Building Multilingual Neural Machine Translation Models Using Cloud LM Algorithm

With the rapid advancement of artificial intelligence technology, machine translation has become an important bridge connecting different languages and cultures. However, in multilingual environments, traditional machine translation models are often limited by high resource consumption and slow response speed, and the accuracy rate is also deficient. Therefore, the study proposes a multilingual neural machine translation model constructed using the cloud language model, which uses the Ken language model as the baseline model, open-source search via Apache Solr, and optimization for caching and block query. The experimental results show that on the WMT2021 dataset, the average bilingual intertranslation evaluation index of the research model reaches $\mathbf{3 4. 0 2 \%}$, which is $\mathbf{1 9. 2}$ percentage points higher than the baseline model. On the Newstest2014 dataset, the memory usage of the research model for translating 100 sentences in 343 MB, which is 105 MB lower than that of the baseline model. Comparative experiment outcomes indicate that on the WMT2021 dataset, the research model outperforms the other models with $\mathrm{P}, \mathrm{R}$, and F 1 scores of $0.877,0.864$, and 0.923. The research model effectively solves the problems of large memory usage and slow response speed of the traditional machine translation model, and is able to realize accurate translation in multilingual environments, which holds significant importance in promoting the development of machine translation technology.