The Construction of Translation Memory and Terminology Database Using Deep Learning and XGBoost Algorithm

Machine translation quality can be significantly enhanced by integrating external information into the translation process. Unlike traditional generative approaches, using external information enables models to access additional data during training, reducing reliance on limited training datasets. Three fusion methods are primarily utilized in this context: those based on bilingual dictionaries, translation memories, and templates. This study investigates techniques for incorporating external information, specifically through the construction of a translation memory and terminology database. The translation memory is optimized using deep learning methods and the XGBoost (eXtreme Gradient Boosting) algorithm. A comparative analysis evaluates the performance of both algorithms on translation accuracy and efficiency. Experimental findings reveal that, with only 10 hours of training on 5,000 data entries, the XGBoost algorithm achieves an 88.5% accuracy rate and covers 75% of the terminology database, demonstrating its advantage in large-scale data processing and predictive tasks. These findings highlight the importance of selecting appropriate algorithms to maximize translation memory applications, offering technical insights to further improve machine translation efficacy.