Memory & Cognition 2002, 30 (6), 850-861 The generation effect: Dissociating enhanced item memory and disrupted order memory NEIL W. MULLIGAN Southern Methodist University, Dallas, Texas Generating stimuli at encoding typically improves memory for occurrence (item memory) but might disrupt memory for order. In three experiments, the relationship between generation and order mem- ory was examined by using familiar stimuli, which give rise to the standard generation advantage in item memory, and unfamiliar stimuli, which do not. The participants generated or read words and non- words in Experiments 1 and 2 and familiar and unfamiliar word compounds in Experiment 3. For the familiar stimuli, generation enhanced item memory (as measured by recognition) but disrupted per- formance on the order-reconstruction test. For the unfamiliar stimuli, generation produced no recog- nition advantage and yet persisted in disrupting order reconstruction. Thus, the positive effects of gen- eration on item memory were dissociated from its negative impact on order memory. Self-generated information is often better remembered memory; although generation may enhance memory for than information that is merely perceived, a phenomenon the occurrence of a stimulus, it can disrupt memory for known as the generation effect (Slamecka & Graf, 1978). serial-order information (e.g., Burns, 1996; Burns, Curti, In a typical study of this effect, participantsgeneratesome & Lavin,1993; Nairne et al., 1991; Serra & Nairne, 1993; study words (perhaps from an antonym or a word frag- cf. Kelley & Nairne, 2001).1 ment) and read others. On a later memory test, the gener- A prominent account argues that the effects of genera- ated items usually produce better performance. The gen- tion on item and order information are central to our un- eration effect has been observed with a variety of materials derstanding the effects of generation on memory in gen- (e.g., singleand compoundwords, sentences, abbreviations, eral (Burns, 1996;Burns et al., 1993; DeLosh & McDaniel, numbers), in which a variety of generationtasks (e.g., gen- 1996;Nairne et al., 1991; Serra & Nairne, 1993).This view eration from antonyms, semantic associates, rhymes, ana- is related to multifactor accounts of generation, which grams, word fragments, second-languagetranslations,de- propose that generation enhances item but disrupts in- finitionshave been used), and on a variety of memory tests teritem relationalprocessing (e.g., deWinstanley,Bjork, & (e.g., free recall, cued recall, recognition,comprehension; Bjork, 1996;Hirshman & Bjork, 1988;Hunt & McDaniel, see Greene, 1992; Mulligan, 2001a, for a review). 1993; McDaniel, Wadill, & Einstein, 1988; Steffens & Despite the impressive generality of this effect, genera- Erdfelder, 1998). Item information refers to features of an tion does not always enhance memory and may even dis- item that are unique or distinctive, whereas relational in- rupt certain aspects of performance. For example,when the formation refers to associations between items within an generation manipulation is implemented in a between- encoding context, including those that underlie serial- subjects or pure-list design, generation typically does not order information (e.g., Hunt & McDaniel, 1993; Mc- enhance recall (e.g., Hirshman & Bjork, 1988; Slamecka Daniel, DeLosh, & Merritt, 2000). Given the importance & Katsaiti, 1987; although, see Mulligan, 2001a, for evi- of both item and interitem relational information to free dence that a between-subjects generation effect emerges recall (Hunt & McDaniel, 1993), this view accounts for over repeated recall tests). Under some conditions, pure- the dependence of generation effects on experimental de- list designs can even produce a negative generation effect sign. According to this view,when generate items are pre- in which the read conditionproduces superior recall (e.g., sented in a pure list, disruption of order and interitem en- Nairne, Reigler,& Serra, 1991;Steffens & Erdfelder, 1998). coding is list-wide and detracts from the typical generation A second negative generation effect, especially important effect in recall, causing generation effects to disappear or in the present context, has been reported for tests of order even reverse (e.g., Hirshman & Bjork, 1988; Kinoshita, 1989; Slamecka & Katsaiti, 1987; Steffens & Erdfelder, 1998), even though the generation advantage persists in recognition, a test with little reliance on interitem pro- This research was partially supported by Grant 1-R03-MH61324-01 from NIMH. Correspondence should be addressed to N. Mulligan, De- cessing (e.g., Hunt & McDaniel, 1993; Kinoshita, 1989). partment of Psychology, University of North Carolina, Chapel Hill, NC In mixed lists, the disruption