Research On English-Chinese Neural Machine Translation Of Social Networks For Military-political Domain

Social network is an indispensable part of social life in the Internet era.It is not only a major platform for domestic and international exchanges and cooperation,but also the main battleground for competition and gaming among countries.Massive military-political information is disseminated in social networking,and its translation task has also become a necessary requirement for serving the national military,political and diplomatic strategies.In the context of big data,the traditional manual translation model can no longer meet the requirements for high-quality and on-demand translation of data in the military-political domain in social networks.On the other hand,the specificity of social network language performance poses new problems and challenges for understanding and translation.Therefore,it is of prominent theoretical significance and application value to study the characteristics of network language and automatic translation methods in military-political domain for social networks.Current English-Chinese machine translation based on deep neural networks has made remarkable progress,even achieving test scores of human translators in the field of journalism.However,the test results of social networking texts in the military-political domain using existing mainstream English-Chinese neural machine translation systems show that there are still many problems in the machine translations,and there is still a considerable gap to meet the practical application needs.To address the outstanding problems caused by the linguistic features of social networks in the military-political domain,this study explores the construction method of a neural machine translation model integrating a priori knowledge by constructing an network informal language expression terminology database and a social media corpus in the military-political domain,and effectively improves the translation quality of English-Chinese neural machine translation in the military-political domain of social networks.The main work of this paper is as follows:1.Analysis of English network language features in the military-political domain and construction network informal language expression terminology database.Network language has its own distinctive features.These features are analysed,summarised and extracted to build a terminology database of social network language features in military-political domain.2.Design and construction of a bilingual English-Chinese parallel corpus of social networks in the military-political domain.The quantity and quality of corpus are the basic conditions for training neural network machine translation model,and important factors affecting the performance of the model.However,the English-Chinese bilingual parallel corpus in the military-political domain of social networks faces great difficulties in both quantity and quality.This study proposes a bilingual vertical domain corpus generation method combining pre-processing means and machine translation engine for constructing,extending and optimizing an English-Chinese bilingual parallel corpus for social networks in the military-political domain,and verifies the effectiveness of the method through experiments.3.The construction of neural machine translation model integrating social network language features in military-political domain.This study tags the linguistic features of social networks in the military-political domain,constructs a word segmentation model based on the BPE algorithm,splices and aligns the linguistic features of English and Chinese training data at the encoding and decoding ends,and incorporates the features into fine-tuned neural network structures based on m Bart pre-trained model and Transformer neural network structures4.Experiments and analysis of results.The method of transfer learning based on the m Bart pre-training model was able to improve the BLEU values,but the overall translation results were poor.In terms of the Transformer structure,the integration of linguistic knowledge improved the BLEU value by 3.35 compared to non-integration of language knowledge,and the translation quality was significantly optimized at the lexical level,which demonstrates that the integration of linguistic features can have a positive effect on English-Chinese machine translation of the military-political domain content in the social networks.