«By Zachary Alexander Rosner A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Psychology ...»
The Value of the Generation Effect Memory researchers have now spent countless hours over more than 30 years studying the generation effect. Why devote so much time to studying a phenomenon that has been practiced since Socrates and preached by our parents? One answer is that students may not be quite convinced enough of the power of active learning to put principle into practice. While pupils might admit the value of active generation, a survey of the metacognitive learning 5 strategies of 177 college students found that in practice the vast majority of students simply reread textbooks or notes despite the limited benefit of this learning strategy (Karpicke, Butler, & Roediger III, 2009). Not only is the memory benefit of repeated study sessions far outweighed by active encoding strategies such as self-testing (Roediger III & Karpicke, 2006a), many students may not recognize instances in which they have insufficiently learned information if that information is explicitly provided, a problem that is eliminated when students test their own memories (Roediger III & Karpicke, 2006b). Thus, in addition to strengthening memories, active generation through self-testing provides an opportunity for students to understand when they sufficiently understand material.
Accruing evidence demonstrating the true power of the generation effect in the classroom could endorse the use of active learning strategies. Self-testing has been shown to greatly enhance learning in the classroom, and active generation is a major component of this effect. For example, after studying prose passages, students who took open or closed-book tests performed better than those who simply restudied the passages (Agarwal, Karpicke, Kang, Roediger III, & McDermott, 2008). Self-testing has even been shown to improve memory relative to commonly practiced mnemonic techniques such as elaborative encoding for material including scientific text (Karpicke & Blunt, 2011).
While the benefits of retrieval practice have been recently demonstrated with educational materials, key differences exist between active generation and retrieval practice that require consideration. The first difference is retrieval mode. In a direct comparison between the generation effect and self-testing, after studying words, some participants re-read the same words while others were shown fragments of those words along with instructions for one of two conditions. In the generate condition, participants were instructed to generate the first word that came to mind, while in the self-test condition participants were instructed to use the fragments as cues to recall the initially presented words. While generating words led to better memory than rereading words, self-testing was superior to both. Self-testing may be more a powerful mnemonic than generation, but its effect can only capitalize on previously learned information. Active generation, on the other hand, may be more practical during the initial encoding session. Indeed, studies have found that in education, generation enhances learning, and the errors students might make by generating incorrect information are not harmful if feedback is provided to correct those errors (Metcalfe & Kornell, 2007). Further, the benefits of generation can continue past the initial encoding phase, as students who generated words a second time showed a memory advantage over those who generated and then read words or simply read words twice (MacLeod, Pottruff, Forrin, & Masson, 2012). To be sure, in a meta-analysis of 17,771 subjects over 445 studies, active generation has been shown to yield an 8.8% advantage over passive learning (Bertsch et al., 2007).
Not only can active generation serve as a valuable tool in student learning, it can aid people with memory impairments. For example, while smaller than that seen in unimpaired populations, people with various causes of traumatic brain injury do show a positive generation effect (Lengenfelder et al., 2007). Additionally, patients exhibiting dementia of Alzheimer type (DAT) and frontal lobe type (FTD) both showed generation benefits for verbal and visuospatial short-term memory (Souliez et al., 1996).
These improvements extend to older people with milder memory impairments (Luo, Hendriks, & Craik, 2007). Studies suggest that older adults lack the self-initiated strategic encoding techniques often employed by younger adults. For example, when using shallow encoding tasks such as rhymes, older adults often fail to display the generation effect observed in 6 younger adults. This finding may be due to the fact that only young adults engage in postgeneration semantic processing, as the use of deeper semantic generation tasks rescues the generation effect in older adults. Several studies further illustrate this point that older adults lack self-initiated strategic encoding processes. For example, dividing attention during semantic encoding disrupts a generation effect in both younger and older adults (Taconnat & Isingrini, 2004). However, younger adults demonstrate a relatively greater generation effect for weakly than strongly associated word pairs (Taconnat, Froger, Sacher, & Isingrini, 2008) as compared to older adults, indicating greater semantic processing. Further, young adults allocate relatively more time to generation (Froger, Sacher, Gaudouen, Isingrini, & Taconnat, 2011), and the magnitude of the generation effect correlates with executive function abilities (Taconnat et al., 2006). Given the apparent value of the generation effect in everyone from students to younger adults to older adults to those with memory impairments, a fuller understanding of the positive and negative effects of generation, the underlying mechanisms of the generation effect, and the universality of this phenomenon is warranted.
Themes of this Dissertation
This dissertation is broken up into three themes investigating the positive and negative effects of generation, the universality of the generation effect, and its underlying neural mechanisms.
Further, these studies test various explanations of the generation effect, and a transferappropriate processing account is considered in detail.
Theme 1: The Positive and Negative Effects of Generation In what ways can actively generating information influence memory for item information, related item information, and contextual information? How resilient is the positive generation effect on item memory over long intervals of retention and in the face of divided attention?
Which aspects of context memory are negatively impacted by active generation? Are there instances in which active generation can impair memory for related items or even the items themselves?
Theme 2: The Universality of the Generation Effect How universal is the generation effect? Is it an effective encoding strategy among people from China who are accustomed to Confucian rather than Socratic learning styles? Further, how does the manner in which Chinese people process information (field-dependent rather than fieldindependent) influence the way that generation impacts memory for contextual information?
Theme 3: Mechanisms Underlying the Generation Effect What are the neural mechanisms that drive the generation effect? Are there specific regions within the brain that are more active when we actively generate rather than passively learn information? If so, do these brain activations drive the mnemonic benefit of active generation, and how can this information inform our understanding of why generation enhances memory?
