A Method for Chinese-Korean Neural Machine Translation Based on Weight Sharing
In addressing the challenges of low-resource neural machine translation,which are characterized by insufficient data and suboptimal translation performance,we propose the utili-zation of weight sharing as a method to optimize the translation process.Firstly,we create a parent model,process the crawled data,and train it using a neural machine translation model.Secondly,we create a unified dictionary to ensure that the child model can effectively utilize the vocabulary knowledge of the parent model.Finally,weight sharing is employed so that the child model can acquire prior knowledge from the parent model.The results showed that the weight-sharing model performed better in translation,with complete translation of key words and semantics.The BLEU score was improved by 3.25 compared to the ordinary neural ma-chine translation model.In summary,by employing weight sharing in transfer learning,we successfully optimized the problem of low-resource neural machine translation and significantly improved translation performance.This method has important application value for addressing machine translation problems for low-resource language pairs.
NETL
NSTL
万方数据
