about language distribution
#11
by
Leon-Leee
- opened
Hi, the blog says during the preprocessing of CC non-English texts were filtered by a detector. So have you done some statistics about the language distribution in the now-processed datasets? I guess the multi-lingual features mean a lot to the open-source community.
Hi @Leon-Leee ,
Regarding the data in the "now-processed" datasets (I guess you mean our released documents?) , the CC part contains English documents only (fastText with top language as "en" and score >= 0.65). Non-English documents were disgarded and not processed.
Thanks! (Yes I mean the released dataset.)
Leon-Leee
changed discussion status to
closed