Generating Adequate Distractors for Multiple-Choice Questions

Closing the Gap: Automated Distractor Generation in Japanese Language Testing

“…These methods have since evolved into more modern versions, which tend to use semantically related words acquired through WordNet [66,67], a thesaurus [22], or n-grams and collocations [23,68]. Recently, word embeddings have been shown to be effective distractors, and the Word2vec method has gained popularity as a generation method [4,25,47], which achieves semantically similar words. Automatic distractors' generation in languages other than English has also gained more attention in the last decade with improvements to NLP methods, allowing for training models in multiple languages [48].…”

“…Specifically, the automatic generation of questions [1][2][3] has garnered a lot of interest in the last decade across a wide variety of educational fields. An efficient method for assessing comprehension, questions are an integral tool for any education setting that benefit both learners and educators [4,5]. We can find automatic question generation in maths [6,7], biology [8], law [9], medicine [10], and languages [11,12].…”

“…Nevertheless, these methods alone do not necessarily produce adequate distractors since there are specific requirements distractors should satisfy [3,4]: (1) Be an incorrect answer to the question; (2) Be related to the correct answer, semantically or from the same category (i.e., nature, color, exercise); (3) Be grammatically correct and consistent with the difficulty of the correct answer, and; (4) Be of the same word class and a similar length to the target word.…”