Collecting Pairs of Word Senses and Their Context Sentences for Generating English Vocabulary Tests

Abstract

This paper describes our ongoing work for automatically generating multiple-choice questions for English vocabulary tests. Vocabulary-type questions consist of four components: a target word, context sentences, a correct choice and distractors. To generate such type of questions, identifying the word sense of a target word in a context is crucial to create a correct choice in the context sentences. We propose a novel Backward method that utilizes the existing search engine API to obtain context sentences for a given target word sense and further applied word sense disambiguation to confirm the retrieved context sentences. Our preliminary experiment showed that the proposed method achieved 0.951 in accuracy in collecting word sense-context pairs.