Menu
Up

Click-Click! Deciphering the Language of the Sperm Whale

Sperm whales have the largest known brains on the planet, and they use their massive minds to communicate with each other in surprisingly complex ways.

The whales use sequences of between three and 40 clicks, known as codas, for social purposes and other sounds for activities such as hunting.

A study in Scientific Reports could help advance and accelerate the study of these sperm whale sounds. The paper, authored by Professor David Gruber (The Graduate Center, Baruch College) and collaborators, is the first time advanced machine learning methods have been used to study these creatures’ bioacoustics.

Not only is sperm whale talk impressive, but the communication methods of cetaceans in general—including porpoises, dolphins, whales, and narwhals—can help scientists test ideas on how communication systems develop. A better understanding of their communication could also have implications for conservation. However, studying wild animals that live underwater is tricky, and progress is slow.

The study showed that it is possible to use “deep” machine learning to study sperm whale sounds, a step that could put the field on a faster track. Machine learning involves training computer programs to perform and learn from tasks on their own. The “deep” version of this involves using systems known as artificial neural networks (ANNs) to start from raw data and gradually learn to do increasingly complex and abstract work.

The authors tested their deep machine learning methods on numerous sperm whale recordings to see if the systems could detect clicks, as well as classify codas by type, whale clan, and even identify individual whales.

The system completed each task with over 93% accuracy, showing that deep machine learning is a valid, useful way of studying whale bioacoustics. The researchers’ ultimate hope, which will require even bigger sets of recordings, is to train an ANN to independently do all of these tasks, as a sort of “all-in-one” package that would simplify analysis even more.