From Symbolic Perception to Logical Deduction: A Framework for Guiding Language Models in Geometric Reasoning
A modular LLM system parsed geometry diagrams into symbols, then used formal deduction to solve them.
The paper says this setup can match state-of-the-art multimodal models on complex plane-geometry problems while being more interpretable. Its pipeline combines a Geometric Vision Parser with a Symbolic Solver to reduce hallucinations and make reasoning steps explicit. The authors tested it on a new benchmark drawn from 2025 Chinese Zhongkao exam problems, meant to avoid stale data and probe deeper deduction. ArXiv · AI/CL/LG's note
The paper says this setup can match state-of-the-art multimodal models on complex plane-geometry problems while being more interpretable. Its pipeline combines a Geometric Vision Parser with a Symbolic Solver to reduce hallucinations and make reasoning steps explicit. The authors tested it on a new benchmark drawn from 2025 Chinese Zhongkao exam problems, meant to avoid stale data and probe deeper deduction. ArXiv · AI/CL/LG's note
score 5