Let be a small corpus of text, check the small count are well computed in the ngrams table. I am looking for data to reproduce it but sometimes, instead of 1 count, I got 0.