Closed HadiZayer closed 5 years ago
Please submit this issue to https://github.com/moses-smt/mosesdecoder/blob/master/scripts/generic/multi-bleu.perl
PyTorch-NLP relies on the official BLEU metric; therefore, by definition, it computes BLEU correctly.
I tried to test the get_moses_multi_bleu metric and it doesn't seem to work properly. I ran the following:
and it prints
0.0
instead of1
Edit: I believe the reason is that because the sentence is too short for 2-gram, 3-gram, etc... So there should be a warning message in that case