基于人工神经网络的C．TEST阅读理解题目难度的预测研究

A Study of Predicting Difficulty of Reading Comprehension Item of C.TEST :Based on Artificial Neural Network

实用汉语水平认定考试(简称C.TEST)是用来测试母语非汉语的外籍人士在国际环境下社会生活以及日常工作中实际运用汉语能力的考试。由于C.TEST的考试题目公开,题库数量较小,所以通过一般标准化考试采用的在部分目标被试中实施预测(field test)的方法来获取考试题目的难度参数存在困难。然而,人工神经网络技术作为现代人工智能研究的成果,在预测(prediction)领域发挥了很大作用。本文选取C.TEST(A-D级)的阅读理解题目作为研究材料,运用人工神经网络技术对其难度进行预测,得到了网络预测难度值与实际考试难度值显著相关的研究结果。这一结果表明,利用人工神经网络模型对语言测验的题目难度等参数进行预测是可行的。

Test of Practical Chinese （C.TEST） is used to test foreigners＇ Chinese language proficiency in the intemation- al environment as well as in their daily work and social life. As its items are not confidential and the numbers of items are not sufficient enough, it is difficult to obtain the difficulty of C.TEST items by field test. Currently, the assessment of the difficulty of the reading comprehension items in C.TEST entirely comes from the test maker＇s experience which is, to a certain extent, subjective. Whereas Artificial Neural Network （ANN）, which can imitate the structure and ways of working and leaming of biological neural networks, plays a significant role in the predicting field. This study selects reading comprehension in C.TEST （A-D） as the basic corpus to predict its items difficulty by the MATLAB neural network software. The result shows that there is a significant correlation between the values of difficulty predicted by ANN and the values of the actual difficulty. This shows that ANN model is suitable for predicting the language tests parameters such as difficulty of items.