基于注意力引导图卷积网络的中英机器翻译模型

Chinese-English Machine Translation Model Based on Attention-guided Graph Convolutional Network

现如今,神经网络在基于句序列的机器翻译模型已占据主流地位。但在中英文互译中,仅对单语句进行翻译不仅仅丢失语义信息,还破坏繁杂的逻辑构造,并不符合当代机器翻译需求。鉴于此,提出一种新型基于注意力引导图卷积网络的机器翻译优化模型,可通过多头注意力机制和图卷积神经网络结构的结合保留词元素特征及段落层次结构信息。为了验证基于注意力引导图卷积网络模型是否优于其他传统算法,在WMT21数据集上进行实验,结果表明各指标均达到理想效果。

Nowadays,neural networks have occupied a dominant position in sentence sequence-based machine translation models.However,in translation between Chinese and English,only translating a single sentence does not only lose semantic infor⁃mation but also destroys complicated logical structures,which does not meet the needs of contemporary machine translation.In view of this,a novel machine translation optimization model based on an attention-guided graph convolutional network(AGGCN)is pro⁃posed,which can retain word element features and paragraph hierarchical structure information through the combination of multi-head attention mechanism and graph convolutional neural network structure.To verify whether the AGGCN is superior to oth⁃er traditional algorithms,experiments are carried out on the WMT21 dataset,and the results show that all the indicators achieve ide⁃al results.