This study examined how linguistic complexity features contribute to second language (L2) processing effort by analyzing the Dutch English-L2 learners’ eye movements from GECO and MECO, two eye-tracking corpora. Processing effort was operationalized as reading rate, mean fixation duration, regression rate, skipping rate, and mean saccade amplitude. In Study 1, the lexical, syntactic, and discoursal indices in 272 snippets of a novel in GECO were regressed against these eye-movement measures. The results showed that the one-component partial least square regression (PLS-R) models could explain 11%–37% of the variance in these eye-movement measures and outperformed eight readability formulas (six traditional and two recent cognitively inspired formulas based on the readers’ perception on text difficulty) in predicting L2 processing effort. In Study 2, the eye-tracking data from MECO were used to evaluate whether the findings from Study 1 could be applied more broadly. The results revealed that although the predictability of these PLS-R components decreased, they still performed better than the readability formulas. These findings suggest that the linguistic indices identified can be used to predict L2 text processing effort and provide useful implications for developing systems to assess text difficulty for L2 learners.

本研究通过分析来自 GECO 和 MECO（两个眼球追踪语料库）的荷兰英语 L2 学习者的眼球运动，探讨了语言复杂性特征如何促进第二语言 (L2) 处理工作。处理工作量被操作化为阅读率、平均注视持续时间、回归率、跳跃率和平均扫视幅度。在研究 1 中，GECO 中一部小说的 272 个片段的词汇、句法和话语索引根据这些眼动测量进行了回归。结果表明，单成分偏最小二乘回归 (PLS-R) 模型可以解释这些眼球运动测量中 11%–37% 的方差，并且优于八个可读性公式（六个传统的和两个最近的基于认知启发的公式）读者对文本难度的看法）来预测 L2 处理工作。在研究 2 中，MECO 的眼动追踪数据被用来评估研究 1 的研究结果是否可以更广泛地应用。结果表明，尽管这些 PLS-R 分量的可预测性有所下降，但它们仍然比可读性公式表现得更好。这些发现表明，所确定的语言指数可用于预测 L2 文本处理工作，并为开发评估 L2 学习者文本难度的系统提供有用的启示。