Question Generation from Text and Multi-hop Question Generation and Answering from Knowledge Base

Abstract

Question answering (QA) is the task of finding out the correct answer given a question and a knowledge source such as unstructured text or structured knowledge base (KB). In contrast, the task of question generation (QG) is a reverse task to generate a corresponding natural language question given knowledge in structured (KB) or unstructured form (text) and optionally a target answer. The motivation for QG is to generate large scale high-quality QA training data, which will help in improving the performance of QA model and also in increasing the efficiency of human annotators in QA dataset construction. In this thesis, we study the problem of automatically generating meaningful, relevant and challenging questions from sentences, paragraphs, and knowledge base.