Machine Reading Comprehension Framework Based on Self-Training for Domain Adaptation

Machine reading comprehension (MRC) is a type of question answering mechanism in which computer reads documents and answers related questions. The accuracies recent MRC systems surpass those humans. However, most exhibit significant performance deteriorations when domains are changed. Hence, we propose self-training framework for MRC. proposed composed pseudo-answer extractor, pseudo-question generator, an system. In the source domain, components pretrained using training dataset. target performances generator system improved through mutual scheme. During self-training, provides new data to obtains rewards from reinforcement learning. experiments with Wikipedia domain (source domain) civil affair (target domain), based on scheme demonstrates better than that automatic augmentation.