臨床前對話模擬：評估會話式人工智慧在失語症治療中回應的可及性

// Preclinical Dialogue Simulation: Evaluating Response Accessibility in Conversational Artificial Intelligence for Aphasia Therapy

本研究開發ABCD模擬方法，評估大型語言模型（LLMs）產生適合失語症患者理解的臨床對話語言｜測試三種模型（Claude、GPT、Gemini），發現少量示範提示與進階推理策略可提升回應可讀性｜結果顯示模型效能異質，Gemini在零示範模式表現最佳｜ABCD框架有助於臨床前評估，促進LLMs在語言治療的安全應用。

// This study developed a simulation method (ABCD) to evaluate how large language models (LLMs) generate clinician language accessible to aphasic patients during therapeutic dialogue. They tested three LLMs (Claude, GPT, Gemini) with different prompting and reasoning strategies, measuring response accessibility using readability metrics. Results showed model-specific differences, with few-shot prompting and advanced reasoning improving accessibility, and Gemini excelling in zero-shot mode. The ABCD framework offers a scalable way to benchmark LLMs for clinical use before real-world application in speech therapy.

Published 2026年6月24日