«By Zachary Alexander Rosner A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Psychology ...»
Results and Discussion In the Read-Read condition, participants correctly recalled the first presented exemplar 60% of the time and the second presented exemplar 49% of the time. In the Read-Generate condition, participants recalled the first exemplar 53% of the time and the second exemplar 71% of the time (Figure 1.4C). Thus, generating the second item enhanced memory for the generated item by 22%, t(59) = 8.89, p.001, while impairing memory for the read item by 7%, t(59) = 3.33, p =.001. In the Generate-Read condition, participants recalled the first exemplar 72% of the time and the second exemplar 46% of the time. Generating the first item therefore improved memory for the generated item by 12%, t(59) = 5.69, p.001, while only impairing memory for the read item by 3%, t(59) = 1.66, p =.1. Overall, memory was 62% in the Generate-Read condition, 59% in the Read-Generate condition, and 55% in the Read-Read condition. These results indicate that active generation, while a powerful mnemonic, has potential the consequence of impairing memory for other related item information. This negative effect is particularly powerful retroactively, and may not exist proactively. Further, it appears that this deficit is less strong than the positive generation effect, as demonstrated by the fact that both the 17 Generate-Read and Read-Generate conditions elicited better overall memory than the Read-Read condition.
General Discussion These experiments demonstrated that while active generation can be an extremely powerful encoding strategy for item information, there are also instances in which it can impair memory for contextual information, related item information, and even the item itself. In general, the positive generation effect was robust, and existed over multiple types of stimuli, including synonyms, antonyms, idioms, pictures and categories, and under different encoding conditions such as covert generation, overt verbal generation, and overt writing. Further, these positive generation effects on item memory persisted over a 24-hour delay, consistent with the work of Roediger III & Karpicke (2006a), who found that active learning techniques such as self-testing can promote long-term retention. Additionally, the benefit of active generation was strong in the face of cognitive distraction, which provides some evidence against attention accounts of the generation effect. These findings are significant, as they extend our knowledge of the generation effect, which is most typically tested without distraction over short time periods, to scenarios more reminiscent of actual academic scenarios.
However, active generation does not result in universally enhanced memory, even for the item itself. When presented with picture stimuli at both encoding and retrieval, generation impaired memory for the item. While one could argue that study and test stimuli were more similar in the read than generate condition, Kinjo & Snodgrass (2000) accounted for this, finding that viewing complete items at study and test resulted in better subsequent recognition memory than viewing fragmented items at study and test. In that same study, however, generating pictures at study led to better subsequent recall memory than did reading. The present study further found that when presented with pictures at test, generating rather than reading words at encoding led to superior memory. Taken together, these results are consistent with Mulligan et al.’s (2006) TAP account. Essentially, generation benefits later memory to the extent that it promotes the appropriate type of processing. If tested with a picture (a perceptual task), studying a complete rather than fragmented picture benefits both item and color memory as the complete (read) condition allows for more perceptual processing. However, studying a fragmented word allows for greater conceptual processing which benefits later item memory as the participant can only rely on the concept of the item as no picture was initially presented. Color memory, however, is still a perceptual retrieval task, and benefits more from the read condition. Further, while the Word-Picture experiment demonstrated the typical positive generation effect for item memory and negative generation effect for context memory tradeoff, the Picture-Picture experiment demonstrated a negative generation effect on both item and color memory. These results suggest that the negative generation effect on source memory is not a necessary consequence of the positive generation effect on item memory, which is again more consistent with a TAP rather than tradeoff account.
Active generation can also impair memory for other items. In the Category Retrieval Blocking experiment, active generation enhanced memory for the generated items at the expense of related non-generated items. This finding is consistent with Bäuml's (2002) finding that generating exemplars impairs the recall of previously presented related exemplars. This result may be driven by a similar mechanism to that of retrieval-induced forgetting (Anderson et al., 1994), in that the strengthening of one memory may impair memory for related information.
Thus, the benefit of generation is not without consequence, even for item information. It should 18 be noted, however, that the overall benefit of generation for generated items was greater than the deficit generation created for read items, and both the Generate-Read and Read-Generate conditions led to better overall memory than the Read-Read condition. Further, a GenerateGenerate condition may be superior to any of these conditions, and should be tested in the future.
Regardless, it is apparent that active generation, while enhancing memory for the generated item, may impair memory for other related information.
Finally, the Idiom experiments replicated and extended Mulligan’s (Mulligan, 2004;
Mulligan et al., 2006) studies by finding that generation benefits item memory, impairs text color memory, and has no effect on text location and background color memory. It should be noted, however, that there was a slight, although nonsignificant, positive generation effect on location memory, which would be consistent with some previous research (Marsh, 2006; Marsh et al., 2001). These results are highly consistent with a TAP account (Jacoby, 1983; Mulligan et al.,
2006) with the caveat that generation impairs memory for intrinsic, but not extrinsic, contextual details (Mulligan, 2011). Typically, generation promotes conceptual processing, while reading promotes perceptual processing. Word recognition is generally considered to be conceptual in nature, and therefore item memory often benefits from generation at encoding. Source memory, on the other hand, is often a perceptual task, and benefits from reading at encoding. However, this holds true only for intrinsic contextual details, while extrinsic contextual details are unaffected by generation.
These experiments demonstrated the strengths, weaknesses, and boundaries of active learning. The positive generation effect on item information is quite powerful, existing in the face of distraction and over long intervals of retention. Under certain conditions, however, generation can impair memory for other items and even the item itself. Further, the negative generation effect was found to occur only for intrinsic contextual details, and coexisted with both positive and negative generation effects on item memory. These results are highly consistent with a TAP processing account.
