An Automatic Error Recognition approach for Machine Translation Results based on Deep Learning

At present, the speedy growth of natural-language-processing(NLP) has produced considerable evolution for the topic of machine-translation. There have been many deep neural network–based machine-translation(MT) methods that have become increasingly broad. Still, there currently lacks reliable automatic-error-detection(AED) systems for MT outcomes. One important metric for gauging the efficacy of an engineering English translation application is the accuracy of the corresponding deep learning model’s translation. This research presents a deep learning-based solution for automatic mistake identification in machine translation results as a means of closing this gap. In this research, we propose a deep generative model to synthesis the training data that will be utilized to train a model to correct grammatical errors in English used in international trade. Next, the learner’s corpus is corrected using the grammatical fault alteration approach, and “error-correct” sentence pairs are created using the exacted aim sentences and by hand interpreted standard sentences and fed back to the error generation technique for retraining. The model’s error detection and correction capabilities are enhanced through the establishment of a connection amid the grammatical error recognition and correction model. The suggested error recognition method has been shown to considerably increase the stealthiest of the trigger, guarantee the efficacy of the backdoor assault, and make the trigger resistant to specific data augmentation operations in experiments using datasets like GTRSB.