Read, Skim, Scan: Gaze-Aware Documents As Feedback For Multimodal Typographical Cuing