Generate-sometimes recognize, sometimes not

We report three experiments testing the extension of Jacoby and Hollingshea d's (1990) revised generate-recognize model of recall to a conceptually dri ven recall test. According to the model, recall involves generation of item s plus recognition of some but not all generated items. Inconsistent with t his model, production of high-frequency words was greater on a test that re quired recognition of all generated items than on a recall test (Experiment s In and Ib), and production of low-frequency words was greater on a recall test than on a generation test that did not require a recognition check (E xperiments la, Ib, and 2). Further, the production advantage for high-frequ ency over low-frequency words was greater on a test that required recogniti on of all generated items than on a recall test. This result is not consist ent with the idea that some items are recalled without a recognition check. If so, then high-frequency words should have exhibited a greater advantage over low-frequency words on a recall test than on a test that required rec ognition of all generated items. The results are interpreted within the vie w that recall can involve using information horn encoding to access target information directly at retrieval. (C) 1999 Academic Press.