From Black-Box to Think-Aloud: Probing the Cognitive Signatures of LLM Translation through Self-Rationalisations

The emergence of large language models (LLMs) has revolutionized machine translation, yet their internal cognitive processes remain largely opaque. This study bridges cognitive linguistics and artificial intelligence by analyzing LLM self-rationalizations as artificial think-aloud protocols, investigating how these systems exhibit cognitive linguistic signatures during Chinese-English translation tasks.