The "AI Recognition of Oracle Bone Divination Signs, Drill Marks, and Prognostication Texts" project extends our team's previous digital humanities research. It integrates expertise from emerging oracle bone scholars and computer scientists with diverse backgrounds, aiming to leverage generative AI for interpreting divination cracks, drilling marks, and auspicious/inauspicious signs on oracle bones from Yinxu. This year's project focuses on two primary objectives:

The first objective involves expanding the existing text-based database to incorporate visual data, including color and black-and-white photographs, rubbings, and hand copies. This enhancement prepares the database for the AI era, establishing a foundation for future cross-referencing analyses. This year, we intend to develop the database architecture and incorporate approximately 500 of the most comprehensive divination turtle shells from the Institute of History and Philology, Academia Sinica.

The second objective centers on training AI for specific reading capabilities. This includes the automatic identification and enumeration of divination cracks on the obverse of oracle bones and their corresponding drilling marks on the reverse. We aim to train AI to interpret various prognostications from photographs and rubbings, such as "auspicious," "greatly auspicious," "vastly auspicious," and "not dark and obscure," as well as numerical sequences. These 500 specimens will serve as both training data and reading targets for the AI system.

Regarding the auspicious/inauspicious nature of oracle bone divinations, previous research by Zhang Bingquan from the Institute of History and Philology hypothesized a correlation between the angle of horizontal cracks and divinatory outcomes. However, his findings disproved this hypothesis. Our research posits that the "head, body, and foot" of horizontal cracks, as mentioned in the "Biographies of Turtles and Milfoil" in the Records of the Grand Historian(Shiji), possess various characteristics that influence the auspicious/inauspicious nature of different divination categories. Given AI's superior sensitivity in image analysis compared to human perception, our team will utilize this capability to analyze the correspondences and patterns among prognostications, drilling marks, and divination cracks. Through this approach, we aim to identify key indicators for determining auspicious/inauspicious signs in oracle bone inscriptions.