| Popis: |
Abstract In the current era of global economic integration and digital economy development, multilingual English translation plays a crucial role in cultural exchange. Traditional translation models have poor adaptability and fitting ability. To improve the translation effect and quality, this article combines the Text-To-Text Transfer Transformer (T5) and Model-Agnostic Meta-Learning (MAML) to study their sustainable improvement and application in multilingual English translation. In this article, the autoregressive learning method is first used to fine tune the pre-trained parameters of T5 model, and a generative multilingual English translation model is constructed. Then, combined with the MAML framework, the model is trained on multiple tasks to achieve rapid adaptation on new task data. Finally, a multilingual parallel corpus is constructed using web crawlers, and the translation model based on T5 and MAML is evaluated for quality using Bilingual Evaluation Understudy (BLEU) and Translation Error Rate (TER) as indicators. The experimental results show that compared to the baseline models Open Neural Machine Translation (OpenNMT), Transformer, and Open Parallel Corpus-Machine Translation (Opus-MT), the mean BLEU Score of the model in this article is 6.05%, 2.59%, and 2.05% higher, respectively. The conclusion indicates that the T5-MAML model can effectively improve the quality of multilingual English translation and achieve more natural and smooth translation output. |
Full text from SpringerLink