Out of Africa: unravelling the mysteries of human generative creativity and the evolutionary origins of language – Fellows’ seminar by Robert Berwick

15 August 2025

“The origins of language are a bit of a mystery – a whodunnit,” said Robert Berwick of the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology and the Department of Computer Science at Virginia Tech. “We know from Darwin and Alfred Russel Wallace that it all came from Africa and we can now estimate when – at the latest, 135 000 years ago. But its origins still pose a mystery. At first glance, Language seems to evade explanation by natural selection and adaptation – as Wallace wrote to Darwin, it appears to be ’more than Nature needs’. This exception horrified Darwin but the mystery remained.

Berwick was presenting the first fellow’s seminar of the second semester of 2025.

He explained that language is part of what makes us human and differentiates us from all other animals.

“At least since Darwin, we have known that in part humanity’s uniqueness arose from our ancestors in Southern Africa, and, as he argued, the origin of language,” continued Berwick. “In our book Why Only Us, Noam Chomsky and I agreed, gathering evidence for a rapidly evolved novel cognitive computation that seemingly first appeared at least 135 000 years ago in Southern Africa, a simple basic operation that underpins the syntax of all the various human languages we see today, along with human language’s apparently unbounded creativity.”

Berwick’s work has involved trying to unpack the what, who, where, when, how and why of this mystery with his STIAS project focusing specifically in on the evidence for where, when and how language arose.

He described it as a controversial field requiring many assumptions.

“The project draws on the rich genomic analyses of those broadly agreed to be the most genetically ancient human lineages preserved in Africa to this day leading to Khoe-Săn populations – to explore exactly when and how language could have led to the emergence of a new human species – Out of Africa.”

“The split into San and groups spanning the rest of Africa and Eurasia occurred roughly 300 000 years ago. Khoe-Săn are genetically diverse and distinctive and remain so. They provide a baseline of what the original split from the full human lineage looked like.

“By doing genomic scans we can work our way backwards through the gene ancestry to arrive at these conclusions.”

Berwick highlighted that language-driven cognition seems to have developed quite quickly after this. Findings like those from the Blombos Cave 300 kms from Cape Town date to about 90 000 years ago and what is believed to be the first horse figurine carved by humans found in the Vogelherd Cave in Germany dated to 35 000 years ago provide evidence of an accelerated pace (in evolutionary terms!), presumably cultural evolution that was faster than evolution by natural selection.

“It was a rapid evolution – not an incremental process – not ’evolution by creeps’ but rather ’evolution by jerks’,” he added.

But what exactly is language?

Berwick explained that Aristotle defined language as sound paired with meaning. This requires internal thought in some primal brain structure plus externalisation via sound or gesture.

“Language requires some kind of computation that relates meaning and sound,” he said. “Vocal learning was already in place – we know that from bird song – and the biological elements were already evolved. The new trait was the computation that linked the two – the central component that builds words into complex structures. Once you had that you could link sounds and meaning.”

He referred to the work of his colleague Angela Friederici looking at brain structure which has found that the human brain wiring ‘ring’ reflects an evolutionary novel division. This is a thick white matter fibre that links three parts of the brain in a circle.

This is absent in other primates like chimps and macaques which means they don’t have the same language capabilities as humans – although some animals can deal with more word-like principles than others. Also in humans this ring circle is not completely formed at birth – it starts to link by about age two which is why babies are not born speaking.

No just beads on a string

Berwick explained that the central processing unit that connects sound to meaning structure is not as simple as that in other non-human animals – it assembles structure not like beads on a string but rather like a non-linear triangles.

This triangle has three components: a single basic computational rule as mentioned above, which, together with lexical items, constitute the basis of the language system, and two interfaces through which mental expressions are connected to the external world (external sensory-motor interface) and to the internal mental world (an internal conceptual-intentional interface).

“At some point there was a change from linear beads on a string which many animals can do, to triangles in the brain that are not linear, and found only in us.”

Using a mobile to demonstrate, he said: “Human language formation is like a mobile – there is no order to it. It has a non-associative order structure. The mobile inside the head enables language.”

He explained that he and Chomsky had labelled this evolved computational operation, forged by a simple rewiring of the brain, and underpinning all human language we observe today as Merge. “The idea behind Merge is simple,” he said. “Everyone has experienced putting two thoughts together to get a new, bigger one — an experience extending to two notes, two brushstrokes, or two footsteps. And typically, once we’ve got hold of a new thought, we can file it away, reusing it. Merge is this principle applied to language.”

“The language CPU puts ‘ate’ and ‘apples’ together to get ‘ate apples’. This Merge computation can then be extended. The linear structure is an illusion. The triangle structure creates different distances between words leading to human language syntax.”

In short: “What reaches the ear is ordered, what reaches the mind is unordered.”

He also pointed out that this is where humans differ from AI and large language models like ChatGPT which predict the next word – like beads on a sting.

“The STIAS part of my work is looking at where, when and how this Merge option came into being. It is new, speculative and tentative,” he added.

Clues as to how Merge evolved?

Berwick explained that we can probe this emergence by looking at the genomics of Neanderthals, who mostly went to Europe, and humans (hominids), using the fact that there was little successful archaic interbreeding between the two, particularly in Africa, but some more recent interbreeding in Europe and Asia.

This implies there are differences in the human genome that make it possible to compare the two species and to ascertain which differences are linked to language. “Thus far we are seeing that those related to language are incompatible with modern humans on Neanderthal side − for example, one of the genes on chromosome 7 – FOXP2 – that has to do with motor articulation for speech, among other properties.”

But it’s no small task.

“About 569 genes so far are known to be different between modern humans and Neandertals. We need to examine each to see how it is related to language.”

“For the language ‘CPU’, Merge, we don’t know as much,” he continued “For now, we are looking at two genomic regions. Fossil remains show that Neanderthals had greater average cranial capacity –bigger brains—but this was to a large extent located in a different region than modern humans – stretching further back in the occipital lobe, relating to other cognitive skills like visual mapping. In humans the frontal cortex is bigger.”

He also pointed to the challenges of acquiring data. “Evidence is the issue,” he said. “The further back we go, the less evidence is available. There are lots of areas where there are no data available.”

And why is it important to know about the origins of language? Berwick emphasised that the development of language is part of how we became a new species – anatomically modern humans.

“It’s a complex problem that we must unravel. It’s not a reductionist view but parts of it can be now simplified and properly analysed.”

He concluded with a quote from 17th century Cartesian rationalists Antoine Arnauld and Claude Lancelot who emphasised that language is what reveals thought: “By combining sounds that have nothing to do with what happens in the head we can convey all sensibilities of thought and feeling. This is the true beauty of language.”

Michelle Galloway: Part-time media officer at STIAS
Photograph: SCPS Photography

 

 

Share this post:

Share on whatsapp
WhatsApp
Share on email
Email
Share on facebook
Facebook
Share on twitter
Twitter
Share on linkedin
LinkedIn

Subscribe to posts like these:

STIAS is a creative space for the mind.