In the field of intelligent translation of foreign language children's books,an optimized interactive neural machine translation model has been designed for application in foreign language children's books,including the translation of children's books between Chinese and German.The system provides different proportions of correct translations during the decoding stage to train the model,in order to cope with different training set sizes.The experimental results show that under different scale training sets,the e-valuation results of the model continue to improve as the proportion of correct inputs increases.Under the condition of a 1 megabyte training set,adding 15％to 45％of correct inputs can result in an average improvement of 16.07％in the results.When training a set of 2 megabytes,an increase in correct input of the same proportion will result in an average improvement of 15.84％in test results.The training set is 3 megabytes,and the average improvement in test results is 15.65％.Therefore,the scale of training data has a significant impact on translation results and model performance,improving the accuracy and fluency of machine translation,and provi-ding a new basis and perspective for the translation of Chinese German children's books.
关键词
交互式机器翻译/童书翻译/多头注意力机制/LSTM网络/解码器
Key words
interactive machine translation/translation of children's books/attention mechanism/LSTM network/decoder
维普
万方数据