A graph-based evolutionary dataset for oracle bone characters from inscriptions to modern Chinese scripts

Abstract

The traditional semantic interpretation of oracle bone characters, the earliest known system of Chinese writing, has relied heavily on expert-driven manual analysis. Therefore, artificial intelligence (AI)-based approaches have been increasingly explored for deciphering these characters. A fundamental prerequisite for AI-driven semantic inference is construction of a high-quality evolutionary dataset. In this study, a graph-based evolutionary dataset is introduced, encompassing 756 groups and 3780 Chinese characters across five historical stages. Unlike existing datasets that primarily represent characters as images, the proposed dataset employs a graph-based representation, wherein nodes correspond to key structural points of a character, and edges define their spatial relationships. Experimental analyses demonstrate that graph representations offer superior capabilities in capturing the structural stability of characters across evolutionary stages compared to image-based representations. The dataset is expected to serve as a valuable resource for the application of AI-driven methodologies in the decipherment of unknown oracle bone characters.