一作ACL 2026Poster· Natural Language Processing
Memorization, Emergence, and Explaining Reversal Failures: A Controlled Study of Relational Semantics in LLM s
Yihua Zhu · Qianying Liu · Jiaxin Wang · Fei Cheng · Chaoran Liu · Akiko Aizawa · Sadao Kurohashi · Hidetoshi Shimodaira
摘要
Autoregressive LLMs perform well on relational tasks that require linking entities via relational words (e.g., father/son, friend), but it is unclear whether they learn the logical semantics of such relations (e.g., symmetry and inversion logic) and, if so, whether reversal-type failures arise fr...