Figure 1.1 – (A) Synonym encoding task for Experiments 1.
1A and 1.1B. (B) Antonym encoding task for Experiments 1.2A and 1.2B. (C) Behavioral data for Experiments 1.1A, 1.1B,
1.2A, and 1.2B.
21 Figure 1.2 – (A) Idiom encoding task for Experiment 1.3A. (B) Idiom encoding task for Experiment 1.3B. (C) Idiom encoding task for Experiment 1.3C. (D) Item and source memory accuracy for Experiments 1.3A-1.3C.
Experiment 2.1 (Idiom Text Color) Cross-cultural Differences in Learning Style The positive generation effect on item memory has been consistent, robust, and convincing, benefitting retrieval by nearly 10 percent as compared to passive study. Outside of the United States, the generation effect has been well-documented in European and North American countries, appearing in France (Taconnat & Isingrini, 2004), Sweden (Lundstrom et al., 2003), Russia (Voskresenskaia, 2010), Poland (Nieznański, 2012), Canada (MacLeod et al., 2012), and other countries. However, the universality of this effect among East Asian populations remains untested. Vast differences in learning styles between East Asian and Western European (and American) cultures have major implications for the effects of active generation.
Tweed and Lehman (2002) argued that Western thought stresses a Socratic learning method based on questioning and evaluating self-directed knowledge. In contrast, Eastern thought stresses a Confucian learning method based on essential knowledge, pragmatic information, and truth learned through collective analysis. It is important to note that while some researchers have mischaracterized this distinction as purely deep versus shallow, Confucian learning does stress effortful learning in combining memorization with understanding (Tweed & Lehman, 2002). Regardless, it appears that the act of self-generating information, reminiscent of the Socratic learning method, may be a more natural practice in America.
The shaping of these Socratic and Confucian learning styles begins early in life. Wang and Brockmeier (2002) observed parent-child memory sharing between cultures. American parents tended to co-recreate stories with their children, emphasizing elaboration and embellishment of self-focused and interesting activities, while Chinese parents stressed the repetition of mundane, socially relevant activities. The influence of these different learning styles on memory is apparent when investigating earliest childhood memories, as American participants recall more specific, self-focused, elaborative, and emotional memories, while Chinese participants recall more skeletal, relationship-centered, routine-related, and unemotional memories (Conway, Wang, Hanyu, & Haque, 2005; Wang & Ross, 2005).
Given this discrepancy in learning style, the present study assessed the universality of the effect of self-generation on memory for both item and context information. To our knowledge, this is the first study to evaluate cross-cultural influences on the generation effect. One possibility is that the generation effect is easily assimilated by Americans because the technique exploits a natural Socratic learning style. East Asians might find the technique less familiar and thus not easily adopt it. By this view, Americans may exhibit stronger positive and negative generation effects (as a product of a potential item-context tradeoff) than East Asians. Yet, another possibility is that the East Asians will be sensitive to the powerful effects of generation, find it to be an effective encoding strategy, and thus exhibit generation effects similar to Americans. In this first experiment, American and Chinese individuals were presented culturebased idioms in either red or green color and asked to generate or simply read the last word of each idiom. Memory for these words and the colors in which they were presented was assessed.
Materials and Methods Participants 31 Fifty UC Berkeley undergraduate students participated 1 one hour of research participation credit for partial fulfillment of a psychology course requirement. Forty-seven undergraduate students from Beijing University and Tsinghua University in Beijing, China participated in exchange for ¥15 RMB (approximately 2 US dollars).
Materials Stimuli consisted of 40 idioms commonly used in their respective language. Half of the stimuli were presented in the read condition, such that idioms were presented in their complete form with the last word bounded by parentheses (e.g., a penny saved is a penny (earned)). The other half were presented in the generate condition, such that the last word was missing and replaced by a blank space bounded by parentheses (e.g., a penny saved is a penny ( )). We will refer to the last words in the idioms as the target words. The idioms were very familiar, as pilot studies showed that individuals from both countries who were presented with their respective idioms could generate the last word with an accuracy level greater than 98%. In both conditions, the target word was marked by parentheses. Half of the idioms were presented in green, and the other half in red. Both encoding strategy (generate vs. read) and source (green vs.
red) were manipulated within participants and counterbalanced such that each idiom appeared in each possible combination of conditions with equal frequency across participants. American idioms were between 3 and 10 words in length with target words consisting of 3 to 9 letters.
Chinese idioms were between 4 and 12 characters in length and all targets were represented as one character.
A distractor task was presented between the study and test phase, which consisted of having individuals complete up to 162 simple math problems (addition, subtraction, multiplication, and division) for a 3-minute period. Thereafter, a recognition test was administered in which 80 words (40 target words, 40 new words) were presented in a random order. New words came from the last words of idioms not presented for study. Stimuli appeared as old and new with equal frequency across participants.
Procedure Participants gave informed consent and were then seated in front of a computer monitor.
They were instructed that they would be presented with idioms and that some of them would have the last word missing. They were also told that some of the idioms would be presented in red and others in green. Regardless of encoding condition (read or generate), participants were asked to write down the last word of each idiom. This ensured that the target word was correctly identified, enabling the elimination of any incorrect words from further analysis. Participants were also instructed to remember both the idiom and its color for a later memory test. Two practice trials were presented (1 read and 1 generate) to ensure that participants understood the task. Participants were then presented a series of 40 randomly ordered read and generate idioms.
Each study trial began with a 1-second fixation interval followed by a 7-second presentation of an idiom (Figure 2.1A, 2.1B). Following the encoding phase, participants performed the 3minute math distractor task.