What is neural machine translation?

What is neural machine translation?

Neural Machine Translation (NMT) is a machine translation approach that uses artificial neural networks to predict the likelihood of a sequence of words. This can be a text fragment, a complete sentence, or an entire document. NMT is a form of end-to-end learning used to automatically produce translations.

NMT systems typically consist of two main sections: an encoder network and a decoder network. Both are neural networks. The encoder network processes the input text and the decoder network generates the translated text. Each word in the input sentence is encoded as a number to be translated by the neural network.

NMT is the dominant approach to machine translation, having replaced Statistical Machine Translation (SMT) as the standard technique by considering sentence context and grammar rather than translating phrase by phrase. A single system can be trained directly on source and target text, without the pipeline of specialized components that SMT required, making NMT a more integrated and efficient approach. Challenges remain, however, particularly for language pairs with limited parallel training data, for domain shift between training data and the text being translated, and for ambiguous or idiomatic expressions, which NMT systems may translate too literally. NMT systems can also struggle with consistency, since they learn patterns from large volumes of data rather than following explicit rules, and there are certain types of content and communication styles where human translators still outperform automated systems.