Machine translation (MT) is a subfield of computational linguistics that automatically transforms source text in one language to text in another language. Neural Machine Translation (NMT) overcomes MT problems, such as less accuracy, context translation difficulty, and word with multiple meanings. NMT has emerged as the advance of machine translation (MT) with the integration of deep learning models. The key significance of NMT is the ability to train the single system on the source and target data to learn a statistical model for machine translation using neural network architecture.

NMT produces high performance with less linguistic knowledge. The advantages of NMT are more fluent, human-like translations, higher translation quality, and accuracy, context is taken into account, and better for under-resourced languages with fewer data. Convolutional neural networks (CNN), Recurrent neural networks (RNN), and self-attention networks (SAN) are the methods of NMT architectures. The key component of NMT architecture is encoder and decoder.

NMT is mainly applied in natural language processing tasks. Some of them are Language modeling, Sequence tagging, Speech recognition, Image to text, Summarization, and so on. Recent developments in NMT are NMT for low resource languages, NMT with domain adaptation and visual interactive NMT, and many more. Future research areas of NMT are NMT with Monolingual Data, NMT by Incorporating Hierarchical Subword Features, NMT with Neural Syntactic Distance, Word Sense Disambiguation in NMT with Sense Embeddings, Neural Sentence Rewriting in NMT, and many more.