AI는 ‘보편 청중’이 될 수 있는가? －페렐만의 신수사학에 기반한 한국어 논증 텍스트 피드백 분석－

초록

This study examines whether six generative AI models can perform the role of Perelman’s “universal audience” in responding to Korean argumentative texts, using the concept as an analytic criterion. The universal audience is a rational judge free from particular group interests, evaluating argument validity by audience agreement rather than logical form. Three Korean texts of varying argumentative quality (high, mid, low) were used as zero-shot stimulus texts for six AI models, with responses analyzed via three criteria: judgment directionality, use of universal grounds, and level-sensitivity. Results show that three general-purpose global models satisfied all three criteria, while the remaining models maintained similar judgments regardless of argumentative quality. The Korean-specialized model’s lower performance suggests that language processing and argument-evaluation are distinct competencies. These findings indicate that AI feedback should be used differentially by model performance, alongside cultivating learners’ feedback literacy. This study offers a rhetorical foundation for designing AI-based Korean writing feedback by applying Perelman’s universal audience concept to AI’s argumentative judgment.

키워드