ACL Anthology · ACL 2026:acl:2026.acl-long.65 · 记录 #1
On the Emergence and Test-Time Use of Structural Information in Large Language Models
Michelle Chao ChenMoritz MillerBernhard Schölkopf ELLIS Institute and Max Planck Institute for Intelligent Systems, Max-Planck InstituteSiyuan Guo Prior Labs
来源日期（仅到月份） 2026-07
来源记录日期 2026-07
雷达观测 2026-11-01 22:06
摘要
Learning structural information from observational data is central to producing new knowledge outside the training corpus. This holds for mechanistic understanding in scientific discovery as well as flexible test-time compositional generation. We thus study how language models learn abstract structures and utilize the learnt structural information at test-time. To ensure a controlled setup, we design a natural language dataset based on linguistic structural transformations. We empirically show that the emergence of learning structural information correlates with complex reasoning tasks, and that the ability to perform test-time compositional generation remains limited.