What language analysis can tell us about brain health
Neurodegenerative diseases are conditions where brain cells (neurons) progressively lose function and die, which can lead to cognitive impairments. Among the most common neurodegenerative conditions are Alzheimer’s (AD) and Parkinson’s disease (PD). “As we age, natural biological changes – like shrinking of brain tissue, loss of energy in cells and reduced ability to repair damage – make the brain more vulnerable,” explains MULTI-LAND project fellow Lucia Amoruso from the Basque Center on Cognition, Brain and Language (BCBL) in Spain. “Factors such as genetics and environmental aspects also play key roles.”
Analysing natural language markers
The MULTI-LAND project, supported by the Marie Skłodowska-Curie Actions programme, sought to develop more accurate and efficient ways of diagnosing and monitoring brain health, through the use of language. Amoruso built on a tried-and-tested method of analysing a patient’s speech – the difference was that she applied artificial intelligence (AI) and machine learning tools to extract far more information than before. “Usually, you might ask a patient to say as many words as possible in a minute beginning with the letter P,” she says. “You would then simply count the number of correct words and make a judgement on that. What I wanted to do was apply AI tools to look for linguistic features like the frequency of the words used, as well as things like speech rate, pitch and pauses.” Amoruso thought that these natural language markers, as they are called, could hold a great deal of information that could help medical staff to characterise disorders and more accurately predict symptom severity. “This approach is low-cost, scalable, and only requires patients to speak,” she adds. “This makes it both accessible and patient-friendly.”
Speech and brain imaging data
In collaboration with the University of San Andrés (website in Spanish) in Argentina, the project analysed speech data from AD and PD patients, as well as speech from healthy people. Machine learning algorithms were then trained to identify whether the speaker was a patient or healthy control subject, based on their speech pattern. Through this AI-driven analysis, the research team was able to identify distinct patterns that differentiated patients from the healthy controls. “For example, AD patients tended to use frequent, less specific words, such as ‘flower’ instead of ‘rose’,” says Amoruso. “These patterns were linked to brain atrophy and reduced connectivity in memory-related networks. PD patients, on the other hand, preferred concrete words, such as ‘piano’ over ‘symphony’, and struggled to shift between ideas.”
Reliable tools for neurodegenerative disease monitoring
Importantly, Amoruso was able to determine that these patterns were the same across Latin American and European Spanish-speaking populations. This underscores the potential of natural language markers (NLMs) as reliable tools for detecting and monitoring neurodegenerative diseases across cultural contexts. “Next steps include testing these markers in more languages and identifying those that remain robust beyond linguistic and cultural differences, and socio-biological variables,” notes Amoruso. “The ultimate goal is to develop user-friendly tools – such as mobile apps – that integrate NLMs into routine clinical evaluations. This would improve diagnosis and monitoring for these conditions.” Amoruso’s ultimate hope is that the project’s work will help to redefine how we detect and manage neurodegenerative diseases moving forward. “By opening the door to affordable, cost-effective and non-invasive tools, we can promote equitable access to diagnostic technologies,” adds Amoruso. “Ultimately, this will empower individuals, reduce health disparities, and improve the quality of life for ageing populations around the globe.”
Keywords
MULTI-LAND, language, brain, neurodegenerative, Alzheimer’s Parkinson’s, genetics
