The staggering amounts of content readily available to us via digital channels can often appear overwhelming. While much research has focused on aiding people at selecting relevant articles to read, only few approaches have been developed to assist readers in more efficiently reading an individual text. In this paper, we present HiText, a simple yet effective way of dynamically marking parts of a document in accordance with their salience. Rather than skimming a text by focusing on randomly chosen sentences, students and other readers can direct their attention to sentences determined to be important by our system. For this, we rely on a deep learning-based sentence ranking method. Our experiments show that this results in marked increases in user satisfaction and reading efficiency, as assessed using TOEFL-style reading comprehension tests.