The Sensitivity of Language Models and Humans to Winograd Schema Perturbations
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
The Sensitivity of Language Models and Humans to Winograd Schema Perturbations. / Abdou, Mostafa; Ravishankar, Vinit; Barrett, Maria; Belinkov, Yonatan; Elliott, Desmond; Søgaard, Anders.
Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, 2020. p. 7590-7604.Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - The Sensitivity of Language Models and Humans to Winograd Schema Perturbations
AU - Abdou, Mostafa
AU - Ravishankar, Vinit
AU - Barrett, Maria
AU - Belinkov, Yonatan
AU - Elliott, Desmond
AU - Søgaard, Anders
PY - 2020
Y1 - 2020
N2 - Large-scale pretrained language models are the major driving force behind recent improvements in perfromance on the Winograd Schema Challenge, a widely employed test of commonsense reasoning ability. We show, however, with a new diagnostic dataset, that these models are sensitive to linguistic perturbations of the Winograd examples that minimally affect human understanding. Our results highlight interesting differences between humans and language models: language models are more sensitive to number or gender alternations and synonym replacements than humans, and humans are more stable and consistent in their predictions, maintain a much higher absolute performance, and perform better on non-associative instances than associative ones.
AB - Large-scale pretrained language models are the major driving force behind recent improvements in perfromance on the Winograd Schema Challenge, a widely employed test of commonsense reasoning ability. We show, however, with a new diagnostic dataset, that these models are sensitive to linguistic perturbations of the Winograd examples that minimally affect human understanding. Our results highlight interesting differences between humans and language models: language models are more sensitive to number or gender alternations and synonym replacements than humans, and humans are more stable and consistent in their predictions, maintain a much higher absolute performance, and perform better on non-associative instances than associative ones.
U2 - 10.18653/v1/2020.acl-main.679
DO - 10.18653/v1/2020.acl-main.679
M3 - Article in proceedings
SP - 7590
EP - 7604
BT - Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
PB - Association for Computational Linguistics
T2 - 58th Annual Meeting of the Association for Computational Linguistics
Y2 - 5 July 2020 through 10 July 2020
ER -
ID: 258374819