Pseudo Zero Pronoun Resolution Improves Zero Anaphora Resolution

Masked language models (MLMs) have contributed to drastic performance improvements with regard zero anaphora resolution (ZAR). To further improve this approach, in study, we made two proposals. The first is a new pretraining task that trains MLMs on anaphoric relations explicit supervision, and the second proposal finetuning method remedies notorious issue, pretrain-finetune discrepancy. Our experiments Japanese ZAR demonstrated our proposals boost state-of-the-art performance, detailed analysis provides insights remaining challenges.