Adversarial Examples for Evaluating Reading Comprehension Systems
Open accessThe DOI matches its Crossref record; 40 of the 40 references it cites were found in a registry or an index.
Abstract
Standard accuracy metrics indicate that reading comprehension systems are making rapid progress, but the extent to which these systems truly understand language remains unclear.To reward systems with real language understanding abilities, we propose an adversarial evaluation scheme for the Stanford Question Answering Dataset (SQuAD).Our method tests whether systems can answer questions about paragraphs that contain adversarially inserted sentences, which are automatically generated to distract computer systems without changing the correct answer or misleading humans.In this adversarial setting, the accuracy of sixteen published models drops from an average of 75% F1 score to 36%; when the adversary is allowed to add ungrammatical sequences of words, average accuracy on four models decreases further to 7%.We hope our insights will motivate the development of new models that understand language more precisely.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The second trend is that neural networks routinely fail on exactly the phenomena where such symbolic models succeed, for example, logical inference [6,9], compositionality [7,23] and out-of-distribution generalization [24–26].”
From Symbols and grounding in large language models (2023) · cites this paper as [25]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The diference between the training dataset and the deployment data is called distribution shift, which often results in a performance drop [24,46,61].”
From Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation (2022) · cites this paper as [46]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“AI performance can signifcantly drop for out-of-distribution examples that are diferent from the training data (in-distribution examples) [9,24,46,61].”
From Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation (2022) · cites this paper as [46]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Glove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 1×
“We replace nouns and adjectives with antonyms from WordNet (Fellbaum, 1998), and change named entities and numbers to the nearest word in GloVe word vector space 2 (Pennington et al., 2014) with the same part of speech.”
MethodLabaca-Castro, R. (2023) · Machine Learning under Malware Attack · doi:10.1007/978-3-658-40442-0_9 · Cited by 19,272
GAN(Generative Adversarial Nets)
(2017) · Journal of Japan Society for Fuzzy Theory and Intelligent Informatics · doi:10.3156/jsoft.29.5_177_2 · Cited by 19,270
WordNet: An Electronic Lexical Database
Kilgarriff, A., & Fellbaum, C. (2000) · Language · doi:10.2307/417141 · Cited by 11,464
Explaining and Harnessing Adversarial Examples
Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014) · arXiv · doi:10.48550/arxiv.1412.6572 · Cited by 10,381 · cited 5×
“Ad- 6 All previous experiments used parameters released by Seo et al. (2016) versarial training (Goodfellow et al., 2015) can be used for any model trained with stochastic gradient descent, but it requires generating new adversarial examples at every iteration; this is feasible for images, where…”
BackgroundMethodInfluentialIntriguing properties of neural networks
Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., & Fergus, R. (2013) · arXiv · doi:10.48550/arxiv.1312.6199 · Cited by 8,143 · cited 4×
“Prior work in computer vision adds imperceptible adversarial perturbations to input images, relying on the fact that such small perturbations cannot change an image’s true label (Szegedy et al., 2014; Goodfellow et al., 2015).”
BackgroundInfluentialThe Stanford CoreNLP Natural Language Processing Toolkit
Manning, C., Surdeanu, M., Bauer, J., Finkel, J., Bethard, S., & McClosky, D. (2014) · Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations · doi:10.3115/v1/p14-5010 · Cited by 7,037 · cited 1×
“We define a set (Manning et al., 2014), plus a few custom categories (e.g., abbreviations), and manually associate a fake answer with each type.”
MethodSQuAD: 100,000+ Questions for Machine Comprehension of Text
Rajpurkar, P., Zhang, J., Lopyrev, K., & Liang, P. (2016) · Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d16-1264 · Cited by 6,208 · cited 4×
“Given a model f that takes in paragraph-question pairs ( p, q ) and outputs an answer ˆ a , the standard accuracy over a test set D test is simply where v is the F1 score between the true answer a and the predicted answer f ( p, q ) (see Rajpurkar et al. (2016) for details).”