of intertarget processing af- 27599-3270 (e-mail: [email protected]). fects read and generate items equally, which in turn per- Copyright 2002 Psychonomic Society, Inc. 850 GENERATION AND ORDER MEMORY 851 mits the superior item encoding in the generate condition Given the similarities between the effects of generation to produce a recall advantage (Hunt & McDaniel, 1993; and perceptual interference, the recent results of Mulligan Nairne et al., 1991). (2000a) are suggestiveof the relationship between gener- A number of variables besides generation have been ation and order memory.Mulligan(2000a)used a limiting proposed to enhance item encoding (see Hunt & Mc- conditionof the perceptual-interference effect (i.e., a ver- Daniel, 1993). Interestingly,a number of these produce a sion of the perceptual-interferencemanipulationthat does disruption to order encoding similar to that produced by not give rise to the typical advantage in item memory) to generation. These manipulations include bizarre versus examine the relationship between perceptual interference common sentences(McDaniel,Einstein,DeLosh, May,& and order. Delayingthe onset of the mask in the perceptual- Brady, 1995), enacted versus observed events (Engelkamp interference condition (thus increasing the presentation & Dehn, 2000), low- versus high-frequency words (De- time of words) produces such a limiting condition.When Losh & McDaniel, 1996), and words presented with or the mask is delayed to a point at which it no longer inter- without perceptual interference (Mulligan, 1999). As De- feres with word perception (e.g., 266 msec), the typical Losh and McDaniel have noted, these manipulationstyp- perceptual-identification effect in item memory is not ically contrast a relativelyunusualencodingcondition(e.g., found in recall (Hirshman, Trembath, & Mulligan, 1994) the generate condition, bizarre sentences, low-frequency or recognition (Mulligan, 2000a). Study items were pre- words) with a more common encodingcondition(e.g., the sented under three conditions—the standard perceptual- read condition,common sentences, high-frequencywords), interference condition(with a 100-msec mask), a delayed leading to the suggestion that the unusual condition at- masking condition(with a 266-msec mask), and an intact tracts greater attention to item characteristics at the ex- control condition (with no mask). On a later recognition pense of processing order and relationalinformation (De- test, the standard perceptual-interference condition pro- Losh & McDaniel, 1996; Engelkamp & Dehn, 2000; duced better performance than either the delayed mask or Nairne et al., 1991; Serra & Nairne, 1993). the intact condition,which did not differ from one another. It is theoreticallyimportant to study memory for order However, on the order-reconstruction task, the standard because disrupted memory for order might be causally re- and delayed conditions produced equivalently worse per- lated to superior item memory in the unusual encoding formances than did the intact condition.Thus, the standard condition (DeLosh & McDaniel, 1996; Engelkamp & perceptual-interferenceconditionenhanceditem memory Dehn, 2000; Nairne et al., 1991; Serra & Nairne, 1993; but disrupted order memory, whereas the delayed masking see Greene, Thepar, & Westerman, 1998, for discussion). condition had no effect on item memory but persisted in However, there are reasons to question whether disrupted disrupting order memory. Given the similarities between order encoding and enhanced item encoding are inex- the perceptual-interference and generation effects, these orably linked in this set of encoding manipulations. Re- results raise the possibilitythat the positiveeffects of gen- cent research on the perceptual-interference effect moti- eration on item memory can likewise be dissociated from vates this question (Mulligan, 2000a). The perceptual- its negative impact on order memory. interference effect is observed when some study words are In the following experiments, the relationship between presented very briefly (e.g., 100 msec) and then are back- generation and order memory was investigated by exam- ward masked (the perceptual-interference condition), ininglimitingconditionsof the generationeffect. One im- whereas others are presented for a longer period (e.g., portant determinant of the generation effect is the type of 2.5 sec; the intact condition).The perceptual-interference materials used. The generation effect in item memory is conditiontypicallyleads to better performance on tests of typically not obtained with unfamiliar materials, such as item occurrence, such as recognitionand free and cued re- nonwords and unfamiliar word compounds, prompting call (e.g., Hirshman & Mulligan, 1991; Mulligan, 1996, claims that the