In the savannas of Kenya, African elephants often emit low rumbles, almost inaudible to humans, which overlap one another to form a kind of chorus. A new study shows that these joint sounds are not random noise but coordinated group messages carrying additional information about behavior and bonds within the family.
Scientists analyzed 902 recordings of rumbles by wild savanna elephants from Amboseli National Park, collected from 1986 to 2022. The sounds were recorded in six social contexts: from "soothing" signals by females to their calves to "greeting" rumbles at reunions and "cadenced" ones during eating or slow movement. Machine learning helped reveal patterns that had previously eluded researchers' attention.
The algorithm predicted the elephants' actions from the choral rumbles better than random guessing, though it did make mistakes. At the same time, in some situations individual rumbles within the chorus became acoustically similar — a phenomenon reminiscent of how people in a conversation begin to adopt the intonations of their interlocutor. This "vocal convergence" was observed in contact, cadenced, and greeting signals and probably helps the family better coordinate joint actions.
The chorus, therefore, carries not only information about who is doing what but also a subtler layer of information inaccessible when analyzing single sounds. This discovery underscores how complex the social life of elephants is and how closely their communication and behavior in the group are intertwined.
The study was published in Scientific Reports and fits into a growing series of works in which artificial intelligence helps study animal sounds — from whales and monkeys to crows. Importantly, the method makes it possible to work with already collected data without disturbing wild populations.
Understanding these signals opens the way to more accurate monitoring of the condition of groups and, possibly, to better protection of the habitats where elephants continue to communicate in their ancient language.
