An international team of researchers, including neuroscientists at HSE University, has developed a computational model for simulating semantic dementia, a severe neurodegenerative condition that progressively deprives patients of their ability to comprehend the meaning of words. The neural network model represents processes occurring in the brain regions critical for language function. The results indicate that initially, the patient's brain forgets the meanings of object-related words, followed by action-related words. Additionally, the degradation of white matter tends to produce more severe language impairments than the decay of grey matter. The study findings have been published in Scientific Reports.
People comprehend the meaning of spoken words because their brains have the ability to encode, store, and use semantic information, ie knowledge about objects, phenomena, actions, and the content of various concepts acquired throughout our lives. Understanding the meanings of words is what enables an individual to engage with the external world: communicate with others, articulate one's thoughts, and effectively participate in society.
Any disruption of this ability can therefore have severe adverse consequences. In semantic dementia, a degenerative neurological condition, individuals progressively lose their ability to comprehend speech addressed to them, forgetting words and their meanings. For instance, a patient may refer to sitting on ‘that thing’ when the word 'chair' eludes their semantic memory, or they may fail to correctly name a dog as opposed to a cat, despite being able to mechanically repeat both words without difficulty.
Meanwhile, the question of how precisely the semantic system of the human brain operates at the neuronal level and how it is impacted by this kind of severe disorder remains unanswered. It is understood that the processes involved in encoding and storing the meanings of words engage at least the frontal, temporal, and occipital lobes of the cerebral cortex, primarily within the left hemisphere.
To develop effective strategies for assisting individuals afflicted by semantic dementia and potentially to prevent its severe progression, it is essential to understand the nature of the condition and its underlying neuronal mechanisms. Given the absence of live biological models, as language is unique to humans, neurocomputational modelling is one of the most objective approaches to studying brain processes under controlled experimental conditions.
Researchers at HSE University, in collaboration with an international team of scientists, have developed a neurocomputational model of the brain designed to mimic language function as accurately as possible, drawing from a biological understanding of the brain regions involved in language processing and their interconnections. The scientists trained the model to recognise and comprehend various words, then proceeded to simulate different types of brain tissue damage to observe what happens to the brain during the progression of semantic dementia, a condition profoundly detrimental to speech and intelligence.
For the first time, we simulated semantic dementia using a realistic computational model of speech function. Studying the disease's progression in actual patients is not always possible due to ethical constraints and the challenges of establishing strict experimental conditions, as we cannot control a disease that varies significantly from patient to patient. Indeed, healthy individuals also exhibit variations in their individual characteristics. We have developed a model of human neuronal networks and their activity, so that it closely mimics the brain's functioning in language processing. Certainly, every model is a simplification, but it's just as essential as, for instance, a map that provides us with a simplified schematic representation of the terrain.
The researchers simulated progressive 'damage' to the model to understand the events occurring at different stages of the disease: initially, the model was only slightly damaged, simulating the early stage of dementia, followed by gradual escalation in damage severity and destruction of ‘tissues’ in the programmatically modelled brain regions. The experiment was conducted in two versions, simulating damage to both grey and white matter.
The results of the experiment confirmed clinical findings indicating that in semantic dementia, tissue degradation occurs in various brain regions, affecting not only areas critical for speech comprehension but also those responsible for motor activity and object perception. The model illustrates that damage can impact both grey matter and neural pathways, with the latter resulting in more severe problems. Furthermore, memory traces of words of various types—those denoting objects, actions, or abstract concepts—are affected differently.
We still lack precise knowledge regarding when and where dementia begins, as physical brain disorders can remain unnoticed for a long time due to cognitive reserve in individuals. However, we have observed that more severe disruption occurs when the connections between different parts of the brain are damaged. Moreover, we have noticed that words associated with objects, ie nouns, tend to fade from the vocabulary more rapidly than those indicating actions, ie verbs. These findings can guide us in the right direction when working with patients, providing insights for a more effective approach to the diagnosis and treatment of semantic dementia.
Yuri Shtyrov
Professor, primary author of the study, Leading Research Fellow, Institute for Cognitive Neuroscience, HSE University
IQ
Yuri Shtyrov
Professor, primary author of the study, Leading Research Fellow, Institute for Cognitive Neuroscience, HSE University