search

kenji-imamura / recycle_bert

A sample code of the paper published at NGMT-2019.
Other
8 stars 4 forks source link

About scoring tools #1

Open Surnal opened 3 years ago

Surnal commented 3 years ago

Hello, I would like to ask which scoring tool is this model. Is it case-sensitive? Or all lowercase scores?

I noticed that the inputs of src and target contain capitals, and the output results also contain capitals. So should I use case-sensitive ref and generated results to score, or turn ref and generated results into all lowercase to score?

Thank you very much for your reply.

@kenji-imamura

kenji-imamura commented 3 years ago

Hi, I used the case-sensitive BLEU for the evaluation because the data used here, which has been preprocessed by Stanford NLP Group contains capitals.