생성형 AI 담화의 미시구조 특성 :프롬프트 유형과 난이도에 따른 통사 및 어휘적 특성 분석 = Analyzing the Microstructure of Generative AI Discourse: The Effects of Prompt Type and Difficulty on Syntactic and Lexical Characteristics

다국어 초록 (Multilingual Abstract)

This study aims to generate discourse tasks required in clinical speech-language pathology using generative AI (ChatGPT-4o) and to systematically analyze their linguistic characteristics. For this purpose, a total of 270 discourse texts were generated by manipulating variables such as discourse type (expository, narrative), prompt condition (basic, structured, example-based), and task difficulty level (low, medium, high). The microstructure (syntactic complexity and lexical diversity) of each generated text was then statistically analyzed. The results showed that the AI-generated discourse exhibited distinct linguistic characteristics depending on its type. Expository discourse scored significantly higher on syntactic complexity indicators, while narrative discourse showed significantly higher values on lexical diversity indicators. The sophistication of the prompts and the level of difficulty had a significant impact on the syntactic and lexical features of the generated discourse. A complex interaction effect between prompt conditions and difficulty levels was also observed, confirming that the optimal prompt strategy varies depending on the context of the task. This suggests that through structured prompt design and difficulty adjustment, generative AI can be utilized as an effective content-assistance tool for creating high-quality, customized intervention materials.