Automatic Chinese Factual Question Generation

Abstract

Question generation is an emerging research area of artificial intelligence in education. Question authoring tools are important in educational technologies, e.g., intelligent tutoring systems, as well as in dialogue systems. Approaches to generate factual questions, i.e., questions that have concrete answers, mainly make use of the syntactical and semantic information in a declarative sentence, which is then transformed into questions. Recently, some research has been conducted to investigate Chinese factual question generation with some limited success. Reported performance is poor due to unavoidable errors (e.g., sentence parsing, name entity recognition, and rule-based question transformation errors) and the complexity of long Chinese sentences. This article introduces a novel Chinese question generation system based on three stages, sentence simplification, question generation and ranking, to address the challenge of automatically generating factual questions in Chinese. The proposed approach and system have been evaluated on sentences from the New Practical Chinese Reader corpus. Experimental results show that ranking improves more than 20 percentage of questions rated as acceptable by annotators, from 65 percent of all questions to 87 percent of the top ranked 25 percent questions.

Publication Title

IEEE Transactions on Learning Technologies

Share

COinS