The Sensitivity of Language Models and Humans to Winograd Schema Perturbations

Mostafa Abdou, Vinit Ravishankar, Maria Barrett, Yonatan Belinkov, Desmond Elliott, Anders Søgaard


Abstract
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.
Anthology ID:
2020.acl-main.679
Volume:
Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
Month:
July
Year:
2020
Address:
Online
Venue:
ACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
7590–7604
Language:
URL:
https://www.aclweb.org/anthology/2020.acl-main.679
DOI:
10.18653/v1/2020.acl-main.679
Bib Export formats:
BibTeX MODS XML EndNote
PDF:
http://aclanthology.lst.uni-saarland.de/2020.acl-main.679.pdf
Dataset:
 2020.acl-main.679.Dataset.zip
Video:
 http://slideslive.com/38929042