Cortical area remixes macaques’ knowledge blocks to solve new problems

Many animals can solve novel problems, often in a single go. For humans, that could be writing the first line of a poem, tackling a complex equation or improvising a jazz solo. For a macaque monkey, it might mean climbing a new tree to snag a delectable fruit. 

A new study, published in May in Nature, adds support to the long-standing idea that the brain accomplishes these feats by piecing together bits of existing knowledge (words, mathematical functions, riffs or tree-climbing moves, for example)—a process called compositional generalization.

Single-neuron recordings in macaques locate the knowledge blocks, according to the study. The brain activity patterns that occur in the ventral premotor cortex when monkeys learn to draw simple symbols recur in concert when the animals are later prompted to draw complex shapes made up of those symbols. 

“We have quite a lot of behavioral evidence for compositional generalization across a wide array of different tasks,” says Charlie Wilson, a tenured researcher at the Institut National de la Santé et de la Recherche Médicale (INSERM) and the Stem Cell and Brain Research Institute in Lyon, who was not involved in the new research. “The interesting element here is the step towards showing a neural basis for that.” 

The new work is part of a growing effort in the field to “bring modern techniques and modern understanding back to bear on this kind of question,” says Tim Buschman, professor of neuroscience and psychology at Princeton University. Buschman was not involved in the study but co-authored a 2025 Nature paper showing how macaques use compositional generalization to respond with specific eye movements to different types of images. “I think it’s really wonderful seeing evidence for these types of components.”

I

n the new study, two macaque monkeys learned to draw simple symbols—circles, zigzags, chevrons, lines at a right angle—in a single stroke on a touch screen. Each monkey developed its own way of creating the forms, like a personal handwriting style. 

In return for a juice reward, the monkeys later attempted to draw a new, more complicated shape featuring combinations of the previously learned ones. The monkeys drew the new shape on the first try, the researchers found, but again used different techniques, remixing their personal repertoire of learned symbols.

Throughout the experiment, implanted electrode arrays recorded the activity in 40 to 50 neurons in eight brain areas with known roles in planning, cognition and motor function. 

Brain areas involved in reusing learned symbols should show three characteristics when the monkeys look at a complex shape before drawing it, the researchers hypothesized. For one, activity there should not vary with the shape’s size or position because “it should not really care about the details of how the muscle movements occur,” says study investigator Lucas Tian, a postdoctoral researcher in the laboratory of Winrich Freiwald, professor of neurosciences and behavior at Rockefeller University. 

Second, neural activity in the region should form a distinct pattern for each simple symbol, and third, a monkey’s overall neural activity in response to a complex shape should comprise the activity that corresponds with each component symbol.

Only one of the eight regions, the ventral premotor cortex, showed “this really striking neural population that encodes these, what we call ‘action symbols,’” Tian says.

T

he drawing task design represents a notable advance because it “has the type of expressivity in the behaviors that hasn’t really been there before,” for example in previous studies in which monkeys combined a push, pull and turn of a lever, Buschman says. The new work “is really beautiful in the sense that it’s expanding out the scope of the types of behaviors that are being generated.”

It’s a surprise that these action symbols were encoded in just the ventral premotor cortex, Wilson says, because previous studies suggested that representing and implementing abstract rules in general is handled by the prefrontal cortex. “It’s striking in this case how specific [the neural activity] seems to be to this area,” he says.

In a way, “we lucked out, or maybe neuroscience at large lucked out” with this localized activity, because it “could have existed in a very distributed fashion,” Freiwald says.

Now the team wants to explore “the entire process of internally recombining these symbols to create new action representations,” Tian says. Freiwald likens this process to a symbolic computer program. In a preprint posted in January that further explores the drawing tasks, the team presented results on how the macaque brain organizes these internal symbols into a type of grammar in the pre-supplementary motor area, in the dorsomedial frontal cortex. 

“We’re beginning to see some components, like loops and clauses, and elementary symbols that you would have in a computer program, in different parts of the brain,” Freiwald says, which suggests that the brain’s inventiveness can be understood mechanistically: “That’s why I’m optimistic that this is really solvable.”