Automatic question generation from text for self-directed learning

Question generation from text is a Natural Language Generation task of vital importance for self-directed learning. Learners have access to learning materials from a wide variety of sources, and these materials are not often accompanied by questions to help guide learning. Prior question generation techniques have focused primarily on generating factoid questions, which are often not the most pedagogically important questions for a learner. Furthermore, prior techniques have not fully leveraged the semantic content of learning materials and have not often been evaluated in a pedagogically-inspired framework. This thesis introduces a novel template-based approach to question generation that combines semantic roles with a method of generating both general and domain-specific questions. We evaluate our approach in a way that is mindful of the context in which the generated questions are to be used. This evaluation shows our approach to be effective in generating pedagogically-useful questions.