thunlp/MultiRD

Question about the method to calculate the score

yipliu opened this issue · 4 comments

Hi Zhang Lei,

Does the following code prevent the model from predicting the token contained in the input sentence?

score = score * mask2 + (-1e6)*(1-mask2)

Yes. The word you are looking for that you don't know can not appear in your words.

Thanks!

Hi Zhang Lei,

I wonder if I could ask you a question. How do you distinguish between NLG and NLU?

Whether a new text is generated?