Can associative information be strategically separated from item information in word-pair recognition?

Subjects learned word pairs either once or three times but were instructed to ignore the association of the two words in each pair and only to memorize the individual words at study and recognize them at test. The test word pairs included intact pairs, rearranged pairs made up of old words exchanged among the studied pairs, and new pairs consisting of one old and one new word. Subjects were instructed to respond yes to both intact and rearranged pairs and no to new pairs. Results showed that the rearranged test pairs, as compared with intact pairs, produced longer reaction times, lower discrimination (d'), and lower remember judgments. In addition, more learning either did not reduce or actually increased the pairing effect. These results are difficult to reconcile with the notion that associative information is retrieved in a slow, controlled, and effortful process but are consistent with the encoding specificity principle and the global activation theory of recognition.