AI could help us "translate" the language of animals

Shane Gero, a biologist specializing in cetaceans from Carleton University in Canada, has been researching whale communication for over 20 years. According to his work, sperm whales have specific sounds that identify their families and use “dialects” depending on their location. Although various species like dolphins, elephants, and birds also use patterned sounds to communicate, understanding these subtleties is a challenging task for humans. This is precisely where artificial intelligence comes into play, as it is capable of distinguishing these small variations.

Nature DOWNLOAD

According to a recent article in Nature, in the past year, several AI-supported studies have revealed that African elephants and common marmosets assign names to their companions. Additionally, various machine learning tools are mapping the calls of crows, allowing for the exploration of animal self-awareness. Despite these advances, translating animal sounds remains a challenge of enormous proportions. “It is a big assumption to think that we can apply that technology to other species and obtain translations,” explains David Gruber, founder of the CETI Project (Cetacean Translation Initiative), focused on the acoustic communication of sperm whales.

Gero, with funding from the CETI Project, has collected data from more than 30 families of sperm whales in the Caribbean. These whales, which forage at depths of up to 2,000 meters, use sound clicks called “codas” to communicate. Each group, led by females, has unique diets, behaviors, and dialects. Differences in the rhythm of the codas create cultural boundaries between clans, according to Gero. Thanks to AI, analyzing these sounds has become more efficient, allowing for the identification of complete conversations and subtle variations like “rubatos” and “ornaments” (terms borrowed from music).

In parallel, Mickey Pardo, a behavioral ecologist, has demonstrated that African elephants emit unique sounds that could be names. Using AI models, he discovered that these sounds have a 27.5% accuracy in identifying individuals. In similar experiments, neuroscientist David Omer found that tamarins from the same family share acoustic characteristics when “naming” each other.

The aim of these studies is not only to understand these languages but also to expand animal vocabularies, such as location terms in elephants. AI tools are not only deciphering patterns but also illuminating the rich world of animal communication, with great potential to promote species conservation.

Nature DOWNLOAD