«By Zachary Alexander Rosner A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Psychology ...»
The Positive Generation Effect Slamecka and Graf (1978) were among the first to delineate the positive effects of active generation on item memory. In a series of experiments that set a foundation for a number of subsequent studies, participants viewed blocks of various stimuli including antonyms, synonyms, associates, categories and rhymes. The read condition consisted of intact cue-target word pairs (e.g., lamp-light), while the generate condition consisted of an intact cue, and only the first letter of the target word (e.g., lamp-l). For incidental and intentional encoding, cued and free recall (retrieval of previously presented targets when presented with and without a cue, respectively), uncued and cued recognition (discrimination between previously presented targets and novel items), experimenter-paced and self-paced timing, and between- and within-subject designs, active generation consistently enhanced memory performance and confidence.
An early argument for this positive generation effect for item memory was the cognitive effort hypothesis (McFarland Jr. et al., 1980; Tyler et al., 1979). In a series of several experiments, increasing the cognitive effort involved in solving anagrams (Tyler et al., 1979), completing sentences (McFarland Jr. et al., 1980; Tyler et al., 1979), and completing rhymes (McFarland Jr. et al., 1980) enhanced subsequent memory. Indeed, a meta-analytic review by Bertsch et al., (2007) found that increasing the amount of generation required, such as generating an entire word rather than simply a portion of a word, tended to increase the size of the generation effect. However, individual studies have revealed mixed results (Nieznański, 2011, 2012). More difficult math problems have not been shown to improve memory as compared to easy math problems (McNamara & Healy, 2000), and other studies have found that while solving more difficult anagrams leads to improved source memory (in this case, memory for whether participants scrambled or solved an anagram), the difficulty of the anagram did not impact later item recognition performance (Foley & Foley, 2007). Further, generation can fail to benefit memory for nonwords when using a letter transposition task (e.g., ralt-lart) (Mulligan, 2002;
Payne et al., 1986), obviating any sort of cognitive effort effect.
The lack of generation effect for nonwords suggests that preexisting representations are necessary for active generation to benefit item memory, and that generation may increase semantic processing (McElroy & Slamecka, 1982). For example, the generation benefit may depend on the semantic relationship of the generated compound, as the generation effect exists when using meaningful rather than meaningless units of letters (e.g., E T vs. E C), unitized rather 2 than nonunitized 2-digit numbers (e.g., 28 vs. 2 8), and familiar rather than unfamiliar noun compounds (e.g,. cheesecake vs. cheese ketchup) (Gardiner & Hampton, 1985). However, while preexisting representations may be necessary for a generation effect, they may not be sufficient.
Even after teaching participants the meanings of nonwords, or when using low frequency words rather than medium and high frequency words, the generation effect is absent (Nairne et al., 1985). Additionally, the generation effect occurs when using nonsemantic phonological tasks such as rhyme generation (Payne et al., 1986). In one set of experiments, however, participants generated homographs using rhymes and synonyms. When presented with cues related to the initially generated word, cues related to the dominant meaning showed a stronger generation effect than did cues related to the weaker meaning. As meaning could not have been encoded before generation took place, the authors argued that participants may spontaneously process semantic information after generation (McElroy, 1987).
Another possible explanation for the positive generation effect is the item distinctiveness account, which proposes that actively generating an item increases its distinctiveness, enhancing later memory for that item (Begg et al., 1989; Kinoshita, 1989). For example, Kinoshita (1989) found that letter transposition enhanced later recognition but not recall, implying that generation enhances the distinctiveness of the word, which facilitates its discriminability among targets while failing to enhance its semantic availability. Threatening this account, however, is the finding that when provided with the target and asked for the cue at retrieval, there was still a positive generation effect (Hunt & McDaniel, 1993).
Threatening all extent accounts, however, was the argument that the generation effect is simply an artifact of displaced rehearsal (Slamecka & Katsaiti, 1987). Slamecka and Katsaiti (1987) found that when the encoding condition was manipulated between subjects, the generation effect vanished. In another experiment, the authors also required participants to rehearse the words aloud, forcing them to study only the words currently on the screen during a within-subjects design, and again found no generation effect. Slamecka and Katsaiti (1987) therefore argued that participants may spend more time rehearsing generated items at the expense of rehearsing read items. However, the generation effect has been found to survive when using pure lists (Begg et al., 1989), and within-subjects tests likely only inflate the size of this effect (Hirshman & Bjork, 1988). The previously mentioned meta-analysis found that while the generation effect is smaller when the generate and read conditions are manipulated betweenrather than within-subjects, it remains robust (Bertsch et al., 2007).
Further research suggested that generation may enhance more than one type of processing. Hirshman and Bjork (1988) found a greater generation effect for cued recall than for free recall, prompting them to propose a 2-factor account in which active generation enhances both item-specific and cue-target information. This idea was later expanded into a 3-factor account in which generation can also enhance whole-list processing (McDaniel et al., 1988). The authors tested participants using several structured categories of related target items (e.g., furniture: chair, table, couch, bed), and found that while generation may impair whole-list informational processing in unstructured lists, generation may strengthen the relationship between targets in structured lists.
The Negative Generation Effect While active learning has proven to be an extremely powerful mnemonic, this benefit may come at a cost. Active generation has been shown to have positive effects on item memory, yet there are instances in which generation may actually impair memory for contextual features 3 of the item, termed the negative generation effect. Multi-factor theories suggest that while generation enhances response-specific and stimulus-response encoding, it may actually disrupt other relational encoding (Burns, 1990, 1992). Adding to this negative generation effect, Nairne et al., (1991) proposed an item-order tradeoff account. In a set of experiments, participants read and generated lists of words. While participants better recognized items that were initially generated, they were better able to place randomly organized read items into the correct order.
