# Using Adversarial Attacks to Reveal the Statistical Bias in Machine Reading Comprehension Models

arXiv: https://arxiv.org/abs/2105.11136 · HTML: https://mlsanity.com/paper/2105.11136/

## Key facts

- Cited by 29 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 27 references that are in the map.
- First posted on arXiv 24 May 2021.

## Builds on (most cited first, 16 of 16)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [RACE: Large-scale ReAding Comprehension Dataset From Examinations](https://mlsanity.com/paper/1704.04683/) (2017)
- [Universal Adversarial Triggers for Attacking and Analyzing NLP](https://mlsanity.com/paper/1908.07125/) (2019)
- [Annotation Artifacts in Natural Language Inference Data](https://mlsanity.com/paper/1803.02324/) (2018)
- [Adversarial Examples for Evaluating Reading Comprehension Systems](https://mlsanity.com/paper/1707.07328/) (2017)
- [Generating Natural Adversarial Examples](https://mlsanity.com/paper/1710.11342/) (2017)
- [Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examples](https://mlsanity.com/paper/1803.01128/) (2018)
- [What does BERT Learn from Multiple-Choice Reading Comprehension Datasets?](https://mlsanity.com/paper/1910.12391/) (2019)
- [Large-scale Cloze Test Dataset Created by Teachers](https://mlsanity.com/paper/1711.03225/) (2017)
- [Dual Co-Matching Network for Multi-choice Reading Comprehension](https://mlsanity.com/paper/1901.09381/) (2019)
- [Benchmarking Robustness of Machine Reading Comprehension Models](https://mlsanity.com/paper/2004.14004/) (2020)
- [STARC: Structured Annotations for Reading Comprehension](https://mlsanity.com/paper/2004.14797/) (2020)
- [Option Comparison Network for Multiple-choice Reading Comprehension](https://mlsanity.com/paper/1903.03033/) (2019)

## Built on by (most cited first, 29 of 29)

- [Advancing the Robustness of Large Language Models through Self-Denoised Smoothing](https://mlsanity.com/paper/2404.12274/) (2024)
- [PertEval: Unveiling Real Knowledge Capacity of LLMs with Knowledge-Invariant Perturbations](https://mlsanity.com/paper/2405.19740/) (2024)
- [A Prompting-based Approach for Adversarial Example Generation and Robustness Enhancement](https://mlsanity.com/paper/2203.10714/) (2022)
- [Divide and Conquer: Text Semantic Matching with Disentangled Keywords and Intents](https://mlsanity.com/paper/2203.02898/) (2022)
- [TextShield: Beyond Successfully Detecting Adversarial Sentences in Text Classification](https://mlsanity.com/paper/2302.02023/) (2023)
- [DSRM: Boost Textual Adversarial Training with Distribution Shift Risk Minimization](https://mlsanity.com/paper/2306.15164/) (2023)
- [Image-to-Text Logic Jailbreak: Your Imagination can Help You Do Anything](https://mlsanity.com/paper/2407.02534/) (2024)
- [ROSE: Robust Selective Fine-tuning for Pre-trained Language Models](https://mlsanity.com/paper/2210.09658/) (2022)
- [LEAP: Efficient and Automated Test Method for NLP Software](https://mlsanity.com/paper/2308.11284/) (2023)
- [Unveiling the Achilles' Heel of NLG Evaluators: A Unified Adversarial Framework Driven by Large Language Models](https://mlsanity.com/paper/2405.14646/) (2024)
- [Rethinking Textual Adversarial Defense for Pre-trained Language Models](https://mlsanity.com/paper/2208.10251/) (2022)
- [Efficient Adversarial Training with Robust Early-Bird Tickets](https://mlsanity.com/paper/2211.07263/) (2022)
- [Are LLMs Stable Formal Logic Translators in Logical Reasoning Across Linguistically Diversified Texts?](https://mlsanity.com/paper/2506.04575/) (2025)
- [A Survey on Measuring and Mitigating Reasoning Shortcuts in Machine Reading Comprehension](https://mlsanity.com/paper/2209.01824/) (2022)
- [Towards Action Hijacking of Large Language Model-based Agent](https://mlsanity.com/paper/2412.10807/) (2024)
- [No Query, No Access](https://mlsanity.com/paper/2505.07258/) (2025)
- [IM-BERT: Enhancing Robustness of BERT through the Implicit Euler Method](https://mlsanity.com/paper/2505.06889/) (2025)
- [AutoEvoEval: An Automated Framework for Evolving Close-Ended LLM Evaluation Data](https://mlsanity.com/paper/2506.23735/) (2025)
- [The Strong, Weak and Benign Goodhart's law. An independence-free and paradigm-agnostic formalisation](https://mlsanity.com/paper/2505.23445/) (2025)
- [Text Adversarial Attacks with Dynamic Outputs](https://mlsanity.com/paper/2509.22393/) (2025)
- [Style Attack Disguise: When Fonts Become a Camouflage for Adversarial Intent](https://mlsanity.com/paper/2510.19641/) (2025)
- [Subspace Defense: Discarding Adversarial Perturbations by Learning a Subspace for Clean Signals](https://mlsanity.com/paper/2403.16176/) (2024)
- [Simpler becomes Harder: Do LLMs Exhibit a Coherent Behavior on Simplified Corpora?](https://mlsanity.com/paper/2404.06838/) (2024)
- [EMBRACE: Evaluation and Modifications for Boosting RACE](https://mlsanity.com/paper/2305.08433/) (2023)
- [SCAT: Robust Self-supervised Contrastive Learning via Adversarial Training for Text Classification](https://mlsanity.com/paper/2307.01488/) (2023)
- [SenTest: Evaluating Robustness of Sentence Encoders](https://mlsanity.com/paper/2311.17722/) (2023)
- [From text to multimodal: a survey of adversarial example generation in question answering systems](https://mlsanity.com/paper/2312.16156/) (2023)
- [Coordinated Robustness Evaluation Framework for Vision-Language Models](https://mlsanity.com/paper/2506.05429/) (2025)
- [Innocence in the Crossfire: Roles of Skip Connections in Jailbreaking Visual Language Models](https://mlsanity.com/paper/2507.13761/) (2025)

## Abstract (arXiv)

Pre-trained language models have achieved human-level performance on many Machine Reading Comprehension (MRC) tasks, but it remains unclear whether these models truly understand language or answer questions by exploiting statistical biases in datasets. Here, we demonstrate a simple yet effective method to attack MRC models and reveal the statistical biases in these models. We apply the method to the RACE dataset, for which the answer to each MRC question is selected from 4 options. It is found that several pre-trained language models, including BERT, ALBERT, and RoBERTa, show consistent preference to some options, even when these options are irrelevant to the question. When interfered by these irrelevant options, the performance of MRC models can be reduced from human-level performance to the chance-level performance. Human readers, however, are not clearly affected by these irrelevant options. Finally, we propose an augmented training method that can greatly reduce models' statistical biases.