-like sounds, including sounds similar to the vowels in “father” and “see,” as well as gliding vowel combinations known as diphthongs. Vowel sounds are considered a defining feature of human speech. Sperm whales produce clicks through nasal passages and fatty tissues in their foreheads rather than vocal cords, meaning the similarity emerged through an independent evolutionary path.
The whales also appear to use codas contextually. The same coda may differ based on the social situation, such as between a mother and calf during nursing, during a group hunt, or when adults greet one another after a separation. This contextual variation is similar to pragmatics in human language, where meaning shifts based on circumstances.
Shane Gero, a researcher who has studied sperm whale families off Dominica for 13 years, has observed exchanges that last about an hour, with multiple whales clicking in overlapping patterns. He described the interactions as resembling conversations between individuals who know each other well.
Sperm whales have the largest brains of any animal known to have lived on Earth. They live in matrilineal family groups, with older females serving as repositories of knowledge about migration routes and feeding grounds. Communication patterns vary between family groups in ways that appear learned rather than genetic, similar to dialects among human communities.
Researchers are currently gathering additional data using suction-cup bio-loggers that record audio while tracking depth, movement, and location. The devices are placed on multiple whales within a family group so that vocalizations can be matched to specific individuals and behaviors. The project team says the analysis is progressing faster than initially projected.
The possibility that sperm whales possess language has raised legal and ethical questions. Some legal scholars have begun discussing whether cetaceans might qualify for legal personhood if their communication is demonstrated to constitute language. The scientific community has not reached a conclusion on the meaning of the whale vocalizations. Project CETI researchers state that translating the codas will require extensive matching of vocalizations to observed behaviors, and that work is underway.