Research On Translation Re-ranking Based On Quality Estimation

With the increase of international communication and the development of the Belt and Road Initiatives,Machine Translation technology has effectively alleviated the language barriers of communication between different countries and regions.In recent years,it is one of the hot spots in the machine translation to research the way about improving the quality of the output translation of machine translation system.In this paper,we propose a method of re-ranking the candidate translations based on neural network quality estimation.We use the latest joint neural network method to improve the quality of the output of machine translation.In order to verify that the unified neural network model(UNQE)quality estimation method can replace the automatic translation evaluation method to accurately re-rank the translation,we conduct experiments on the WMT19 Metrics Shared Task.The experimental results show that the UNQE can achieve the performance comparable to the automatic translation evaluation method with reference translation,which shows that the joint neural network translation quality estimation method can achieve the performance comparable to the automatic translation evaluation method with reference translation The translation quality estimation method can rank the quality of multiple output translations of the same source language sentence.In this paper,we propose a method of translation re-ranking based on neural network quality estimation.At the same time,in order to further improve the effect of translation re-ranking,we approximately take the score of translation quality estimation as the likelihood probability,and the performance of translation system in the development set as the prior probability to calculate the Bayesian posterior probability of translation,and use the posterior probability to output the translation of multiple translation systems Reorder and select the best machine translation.In order to verify the performance of this method,we conducted experiments on CCMT19 EnglishChinese and WMT18 English-German language pairs translation tasks respectively.The experimental results show that the proposed method significantly improves the quality of the output translation.Further experimental analysis reveals the advantages of the UNQE method in estimating the quality of multiple output translations.