The evaluation criteria says that we could only use the corresponding dataset for task-1 to build models for task-1 and dataset for task-2 to build models for task-2 to ensure fairness.
So could we use pre-trained models, such as pre-trained word2vec embeddings or BERT, which are trained on a large amount of unannotated data, and build our models upon them using only the provided annotated data?
Sorry for confusing, the labeled data you use should be our provided data, while of course you could use other unlabeled data or pre-trained models like BERT.Posted by: Ariel_yang @ Nov. 18, 2019, 5:03 p.m.
OK. Thanks!Posted by: kliao @ Nov. 20, 2019, 2:09 a.m.
So, if I understand correctly, e.g. for task2; data augmentation / using annotation from similar tasks is allowed, and we just cannot use data from task1.
Is that right?