贅詞減量的語言測驗：克漏字與C測驗的重新檢驗

Language Reduced Redundancy Tests: A Reexamination of Cloze Test and C-test

本研究的主要目的：(1)探究採用由Farhady與Keramatic(1996)兩位學者所提出的由文本決定挖空方法(text-driven method)的克漏字測驗，與標準克漏字測驗(standard cloze test)比較起來，前者是否會產生較好的測量特質(psychometric propenies)；(2)比較克漏字測驗和C-test的測量特質；(3)探究採用不同的挖空比例是否會導致考生的考試成績表現高低不同；(4)驗證「考生不會因採用不同文本而產生不同的考試成績表現」(test-takers' performance invariance across different texts)這個假定(assumption)是否適用於克漏字測驗和C-test。本研究的受試者是來自北臺灣一所大學的二百三十七位大一學生。本研究的結果顯示從信、效度的層面來看，由文本決定挖空方法的三個克漏字測驗版本及C-test，並沒有比標準克漏字測驗版本來得好。針對不同的挖空比例是否會導致考生的考試成績表現高低不同這個研究問題，本研究似乎未能得到具有結論性的證據。本研究也未得到強而有力的證據來支持「考生不會因採用不同文本而產生不同的考試成績表現」這個假定。

The purposes of this study are: (1) to investigate whether cloze forms with text-driven deletion method, proposed by Farhady and Keramati (1996), will produce better psychometric properties than the standard cloze form; (2) to compare the psychometric properties of cloze test and C-test, both of which belong to the family of language reduced redundancy test; (3) to examine whether different deletion rates lead to difference in test-takers' performance; and (4) to test whether the assumption of test-takers' performance invariance across different texts hold for both cloze test and C-test. Based on two authentic texts with different rhetoric modes, three cloze forms with text-driven deletion method, along with one standard cloze and one form of C-test, were constructed and randomly administered to 237 student subjects at one private university in northern Taiwan. Furthermore, each subject was required to take three subtests (from a sample TOEFL test) as criterion measures for empirical validity. The results of the study indicated that neither the three cloze forms nor the C-test was substantially superior to the standard cloze in terms of reliability and validity. In addition, the findings of the study were inconclusive with regard to whether different deletion rates result in different test-takers' performance. Finally, no strong evidence was found to substantiate the claim that both cloze test and C-test meet the assumption of test-takers' performance invariance across different texts.