Shanghai team develops AI screener for language disorder in seniors

A team from Shanghai Jiao Tong University announced on Friday the successful development of China's first intelligent rapid screening system for language disorders in the elderly, which can help detect early signs of neurological and cognitive diseases, and thus promote early screening and diagnosis.

These diseases include dementia, stroke, and Parkinson's disease.

By answering questions based on audio content and describing pictures, an individual usually completes a test from the system in about seven minutes, and can obtain the result instantly.

The team said the system, which offers the tests at a cost of 0.01 yuan each time, is intended for routine checks in hospitals and community centers for the elderly, as well as for daily use in nursing homes and family settings via smart phones, tablets, and computers.

Unlike some developed countries, including the United States, Canada, and the United Kingdom, where such systems for assessing elderly language health are advanced and comprehensive, China has not previously explored this area, said experts. One of the core challenges is developing a scientific and appropriate set of test questions.

Led by Chang Hui, the team from the National Center for Language and Well-being at Shanghai Jiao Tong University leveraged its expertise in pathological linguistics and language testing to create a system that includes constructs, test items, and scoring standards.

They developed this screening system by also drawing on the university's experience of organizing the national College English Test.

To establish scientific evaluation standards, the team collected data from over 1,000 elderly individuals aged 50 to 79 across various regions in the country, considering factors such as age, education level, cognitive ability, gender, occupation, and dialect. Using this data, they applied regression norm-setting methods to build a hierarchical regression model and adhered to norm-referenced testing procedures for standard setting.

In order to develop an intelligent scoring and feedback system, the team collaborated with the School of Medical Devices at the University of Shanghai for Science and Technology.

The system they developed uses a language evaluation framework and large-scale pre-trained language models to address the inefficiencies and inconsistencies of traditional manual evaluations, enabling automated, multi-dimensional, and highly stable intelligent scoring of language expression abilities.

"To address the inherent volatility of large models in generative scoring, we developed a composite intelligent scoring method that combines 'multi-expert parallel scoring' with 'iterative self-calibration'," said Zhang Tianyi, a lecturer at the University of Shanghai for Science and Technology.

"This approach allows the system to learn from both correct and incorrect samples, enhancing scoring quality," she said.

Chang said, "The system is currently being piloted in several hospitals across China, including in the Xinjiang Uygur autonomous region as well as Gansu and Hainan provinces. The next step is to expand the question bank, so that an individual can take multiple tests in the system."

zhouwenting@chinadaily.com.cn