Are We Modeling the Task or the Annotator? An Investigation of Annotator Bias in Natural Language Understanding Datasets https://arxiv.org/abs/1908.07898 Mor Geva, Yoav Goldberg, Jonathan Berant Text annotation,