U.S. research team “Can understand the context of a group’s behavior by analyzing overlapping sounds”

U.S. research team “Can understand the context of a group’s behavior by analyzing overlapping sounds”

(Seoul = Yonhap News) Reporter Jooyoung Lee = Several African elephants are often seen making low sounds (rumbles) at the same time. What meaning does this low-frequency chorus of elephants contain?

The overlapping chorus of different elephant voices contains information about the group's situation and behavior, and research shows that in certain situations, such as when they meet again or coordinate what to do next, their sounds become more similar to each other.

Research teams including the U.S. Earth Species Project and ElephantVoices announced in the scientific journal Scientific Reports on the 2nd that they were able to predict in what behavioral context the sounds came from by analyzing 902 sounds made by African elephants singly or in chorus using machine learning.

The research team said that they were able to identify the behavioral context of the entire chorus without separating individual elephant voices, and that the significance of this study was that they analyzed the overlapping elephant sounds themselves, which had been difficult to analyze, as a single signal.

African savannah elephants (Loxodonta africana) are highly social animals, often communicating using low-frequency 'rumbles', some of which reach frequencies below the human hearing range.

The research team pointed out that about 60% of the growling sounds made by elephants are in the form of a chorus where the sounds made by two or more elephants overlap, and that such choruses often appear during intense social interactions such as greeting ceremonies, but it was difficult to analyze them because the sounds were mixed.

The research team used machine learning to analyze 902 sounds made singly or in chorus by 84 females from 39 African savannah elephant families collected at various times from 1986 to 2022 in Amboseli National Park, Kenya.

As a result, the machine learning model was found to be able to predict the context of vocalization behavior at a significantly higher level than chance by analyzing the entire chorus as a single signal without separating the overlapping elephant sounds one by one.

In particular, the research team said that when the model learned from the sounds made by elephants alone was applied to the chorus, it predicted the behavioral context significantly higher than chance. This shows that the information is not lost because the individual sounds are completely mixed.

This study also confirmed the phenomenon of 'acoustic convergence', in which the sounds made by elephants become acoustically similar to each other over time in some situations.

Acoustic convergence was confirmed in the 'Contact' made by separated elephants to maintain contact, the 'Cadenced' sound exchanged while eating, resting, or moving slowly, and the 'Greeting' sound made when separated family or group members meet again.

The research team explained that 'contact' and 'regular beat' sounds often appear when several elephants exchange sounds and coordinate joint actions, and that the phenomenon of elephants' sounds coming closer to each other while talking is likely related to behavioral coordination.

However, because this study did not conduct an experiment to compare the responses of elephants by listening to convergent vocalizations and non-convergent vocalizations, it was not possible to conclude that 'when voices are matched, the herd decides the next action'.

He went on to say that although it is not yet known exactly what words the elephants exchange while singing in unison, this study suggests that the elephants' chorus may not be noise that interrupts each other's voices, but may be a signal created in the process of multiple individuals acting together.

He also emphasized that the context of elephants' behavior can be understood through sound, and that the distribution and behavior of elephants can be identified through acoustic surveillance even in places where direct observation is difficult, and that this study will help develop an approach for the coexistence of humans and elephants.

◆ Source: Scientific Reports, Sara Keen et al., ‘Social context information and acoustic convergence in elephant vocal choruses’, https://www.nature.com/articles/s41598-026-68525-3