HIT-SCIR at SemEval-2020 Task 5: Training Pre-trained Language Model with Pseudo-labeling Data for Counterfactuals Detection