On the Role of Pre-trained Language Models in Word Ordering: A Case Study with BART

4657

Knowledge Triples