# Improving Machine Reading Comprehension via Adversarial Training

arXiv: https://arxiv.org/abs/1911.03614 · HTML: https://mlsanity.com/paper/1911.03614/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 9 Nov 2019.

## Built on by (most cited first, 7 of 7)

- [Enhancing Knowledge Tracing via Adversarial Training](https://mlsanity.com/paper/2108.04430/) (2021)
- [Model Agnostic Answer Reranking System for Adversarial Question Answering](https://mlsanity.com/paper/2102.03016/) (2021)
- [To Answer or Not to Answer? Improving Machine Reading Comprehension Model with Span-based Contrastive Learning](https://mlsanity.com/paper/2208.01299/) (2022)
- [Using calibrator to improve robustness in Machine Reading Comprehension](https://mlsanity.com/paper/2202.11865/) (2022)
- [Improving Lexical Embeddings for Robust Question Answering](https://mlsanity.com/paper/2202.13636/) (2022)
- [A Lightweight Method to Generate Unanswerable Questions in English](https://mlsanity.com/paper/2310.19403/) (2023)
- [QLSC: A Query Latent Semantic Calibrator for Robust Extractive Question Answering](https://mlsanity.com/paper/2404.19316/) (2024)

## Abstract (arXiv)

Adversarial training (AT) as a regularization method has proved its effectiveness in various tasks, such as image classification and text classification. Though there are successful applications of AT in many tasks of natural language processing (NLP), the mechanism behind it is still unclear. In this paper, we aim to apply AT on machine reading comprehension (MRC) and study its effects from multiple perspectives. We experiment with three different kinds of RC tasks: span-based RC, span-based RC with unanswerable questions and multi-choice RC. The experimental results show that the proposed method can improve the performance significantly and universally on SQuAD1.1, SQuAD2.0 and RACE. With virtual adversarial training (VAT), we explore the possibility of improving the RC models with semi-supervised learning and prove that examples from a different task are also beneficial. We also find that AT helps little in defending against artificial adversarial examples, but AT helps the model to learn better on examples that contain more low-frequency words.