Clause-Internal or Clause-External? Testing Turkish Reflexive Binding in Adapted versus Chain-of-Thought Large Language Models

Abstract

This study evaluates whether state‑of‑the‑art large language models capture the binding constraints that govern Turkish reflexive pronouns. We construct a balanced set of 100 sentences that pit local against non‑local antecedents for the reflexives kendi and kendisi and test two contrasting systems: an OpenAI chain‑of‑thought model designed for multi‑step reasoning and a LLaMA‑2 derivative extensively fine‑tuned on Turkish data. Antecedent choice is assessed via a combined sentence-level-perplexity and forced-choice paradigm. Trendyol places ≈70 % of its probability on local bindings, having a strong locality bias whereas O1 Mini splits probabilities almost evenly between local and long-distance readings. Thus, massive in-language fine-tuning can exaggerate structural cues, while reasoning-oriented supervision may ensure grammatical fidelity.