Human and AI Preferences in Turkish Subject Realization: Linguistic and Educational Perspectives
Human and AI Preferences in Turkish Subject Realization: Linguistic and Educational Perspectives

Summary/Abstract: This study investigates whether ChatGPT exhibits human-like sensitivity to the discursive constraints governing the distribution of null and overt subjects in Turkish. The study compares acceptability judgments from 27 native speakers with corresponding evaluations produced by ChatGPT 5.0. Participants and the model evaluated 24 contextualized Turkish sentences representing four conditions: Topic continuity with null subject (TC-Null), topic continuity with overt subject (TC-Overt), contrastive focus with null subject (CF-Null), and contrastive focus with overt subject (CF-Overt). The results showed that TC-Null and CFOvert sentences were judged highly acceptable, while CF-Null sentences received the lowest ratings, and TC-Overt sentences were moderately acceptable by human participants. ChatGPT closely mirrored these patterns, with negligible-to-small effect size differences in three of the four conditions and a small-tomoderate difference only in the CF-Overt condition. Regarding the results, the findings show that ChatGPT is sensitive to subject distribution in Turkish. Further, the study evaluates the implications for language education and highlights the role of large language models (LLMs) as tools for teaching context-sensitive grammatical features in Turkish.

• Details
• Contents