An Automatic Error Detection Method for Machine Translation Results Via Deep Learning

Nowadays, the rapid development of natural language processing has brought great progress for area machine translation. Various deep neural network-based translation approaches have been more and general. However, there still lacks effective automatic error detection results. To bridge such gap, this paper proposes an method results via learning. The training data is synthesized using generative model proposed in paper, which used foreign trade English grammatical correction model. Then, to correct source sentences learner’s corpus, corrected target manually annotated standard are formed into “error-correct” sentence pairs, fed back generation alternate training. By establishing a link between model, capability improved. Experiments on datasets as GTRSB show that significantly improves stealthiness trigger while ensuring effectiveness backdoor attack, at same time enables resist certain augmentation operations.