The Sensitivity of Language Models and Humans to Winograd Schema Perturbations
The Sensitivity of Language Models and Humans to Winograd Schema Perturbations
复制标题
DOI：
10.18653/v1/2020.acl-main.679
复制
发表时间：
2020-05
期刊：
影响因子：
--
通讯作者：
Mostafa Abdou;Vinit Ravishankar;Maria Barrett;Yonatan Belinkov;Desmond Elliott;Anders Søgaard
中科院分区：
文献类型：
--
作者：
Mostafa Abdou;Vinit Ravishankar;Maria Barrett;Yonatan Belinkov;Desmond Elliott;Anders Søgaard
大规模预先训练的语言模型是最近Winograd模式挑战性能改进的主要驱动力，Winograd模式挑战是一种广泛使用的常识推理能力测试。然而，我们用一个新的诊断数据集表明，这些模型对Winograd例子的语言扰动很敏感，这些扰动对人类的理解影响最小。我们的结果突出了人类和语言模型之间的有趣差异：语言模型比人类对数字或性别变化和同义词替换更敏感，人类在预测方面更稳定和一致，保持更高的绝对表现，在非联想实例上的表现比联想实例更好。
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.