Automatic Generation of English Vocabulary Tests

Abstract

This paper presents a novel method for automatically generating English vocabulary tests using TOEFL vo- cabulary questions as a model. English vocabulary questions in TOEFL is a multiple-choice question con- sisting of four components: a target word, a reading passage, a correct answer and distractors. Given a target word, we generate a reading passage from Web texts retrieved from the Internet, and then employ that read- ing passage and the WordNet lexical dictionary for generating question options, both the correct answer and distractors. Human evaluation indicated that 45% of the responses from English teachers mistakenly judged the automatically generated questions by the proposed method to be human-generated questions. In addition, half of the machine-generated questions were received average rating more than or equals than 3 in 5 point scale. This suggests that our machine-generated questions succeeded in capturing some characteristics of the human-generated questions, and half of them can be used for English test.