San Antonio researchers outline AI study of veterans’ speech after mild brain injury

San Antonio researchers are investigating whether AI can help measure communication problems in veterans with mild traumatic brain injury. In a November 5 account, UT Health San Antonio detailed a $200,000, yearlong study that will analyze recorded stories for language patterns and compare them with clinical symptoms. The goal is evidence clinicians can use, not an automated replacement for their judgment.

The San Antonio Medical Foundation funded the proof-of-concept project, led by speech-language researcher Rocio Norman, psychologist Vivian Ta-Johnson and rehabilitation researcher Alicia Swan. The collaboration brings together expertise at the University of Texas at San Antonio and the South Texas Veterans Health Care System. Its ambition is to make detailed language analysis practical during a routine clinic visit.

When words and grammar are not the problem

Norman says the project grew out of a gap she encountered while treating veterans of the Iraq and Afghanistan wars. Some described trouble finding words, organizing thoughts or following fast-moving and noisy conversations, even when standard assessments did not show a clear impairment. Those difficulties could become apparent when veterans returned to college, work or conversations with loved ones.

Many adult language assessments were developed for people with stroke-related aphasia—a disorder affecting communication—or severe brain injury, Norman explains. Veterans with mild injuries may retain vocabulary and grammar but struggle to sequence ideas, keep a story coherent, respond to a listener or provide enough context. The distinction shifts attention from knowing words to using them effectively.

There are limits to other assessments, too. In an October 24 San Antonio Report article, Monica Verduzco-Gutierrez, who is not involved in the project, explained that concussion assessments do not show how patients function outside the clinic. CT and MRI scans can also appear normal when treatment is needed, she said.

The work is rooted in clinical experience with patients whose concerns could not be validated objectively.

Rocio Norman, study co-principal investigator

Recorded stories become measurable features

The team will use recordings already collected through a targeted auditory plasticity study Norman leads at the San Antonio VA. Veterans described how they were injured and recounted an important life event. Those recordings were gathered at multiple points during treatment, providing language samples from more than one moment in their care.

After transcription, natural language processing—AI methods for analyzing language—will convert patterns into measurable features. The planned measures include coherence, emotional tone, organization and level of detail. Ta-Johnson says the approach allows researchers to quantify how people organize and connect ideas, then examine those measurements alongside clinical and other data.

The rationale is that telling a story draws on several abilities at once. Norman points to attention, memory, planning and executive function—the skills used to organize and carry out tasks. A person may still have words available while finding it harder to arrange thoughts or adapt to a listener’s questions and nonverbal feedback.

A clinical aid still has to earn its place

The immediate test is whether automated analysis can identify communication differences and show how they relate to anxiety, post-traumatic stress, depression and sleep difficulties. This is a research question, not a reported finding or a validated diagnostic capability.

Detailed language analysis currently takes time-consuming transcription and coding, often involving multiple reviewers. The researchers envision an automated system that flags patterns for clinicians to consider alongside their observations and standard assessments. If the pilot produces meaningful findings, they plan to seek larger federal grants, study broader populations and investigate more specific cognitive difficulties reflected in language.