Automatic Multiple-Choice Question Generation based on Coreference Resolution

In this paper, we propose a multiple-choice question generation program based on coreference resolution for measuring learners’ comprehension of the article. The coreference of the entire article is accomplished by the connection of noun phrases referring to the same entity in the real world. We apply the coreference resolution to the issue of automatic question generation. ere we have three types of target key: pronoun, pleonastic pronoun, and NP. In order to improve question difficulty and discrimination, we employ clusters’ relation of the coreference to generate the answer and distractor. If readers understand the article, they should know which noun phrases refer to the same entity in the real world. We generate the answers to the questions that are the closest NP of target words in a coreference chain. For discriminating non-proficiency readers form proficiency readers, the answer and distractor of the question are in the similar agreement features (e.g., Number, gender et al) to confuse readers. We generate the distractors of the questions occurring in coreference chains but the one target word occurs in are as similar as possible to the target word in the agreement features.