Abstract This study examined effects of n-gram measures (i.e., range, frequency, and association strength) on the rated L2 writing quality. A corpus of 1643 expository essays were analyzed with a natural language processing tool. Results of separate regression trees showed that spoken bigram range, spoken bigram proportion 10K, academic trigram frequency, and spoken bi- and-trigram association strength (ΔP and MI) were reliably predictive of the variance in L2 writing quality. However, when indices of the range, frequency and association strength were combined, results of the regression tree indicated that spoken bigram proportion 10k, spoken trigram ΔP, spoken bigram MI, and spoken bigram range were reliable predictors for the rated L2 writing quality. These observations differed from previous ones with respect to effects of n-gram measures on the rated L2 writing quality. Such differences call for the need to systematically assess the predictive power of n-grams on L2 writing quality across genres.

中文翻译：

---

n-grams对说明性论文二级写作质量的影响：概念复制和扩展

摘要 本研究检验了 n-gram 度量（即范围、频率和关联强度）对 L2 写作质量的影响。使用自然语言处理工具分析了 1643 篇说明性文章的语料库。单独回归树的结果表明，口语二元组范围、口语二元组比例 10K、学术三元组频率和口语二元组和三元组关联强度（ΔP 和 MI）可以可靠地预测二语写作质量的变化。然而，当范围、频率和关联强度的指标结合起来时，回归树的结果表明，口语二元组比例 10k、口语三元组 ΔP、口语二元组 MI 和口语二元组范围是二语写作质量的可靠预测指标。在 n-gram 测量对额定 L2 写作质量的影响方面，这些观察结果与之前的观察结果不同。这种差异要求系统地评估 n-gram 对跨流派的 L2 写作质量的预测能力。