Zhan, W., Cao, X., Cui, W., & Chang, B. (2019). An Experimental Study on Discriminating Chinese Near Synonyms: Contrasts between Machine Learning Systems and Second Language Learners. Journal of Technology and Chinese Language Teaching, 10(1), 1-25.
[詹卫东, 曹晓玉, 崔巍, & 常宝宝. (2019). 中文近义词辨析实验——机器学习程序与二语学习者的对比. 科技与中文教学 (Journal of Technology and Chinese Language Teaching), 10(1), 1-25.]

Full paper

Abstract/摘要：

In this article, machine learning technology is introduced into fill-in-the-blank (FITB) tasks involving the discrimination of Chinese near synonyms. A preliminary experimental study was carried out on said tasks between machines and L2 learners of Chinese. This study adheres to principles of balance and comprehensiveness in selecting synonyms and making test sets for the experimental research. The test results show that the performance of machines in discriminating Chinese near synonyms in FITB tasks is significantly comparable to that of human L2 learners. The score of the machine was also positively correlated with that of intermediate-level Chinese learners. In addition to the sets of near synonyms varying in difficulty, the difference of test question types also has a significant impact on test scores. The influence of lexical meaning features on the discrimination of near synonyms is no less than that of its formal features. Meanwhile, it is more effective for machines and L2 learners to exploit syntactic formal features rather than distinguishing collocation features in FITB tasks.

本文将机器学习技术引入中文近义词辨析任务，与二语学习者在近义词辨析任务上展开了初步的实验对比研究。在近义词集的选取、测试题制作方面，本文遵循平衡与周全原则。测试结果显示：机器在中文近义词辨析任务上的表现与二语学习者有明显可比性，机器测试成绩与中级水平的汉语学习者测试成绩呈正相关。除近义词本身难度有别外，题型差异对测试成绩有显著影响。词语意义特征对机器近义词辨析的影响并不低于形式特征的影响，在辨析时机器和二语者对句法形式特征的把握比对搭配区别特征的把握更有效。