Storytelling language sheds light on Parkinson’s memory issues

Language-based analyses of how people with Parkinson’s disease tell stories may provide a more nuanced look at how the disease affects memory and cognition, according to a study.

In fact, the data suggest that a language-based approach can distinguish between people with or without Parkinson’s more accurately than standard cognitive tests. The researchers called for further studies on how this type of language analysis might be deployed to monitor cognitive changes in patients with Parkinson’s.

An early-access version of the study, “Digitizing episodic memory assessments in Parkinson’s disease via natural language processing,” was published in npj Parkinson’s Disease.

Parkinson’s is a neurological disorder that often causes memory and cognition problems. One of the more common manifestations of Parkinson’s-related cognitive challenges is problems with episodic memory — that is, the ability to recall and recount specific events, such as a trip or a party.

A range of standardized cognitive tests is routinely used to check for memory issues in Parkinson’s. Most of these tests assess how well a person can recall facts — for example, by providing a list of numbers and later asking the person to recite it.

Recommended Reading

June 12, 2026 News by Andrea Lobo

New research shows Parkinson’s off episodes go beyond movement issues

Taking a nuanced approach

But memory isn’t just about recalling discrete facts. This is especially true for episodic memory: A person’s memory of an event includes not only what happened, but also their emotions and sensory experiences. Standard cognitive tests that assess only how well people recall facts may miss key nuances in how memory is affected in people with Parkinson’s.

An international team of scientists tested a different approach, using a computer-based analytical technique called natural language processing. They had patients read two stories — one action-based, focused on movement, and one more focused on internal emotions — and then recount both. A computer then analyzed the patients’ words.

The computer looks at three factors: verbosity, or how many different types of words the person uses; semantic acuity, which refers to how accurately the facts of the story are recalled; and organizational similarity, which evaluates whether the patient’s retelling matches the structure of the original story.

To test this approach, the researchers recruited 35 people with Parkinson’s and 39 control participants without the disease. They found that, for both types of stories, those with Parkinson’s showed reduced semantic acuity and lower organizational similarity compared with controls. Parkinson’s patients were also less verbose than controls on the action story, though verbosity was similar for the emotion-focused story.

The researchers then tested how well these language-based scores distinguished between people with Parkinson’s and controls. They used a statistical test called the area under the receiver operating characteristic curve (AUC), which measures how well a set of criteria can distinguish between two groups. Scores range from 0.5 to one, with higher numbers indicating better accuracy.

When the researchers used language-based scores, the AUC for distinguishing Parkinson’s patients from controls was 0.79, suggesting reasonably high accuracy. By contrast, when the researchers employed standard cognitive tests, the AUC was lower at 0.7. The language-based scores also accurately distinguished between Parkinson’s patients with or without diagnosed cognitive impairment, with an AUC of 0.87.

“Our full model yielded higher AUC values than standard cognitive tests,” the researchers wrote. “This pattern reinforces previous studies gauging the potential of [natural language processing] metrics to detect early, subtle cognitive changes before general screening tests.”

The scientists noted that the study was limited to a few dozen patients, so further work is needed to expand on the results and determine how best to translate the findings into clinical practice.