1. get word embeddings from larger corpora 2. rotate embeddings space of two languages for optimal alignment, assuming frequency and neighborhoods of word embe

1. get word embeddings from larger corpora 2. rotate embeddings space of two languages for optimal alignment, assuming frequency and neighborhoods of word embeddings are more-less the same in any language 3. iteratively minimize difference in bidirectional translations