I’m using OpenNMT python version for translating Arabic to Persian, and I wondered how I could calculate the belu score to evaluate translation? However, I used [multi-bleu-detok.perl] but I guess the score is not correct. (is the multi-bleu-detok.perl supports utf8 encoding?)
In addition, Is there a way to change the prediction score from likelihood to something else?
Hello!
You can use sacreBLEU. Please note that the BLEU must be calculated on a detokenized text, both the target and reference (if you tokenized the text during data preprocessing).
Kind regards,
Yasmin