Research on English Chinese Neural Machine Translation based on Improved Deep Q-Network Approach

Currently, development of Deep Learning (DL), Neural Machine Translation (NMT) with neural networks has progressively become conventional method in academia. The past English-Chinese NMT models usually accept Deep Neural Network (DNN) architecture assisted by an attention mechanism. However, there is a challenge in existing models for learning and translating short and long sentences continuously from one language to other. To overcome this problem a Deep Reinforcement Learning (DRL) based Improved Deep QNetwork (IDQN) method is proposed for enhancing EnglishChinese translation. The dataset utilized for research is the International Workshop on Spoken Language Translation (IWSLT) dataset and preprocessing is performed to generate word vector. After that, an optimized Convolutional Neural Network (CNN) is utilized for extracting informative features from texts and they are translated from English to Chinese language by neural translation method based on IDQN approach. The experimental results stipulate better translation of words efficiently with $95.2 \%$ precision, $95.3 \%$ accuracy, $\mathbf{9 5. 3 \%}$ F1-score and $\mathbf{4 2. 1 3 \%}$ BLEU which is superior to existing methods like Random Decision Forest (RDF), Transformer, Self-Attention-Transformer (SA-Transformer), Grouping Transformer (GTrans), and Multi Teacher DistillationTransformer (MTD-Transformer).