Chatty little marmosets call each other by name
Marmosets, small primates native to South America, are known to be highly social primates. Living in family-based groups of up to 15 individuals, they use a complex system of vocalizations for communication. These sounds range from high-pitched calls to trills and whistles that are referred to as “phee” calls. According to The Guardian, this behavior, identified for the first time in non-human primates, aids social cohesion.
A recent study by Hebrew University in Jerusalem showed that marmosets can communicate with one another by name and know when they are being addressed. These adorable primates join a very short list of species exhibiting such behavior. Dolphins, parrots, and elephants are the other species known to use names.
The study was published in the journal Science.
The scientists studied ten captive marmosets from three family groups, analyzing the phee calls between different pairs of monkeys.
“The experiment was very simple,” David Omer, lead author and assistant professor at the university’s Safra Center for Brain Sciences (ELSC), told CNN. “We just positioned two marmosets in the same room and positioned a visual barrier between them. When you do this, they spontaneously start to engage in dialogue.”
“This past decade, quantitative behavioral analysis has advanced as much as neural recording technologies in the neuroscience field”, says Vijay Iyer, principal academic discipline manager for neuroscience at MathWorks. “While I enjoy supporting applications pushing technology boundaries, this report is a delightful reminder that science is about discovery: asking and answering questions that haven’t been addressed before.”
Analyzing the calls with machine learning
The team recorded natural conversations between pairs of marmosets, as well as the primates’ interactions with a computer system. The sound data was collected at 96 kHZ with a custom MATLAB program. They recorded 54,000 calls.
Here is what a phee call sounds like:
The sound data was fed into a machine learning system created with the MATLAB algorithm, TreeBagger. TreeBagger is an ensemble method that reduces the effects of overfitting and improves generalization.
The team used time-frequency analysis in MATLAB for signal feature extraction. Based on acoustic features alone, the machine learning system could accurately predict which monkey a particular call had been addressed to. The researchers found the marmosets used distinct phee calls for each monkey on the other side of the visual barrier, similar to how humans use names.

A and B: Average classification accuracy of 100 random-forest models trained and tested on calls from Monday A) Adonis and B) Ella. The left panel is a confusion matrix, and the center panel shows violin plots. In C and D the same calls are shown as spectrograms. E through G show results for all of the monkeys in the study. Image credit: David Omer’s Lab, Hebrew University of Jerusalem.
Vocal learning
The team concluded that the marmosets addressed specific individuals with the phee calls, or names. They found that not only do these tiny primates use names for each other, but they are more likely to respond if their name is called. And these aren’t unique to given pairs: the other members of the family groups use the same name when calling to a specific family member.
“This is evidence for vocal learning. They learn vocal labels from their family members,” says Omer.

A mother and daughter pair, Bhumi and Belle, from the study. Image credit: David Omer’s Lab, Hebrew University of Jerusalem.
The code and data from this study is available here. More details on the analysis can be found in the supplemental information.
To read the full research paper, see DOI: 10.1126.
Comments
To leave a comment, please click here to sign in to your MathWorks Account or create a new one.