摘要
针对基于编码/解码的中英文机器翻译收敛速度慢和准确率不高的问题,提出了一种改进的机器翻译模型。该模型采用长短时记忆循环神经网络实现词向量生成,然后在编码阶段利用组嵌入方法提高模型训练效率,最后在解码阶段加入权值衰减因子提高模型翻译准确性。实验结果表明,改进模型能够有效降低机器翻译训练的迭代次数,且具有较高的翻译准确率。
Aiming at the problem of slow convergence and low accuracy of Chinese-English machine translation based on cod⁃ing-decoding,an improved machine translation model is proposed.The model uses long-term and short-term memory cyclic neural network to generate word vectors,then group embedding method is used to improve the training efficiency of the model in the coding stage,and finally weight attenuation factor in the decoding stage is added to improve the accuracy of model translation.The experi⁃mental results show that the improved model can effectively reduce the number of iterations in machine translation training,and has higher translation accuracy.
作者
董斌
DONG Bin(Mingde College,Northwestern Polytechnical University,Xi'an 710124)
出处
《计算机与数字工程》
2021年第6期1253-1257,共5页
Computer & Digital Engineering
关键词
机器翻译
循环神经网络
组嵌入
权值衰减
machine translation
cyclic neural network
group embedding
weight decay