Beyond English-centred aphasia technology: exposing the promise and perils of generative AI for Chinese aphasia communication

Background Aphasia and acquired dysgraphia can limit written communication and participation in daily life. AI chatbots can support functional writing by organizing and clarifying input. However, evidence on AI for aphasia is predominantly English-language, with little evidence of chatbot use in Chinese written communication.Aims This study evaluated generative AI chatbot support for functional written communication in native Chinese speakers with aphasia, assessing meaning preservation, linguistic accuracy, functional appropriateness, aphasia accessibility, cultural-pragmatic appropriateness, clinical safety, and recurring errors.Methods & procedures Anonymised written responses from 32 native Chinese speakers with aphasia were entered into six generative AI chatbots: ChatGPT, Gemini, Claude, ERNIE Bot, Kimi and DeepSeek. All chatbots were asked to rewrite the fragmented input into comprehensible Chinese, but not to add new information. The outputs were evaluated by four bilingual expert raters for meaning preservation, linguistic accuracy, functional appropriateness, aphasia accessibility, cultural-pragmatic appropriateness and clinical safety.Outcomes & results Overall expert ratings were generally high, with all systems receiving ratings above 4.0 on a 5-point scale for the usability of Chinese messages generated from fragmented aphasic input. The highest overall ratings were observed for ChatGPT, Claude, and Kimi. Meaning preservation, linguistic accuracy, and functional appropriateness received the highest ratings, while aphasia accessibility received the lowest. Higher CRRCAE (Chinese Rehabilitation Research Center Standard Aphasia Examination) language performance was associated with higher expert-rated output quality, whereas ratings were lower for participants with greater aphasia severity. Unsupported information and excessive changes in language and meaning were the most common errors.Conclusions Chatbots using generative AI have the potential to serve as complementary aids for Chinese written communication among persons with aphasia. However, direct user-centred evaluation is required before conclusions can be drawn regarding clinical usefulness, user accessibility, or preservation of communicative ownership.