7 Chapter 2: The Positive and Negative Effects of Generation Introduction As stated earlier, the mechanisms driving the generation effect are still not fully understood. Theories which attempt to account for positive generation effects on item memory have argued that active generation increases cognitive effort (McFarland Jr. et al., 1980; Tyler et al., 1979), conceptual processing (Jacoby, 1983), item distinctiveness (Begg et al., 1989; Hunt & McDaniel, 1993; Kinoshita, 1989), and semantic processing (McElroy, 1987; McElroy & Slamecka, 1982). Other theories posit that generation may enhance memory along multiple dimensions, including item-specific, cue-target (Hirshman & Bjork, 1988), and inter-target relational information (McDaniel et al., 1988). Still, there are instances in which the generation effect is reduced or absent, for example when manipulating the read and generate condition between subjects (Slamecka & Katsaiti, 1987), or when using nonwords as stimuli (McDaniel et al., 1988; Payne et al., 1986).
For all of the documented positive generation effects, negative effects have been found for contextual information such as the order (Nairne et al., 1991), color, and font of items (Mulligan, 2004; Mulligan et al., 2006), and the person who presented information (Jurica & Shimamura, 1999). To explain these negative generation effects, item-context tradeoff (Jurica & Shimamura, 1999) and transfer-appropriate processing accounts (Jacoby, 1983; Mulligan et al.,
2006) have been proposed. However, other researchers have found positive generation effects for item color and location, and proposed that active generation can enhance both item and context memory (Marsh, 2006; Marsh et al., 2001). As evidenced by these contradictory results, the effects of generation are sensitive to slight experimental manipulations, leaving various findings open to interpretation and multiple accounts plausible. This first series of studies sought to characterize the boundaries of the generation effect by addressing several issues. How resilient is the positive generation effect on item memory to decay over time and in the face of distraction?
Which aspects of context memory are negatively impacted by active generation? Are there instances in which active generation can impair memory for related item information or even the item itself?
Experiments 1.1A and 1.1B (Synonym Immediate Recognition; Synonym Delayed Recognition) The purpose of these two experiments was to test the resistance of the generation effect to decay over time. Active generation has proven beneficial after short periods of delay. Positive generation effects over longer retention intervals, as seen in self-testing experiments (Roediger III & Karpicke, 2006a), would demonstrate more value as an effective learning strategy. In these first experiments, participants read (e.g., STUDENT-PUPIL), or generated (e.g., STUDENTP_P_L) target synonyms and were tested either immediately or 24 hours later.
Materials and Methods Participants Seventy one UC Berkeley undergraduate students participated in these experiments.
Forty of the students participated in the Synonym Immediate Recognition experiment for 1 hour of research participation credit for partial fulfillment of a psychology course requirement. Fortyone of these students participated in the Synonym Delayed Recognition experiment for $12 for 1 hour.
Design and Materials Encoding stimuli consisted of 48 synonym word pairs. Half of the stimuli were presented in the read condition, meaning that synonym pairs were presented in complete form (e.g., STUDENT - PUPIL). The other half of the stimuli were presented in the generate condition, meaning that the vowels of the second word were removed (e.g., STUDENT – P_P_L).
Encoding strategy (generate vs. read) was manipulated within participants and counterbalanced such that each synonym pair appeared in each condition with equal frequency. Only synonym pairs in which participants were able to correctly generate the second word with at least 99 percent accuracy (demonstrated through prior experiments to pilot stimuli) were used. The distractor task consisted of a worksheet of 162 simple arithmetic problems, including addition, subtraction, multiplication, and division. The recognition portion of the experiment contained 96 randomly ordered items. Forty-eight of these items were old, consisting of the second, target word of each synonym pair from the encoding phase. The other 48 items were new, consisting of the second, target word of unused synonym pairs. Stimuli appeared as old and new with equal frequency over all participants.
Procedure Participants were seated facing a computer and told that they would see a series of synonym pairs, and while some pairs would be complete, the second word of other pairs would have the vowels removed. Regardless of condition, they were asked to say the second word of each pair aloud for the experimenter to record. This ensured that the correct word was generated, enabling the elimination of incorrectly identified words from future analyses. Participants were also told to remember the word for a later memory test. Before beginning the encoding phase, 2 practice encoding trials (1 read and 1 generate) were performed to ensure that participants sufficiently understood the task. Participants then viewed a series of 48 randomly ordered read and generate synonym pairs. Each trial began with a 2-second fixation inter-trial interval, followed by the presentation of a synonym pair for 3 seconds (see Figure 1.1A). Following the encoding portion of the experiment, participants performed the math distractor task. They were asked to answer as many of the problems as they could in 2 minutes. The purpose of the distractor task was to prevent the rehearsal of recently presented word pairs, and ensure that long-term memory would be tested.
Participants in the Synonym Immediate Recognition experiment performed the retrieval task immediately following the distractor task, while participants in the Synonym Delayed Recognition experiment performed the retrieval task 24 hours following the onset of the encoding session. Participants viewed a series of 96 randomly ordered words. Forty-eight of these words were old, consisting of the second, target word of each synonym pair from the encoding phase. The other 48 words were new, consisting of the second, target word of unused synonym pairs. Participants decided if each word was old or new with a confidence rating (1 = definitely old, 2 = probably old, 3 = probably new, 4 = definitely new. Words were presented one at a time, in black, in the center of the screen.
Results and Discussion
9 There was a positive generation effect for item memory in both experiments (Figure