A common drawback in traditional language education is that all students in the same class use the same content. Since students may have different backgrounds such as prior knowledge and learning speed, one single curriculum may not be able to accommodate every student. Unfortunately, most students cannot afford personalized language learning, since preparing personalized learning content can be very time-consuming and potentially requires a significant amount of expert labor. Recently, researchers have proposed automatic systems to assist language education, such as Computer-based Assessment Systems (CAT) and Intelligent Tutoring Systems (ITS). However, previous work usually characterizes the student's knowledge and the difficulty of learning content using numeric scores, which may not be comprehensive. To improve on this, this thesis introduces hierarchical knowledge structures to assist in multiple tasks in language education. First, this structure multidimensionally characterizes the difficulty of each learning material by its relative difficulty to other materials and models the whole corpus with a graph structure. Additionally, we can utilize the hierarchical knowledge structure to multidimensionally assess a student's prior knowledge, predict the student's future performance on a specific task, and recommend learning content that is appropriate for each student. Furthermore, the hierarchical knowledge structure enables us to build a framework to characterize existing learning curricula extracted from textbooks and online learning tools, and apply expert wisdom that we have discovered to automatically design learning curricula. The hierarchical knowledge structure reduces the cost of expert labor and potentially makes language education more affordable and more engaging.