Post: Rare Words for NMT

In the work “Continuous Learning in “Neural Machine Translation using Bilingual Dictionaries” we analysed the ability of NMT systems to translate rare terms and presented techniques to improve their ability to translate morphological variants. The propsed methods is based on creating a new test set using a different split of the training and test data concentrating on these terms. You can easily work on the proposed testset by downloading the newly created “splits of the data set” or split other copora in the same way using the provided code. For more information