neulab / BARTScore

BARTScore: Evaluating Generated Text as Text Generation
Apache License 2.0
318 stars 37 forks source link

question before using it #15

Closed xinli2008 closed 2 years ago

xinli2008 commented 2 years ago

Hello, I want to use the modified code to test the text generation task, because I haven't read the paper, I want to ask two questions, first, is the higher the bartscore the better, and second, what does the negative bartscore mean?

yyy-Apple commented 2 years ago

Hi, To your questions: (1) The higher score corresponds to a better generation (2) The negative score is the average log-likelihood of each generated token.