Study on the connection between language and thinking by neuroscientists at Freie Universität Berlin published
People can learn one or more languages almost effortlessly. To do so, they not only have to learn how to pronounce words, but also to associate these words with content - with concepts such as ’coffee’, ’Monday’ and ’beauty’. But how does this work in the neural networks of the brain? And can words help in learning concepts?
To answer these questions, Friedemann Pulvermüller and his research team are developing neuronal networks that are structurally modeled on the human brain and based on findings from neurobiology. These networks are not only divided into ’areas’ that resemble those of the human brain, but also the connection structure between these areas has been modeled on the cerebral cortex of humans. The areas consist of groups of artificial ’nerve cells’, which in turn communicate with each other via local connections. These individual ’neurons’ can strengthen their connections when they are active together, or weaken them when they are active independently. This learning principle, known as Hebbian learning, has been well studied in biological systems. Using these brain-constrained networks, researchers can test neurobiologically based theories of language and cognition and explain cognitive phenomena.
For example, they can make these networks perceive "objects" in simulated "perception experiments." In other learning tasks, they are provided with linguistic information. It has been shown that the networks effortlessly learn which words can be used for which objects.
What is particularly interesting and unexpected for the researchers is that within the brain-like networks, highly interconnected nerve cell populations emerge that act as the biological basis of concepts. These nerve cell populations are active not only for specific objects, but also for entire classes of similar objects and entities, such as ’robots’, ’cats’ or ’sunrises’. Even for new, previously unobserved objects, the networks activate the relevant conceptual ’nerve cell circuitry’. When learning linguistic expressions simultaneously, this concept formation is even more efficient and faster than in non-linguistic learning: "These results suggest that language can support and accelerate concept formation at the biological level," emphasizes neurobiologist Friedemann Pulvermüller.
The influence of language on the formation of abstract concepts such as "beauty" or "peace" is particularly pronounced. These comprise many different sensory impressions that cannot be grasped by the biological learning mechanism because of their diversity - a painting, a sunset, and a concert do not have much in common, but can all be "beautiful. The diversity of these sensory impressions initially makes it impossible for the biological learning mechanism to grasp certain similarities until the network is taught a linguistic basis for each concept: Only then does it also form abstract concept representations, but these are closely fused with the linguistic representations.
This new research suggests that the influence of language on our thinking is much stronger and more important than previously thought. Although Wilhelm von Humboldt and a number of linguists have pointed out that thinking and language are interrelated, the idea that language strongly influences our thinking has been rather unpopular or at least highly controversial among linguists. The new results with brain-like networks now show a strong influence of language on concept formation in simulation experiments. They also suggest a neurobiological mechanism for the causal influence of language on thinking.
The research was conducted as part of an ERC Advanced Grant project funded by the European Research Council entitled ,,Material Constraints Enabling Human Cognition" or ,,MatCo" (ERC-AdG 883811), which is led by Friedemann Pulvermüller in the Department of Philosophy and Humanities at Freie Universität. (cxm)
Abstract concepts in the brain-like network. The neural network is modeled on the human cerebral cortex. The researchers have simulated areas of the cerebral cortex that are relevant for hearing (blue) and articulation (red to pink). These areas are very important for speech. In addition, they have simulated areas that are important for vision (green) and hand movements (brown to yellow) and contribute to the processing of action and object words. Shown here is the ’neuron circuit’ of one of the abstract sensory inputs learned by the network in the study. Each colored square corresponds to an active neuron. Without a word, only a few neurons are active in the ’nerve cell circuit’, and they respond only briefly to the entire concept: the concept has not been learned. With the word learned, there are significantly more. The ’conceptual neuron circuit’ also remains active longer with the word - and thus longer ’in the memory’ of the network.
