This study proposes a cross-cultural semantic association computational model for German literary translation, aiming to address the shortcomings of traditional machine translation in terms of cultural differences and semantic complexity. By fusing RNN and Self-Attention Network SAN encoder, the semantic associations of German words, sentences and paragraphs are quantified layer by layer, and a cross-language semantic ontology structure model is constructed to realize accurate semantic mapping between German and Chinese. The hybrid translation model is further designed to combine four pre-trained encoders with Marian decoder to optimize the translation generation in literary context. The experiments are based on the German-Chinese parallel corpus of WMT2018 and WMT2022.In the German-Chinese translation task, the correlation misalignment rate of this paper’s model is only 4.82%, which is 49.05% lower than that of the baseline model, QE-BERT.The BLEU value is up to 26.21, which is significantly better than that of the comparison model, e.g., 20.05 for Gen-Det.In addition, the model performs stably in the multi-language task. The highest BLEU value of 31.86 in German-English language translation and content word ratio experiments show its robustness to complex semantic elements. The study shows that cross-cultural semantic association computation and hybrid model design can effectively improve the accuracy and cultural appropriateness of literary translation
