在語言的教學與學習中,教材扮演着影響教與學成效的關鍵角色。臺灣現有華語文教材的內容雖豐富多元也提供具體級別,但是這些教材對以華語為外語學習者的可讀性仍未有系統性研究。此外,由於文本可讀性的研究需整合跨領域的研究專長,因而國際間有關華語教材分級的科學化研究為數甚少。本研究以宋曜廷等人(2015)開發的CRIE-CFL文本自動分級系統進行華語文教材的可讀性分析,CRIE系統之華語文可讀指標包含詞彙、語意、語法與篇章凝聚等多層次特徵,其算則是以支向量機(support vector machine, SVM)機器學習法為本,以對應CEFR等級的華語文本為可讀性效標,此系統對於華語為外語教材之可讀性預測正確率為89.86%。本研究以《幼童華語讀本》、《遠東生活華語》、《實用中文讀冩》等臺灣現行流通之華語教材約六百篇,以CRIE-CFL文本分析系統進行文本分析,結果顯示部分華語教材冊別間的分級較不明確,編撰者所撰寫的教材內容與所設定的學習者存在落差。
Readability has been of long-standing research interest to educational psychologists, it has well established that texts with high readability facilitate comprehension and learning efficiency. However, few readability studies focus on Chinese or texts that designed for learner of Chinese as foreign language. Sung (2015) proposed an approach for constructing and validating readability formulae by integrating multilevel linguistic features with the machine learning(support vector machine, SVM) model, and developed a tool for the automated analysis of Chinese texts called the Chinese Readability Index Explorer for Chinese as a Foreign Language (CRIE-CFL). The CRIE-CFL provides linguistic information, readability-level prediction, and writing diagnosis using multilevel linguistic features as predictors and proficiency level of texts that match CEFR classified by expert teachers as criterion. The predicting accuracy of CRIE-CFL is 89.86 %. This study used 597 texts from current text books published in Taiwan, using CRI-CFL as a tool for text analysis. The results suggest that only some textbooks can be categorically distinguished in terms of the readability of texts. Furthermore, the content of the textbooks pre-determined by the text authors do not concur with the target readers' proficiency levels. In other words, the predefined reading levels of the textbooks mismatch the intended audience's reading proficiency.