A Global-Features and Local-Features-Jointly Fused Deep Semantic Learning Framework for Error Detection of Machine Translation

Error detection in machine translation is very important for users, but current methods often only focus on one of the local or global features, neglecting the fusion of the two, thereby reducing the detection effect. To address this issue, this study proposes a deep semantic learning framework based on deep learning mechanisms. This study utilizes convolutional neural networks to extract local features at the sentence level, and then uses recurrent neural networks to capture global features at the sentence sequence level. By integrating local features with global features, we have established a deep semantic learning framework. To evaluate the performance of this method, we conducted extensive detection experiments on multiple publicly available datasets, including translation error detection for words, grammar, and punctuation. Each detection metric was tested 2000 times, and 5 cyclic tests were conducted to reduce errors in deep semantic learning. The experimental results indicate that our method has made significant improvements in the error detection task of machine translation. Compared with the traditional methods, this method can better detect and repair errors in translation, and its stability and applicability are broader.