BackgroundMethodInfluentialSelf-citationPractical Black-Box Attacks against Machine Learning
Papernot, N., McDaniel, P., Goodfellow, I., Jha, S., Celik, Z. B., & Swami, A. (2017) · Proceedings of the 2017 ACM on Asia Conference on Computer and Communications Security · doi:10.1145/3052973.3053009 · Cited by 3,495
Universal Adversarial Perturbations
Moosavi-Dezfooli, S.-M., Fawzi, A., Fawzi, O., & Frossard, P. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.17 · Cited by 2,680
Generating Sentences from a Continuous Space
Bowman, S. R., Vilnis, L., Vinyals, O., Dai, A., Jozefowicz, R., & Bengio, S. (2016) · Proceedings of The 20th SIGNLL Conference on Computational Natural Language Learning · doi:10.18653/v1/k16-1002 · Cited by 2,061 · cited 1×
“Bowman et al. (2016) and Li et al. (2017) used such a setup for sentence and dialogue generation, respectively.”
Bidirectional Attention Flow for Machine Comprehension
Seo, M. J., Kembhavi, A., Farhadi, A., & Hajishirzi, H. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.01603 · Cited by 1,445 · cited 15×
“Ad- 6 All previous experiments used parameters released by Seo et al. (2016) versarial training (Goodfellow et al., 2015) can be used for any model trained with stochastic gradient descent, but it requires generating new adversarial examples at every iteration; this is feasible for images, where…”
MethodInfluentialDalvi, N., Domingos, P., Mausam, Sanghai, S., & Verma, D. (2004) · Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/1014052.1014066 · Cited by 887 · cited 1×
“Dalvi et al. (2004) formulated such tasks as a game between a classifier and an adversary, and analyzed optimal strategies for each player.”
BackgroundAdversarial Learning for Neural Dialogue Generation
Li, J., Monroe, W., Shi, T., Jean, S., Ritter, A., & Jurafsky, D. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1230 · Cited by 873 · cited 1×
“Bowman et al. (2016) and Li et al. (2017) used such a setup for sentence and dialogue generation, respectively.”
Lowd, D., & Meek, C. (2005) · Proceedings of the eleventh ACM SIGKDD international conference on Knowledge discovery in data mining · doi:10.1145/1081870.1081950 · Cited by 755 · cited 1×
“Lowd and Meek (2005) described an efficient attack by which an adversary can reverse-engineer the weights of a linear classifier, in order to then generate adversarial inputs.”
BackgroundData Recombination for Neural Semantic Parsing
Jia, R., & Liang, P. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p16-1002 · Cited by 441 · cited 1×
“We could also adversarially generate new examples by combining multiple existing ones, in the spirit of Data Recombination (Jia and Liang, 2016).”
MethodSelf-citationMachine Comprehension Using Match-LSTM and Answer Pointer
Wang, S., & Jiang, J. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1608.07905 · Cited by 408 · cited 4×
“Match-LSTM has a similar structure, but also has a hard-coded rule that stops it from predicting very long answers.”
MethodInfluentialPractical Black-Box Attacks against Deep Learning Systems using Adversarial Examples.Preprint
Papernot, N., McDaniel, P. D., Goodfellow, I., Jha, S., Celik, Z. B., & Swami, A. (2016) · arXiv (Cornell University) · Cited by 350
Nightmare at test time: robust learning by feature deletion
Globerson, A., & Roweis, S. (2006) · Proceedings of the 23rd international conference on Machine learning - ICML '06 · doi:10.1145/1143844.1143889 · Cited by 309 · cited 1×
“For certain classes of models and adversaries, efficient training strategies exist: for example, Globerson and Roweis (2006) train classifiers that are optimally robust to adversarial feature deletion.”
BackgroundGenerating Phrasal and Sentential Paraphrases: A Survey of Data-Driven Methods
Madnani, N., & Dorr, B. J. (2010) · Computational Linguistics · doi:10.1162/coli_a_00002 · Cited by 298
ReasoNet: Learning to Stop Reading in Machine Comprehension
Shen, Y., Huang, P.-S., Gao, J., & Chen, W. (2017) · Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/3097983.3098177 · Cited by 262 · cited 2×
“We also validate our major findings on twelve other published models with publicly available test-time code: ReasoNet Single and Ensemble versions (Shen et al., 2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble…”
ResultAdversarial Learning for Neural Dialogue GenerationPreprint
Li, J., Monroe, W., Shi, T., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1701.06547 · Cited by 211 · cited 1×
“Bowman et al. (2016) and Li et al. (2017) used such a setup for sentence and dialogue generation, respectively.”
Making Neural QA as Simple as Possible but not Simpler
Weissenborn, D., Wiese, G., & Seiffe, L. (2017) · Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017) · doi:10.18653/v1/k17-1028 · Cited by 164 · cited 1×
“Weissenborn et al. (2017) argue that many SQuAD questions can be answered with heuristics based on type and keyword-matching.”
