Research On Neural Machine Translation Based On Re-decoding

Machine Translation(MT)refers to the process of automatically converting one language(source language)into another natural language(target language)by using a machine(computer).Theoretically speaking,the research of translation between different languages involves numbers of disciplines and technologies,such as computer science,linguistics,mathematics,and logic,etc.It is of great research significance.With the increasing frequency of international exchanges and the promotion of China’s Belt and Road strategy,machine translation technology has effectively alleviated the language barriers in communication between people in different regions and countries.Research on ways to improve the quality of machine translation is one of the hot topics in the field of machine translation.In recent years,deep learning technology has made great progress in the field of machine translation,and has surpassed traditional machine translation methods in many language pairs.The sequence transduction models based on Transformer is one of the best performing machine translation models.The model usually generates tokens one by one from left to right;therefore,it lacks the guidance of future contextual information,which will lead to under-translation and the decreasing of the machine translation quality.To alleviate this issue,we propose a neural machine translation model based on re-decoding.The model treats the generated machine translation outputs as approximate contextual environment of the target language,and then re-decodes each token in the machine translation output successively.The masked multi-head attention of the Transformer decoder only masks the current position token in the generated translation output.As a result,every token re-decoded can make full use of its contextual information.Experimental results on several test sets of the WMT machine translation evaluation task show that the quality of machine translation improved significantly by leveraging the neural machine translation method based on re-decoding.In this paper,the research on neural machine translation(NMT)based on re-decoding is validated and analyzed on the WMT2019,WMT2018 and WMT2017 news field machine translation datasets.In addition,ablation experiments are carried out for different occlusion methods and the number of decoder layers to verify the effectiveness of the TransRedecoder model in improving the quality of re-decoded translation.The experimental results show that the neural machine translation method based on re-decoding makes the target word generated at each moment depend not only on the above information of the word,but also on the following information of the word in the process of generating the re-decoded translation.,to alleviate the problem of under-translation caused by the lack of the following text,thus significantly improving the quality of machine translation in the target language.