The Voices That Match

A study published October 1 in Scientific Reports, also covered by New Scientist, finds acoustic convergence in particular social contexts among wild African savannah elephants. It also shows that overlapping choruses retain acoustic information associated with what the animals are doing. Everyone speaking at once is not necessarily the opposite of communication. Sometimes it is the material.

Listening to the overlap

Louis Mahon and colleagues analyzed 902 recordings of single and chorused rumbles from adult females in Kenya’s Amboseli National Park. The recordings came from periods between 1986 and 2022, drawing on decades of field observation rather than a computer briefly eavesdropping on an unfamiliar species.

Researchers assigned calls to six behavioral contexts, including contact between separated elephants, greetings and situations associated with coordinating movement. They then trained statistical classifiers to predict those contexts from acoustic features.

Both single rumbles and overlapping choruses carried enough structure for predictions above chance. Models trained on single rumbles could also classify choruses above chance, suggesting that the overlapping voices preserve features associated with the same context.

That is not an elephant translation app. The authors report that more rumbles were classified incorrectly than correctly. Behavioral situations can shade into one another, and the categories a field observer assigns need not map perfectly onto distinctions meaningful to elephants. Recognizing that a recording resembles a greeting is not the same as knowing what anyone said.

The useful surprise is that overlap need not erase the signal. Studying only neatly isolated calls risks leaving out much of the interaction we hoped to understand.

Getting closer in sound

The convergence analysis asked a different question: do successive rumbles become acoustically more similar during an exchange?

For this, the team examined 92 multi-caller exchanges from one family group, measuring the non-overlapping portions of individual rumbles. They found evidence of convergence in Contact and Greeting exchanges under both of their statistical tests. Cadenced exchanges, associated with coordinating activity, showed convergence under one test but did not meet the conventional significance threshold under the other. The remaining three contexts showed no significant convergence, though some samples were small.

The pattern suggests a possible role in coordinating group activity, the authors argue. It does not establish that matching sounds causes agreement, or that every chorus involves deliberate imitation. The distinction leaves a good scientific question standing: what does becoming more alike in sound allow these animals to do together?

Nor did shared context make individual variation disappear. Classifiers performed worse when tested on unfamiliar callers or across family groups. Apparent family differences could reflect differences between individuals rather than a distinctive family-wide vocal style.

For a coherenceist reading, that is the interesting scale change: a group signal need not mean that the group has become one speaker. Coordination can live in relationships among distinct voices. But the philosophy cannot decide what the matching is for; that remains work for observation and experiment.

There is warmth in the restraint. We do not need to promote elephants into honorary humans to find their social lives remarkable. Their rumbles already contain patterns we are learning to distinguish, including patterns that unfold between callers rather than residing in one call alone.

We have not translated the gathering. We have learned a little more about how to listen when the gathering refuses to take turns.