BackgroundSimple Black-Box Adversarial Perturbations for Deep Networks
Narodytska, N., & Kasiviswanathan, S. P. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1612.06299 · Cited by 150 · cited 1×
“In contrast with prior work in computer vision (Papernot et al., 2017; Narodytska and Kasiviswanathan, 2016; Moosavi-Dezfooli et al., 2017), A DD O NE S ENT does not require any access to the model or to any training data: it generates adversarial examples based solely on the intuition that…”
BackgroundLevesque, H. J. (2014) · Artificial Intelligence · doi:10.1016/j.artint.2014.03.007 · Cited by 144 · cited 2×
“Quantifying the extent to which a computer sys-tem exhibits intelligent behavior is a longstanding problem in AI (Levesque, 2013).”
BackgroundLearning Recurrent Span Representations for Extractive Question Answering
Lee, K., Salant, S., Kwiatkowski, T., Parikh, A. P., Das, D., & Berant, J. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.01436 · Cited by 108 · cited 2×
“We also validate our major findings on twelve other published models with publicly available test-time code: ReasoNet Single and Ensemble versions (Shen et al., 2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al.…”
MethodMulti-Perspective Context Matching for Machine Comprehension
Wang, Z., Mi, H., Hamza, W., & Florian, R. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1612.04211 · Cited by 96 · cited 2×
“, 2017), Ruminating Reader (Gong and Bowman, 2017), MultiPerspective Context Matching (MPCM) Single version (Wang et al., 2016), RaSOR (Lee et al.”
BackgroundThe LAMBADA dataset: Word prediction requiring a broad discourse context
Paperno, D., Kruszewski, G., Lazaridou, A., Pham, N. Q., Bernardi, R., Pezzelle, S., Baroni, M., Boleda, G., & Fernandez, R. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p16-1144 · Cited by 87 · cited 2×
“However, models can succeed in this paradigm by recognizing patterns that happen to be predictive on most of the test examples, while ignoring deeper, more difficult phenomena (Rimell et al., 2009; Paperno et al., 2016).”
BackgroundUnbounded Dependency Recovery for Parser Evaluation
Rimell, L., Clark, S. C., & Steedman, M. J. (2009) · Cited by 82
Making Neural QA as Simple as Possible but not SimplerPreprint
Weissenborn, D., Wiese, G. K., & Seiffe, L. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.04816 · Cited by 73 · cited 1×
“Weissenborn et al. (2017) argue that many SQuAD questions can be answered with heuristics based on type and keyword-matching.”
BackgroundEnd-to-End Answer Chunk Extraction and Ranking for Reading ComprehensionPreprint
Yu, Y., Zhang, W., Hasan, K. F., Yu, M., Xiang, B., & Zhou, B. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1610.09996 · Cited by 54 · cited 2×
“We also validate our major findings on twelve other published models with publicly available test-time code: ReasoNet Single and Ensemble versions (Shen et al., 2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al.…”
MethodExploring Question Understanding and Adaptation in Neural-Network-Based Question AnsweringPreprint
Zhang, J., Zhu, X., Chen, Q., Dai, L.-R., Wei, S., & Jiang, H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.04617 · Cited by 46 · cited 2×
“, 2017), jNet (Zhang et al., 2017), Ruminating Reader (Gong and Bowman, 2017), MultiPerspective Context Matching (MPCM) Single version (Wang et al.”
BackgroundRuminating Reader: Reasoning with Gated Multi-hop Attention
Gong, Y., & Bowman, S. R. (2018) · Workshop on Machine Reading for Question Answering (MRQA) · doi:10.18653/v1/w18-2601 · Cited by 46 · cited 2×
“We also validate our major findings on twelve other published models with publicly available test-time code: ReasoNet Single and Ensemble versions (Shen et al., 2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al.…”
BackgroundStructural Embedding of Syntactic Trees for Machine Comprehension
Liu, R., Hu, J., Wei, W., Yang, Z., & Nyberg, E. (2017) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d17-1085 · Cited by 45 · cited 1×
“…2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al., 2017), jNet (Zhang et al., 2017), Rumi-nating Reader (Gong and Bowman, 2017), Multi-Perspective Context Matching (MPCM) Single version…”
BackgroundUnbounded dependency recovery for parser evaluation
Rimell, L., Clark, S. C., & Steedman, M. J. (2009) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/1699571.1699619 · Cited by 38 · cited 2×
“However, models can succeed in this paradigm by recognizing patterns that happen to be predictive on most of the test examples, while ignoring deeper, more difficult phenomena (Rimell et al., 2009; Paperno et al., 2016).”
