Hi, I'm interested in your paper Bilingual Lexicon Induction through Unsupervised Machine Translation. I understand that the paper is based on the code in this repository, but I think there might be a minor difference. In the paper, frequent n-grams are assigned the centroid of the words they contain, whereas the SMT paper uses phrase2Vec, correct? Is there a difference in performance if phrase2Vec is used for BLI?
Hi, I'm interested in your paper Bilingual Lexicon Induction through Unsupervised Machine Translation. I understand that the paper is based on the code in this repository, but I think there might be a minor difference. In the paper, frequent n-grams are assigned the centroid of the words they contain, whereas the SMT paper uses phrase2Vec, correct? Is there a difference in performance if phrase2Vec is used for BLI?