本研究檢驗由語意透明度所引發的再認記憶鏡像效應之中，對於語意不透明詞及語意透明詞的再認表現如何受到不同類型的記憶登錄作業調節，並藉此探討不透明詞擁有較佳再認記憶表現的原因。實驗一使用具體度判斷作為登錄作業，結果發現受試者對不透明詞仍然展現較佳的正確再認率及較多的「記得」反應，顯示在學習階段二者處理深度的不同無法完全解釋不透明詞再認表現較佳的現象。實驗二採用透明度判斷作為登錄作業。結果透明詞不論新舊，皆比不透明詞引起較多的「舊」及「記得」反應，表示不透明詞的再認記憶優勢已經被此種強調部件意義及其與全詞意義一致性的作業削弱。實驗三採用與實驗二相同的實驗設計，並紀錄受試者進行再認時的腦電波。結果發現不論透明詞或不透明詞引發大小相當的左側頂葉新舊效果及中額葉區新舊效果，而且此二效果的分布狀況不因語意透明度高低而有所不同，這樣的結果不支持提取透明詞與不透明詞時的神經來源有所不同。由於前述實驗的正確再認率非常高，因此實驗四延長學習階段與測驗階段之間的時距，藉以排除實驗二及三的結果為天花板效應。實驗五測量受試者在不同登錄作業下，對透明詞及不透明詞記憶難易程度的主觀評量。結果發現受試者在使用詞彙判斷做為登錄作業的情況下，主觀評定不透明詞比透明詞容易記憶；然而在受試者採用透明度判斷做為登錄作業時，主觀評定的記憶難易度在二類詞之間則沒有差異。此結果表示在語意透明度引起的鏡像效應中，由於受試者根據記憶難易度對不透明詞採用比較嚴格的決斷標準，因此比起透明詞有較低的錯誤再認比率。本研究根據以上結果，指出不透明詞的記憶優勢及引發較多的回想歷程（recollection）來自於全詞意義及部件意義的不一致性，然而此優勢可藉由使受試者在登錄時專注在部件意義的方式削弱。 This study investigated the mirror effect in recognition memory for semantic transparency: why semantically opaque words are better remembered than transparent ones. To examine whether the recognition advantage for opaque words results from a more elaborative encoding processing in comparison to that for transparent ones, the types of encoding tasks were manipulated among the experiments. Experiment 1 employed concreteness judgments as the encoding task. The hit rate and the proportion of “Remember” response remained to be higher for opaque words than for transparent ones. These results indicated that the elaborative encoding processing on the opaque words cannot fully account for the recognition advantage of these words. A transparency judgment task was employed as the encoding task in Experiment 2. Both old and new transparent words elicited more “old” r and “Remember” responses than opaque words. The results suggested that the recognition advantage of opaque words was decreased by the encoding task that demanded participants to process the meanings of each constituent. In the encoding task of transparency judgment, the representations of the transparent words might have been marked because of the consistency between the meanings of the transparent words and their constituents. These words might therefore be better remembered than opaque ones. Experiment 3 employed a similar procedure of Experiment 2, during which ERPs were recorded at test. Both transparent and opaque words showed reliable mid-frontal and left-parietal effects. The magnitudes and the topographic distributions of these two effects were not differed by transparent words and opaque words. The ERP data revealed no evidence that there are non-overlapping neural substrates underlying the retrievals of semantically transparent and opaque words. The hit rates for both kinds of words were high (above 90%) in Experiments 2 and 3, leading to the concerns of ceiling effects. The possibility was excluded in Experiment 4 by deferring the test phase to one day after the study phase, which resulted in lower but equivalent hit rates for both types of words. By measuring the subjective memorability of transparent and opaque words following different encoding conditions, Experiment 5 revealed that the false-alarm portion of the semantic transparency mirror effect was a consequence of criterion shifting. Participants tend to adopt a more stringent criterion for opaque words than that for opaque words when lexical decision was employed as the encoding task. In sum, the results of the five experiments suggest that the incongruence between the meanings of opaque words and their constituents resulted in the better recognition performance and greater recollection than transparent words. However, this advantage can be decreased when the encoding task emphasized the processing of the meanings of the individual constituents of both transparent and opaque words.
