Details of evaluation

[HarlynDN]

Hi, I'm following your work recently but I can't replicate some of your results especially on several dimensions. I guess it might have something to do with post-processing the model's outputs. However, I'm unable to look at your submission file since the file paths in evaluate_GLUCOSE_submissions.ipynb are inaccessible. Therefore, could you share some submission examples on this repository for reference?

Also, I'm curious about some details of the evaluation. For one-sided generation, did you compute the bleu score using full rule references in test files? If so, did you turn the one-sided generation into full rule format by adding the other side and connective using the original sentence X and predefined connective keywords?

Thanks a lot.