Chimpanzee “accents” become more recognizable as they grow

Chimpanzees from different places produce acoustically distinct vocalizations—what the researchers operationally call dialects. This is not language, vocabulary, or a human accent, but a set of sound features that helps a statistical model identify population of origin. Kassandra Giragosian, Marina Davila-Ross, and Derry Taylor compared two geographically separated populations: 17 chimpanzees at Chimfunshi Wildlife Orphanage in Zambia and 13 at Gombe National Park in Tanzania.

The records were also divided by development: 21 individuals were observed as infants aged 0–4 and 10 as juveniles aged over 4–10; one chimpanzee appeared at both stages. Population and age are therefore separate dimensions. The dataset contained 11,225 vocal units: 3,694 grunts, 2,798 whimpers, and 4,733 laughs. These are three call types—bioacoustic categories recognized by sound form and behavioral use—not direct classifications of the animal’s mental intention.

The team extracted 15 temporal and spectral measurements from each sound, including duration, element rate, frequencies, and energy distribution. A permuted discriminant function analysis then tried to assign each vocalization to Gombe or Chimfunshi. Classification accuracy is the proportion of sounds assigned to the correct population; it is not the proportion of animals identified as infants or juveniles. Individual identity was controlled so that hundreds of calls from one chimpanzee were not treated as hundreds of independent animals.

A later analysis asked whether population became easier to recognize in juvenile sounds. In this age comparison, estimated correct classification for grunts rose from 76.1% in infants to 91.9% in juveniles. Whimpers rose from 76.6% to 84.9%, and laughs from 92.9% to 96.5%. For every call type, the infant–juvenile difference had p < 0.001. Under a statistical model in which age had no association with accuracy, a result at least this extreme would have a probability below 0.1%, assuming the model’s conditions hold.

A separate analysis asked a different question: within each age group, did acoustic features distinguish the two populations better than shuffled labels? Infant grunts and laughs showed significant discrimination. Infant whimpers were classified correctly 71.81% of the time, compared with a mean of 55.31% across randomized permutations. That mean is not a significance threshold. Because 12.3% of the permutations reached a result at least as high as the observed one, p = 0.123, above the conventional 0.05 criterion. The study therefore lacked sufficient evidence of a population signature in infant whimpers; it did not prove the populations identical.

The 71.81% and 76.6% figures are not contradictory versions of one calculation. The first is cross-classification accuracy from the infant-whimper discriminant analysis; the second is a mean predicted by the beta regression comparing ages. Likewise, p = 0.123 tests infant whimpers against a shuffled-label distribution, whereas p < 0.001 tests whether accuracy differs between infants and juveniles. An age-group difference can be statistically clear even when one group alone does not pass a separate test against randomization.

The authors interpret the developmental strengthening of acoustic signatures as compatible with social and ecological influence. Call-type specificity—especially the inconclusive result for infant whimpers—weighs against a purely genetic explanation but does not rule out genetic contributions. The populations have no migration between them and probably differ genetically; Gombe is also wild whereas Chimfunshi is semi-wild, with different habitats, social histories, and recording conditions.

The retrospective design also combined recordings made with different equipment and protocols. The authors argue that equipment alone is unlikely to create a pattern that varies by call type and age, but the limitation remains. The study shows that population can be recognized from vocal acoustics more accurately in juveniles than infants. Longitudinal studies and comparisons that separate social learning, anatomical maturation, ecology, and genetics are still needed to identify the causes.

Correction — September 4, 2026: the article was rewritten to distinguish population from age group, explain what was classified, separate the two statistical tests, and qualify the interpretation of learning, environment, and genetics. The numerical data and central finding remain unchanged.