A bone to pick with ancient Chinese: AI Analysis of Handedness in Bone Inscriptions

Tracing the evolution of human habits and cognition through ancient artifacts offers unique insights into our past. This paper explores the handedness of ancient Chinese individuals through the study of oracle bone inscriptions—some of the earliest forms of Chinese writing, dating back to the Shang Dynasty approximately 3,000 years ago. Our research utilizes manually scanned real images of genuine oracle bone rubbings provided by National Museum of Chinese Writing. We have constructed the largest genuine oracle bone inscriptions dataset currently used in the field of computer technology, which presents unique challenges due to their variable and pictographic nature. Employing unsupervised deep learning techniques, we analyze the subtle stylistic differences in these images to discern whether the inscriptions were crafted by left-handed (sinistromanual) or right-handed (dextromanual) individuals. Our novel computational method, Bone2Vec, treats each pixel of the oracle bone image as a word in text, enabling us to embed and cluster these images to determine handedness patterns. Our findings not only advance our understanding of early Chinese script and its creators but also contribute to anthropological research by providing new evidence of handedness in ancient civilizations. This interdisciplinary approach underscores the potential of artificial intelligence in historical linguistics and archaeology, offering a fresh perspective on the cognitive behaviors of ancient societies.