A Research on Machine Translation Error Analysis and Correction Method Based on Computer Vision and Machine Learning

There are often subtle errors in English machine translation. If these errors can be detected in a timely manner, it would greatly improve the quality and effectiveness of English translation, reduce barriers to human communication, and promote international communication and cooperation in English. Although the vast majority of grammar errors occur in a certain part of the discourse, there are also a few grammar errors that run through the entire discourse. This article intended to use an encoder-decoder (Transformer) model to perform syntax error correction, so that it can simultaneously consider the context and distance correlation in the text. This article adopted an adversarial learning framework of computer vision for correcting grammar errors to address the issues of exposure bias and loss evaluation mismatch in neural machine translation, including discriminators and generators. Due to the stronger data sparsity faced by syntax error correction, machine translation algorithms are difficult to apply. This article was based on the back translation data augmentation method to synthesize pseudo “error correction” parallel sentence pairs. This article presented a machine translation error analysis and correction method based on computer vision and machine learning, with a BLEU (Bilingual Evaluation Understudy) score of 90.5 points and a rule-based machine translation method with a BLEU score of 85.3 points. The performance of the machine translation error analysis and correction method in this article was good.

I. Introduction

Due to the prevalence of computer technology, communication between countries needs to increase, and communication barriers between people of different countries are becoming more prominent, leading to higher requirements for translation quality. Machine translation can mimic human language through computers for language conversion. With the development of the economy, many countries are paying more and more attention to machine translation. However, on account of shortcomings existing in any machine translation method, relying solely on one method cannot guarantee the correct translation, and ideal translation results cannot be achieved. This article proposes a machine translation error analysis and correction method based on computer vision and machine learning, which can improve the quality of translation.