Completing missing portions of ancient Chinese texts is a critical task in textual research and cultural heritage preservation. This study proposes a pretraining model based on the fusion of phonetic and character shape features to improve the accuracy of ancient text completion. The proposed method decomposes the shape and pronunciation of Chinese characters to generate sub-character structures and integrates this information through a cross-attention mechanism, thereby enhancing the model’s semantic understanding of ancient texts. The methodology involves modelling Chinese characters using Wubi input codes and phonetic systems, encoding each character as a short sequence, and constructing a vocabulary through a sub-word segmentation algorithm. Experimental results show that the proposed method achieves a completion accuracy of 70.02% on the transmitted literature dataset and 63.76% on the unearthed literature dataset, demonstrating its effectiveness for the intelligent completion of ancient texts.