An Intelligent Error Detection Model for Machine Translation Using Composite Neural Network-Based Semantic Perception

Although machine translation has received great progress in recent years, machine translation results usually existed some errors due to the complex relationship between sentence structure and semantics. Currently, the automatic error detection techniques towards machine translation errors have not been deeply investigated. To deal with the challenge, this paper proposes an intelligent error detection model for machine translation using composite neural network-based semantic perception. Firstly, integrating attention mechanism into Bi-GRU encoder can effectively learn contextual information of sentences and generate high-quality global feature representations. Then, multiscale CNN can extract local features at different scales, thereby capturing finer grained semantic information. Experiments are conducted on datasets containing a large amount of English text and machine translation errors, in which the proposed model is compared several benchmark methods. The experimental results indicate that the proposal has achieved significant improvements in machine translation error detection tasks. It comparison, it can more accurately identify common problems such as grammar errors, semantic errors, and word spelling errors in translation results, verifying its effectiveness and practicality.