Multi-Task Transfer Learning with Data Augmentation for Recognizing Question Entailment in the Medical Domain

Recognizing Question Entailment (RQE) is an important task for Question Answering (QA) as it infers the logical relation between two questions. Existing research efforts on this topic using multi-task transfer learning models have proven to be successful. However, the relative scarcity of labeled data has impeded the exploitation of this family of high-capacity approaches. In this paper, we propose a multi-task transfer learning-based method with data augmentation for RQE in medical QA. The proposed method first generates new training data from existing RQE examples based on contextual word embeddings and QA data. It then uses multi-task transfer learning that combines multi-task learning and language-model pre-training. Experimental evaluations performed on the RQE test set of the 2019 MEDIQA challenge show that our data augmentation method led to a substantial increase of an average of 18.5% and that the proposed RQE approach is more effective than the state-of-the-art systems.