Designing Templates for Eliciting Commonsense Knowledge from Pretrained Sequence-to-Sequence Models

4052

Knowledge Triples