A substantive body of research has been revolving around the linguistic features that distinguish different levels of students’ writing samples (e.g. Crossley and McNamara 2012; McNamara et al. 2015; Lu 2017). Nevertheless, it is somewhat difficult to generalize the findings across various empirical studies, given that different criteria were adopted to measure language learners’ proficiency levels (Chen and Baker 2016). Some researchers suggested using the Common European Framework of Reference for Languages (CEFR) (Council of Europe 2001) as the common standard of evaluating and describing students’ proficiency levels. Therefore, the current research intends to identify the linguistic features that distinguish students’ writing samples across CEFR levels by adopting a machine-learning method, decision tree, which provides the direct visualization of decisions made in each step of the classification procedure. The linguistic features that emerged as predicative of CEFR levels could be employed to (i) inform L2 writing instruction, (ii) track long-term development of writing ability, and (iii) facilitate experts’ judgment in the practice of aligning writing tests/samples with CEFR.

中文翻译：

---

区分学生写作能力与 CEFR 水平一致的语言特征

大量研究一直围绕着区分不同水平学生写作样本的语言特征（例如 Crossley 和 McNamara 2012；McNamara 等人 2015；Lu 2017）。然而，鉴于采用不同的标准来衡量语言学习者的熟练程度，将各种实证研究的结果概括起来有些困难（Chen 和 Baker 2016）。一些研究人员建议使用欧洲共同语言参考框架（CEFR）（欧洲委员会2001）作为评估和描述学生熟练程度的通用标准。因此，当前的研究旨在通过采用机器学习方法、决策树、它提供了分类过程每个步骤中所做决策的直接可视化。作为 CEFR 水平预测的语言特征可用于 (i) 为 L2 写作教学提供信息，(ii) 跟踪写作能力的长期发展，以及 (iii) 促进专家在调整写作测试/实践中的判断具有 CEFR 的样本。