The goal in Task1 is to predict "DA prediction agains human DA (z-standardised mean DA score, i.e. z_mean)".
Is the goal in Task1 to predict sentence DA score (training target is the average of the DA scores) or z-standardized mean DA score (which is also available in the train and dev sets)?
In the first case, I received MAE around 85 in en-de and in the second case MAE around 0.5 but with lower correlation.
So, is the training target the mean DA scores or z_mean? In qet 2020, I used z_mean but this year the Pearson's r is lower for some reason for the same target so I changed it to DA scores and obtained higher correlation but MAE and RMSE are larger now.
Thank you.
z_mean is in the range [-7.542, 3.178] and DA score is in the range [1, 100].
Posted by: bicici @ July 28, 2021, 12:38 p.m.