[x] Is "unk-N" really needed for the fairseq implementation of a pointer-generator model?
[x] Inquire whether the replace-unk option is needed when evaluating the subword models (related with #49) --> no, since subwords should have no unk due to correct setup
Change template file to include less duplicates of the same question
[x] Integrate Copy Script with Data pipeline (#54)
[x] Check whether correct dictionaries are copied in copy script
[x] Adapt train scripts to Sacrebleu
[x] Use 1 question for query for models and XLMR --> do this for best performing model
[x] Add extra test set with same templates, only different in names
[x] Check file directory for every model script, adapted to folder?
Change template file to include less duplicates of the same questionAll of the above done?