# Avoiding Reasoning Shortcuts: Adversarial Evaluation, Training, and Model Development for Multi-Hop QA

arXiv: https://arxiv.org/abs/1906.07132 · HTML: https://mlsanity.com/paper/1906.07132/

## Key facts

- Cited by 63 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Jun 2019.

## Built on by (most cited first, 30 of 63)

- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies](https://mlsanity.com/paper/2101.02235/) (2021)
- [Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned](https://mlsanity.com/paper/2209.07858/) (2022)
- [QuALITY: Question Answering with Long Input Texts, Yes!](https://mlsanity.com/paper/2112.08608/) (2021)
- [The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning](https://mlsanity.com/paper/2205.03401/) (2022)
- [Knowledge Graph Prompting for Multi-Document Question Answering](https://mlsanity.com/paper/2308.11730/) (2023)
- [Unlimiformer: Long-Range Transformers with Unlimited Length Input](https://mlsanity.com/paper/2305.01625/) (2023)
- [MultiHiertt: Numerical Reasoning over Multi Hierarchical Tabular and Textual Data](https://mlsanity.com/paper/2206.01347/) (2022)
- [An Empirical Survey of Data Augmentation for Limited Data Learning in NLP](https://mlsanity.com/paper/2106.07499/) (2021)
- [Trustworthiness in Retrieval-Augmented Generation Systems: A Survey](https://mlsanity.com/paper/2409.10102/) (2024)
- [Efficient Long-Text Understanding with Short-Text Models](https://mlsanity.com/paper/2208.00748/) (2022)
- [Natural Language Reasoning, A Survey](https://mlsanity.com/paper/2303.14725/) (2023)
- [Agentic Web: Weaving the Next Web with AI Agents](https://mlsanity.com/paper/2507.21206/) (2025)
- [Attacking Open-domain Question Answering by Injecting Misinformation](https://mlsanity.com/paper/2110.07803/) (2021)
- [Why Machine Reading Comprehension Models Learn Shortcuts?](https://mlsanity.com/paper/2106.01024/) (2021)
- [MoreHopQA: More Than Multi-hop Reasoning](https://mlsanity.com/paper/2406.13397/) (2024)
- [EX-FEVER: A Dataset for Multi-hop Explainable Fact Verification](https://mlsanity.com/paper/2310.09754/) (2023)
- [CLIP-TD: CLIP Targeted Distillation for Vision-Language Tasks](https://mlsanity.com/paper/2201.05729/) (2022)
- [Break the Chain: Large Language Models Can be Shortcut Reasoners](https://mlsanity.com/paper/2406.06580/) (2024)
- [Multimodal Adaptive Distillation for Leveraging Unimodal Encoders for Vision-Language Tasks](https://mlsanity.com/paper/2204.10496/) (2022)
- [Rissanen Data Analysis: Examining Dataset Characteristics via Description Length](https://mlsanity.com/paper/2103.03872/) (2021)
- [Connecting Attributions and QA Model Behavior on Realistic Counterfactuals](https://mlsanity.com/paper/2104.04515/) (2021)
- [Evolving Diverse Red-team Language Models in Multi-round Multi-agent Games](https://mlsanity.com/paper/2310.00322/) (2023)
- [FactCG: Enhancing Fact Checkers with Graph-Based Multi-Hop Data](https://mlsanity.com/paper/2501.17144/) (2025)
- [Robustifying Multi-hop QA through Pseudo-Evidentiality Training](https://mlsanity.com/paper/2107.03242/) (2021)
- [Graph-free Multi-hop Reading Comprehension: A Select-to-Guide Strategy](https://mlsanity.com/paper/2107.11823/) (2021)
- [From Easy to Hard: Two-stage Selector and Reader for Multi-hop Question Answering](https://mlsanity.com/paper/2205.11729/) (2022)
- [Seemingly Plausible Distractors in Multi-Hop Reasoning: Are Large Language Models Attentive Readers?](https://mlsanity.com/paper/2409.05197/) (2024)
- [Which Shortcut Solution Do Question Answering Models Prefer to Learn?](https://mlsanity.com/paper/2211.16220/) (2022)
- [Chain-of-Questions Training with Latent Answers for Robust Multistep Question Answering](https://mlsanity.com/paper/2305.14901/) (2023)

## Abstract (arXiv)

Multi-hop question answering requires a model to connect multiple pieces of evidence scattered in a long context to answer the question. In this paper, we show that in the multi-hop HotpotQA (Yang et al., 2018) dataset, the examples often contain reasoning shortcuts through which models can directly locate the answer by word-matching the question with a sentence in the context. We demonstrate this issue by constructing adversarial documents that create contradicting answers to the shortcut but do not affect the validity of the original answer. The performance of strong baseline models drops significantly on our adversarial evaluation, indicating that they are indeed exploiting the shortcuts rather than performing multi-hop reasoning. After adversarial training, the baseline's performance improves but is still limited on the adversarial evaluation. Hence, we use a control unit that dynamically attends to the question at different reasoning hops to guide the model's multi-hop reasoning. We show that this 2-hop model trained on the regular data is more robust to the adversaries than the baseline model. After adversarial training, this 2-hop model not only achieves improvements over its counterpart trained on regular data, but also outperforms the adversarially-trained 1-hop baseline. We hope that these insights and initial improvements will motivate the development of new models that combine explicit compositional reasoning with adversarial training.