ukplab / acl2016-supersense-embeddings Goto Github PK

I have parsed the entirety of the Wikipedia/Babelnet dataset you've provided and seek to use it to train embeddings. However, I've found that the dataset is three times larger than what is reported in the paper: it has about 1.5 billion words, where about 400 million have supersense labels (in the paper, you report 500 million words and 100 million labels). I was wondering, is there something I'm missing, or did you update the dataset link to be larger than what is reported in your paper?