Neural Machine Translation
There is general agreement that Neural Machine Translation (NMT) outperforms Statistical Machine Translation (SMT) in terms of fluency and appropriateness when humans read software-generated text. NMT uses a large artificial neural network, similar to what happens in the human brain with thousands of connections. One of the main advantages of NMT is that the context of the translation is much longer than SMT (phrase-level translation). Currently, developers mainly use a sequence-to-sequence approach, where the full context of the sentence is taken into account. The use of NMT improves the accuracy and fluency of the translation. The other advantage of NMT with respect to SMT is that NMT requires only a portion of the memory needed for SMT, and all parts of the NMT model are trained jointly to maximize the target translation performance (end-to-end approach).