AI Model Decodes Sperm Whale 'Phonetic Alphabet', Revealing Language-Like Complexity

Deep in the waters of the Eastern Caribbean, an artificial intelligence model is listening to the ocean and learning to speak whale. For decades, marine biologists have recorded the rhythmic, Morse code-like clicks of sperm whales, known as codas, treating them primarily as simple identification signals. But a multidisciplinary initiative known as Project CETI (Cetacean Translation Initiative) is fundamentally challenging that assumption. By applying advanced machine learning to massive acoustic datasets, researchers have discovered that sperm whale communication possesses a structural complexity previously thought to be exclusive to human language.[1][3]

The breakthrough centers on the identification of a "phonetic alphabet" within the whales' vocalizations. When researchers sped up the codas and removed the silences, they found distinct acoustic elements that function remarkably like human vowels and diphthongs. Just as the English word "cow" glides seamlessly from an "ah" to an "oo" sound, sperm whale clicks exhibit systemic shifts in length, resonance, and frequency. This suggests that the whales are not merely repeating static signals, but actively combining and modifying phonetic units to convey nuanced, context-dependent information.[1][2][6]

To decode these patterns, Project CETI developed WhAM (Whale Acoustic Model), an AI system originally inspired by music-generation algorithms. The team, which includes researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), trained the model on an "ark's worth" of animal audio from open databases before fine-tuning it exclusively on thousands of sperm whale codas. WhAM operates like an acoustic game of Mad Libs, analyzing incomplete sequences of clicks and predicting the missing elements until its outputs sound convincingly whale-like.[2][5]

Through this predictive training, the AI began detecting subtle acoustic features that human ears—and traditional spectrograms—had entirely missed. Dr. Daniela Rus, director of MIT CSAIL, noted that the whales modify their codas through "ornamentation" (the addition of extra clicks) and "rubato" (expressive, slight variations in the timing between clicks). These structural tweaks indicate that the sperm whale communication system is capable of representing a vast space of possible meanings, far exceeding the complexity of most other animal vocalizations.[2][3][5]

For the linguists involved in the project, the process of deciphering these codas is akin to studying ancient, dead languages like Hittite or Sanskrit. However, instead of being separated by thousands of years of history, the researchers are separated by the profound environmental divide of the deep ocean. This "massive domain gap" is the greatest hurdle in moving from structural recognition to actual translation. Because sperm whales live in an environment utterly alien to humans—navigating the pitch-black abyss via echolocation and hunting giant squid—much of their vocabulary likely references experiences for which no human words exist.[1][2]

Despite this semantic barrier, the structural revelations are already sending shockwaves beyond the scientific community. The discovery that cetaceans may possess a capacity for true language is challenging long-held anthropocentric views in both philosophy and law. If sperm whales are not just signaling, but actively conversing using a phonetic alphabet, it fundamentally alters the ethical calculus of how humans interact with the marine environment.[4]

In response to these findings, legal scholars are exploring how AI-assisted bioacoustics could reshape environmental policy. New York University's More-Than-Human Life (MOTH) Program has partnered with Project CETI to investigate the legal impacts of these discoveries. Their collaborative research argues that proving cetaceans have language could disrupt current legal frameworks, providing a foundation to grant whales stronger, legally binding protections as sentient, communicating beings rather than mere biological resources.[4]

While WhAM cannot yet translate a sperm whale's clicks into English, the model's ability to map the architecture of their communication marks a historic milestone in interspecies relations. By leveraging artificial intelligence to bridge the cognitive divide, researchers are moving closer to a future where conservation is driven not just by human observation, but by a genuine understanding of the voices that inhabit the deep.[1][2][4]