Not Logged In

[Zoom] You need to be logged in, be a member, and get a ticket to attend this session.

Paper Presentation Regular Presentation

Question Generator: Preliminary evaluation

Sun, Jun 6, 10:45-11:15 Asia/Tokyo Room F

This presentation describes the preliminary evaluation of the accuracy and usability of a prototype online tool. The Question Generator automatically creates questions from an input sentence. Learners submit any sentence (simple, compound, complex or compound-complex) and select the type or types of questions to be generated. The system can create open-ended, closed-ended and tag questions for any declarative statement. This educational language app is designed to provide unlimited question practice. One way in which it can be used is for learners to create questions for a stimulus sentence (e.g. Tom loves Jane) and then compare their questions with those generated automatically (e.g. Who loves Jane? Who does Tom love?). The web app uses a combination of machine learning and rule-based parsing to select appropriate question words, pronouns and appropriate syntax needed to generate appropriate questions. Initial feedback from focus groups of university, high school and junior high school students was positive. The accuracy rate for sentences randomly extracted from MEXT-approved textbooks was high at approximately gold-standard. The accuracy of the questions is the highest for simple sentences; but as sentence complexity and ambiguity increases, accuracy decreases. Practical suggestions on how to use the app will be given. This open-access tool is developed in-house and is non-commercial.

  • John Blake

    My current focus is on creating practical online tools that help people learn English. The tools created often detect and/or visualize particular language features. Some language features are easy to detect automatically while others are much more challenging. This research draws on corpus linguistics to analyze texts and computational linguistics to create rule-based and probabilistic-based pattern-searching tools or pipelines.