Literature Survey: Neural Machine Translation

Neural Machine Translation (NMT) is a new highly active approach for machine translation, which has showed promising results and due to its success it has attracted many researchers in the field. In this paper we investigate how NMT architecture changed over very short span of time. Starting with basic encoder-decoder architecture that suffered two problems, poor performance with longer sentences and out-of-vocabulary (OOV) problem. Attention based NMT performs better with longer sentences but it still faces the OOV problem. To deal with OOV problem attention based NMT along with subword segmentation called Subword NMT are being used. With word segmentations NMT systems are able to cover larger range of vocabulary. Later factored NMT were proposed which shows improvement in performance, but requires linguistic features annotated data, which may not be available easily for most of the language pairs.