Closed obo closed 3 years ago
NLTK is quite a heavy dependence. All we want from it is tokenization. We would probably be better off with sacreMoses (which also adds some useless bells and whistles like the progress bar). Please discuss here.
Hi Ondrej, I have dropped NLTK in the commit by the name " NLTK dropped in quality calculation". Also, I have used mosestokenizer for tokenization.
NLTK is still in requirements.txt
NLTK is quite a heavy dependence. All we want from it is tokenization. We would probably be better off with sacreMoses (which also adds some useless bells and whistles like the progress bar). Please discuss here.