Research on Automatic Identification of Machine English Translation Errors Based on Improved GLR Algorithm

Abstract

Machine translation is a powerful tool for overcoming linguistic obstacles, but it often introduces errors that lower the overall translation quality. This research project aims to enhance machine-translated documents by identifying and classifying translation faults. To identify errors, the traditional Generalized LR (GLR) technique is modified and enhanced, incorporating linguistic and statistical elements from the machine-translated texts. Contextual information from GLR parsing is utilized to improve error detection, and additional parsing algorithms are integrated to handle the complexities of machine translation. The proposed improved GLR algorithm is compared with three baseline models: the statistical algorithm, dynamic memory algorithm, and traditional GLR algorithm. The evaluation is based on two key metrics: accuracy and recognition speed, with a focus on renewal capability. The improved GLR algorithm achieves a significantly higher accuracy of 92.5% compared to the baseline models: statistical algorithm (85.2%), dynamic memory algorithm (88.9%), and traditional GLR algorithm (80.6%). Additionally, the improved GLR algorithm demonstrates a recognition speed of 1200 words per second, showcasing its efficiency in real-time translation scenarios. The results show that the enhanced GLR algorithm outperforms the baseline models in accurately detecting translation errors while maintaining an efficient recognition speed. Its high renewal capability ensures adaptability to changing translation challenges and continuous improvement over time.