Self-attention Convolutional Long Short Term Memory for Intelligent Detection System in English Translation Errors

The English translation error detection and learning, intelligent system software is required to transform traditional English teaching as well as learning. However, English language processing faces challenges in finding and correction errors, especially with small data due to the risk of overfitting. In this paper, Self-attention Conv Long Short-Term Memory (SA-ConvLSTM) network, which enables the model to focus on relationships between distant words. This capability is crucial for detecting errors in complex translations, where issues with word order. The Conv LSTM capture local feature like phase structures and LSTM maintains sequence context allowing it to learn how errors develop over time in the sentence. The system ability to capture both local and global sentence learn efficiently for detecting English translation of errors improving its effectiveness to evaluated using BEA-2019 dataset. The proposed SA-ConvLSTM network achieve better accuracy of 91.02%, recall of 89.59%, precision of 91.06%, and F1-score of 89.12% on BEA-2019 dataset. The existing method such as Long Short Term Memory -Conditional Random Field model (LSTM-CRF) and Detection-Correction Structure named (DeCoGLM) based on General Language Model (GLM) are evaluated proposed method.