BackgroundMnemonic Reader for Machine Comprehension.Preprint
Hu, M., Peng, Y., & Qiu, X. (2017) · arXiv (Cornell University) · Cited by 28
Adversarial Evaluation for Models of Natural LanguagePreprint
Smith, N. A. (2012) · arXiv (Cornell University) · doi:10.1184/r1/6473021 · Cited by 19
Structural Embedding of Syntactic Trees for Machine ComprehensionPreprint
Liu, R., Hu, J., Wei, W., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.00572 · Cited by 17 · cited 1×
“…2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al., 2017), jNet (Zhang et al., 2017), Rumi-nating Reader (Gong and Bowman, 2017), Multi-Perspective Context Matching (MPCM) Single version…”
BackgroundRuminating Reader: Reasoning with Gated Multi-Hop AttentionPreprint
Gong, Y., & Bowman, S. R. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1704.07415 · Cited by 14 · cited 2×
“We also validate our major findings on twelve other published models with publicly available test-time code: ReasoNet Single and Ensemble versions (Shen et al., 2017), Mnemonic Reader Single and Ensemble versions (Hu et al., 2017), Structural Embedding of Dependency Trees (SEDT) Single and Ensemble versions (Liu et al.…”
BackgroundAdversarial Evaluation for Models of Natural Language
Smith, N. A. (2012) · arXiv · doi:10.48550/arxiv.1207.0245 · cited 1×
“Generative adversarial models pit a generative model, whose goal is to generate realistic outputs, against a discriminative model, whose goal is to distinguish the generator’s outputs from real data (Smith, 2012; Goodfellow et al., 2014).”
Background
Cited by
Explainable AI: a review of machine learning interpretability methods
Linardatos, P., Papastefanopoulos, V., & Kotsiantis, S. (2020) · Entropy · doi:10.3390/e23010018 · Cited by 2,922
“In their study, Jia and Liang [152] investigated generating examples for reading comprehension tasks: given a paragraph and a related question, the model has to generate an answer.”
Explaining Explanations: An Overview of Interpretability of Machine Learning
Gilpin, L. H., Bau, D., Yuan, B. Z., Bajwa, A., Specter, M., & Kagal, L. (2018) · 2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA) · doi:10.1109/dsaa.2018.00018 · Cited by 2,507
“Adversarial examples are not confined to images; natural language networks can also be fooled [6].”
BackgroundKnow What You Don't Know: Unanswerable Questions for SQuAD
Rajpurkar, P., Jia, R., & Liang, P. (2018) · arXiv · doi:10.48550/arxiv.1806.03822 · Cited by 2,242
“Jia and Liang (2017) created adversarial test examples that fool models trained on SQuAD 1.1.”
BackgroundMethodInfluentialSelf-citationKnow What You Don’t Know: Unanswerable Questions for SQuAD
Rajpurkar, P., Jia, R., & Liang, P. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) · doi:10.18653/v1/p18-2124 · Cited by 2,240
“Jia and Liang (2017) created adversarial test examples that fool models trained on SQuAD 1.1.”
BackgroundMethodInfluentialSelf-citationShortcut learning in deep neural networks
Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., & Wichmann, F. A. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-00257-z · Cited by 2,224
“Feature combination crucially depends on shortcut features like word length [59, 60, 14, 61], and consequently leads to a severe lack of robustness such as an inability to generalise to more challenging test conditions [62, 63, 64, 65].”
Background
Metadata
- Authors
- Jia, R., & Liang, P.
- Affiliations
- Stanford University (US)
- Venue
- Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing
- Published
- 2017
- Pages
- 2021-2031
- OpenAlex
- W2963969878
- Cited by
- 1,410
- References
- 39
- Citations by year
- 2026: 33 · 2025: 72 · 2024: 67 · 2023: 156 · 2022: 189 · 2021: 270 · 2020: 291 · 2019: 198 · 2018: 124 · 2017: 10
- FWCI
- 76.16
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation ((DGE-114747, 114747)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- adversarial examples, reading comprehension assessment, question answering, SQuAD, adversarial perturbations, language comprehension, model robustness
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.