Machine Translation Post-Editing (MTPE) suffers from challenges such as complex grammatical error types, strong cross-language transfer interference, and insufficient coverage of manual rules, limiting the accuracy of automatic detection and correction. This paper proposes an automatic grammatical error detection and correction algorithm based on multi-task learning : First, a joint representation including lexical, syntactic, and contextual semantic features is constructed, and a bidirectional encoder is introduced to obtain alignment information between the source language and the target text; second, the grammatical error boundary is located through an error detection subtask, while candidate correction results are generated using a sequence-to-sequence correction subtask; third, a confidence fusion mechanism is used to jointly optimize the detection and correction results; finally, the optimal translated text is output through language model re-ranking. Experimental results on a Chinese-English MTPE dataset show that the proposed method achieves a high error recall rate of 88.7% in the grammatical error detection task and a grammatical correction F1 score of 88.8 %, effectively reducing the cost of manual editing and improving the quality of the translation.
Xuefei Meng (Thu,) studied this question.