Unsupervised English-Chinese word translation using various retrieval methods

Because it is essential for improving the user experience, controlling styles in neural machine translation (NMT) has garnered a lot of interest in recent years. The majority of the earlier research on this subject focused on managing the amount of formality, and it was successful in making some headway in this particular area. The purpose of this study is to tackle each of these difficulties by presenting a new benchmark and strategy. A benchmark for multiway stylistic machine translation (MSMT) is presented, which incorporates a wide variety of styles that span four different language domains. Following that, we offer an approach that we call style activation prompt (StyleAP), which involves extracting prompts from a styled monolingual corpus and does not need any more fine-tuning alterations. Experiments demonstrate that StyleAP is capable of exerting a significant amount of control on the translation style and achieving extraordinary levels of performance.