過往研究顯示，測驗所引起的記憶提取練習歷程相較於重複學習，更利於長期記憶之保存，此結果通稱為測驗效應。然而，測驗效應是否或如何受到測驗時產生之錯誤影響，仍有待研究。關於此議題的少量研究結果並不一致，因此，本研究使用中文的假字-注音配對探討下列問題: (1)測驗產生錯誤和沒有產生錯誤，是否比直接學習更有助於長期記憶? (2)中文假字的表音限制性如何影響試誤(測驗產生錯誤)和無誤(測驗無產生錯誤)學習? (3) 相關之電生理訊號會如何受到文字傳達的訊息量調控? 實驗中運用字型和字音之間的表音一致性，將中文假字-注音配對分為高限制性和低限制性兩組，組內一半的材料讓受試者直接學習，另一半則讓受試者透過測驗學習。分析時，測驗學習組可進一步分為學習階段答對而最終測驗也答對(無誤學習)，或是學習階段答錯而最終測驗答對(試誤學習)兩種學習型態。 實驗一至三皆顯示，在高或低限制性的假字-注音配對中，受試對高限制性組表現較好。實驗二顯示，直接學習組和測驗學習組比二擇一組表現的好。實驗二¬之二和實驗三皆顯示，無誤學習組的表現較直接學習組好，而直接學習組的表現又比試誤學習組好。事件相關腦電位結果顯示，高限制性假字比低限制性假字引起更大的N400，而且，在假字出現後的1000~1900ms中，低限制性假字比高限制性假字引起較正向的振幅。在低限制性組中，常見音將近比罕見音引起更大的P300，高限制組中則無顯著差異。此外，在注音出現後的1200~1900ms中，常見音及罕見音兩者間的振幅差異在高限制性組中較大。本研究結果指出，測驗時產生的錯誤不利於學習。不論中文假字的表音限制性高低，受試者在試誤學習中表現最差，而在無誤學習的表現比直接學習好。此外，除了N400和P300效果外，研究中還發現類似於延伸的pN400FP效果，此效果可能暗示當預期與結果不一致時，若對於事件的預期越強，需要花更多的努力去推翻原先的預期。 ;According to previous studies, testing produced better long-term retention than repeated studying because retrieval practices are needed in testing which is known as testing effect. However, it is not clear whether and how retrieval errors affect the mnemonic benefits of testing. The very few studies that examined this issue gave rise to inconsistent results. Therefore, this thesis employed Chinese pseudocharacters- ZhuYin FuHao associations to examine the following issues in the same task: (a) Does test-errorless learning and test-errorful learning benefits memory retention compared with study-only (no-test) learning? (b) How does the errorful and errorless learning affect by cue-target constraint levels? (c) How does the information conveyed by words modulate electrophysiological correlates of the testing effect? In the current study, orthography-to-phonology consistency was applied to define the constraints between Chinese pseudocharacters-ZhuYin FuHao associations. In both high-constraint and low-constraint associations, half of associations was assigned to no-test learning condition, the other half was assigned to testing condition. During analysis, the recall rates of testing condition were further calculated for errorful and errorless trials. In Experiment 1 to 3, participants’ performance on high-constraint characters was always better than low-constraint characters. The results of Experiment 2 revealed that no-test (study-only) learning group and test self-generated learning group led to better performance than test with 2 alternatives learning group. The results of the Experiment 2-2 and 3 both revealed that errorless learning led to better performance than no-test (study-only) learning, and no-test learning led to better performance than errorful learning. The ERP results revealed that high-constraint characters elicited a more negative N400 wave than low-constraint characters, and the mean amplitudes of low-constraint characters were more positive than high-constraint characters in the time windows from 1000 to1900ms. The ERPs elicited by the ZhuYin FuHao revealed that common pronunciations marginally elicited more positive P300 than uncommon pronunciations for low-constraint condition but not for high-constraint condition. In addition, mean amplitudes difference between common and uncommon pronunciations was greater on high-constraint characters in the time windows from 1200 to1900ms. The current findings suggested that test with error occurred decreased the benefit of testing on learning. Errorful learning led to the worse retention, and errorless learning led to better retention than no-test learning whether the cue-target constraint was high or low. In addition, except for ERP results revealed consistency (cue-target constraint) and oddball effects at N400 and P300, the novel prolonged pN400FP like effects were found, which might suggested that more efforts were needed to override a strongly held prediction than weakly held prediction.
