Adversarial Examples for Evaluating Reading Comprehension Systems

Robin Jia, Percy Liang


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.
Anthology ID:
D17-1215
Volume:
Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing
Month:
September
Year:
2017
Address:
Copenhagen, Denmark
Editors:
Martha Palmer, Rebecca Hwa, Sebastian Riedel
Venue:
EMNLP
SIG:
SIGDAT
Publisher:
Association for Computational Linguistics
Note:
Pages:
2021–2031
Language:
URL:
https://aclanthology.org/D17-1215
DOI:
10.18653/v1/D17-1215
Bibkey:
Cite (ACL):
Robin Jia and Percy Liang. 2017. Adversarial Examples for Evaluating Reading Comprehension Systems. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2021–2031, Copenhagen, Denmark. Association for Computational Linguistics.
Cite (Informal):
Adversarial Examples for Evaluating Reading Comprehension Systems (Jia & Liang, EMNLP 2017)
Copy Citation:
PDF:
https://aclanthology.org/D17-1215.pdf
Video:
 https://aclanthology.org/D17-1215.mp4
Code
 worksheets/0xc86d3ebe +  additional community code
Data
SQuAD