Research On Ancient Chinese Translation Method Based On Deep Learning

Traditional Chinese culture is an important spiritual treasure of the Chinese nation and a crucial component of Chinese culture.Ancient texts are the primary carrier of traditional Chinese culture,yet significant linguistic and temporal differences between ancient and modern Chinese make it increasingly challenging for modern readers to directly comprehend these texts.Currently,manual translation remains the dominant method for translating ancient texts,but this approach is both expensive and insufficient due to the lack of digital translations for a considerable amount of ancient texts into modern Chinese.With the rapid development of the Internet and artificial intelligence,machine translation technology based on deep learning has been widely applied.It is important to realize the automatic translation of ancient texts through machine translation technology to facilitate the dissemination and preservation of traditional Chinese culture.Therefore,this study aims to construct a deep learning-based machine translation model capable of performing automatic translation from ancient to modern Chinese.The main research content of this paper is as follows:(1)This paper proposes a machine translation method for classical Chinese based on semantic information-sharing Transformer.Building upon the Transformer,the proposed method achieves sharing of semantic information for identical vocabulary between classical and modern Chinese by sharing vocabulary and embedding layer parameters.The shared vocabulary enables the encoder and decoder to utilize semantically similar words,effectively reducing low-frequency words and improving the model’s generalization ability.Sharing embedding layer parameters can better model the consistency of semantically equivalent words during translation by taking advantage of the fact that classical and modern Chinese belong to the same language family.It also reduces the number of model parameters and speeds up convergence.Experimental results show that the proposed model achieves a BLEU score of31.43,which improves the traditional seq2 seq models based on GRU and LSTM by 26.94 and15.19 BLEU scores respectively,and outperforms the baseline Transformer model by 13.41 BLEU scores,demonstrating its effectiveness.(2)This paper proposes a classical Chinese translation method based on a pre-trained language model.Traditional translation models struggle with the significant contextual differences between classical and modern Chinese,so this paper introduces pre-trained language models.Pre-trained language models can learn language structures and cultural background knowledge from large-scale text corpora,enabling better understanding of cultural connotations and contextual information in classical Chinese.This is especially helpful for accurately translating cultural features,idioms,and allusions in classical Chinese.This paper utilizes Guwen-BERT to learn classical Chinese features and Chinese Ro BERTa to learn modern Chinese features,then builds a seq2 seq model that is compatible with the pre-trained language models.The model is fine-tuned using parallel corpora of classical and modern Chinese.Experimental results show that the proposed model achieves a highest BLEU score of39.19,effectively addressing the translation problem caused by differences in classical Chinese background knowledge,producing higher-quality translations.