Specializing Large Models for Oracle Bone Script Interpretation via Component-Grounded Multimodal Knowledge Augmentation

2026-04-08

AI Summary

This study addresses the persistent “interpretation gap” in oracle bone script decipherment, stemming from the rarity of characters and their unique structural compositions, which hinder existing methods from effectively leveraging shared, semantically informative components. To bridge this gap, we propose a multimodal agent-based vision-language framework that, for the first time, integrates component-level semantic structures into oracle bone interpretation. Our approach combines fine-grained visual grounding, knowledge graph retrieval, and collaborative reasoning with large language models to construct interpretable and transferable interpretation chains. We introduce OB-Radix, a new expert-annotated dataset, and demonstrate significant improvements over current baselines across three benchmark tasks, achieving notable advances in both accuracy and descriptive richness, thereby substantially enhancing the capacity for automated oracle bone script decipherment.

Abstract

Deciphering ancient Chinese Oracle Bone Script (OBS) is a challenging task that offers insights into the beliefs, systems, and culture of the ancient era. Existing approaches treat decipherment as a closed-set image recognition problem, which fails to bridge the ``interpretation gap'': while individual characters are often unique and rare, they are composed of a limited set of recurring, pictographic components that carry transferable semantic meanings. To leverage this structural logic, we propose an agent-driven Vision-Language Model (VLM) framework that integrates a VLM for precise visual grounding with an LLM-based agent to automate a reasoning chain of component identification, graph-based knowledge retrieval, and relationship inference for linguistically accurate interpretation. To support this, we also introduce OB-Radix, an expert-annotated dataset providing structural and semantic data absent from prior corpora, comprising 1,022 character images (934 unique characters) and 1,853 fine-grained component images across 478 distinct components with verified explanations. By evaluating our system across three benchmarks of different tasks, we demonstrate that our framework yields more detailed and precise decipherments compared to baseline methods.

Problem

• Oracle Bone Script
• interpretation gap
• component-based decipherment
• multimodal knowledge
• ancient script interpretation

Innovation

• Oracle Bone Script
• Component-Grounded Reasoning
• Vision-Language Model
• Multimodal Knowledge Augmentation
• OB-Radix Dataset