Do Neural Language Models Overcome Reporting Bias?
Do Neural Language Models Overcome Reporting Bias?
复制标题
DOI：
10.18653/v1/2020.coling-main.605
复制
发表时间：
2020-12
期刊：
影响因子：
--
通讯作者：
Vered Shwartz;Yejin Choi
中科院分区：
文献类型：
--
作者：
Vered Shwartz;Yejin Choi
从语料库中挖掘常识性知识会受到报告偏见的影响，过度代表罕见的知识而忽视了琐碎的知识（Gordon 和 Van Durme，2013）。我们研究预训练的语言模型在多大程度上克服了这个问题。我们发现，虽然他们的泛化能力使他们能够更好地估计频繁但不言而喻的行为、结果和属性的合理性，但他们也倾向于高估非常罕见的行为、结果和属性，从而放大了训练语料库中已经存在的偏见。
Mining commonsense knowledge from corpora suffers from reporting bias, over-representing the rare at the expense of the trivial (Gordon and Van Durme, 2013). We study to what extent pre-trained language models overcome this issue. We find that while their generalization capacity allows them to better estimate the plausibility of frequent but unspoken of actions, outcomes, and properties, they also tend to overestimate that of the very rare, amplifying the bias that already exists in their training corpus.