Adversarial Networks for Machine Reading

Deep machine reading models have recently progressed remarkably with the help of differentiable reasoning models. In this context, deep end-to-end trainable networks enhanced with memory and attention have demonstrated promising performance on simple natural language based reasoning tasks. However, the training of machine comprehension models commonly requires a large annotated question-answer dataset for learning. In this paper, we explore the paradigm of adversarial learning and self-play for machine reading comprehension. Inspired by the success in the domain of game learning, we propose a novel approach to train machine comprehension models based on a coupled attention-based model. In this approach, a reader network is in charge of finding answers to the questions regarding a passage of text, while an obfuscation network tries to obfuscate spans of text in order to minimize the proba- bility of success of the reader. The model is evaluated on several question-answering corpora. The proposed learning paradigm and associated models show promising results.