Who is Mistaken?

Images: We take advantage of abstract images pioneered by Zitnick et al. [38], which have received wide interest in computer vision for studying high-level vision tasks. Chandrasekaran et al. [7] use abstract images to detect visual humor. Zhang et al. [36] explore binary questionanswering in abstract scenes, and Fouhey et al. [12] learn to predict object dynamics in clip art. While these approaches reason about image-level features and semantics, our approach looks at character-level features. Importantly,