以語文特徵為基之中學閱讀測驗短文分級

Using Linguistic Features to Classify Texts for Reading Comprehension Tests at the High School Levels

短文閱讀是中階語文學習者的重要功課，閱讀測驗也是語文測驗中的重要項目。掌握文本的分級技術，是電腦輔助閱讀測驗選題和電腦輔助閱讀的重要基礎。雖然閱讀測驗的文本部分，並不能全然決定整體試題難易度，但是文本部分的分級，仍為一個相關的基石研究。本研究以國內高級中學程度的真實閱讀測驗文本為研究語料，考慮詞彙、句子表面特徵與句法相關訊息等特徵，搭配不同的機器學習技術進行分類工作。我們分析了不同類別文字資訊對於分類正確性的貢獻度，目前最高可達53.6%的分類正確性。

We investigate the issue of classifying short essays based their linguistic issues, for English at the high school levels. A good selection of appropriate essays is crucial for the language learners and for the reading comprehension tests, which is an important type of tests for language competence examinations. Although the text alone does not allow us to judge the difficulty of reading comprehension tests, the capability to identify the levels of high school students for whom the texts were used in the reading comprehension can be an important step toward computer assisted selection of reading comprehension test items. We employed word-level statistics, sentence-level statistics, and syntactic-level information of the text, and applied several machine learning techniques for this text classification problem. Experimental results show that, with the best performing combination of features and learning method, we achieved 53.6% in accuracy.