OracleAnalyser: Analysing Implicit Semantics of Oracle Bone Scripts through MLLMs with Post-training

2026-06-24

AI Summary

This work addresses the limited semantic modeling in existing oracle bone script research, which has predominantly focused on character recognition. To bridge this gap, the authors propose a multi-stage post-training inference framework that fine-tunes the Qwen2.5-VL-3B-Instruct model and incorporates a novel Stable Focal Preference Optimization (SFPO) algorithm tailored to the unique characteristics of oracle bone script data. The study introduces the first reasoning and preference dataset along with an evaluation benchmark specifically designed for oracle bone script. Remarkably, the proposed approach achieves superior semantic analysis performance—even outperforming significantly larger models—despite utilizing only a 3-billion-parameter architecture, thereby demonstrating exceptional efficacy in capturing the deep semantics of oracle bone inscriptions.

Abstract

With the advancement of artificial intelligence, research on oracle bone scripts has entered a new era. However, existing methods and benchmarks remain largely confined to recognition tasks, overlooking the equally crucial aspect of oracle bone analysis. To address this gap, we propose OracleAnalyser, a reasoning framework for oracle bone analysis based on post-training techniques. Specifically, we fine-tune Qwen2.5-VL-3B-Instruct through multiple post-training stages and introduce a new preference optimization algorithm, Stable Focal Preference Optimization (SFPO), tailored to the characteristics of oracle bone datasets. In addition, we release both an oracle bone reasoning dataset and an oracle bone preference dataset, and further construct a new benchmark to evaluate models' analytical capabilities for oracle bone scripts. Extensive experiments validate the superior analytical performance of OracleAnalyser, which achieves remarkable results with only 3B parameters, surpassing models with substantially larger scales.

Problem

• oracle bone scripts
• semantic analysis
• artificial intelligence
• reasoning framework
• post-training

Innovation

• Oracle Bone Script Analysis
• Post-training
• Preference Optimization
• Multimodal LLMs
• SFPO