These results indicated that increased engagement with generated items improves memory for the word at the expense of processing relational information, such as order. Indeed, this idea is consistent with a more recent experiment in which Hendry & Tehan (2003) found that increasing word length promotes better item memory but worse order memory.
Subsequent experiments extended these negative generation effects. Jurica and Shimamura (1999) had participants view faces associated with either statements (e.g., Basketball is a fun sport) or questions that required generating an answer (e.g., Which sport do you think is the most fun?). Participants remembered topics presented as questions better than those presented as statements (i.e., positive generation effect for item information), but they had better source memory (i.e., memory for the face that presented the statements or posed the questions) for items presented as statements. Jurica and Shimamura (1999) therefore expanded the item-order tradeoff account to an item-context tradeoff account in which actively generating information forces one to attend to generated items at the expense of forming contextual associations.
In contrast, however, some researchers have found that generation can enhance both item and context memory. Consistent with such an idea are dual-process theories of memory such as remember-know and recollection-familiarity dissociations. For example, recollection may be characterized as a vivid re-experiencing of a prior event, rich with contextual information, while familiarity may be characterized an acknowledgment of recognition devoid of contextual details (Yonelinas, 2002). As generation enhances subsequent recollection, dual-process theories of memory suggest that this recollection should entail greater contextual binding. In a similar experiment to the one performed by Jurica and Shimamura (1999), participants answered questions or made statements to faces presented on a computer screen by either reading, unscrambling, or filling in words (Geghman & Multhaup, 2004). Rather than being tested for the prompts as in Jurica and Shimamura (1999) study, participants responded directly to the information that was actually read and generated, which resulted in positive generation effects for both item and context memory. Also finding uniformly positive generation effects, Marsh, Edelman, and Bower (2001) varied color, computer screen location, and room location of category-exemplar word pairs. Regardless of manipulation, active generation consistently benefited both item and context memory. Consequently, the authors concluded that any operation that strengthens item memory should strengthen memory for contextual associations as well (Marsh et al., 2001).
Upon attempting to replicate these results, however, Mulligan (2004) found generation to have a consistent positive effect on item memory, yet varying effects on context memory including a negative effect on color memory and no effect on location, background color, and cue-word color memory. Claiming that an item-context account fails to explain this pattern of effects, Mulligan (2004) adopted a TAP account (see Blaxton, 1989; Jacoby, 1983) in which active generation encourages the use of conceptual processing, while passive reading encourages the use of perceptual processing. Previously, Jacoby (1983) had participants read a word out of context, read a word in context, or generate a word. Consistent with a TAP account, generation provided the greatest memory benefit for item recognition, trailed by reading a word out of 4 context, and then reading a word in context, while perceptual identification followed the opposite pattern. Indeed, Blaxton (1989) found that conceptually driven tasks at retrieval such as cued and free recall benefit from conceptually driven tasks at encoding (i.e., generating), while data driven tasks at retrieval such as word fragment completion using graphemic cues benefit from perceptually driven tasks at encoding (i.e., reading). As item retrieval is generally assessed in a conceptual manner, TAP predicts positive generation effects on most standard tests of memory.
However, memory for perceptual features such as memory for the color of presented items should benefit more from passive study. Such dissociations between item and context memory fit well within a source-monitoring framework (Johnson, Hashtroudi, & Stephen, 1993) in which perceived events contain more sensory and spatiotemporal information, while imagined events contain more information about the mental processes engaged in their creation.
To investigate the apparent discrepancy between studies that find various positive and negative effects of generation on context (specifically location) memory, Marsh (2006) compared her previous study with Mulligan's (2004) study by manipulating three differing conditions. In the Marsh et al., (2001) study, items were generated covertly, the distractor task involved visuospatial puzzles, and people responded at test with pen and paper. In the Mulligan (2004) study, participants wrote down the generated and read words, completed a state-name fragment distractor task, and made their responses on a computer. Marsh (2006) found that when conditions were most similar to her own work, there was a positive generation effect on location memory, while when conditions were most similar to Mulligan's (2004) work, there was a null effect. If anything, these results illustrate the malleability and susceptibility of generation effects to slight experimental manipulations.
Amassing evidence of the negative generation effect on context memory, Mulligan et al., (2006) manipulated several aspects of generation. Using an antonym generation task, the authors again found positive generation effects for item recognition, a negative effect for color, a negative effect for font type, and a null effect for location. Additionally, they used a letter transposition task with real words (e.g., anger-rage) in which generation should increase the amount of conceptual processing while equating the amount of perceptual processing with the read condition. In support of a TAP account, this study eliminated the negative generation effect on color memory while preserving the positive generation effect on item memory. Further, they performed a nonperceptual rhyming task with nonwords (e.g., rart-lart), and found a negative generation effect on color memory, yet no generation effect on item memory. Dissociating these positive and negative item and source memory generation effects and independently manipulating each without impacting the other type of memory directly disputed any trade-off account that posits that the negative effect on source memory is a necessary byproduct of the positive effect of generation. In an attempt to explain the various generation effects for context memory, Mulligan (2011) recently argued that generation disrupts memory for intrinsic contextual details such as color or font type while ignoring extrinsic contextual details such as location.