
Pratyusha Sharma, an IIT Delhi alumna and a PhD student at the MIT Computer Science and Artificial Intelligence Laboratory, has been listening to sperm whales as part of her research. As a computer scientist, she works on developing algorithms to understand the structure of artificial neural networks, understanding the complexity and structure of naturally arising animal communication systems in the wild, and how language and natural-language-like structures can support effective reasoning and planning in embodied agents and robots. As part of her doctoral research, she has been collaborating with the Dominica-based Cetacean Translation Initiative (CETI) to decipher sperm whale vocalisations.
Sperm whales are not easy to observe. They spend much of their lives foraging and hunting fish and squids at depths of the oceans beyond the reach of sunlight. While the males roam the open oceans, visiting other groups to propagate, the females live in complex, multi-generational, matrilineal societies — in groups of daughters, mothers, and grandmothers — forming close-knit pods to nurse and rear calves. They are known to exhibit complex social behaviour and collective decision-making, which requires sophisticated communication between individuals.
Scientists have known that whales communicate with each other since the mid-20th century when, in the 1960s, marine biologist Roger Payne (CETI's principal advisor) noticed that humpback whale calls repeated in patterns. Unlike the eerie songs of humpbacks, however, sperm whales communicate with each other using rhythmic sequences of clicks and buzzes known as 'codas'. Past research showed that sperm whales had at least 21 types of codas; but with recent advances in deep sea monitoring, recording, and machine learning, Sharma and CETI were able to identify 156 distinct codas based on almost 9,000 recordings. Using machine learning technology, they were also able to identify the basic building blocks of these vocalisations which they describe as a "sperm whale phonetic alphabet" — kind of like phonemes, or the units of sound which combine to form words in human languages.
To study sperm whale communications, Sharma and her team use advanced non-invasive suction-cup-like devices known as Digital Acoustic Recording Tags (DTAGs) that are attached to the whales temporarily and record their vocalisations. In a recent interview to Telegraph India, she explained that these devices help in mapping whale behaviour and positions while simultaneously recording vocalisations using microphones to capture underwater vocalisations, accelerometers, gyroscopes, and magnetometers to record movement and orientation, and pressure sensors to track diving depth. Then, they study this unstructured data with the help of machine learning algorithms to identify patterns and structures in the data and develop sequence models that capture the nuances of whale calls and behaviours.
Sharma's research suggests that sperm whale communications operate as a kind of "combinatorial coding system". Each coda consists of between three and forty rapid-fire clicks and features variations in rhythm, tempo, rubato, and ornamentation. This combinatorial character is a prerequisite for the linguistic phenomenon known as "duality of patterning" — previously thought to be unique to human language — which refers to the combination of meaningless elements to form meaningful words.
While we do not yet know exactly what this means for sperm whales' intelligence or consciousness, this finding does point to the possibility that sperm whale language maybe more advanced that previously thought. But before we can draw conclusions — and maybe communicate with sperm whales someday in the near or distant future — Sharma and her team at CETI must first build a fundamental understanding of how sperm whales communicate and what's meaningful to them.
Learn more about Sharma's research here:
To learn more about sperm whale codas, read 'Contextual and Combinatorial Structure in Sperm Whale Vocalisations' by Pratyusha Sharma, Shane Gero, Roger Payne et al here.