Automatic generation of multiple-choice items for prepositions based on Word2vec

Abstract

The automatic generation of multiple-choice item (MI) has attracted amounts of attention. However, only a limited number of existing research address automatic MI generation for prepositions, and even fewer consider learners’ need in the generation process. In this paper, we propose an approach to generate preposition MIs suitable for non-native English learners of different language proficiency. First we select sentences with similar difficulty level to that of a given textbook as stems by using the sentence difficulty model we constructed. Then, we use the Word2vec model to retrieve a preposition list of distractor candidates where three of them are chosen as distractors. To validate the effectiveness of our approach, we produce four tests of preposition MIs at different difficulty levels and conduct a series of experiments regarding evaluations of stem difficulty, distractor plausibility and reliability. The experimental results show that our approach can generate preposition MIs targeting learners at different levels. The results of distractor plausibility and reliability also point to the validity of our approach.