Separate train script into one to train directly one classifier and one to train one binary classifier (accept/reject) for each discursive function