Whale Said: How Far Has Science Come in Decoding Animal Communication?

In Kenya, a group of adult female elephants and their offspring heard a recording played back to them. It was a call, and it belonged to one of them. Certain individuals perked up their ears and rumbled in reply. For the researchers, that was the sign of something like a name: a sound the family uses for one particular elephant.

That finding, published last June in Nature Ecology & Evolution, came from a team led by behavioural ecologist Mickey Pardo of Colorado State University. Using machine learning, they analysed 469 vocalisations made by the elephants over 14 months. The analysis suggested that the animals appear to have unique calls for specific individuals.

For most of human history, the buzz of bees, the calls of birds, the roar of a lion and the screech of a dolphin were background noise in a drama where people were the only characters with lines. That began to change in the early 1900s, when Karl von Frisch showed that honeybees dance inside the hive to tell one another where the nearest food is. His work opened the study of how animals communicate. By the 1960s it was clear that some species call and sing in notes that shift with the message, whether a summons, a warning or an invitation to mate. Recordings by Roger Payne, Katy Payne and Scott McVay became the 1970 best-selling album Songs of the Humpback Whale.

Researchers are now starting to decode how some of this communication works. In February, an international study in the journal Science analysed eight years of humpback recordings. It found that shorter sounds are used far more often than complex ones, so that whale songs follow a Zipfian distribution. In human language, the most common word, such as "the", appears about twice as often as the second most common and three times as often as the third. The pattern holds across all human languages, and its presence in whale song suggests the songs carry more than innate, reflexive responses to stimuli. The discovery came about because behavioural ecologists, other scientists and linguists worked alongside marine biologists. That kind of collaboration, together with machine learning and AI, is pushing the field forward.

Other efforts are more ambitious. Since May last year, marine biologist David Gruber and his organisation Project CETI (Cetacean Translation Initiative) have worked in the Dominican Islands to use AI to decode sperm whales, whose Morse-code-like click patterns are called coda. The plan is then to have AI generate clicks of its own, in an attempt to "speak back".

Off the coast of Florida, marine biologist Laela Sayigh and her team at the Woods Hole Oceanographic Institution studied the clicks and whistles of wild bottlenose dolphins. They believe they have worked out how two of the 22 whistle types are used: one is an alarm and the other a query. Their study, posted as a preprint on bioRxiv in April, says it offers the first evidence in dolphins of a wider repertoire of shared, context-specific signals that could form the basis of a language-like system. In May, Google DeepMind released DolphinGemma, a large language model meant to help researchers gather and analyse dolphin sounds.

Does this mean humans might communicate, even in limited ways, with some species? Two hurdles remain, even if every whale click and dolphin whistle were decoded. Most animals appear to communicate mainly to signal things like food, danger or readiness to mate, a narrower skill than human language. And many animals also use visual, chemical and mechanical signals alongside sound, in ways humans do not, which could make speaking back both difficult and meaningless.

What can be expected for now is a better understanding of how meaningful, layered and intentional animal communication can be, as with the elephant calling across the Kenyan grassland and her calf turning toward the sound his family uses only for him.