大规模大学英语口语测试朗读题型机器阅卷可行性研究与实践  

Machine scoring of reading aloud item of large-scale college English oral tests

作　　者：李萌涛[1] 杨晓果[1] 冯国栋[1] 吴敏[1] 陈纪梁[1] 胡国平[2] 

机构地区：[1]中国科学技术大学,安徽合肥230026 [2]安徽科大讯飞信息科技股份有限公司,安徽合肥230088

出　　处：《外语界》2008年第4期88-95,共8页Foreign Language World

摘　　要：本文是关于大规模大学英语口语朗读题型测试及其机器阅卷的研究。我们从中国科学技术大学2005—2007年期末口语考试中随机抽取了836份有效数据，并进行了人工阅卷与机器阅卷的比对实验。研究结果表明，机器阅卷与人工阅卷所得成绩的相关度为0．713，在5分制评分标准下平均分差为0．335，分差大于1的仅占3．2％。随着口语考试机器阅卷数据库的不断扩大，机器阅卷的信度还将进一步提高。因此，在大规模英语口语测试中，朗读题型采用机器阅卷已具备较高的评分信度与实用性，可替代人工阅卷。The current research paper is about the investigations into the reading aloud item of large-scale college English oral tests and its possible machine scoring. 836 valid reading aloud samples are collected from college English final oral tests in USTC in 2005 and 2007. The reading aloud samples are scored by both trained scorers and the Machine Scoring System. The results are as follows： the correlation coefficient between scorers＇ mean scores and the machine scores is 0.713, and the two series of scores differ at a 0.335 level, with the full mark 5 points. The difference larger than 1 point only counts for 3. 2%. With the growth of the machine scoring database, the scoring reliability will increase. So it is concluded that in scoring reading aloud item of large-scale college English oral tests, the machine scoring can function as a substitute for human scorers with high reliability and efficiency.

关 键 词：朗读 大规模英语口语测试 机器阅卷 

分 类 号：H319[语言文字—英语]