HMN 2026: How AI is unlocking new paths to recovery for bilingual aphasia patients

According to the National Aphasia Association, about 2 million U.S. citizens live with aphasia, but only two-thirds of Americans are aware of the condition, a communication disorder that often occurs after a stroke. It can affect many different neurophysiological processes related to communication, such as reading, speaking and gesturing.

June is National Aphasia Awareness Month, and BU’s Aphasia Resource Center helped advocate for Massachusetts Gov. Maura Healey to issue a proclamation recognizing it.

For bilingual people, aphasia presents additional challenges compared with those for people who are monolingual because “when someone knows more than one language, these languages share resources within the brain, but they also have individual and separate control mechanisms”, making rehabilitation more difficult. Some people may experience differences in how aphasia affects one language over the other or in the speed at which one language recovers after a stroke, for example.

Swathi Kiran is the founding director of the Center for Brain Recovery (CBR) at Boston University, and her work focuses on neuroscience, brain plasticity, language recovery and bilingualism. She and her team publish more than 30 papers each year.

In a study published in npj Digital Medicine, Kiran used an artificial intelligence model to predict which language would be most effective for recovery in bilingual aphasia patients, based on the patient’s “digital twin.”

In a Q&A, Kiran explains what aphasia is, discusses the recent research findings and explains how the use of artificial intelligence may help guide rehabilitation decisions in bilingual patients.

Boston University: Your study looked at people who have aphasia. Could you briefly explain what aphasia is and why it can be difficult to treat, especially for those who speak multiple languages?

Swathi Kiran: The difficulty or inability to speak or understand language is called aphasia. Aphasia usually occurs due to a stroke or brain injury. The most common type of stroke is ischemic stroke, when a vessel supplying blood to the brain is obstructed. Typically, a stroke occurs in the left middle cerebral artery, which supplies blood to parts of the brain involved in speech and language.

In multilingual individuals with aphasia, impairment typically affects all languages rather than just one, possibly because stroke disrupts the network that enables switching between languages.

This study used an AI model called BiLex to create ‘digital twins’ of patients. What is it, and what makes it different from the AI tools most people are familiar with?

BiLex is an AI system designed to act like a “digital twin”—a detailed, computer-based copy of an individual patient’s language system in the brain. Instead of just analyzing data, it tries to simulate how that person’s brain organizes and uses words across languages, based on lifetime language experience and specific impairment after stroke.

BiLex is a brain-inspired model of language, with separate systems for each language and shared meaning representations. It is personalized because it is trained to match one individual’s language history and language difficulties after stroke. And it can be “experimented on” safely—researchers can simulate brain damage and try different therapies to see what works best for that specific patient.

It is different from AI tools most people are familiar with in that it is not just pattern recognition from large data sets. Typical AI generates answers or images, but digital twin models like BiLex help researchers understand why language breaks down and how it can recover.

Before this study, how did clinicians decide which language a bilingual aphasia patient should focus on in therapy?

Before this study, clinicians usually decided which language a bilingual aphasia patient should focus on by (a) asking the patient what language they wanted to focus on and providing therapy in that language whether or not that was an optimal language, or (b) if they did not speak the languages the patient spoke, simply providing therapy in English, whether or not that was an optimal language.

What happened when patients followed the language the model recommended versus when they didn’t?

In this study, we made two observations. First, when simply looking at group-level differences between the patients who followed the model prescription versus the placebo, there were no significant differences between the groups.

However, when we subcategorized the patients into smaller cohorts organized by language history or by the severity of their aphasia, we found that the digital twin simulations accurately captured these differences between people. In other words, the digital twin was able to accurately capture differences in an individual’s language profile.

What are the next steps in this research? Are there areas that need to be explored further before your findings can be put into practice?

These results really showed us that bilingual aphasia recovery is complex and demonstrated the potential of computational models to guide rehabilitation strategies. As the field of computational modeling advances, these tools will become increasingly valuable for developing personalized therapy plans that account for the unique linguistic profiles of patients.

The content is provided for information purposes only.