Answering Word Sense Judgement Questions In Chinese Reading Comprehension

Reading comprehension is to ask the computer to answer the corresponding questions based on a given text,which is the most popular research in the field of artificial intelligence,involving a variety of complex natural language processing techniques.This paper studies the reading comprehension in the Chinese native language subject test of Beijing College Entrance Examinations(Beijing GaoKao),which consists of two types of reading comprehension materials: science articles and literary articles.The types of questions are multiple-choice questions and description questions.Multiple-choice questions include literary comprehension,word sense comprehension,and coreference resolution questions,and description questions include sentence comprehension,title comprehension,and summarization questions.This article focuses on the study of word sense judgment questions in multiple-choice questions in Chinese reading comprehension.The main tasks are as follows:1)A strategy is proposed to calculate support value for word sense judgment questions.The idea is that: the support value is calculated according to the support degree of the relevant sentences in the document,and then the candidate answers are ranked based on the support value.2)Three calculation methods are proposed,which are according to the word order and word senses.The three methods are respectively based on the language models,PMI and the sentence similarity,which judge sentence coherence and semantic similarity of the sentence which the word belongs to.3)The method of synonym replacement is tried to automatically generate the dataset in order to make up for the deficiency of the small size of the dataset.4)A system of answering the word sense judgment questions is implemented.We test the system in the dataset,which is collected from real Beijing GaoKao examinations and the Beijing GaoKao examination simulations.The accuracies of the three methods are respectively 62.5%,75%,and 62.5%.We also test the system in the generated dataset,and However,The accuracies of the three methods are respectively 67.5%、65.5% and 64%.It can be seen that the method proposed in this paper effectively solves the problem of word sense judgment and promotes the solution of other reading comprehension.However,there are still many deficiencies in the proposed strategy,for example,the strategy is mainly studied from the perspective of word matching,and it does not answer this question from a semantic perspective.