词汇量化特征对作文机评分数的预测能力分析

An Analysis of the Predictive Power of Quantitative Lexical Features on Machine-graded Essay Scores

为探究词汇量化特征对作文机评分数的预测能力,本研究以120篇学生作文为研究样本,通过人工和软件相结合的方法分析每篇作文的词汇复杂性及准确性的定量特征;同时采用SPSS18.0对作文分数及词汇量化结果进行多元回归分析。结果显示,副词比率(Adverb incidence)、代词比率(Pronoun incidence)、U指数(Uber index)以及名词上义度(Hypernymy for nouns)四项指标进入回归模型,共解释27.8%的方差。本文指出,大学英语学习者与教师应充分了解作文自动评阅系统关注的词汇特征,以应对未来大规模英语考试采用机器评阅作文这一大趋势。

To explore the predictive power of lexical quantitative features on machine-graded essay scores,the present study took 120 student essays as the research samples and analyzed the quantitative features of lexical complexity and accuracy of each essay through both manual and computational methods.Meanwhile,multiple regression analysis of the essay scores and quantitative lexical data was run by SPSS 18.0.Results indicated that four indices,namely Adverb incidence,Pronoun incidence,Uber index and Hypernymy for nouns entered the regression model,explaining 27.8% of the variance.It is concluded that college English learners and teachers should be fully aware of the lexical features attended to by Automated Writing Evaluation systems in response to the trend that machine scoring will be adopted to evaluate essays in large-scale English tests.