A Cloze Test Authoring System and Its Automation

This paper presents a pilot system and discusses its possible extensions. In the first sections, we present a web-based test authoring system for English grammar and vocabulary. It assists language teachers to make questions on arbitrary input texts, such as online news. The system employs the techniques of NLP (Natural Language Processing) and is compared with other studies in the same field. In the latter sections we discuss some extensions of the system. First, we show a method of automatically acquiring the grammar targets. Secondly, we discuss an extension to a computer-adaptive question generator which interacts with students. We describe the design of the extended system and focus on the difficulty predictor, one of the major components needed. Then we present some primary results of the difficulty prediction task. Finally, we compare existing systems and discuss their use and evaluations.