Augmenting Training Data for Low-Resource Neural Machine Translation via Bilingual Word Embeddings and BERT Language Modelling
