作者位次 #5/7ACL 2026Poster· Natural Language Processing
Revealing Procedural Reasoning Structures in Chain-of-Thought Training via Span-Level Gradient Organization
Jia Liu · Jiaxin Luo · Weiwen Xu · Jonathan M. Garibaldi · Xiao-Kun Wu · Yixue Hao · Min Chen
摘要
Chain-of-Thought (CoT) prompting enables large language models to produce multi-step reasoning, yet how such reasoning-related structure is expressed during training remains poorly understood. We present Gradient-based Structural Developer (GSD), an unsupervised framework with a principled gradie...