Mental-Imagery-Based Mnemonic Training: A New Kind of Cognitive Training

Luan, Xiaoyu; Kawasaki, Yayoi; Chen, Qi; Sugimori, Eriko

doi:10.3389/fpsyg.2021.740829

ORIGINAL RESEARCH article

Front. Psychol., 09 February 2022

Sec. Cognitive Science

Volume 12 - 2021 | https://doi.org/10.3389/fpsyg.2021.740829

Mental-Imagery-Based Mnemonic Training: A New Kind of Cognitive Training

$\r\nXiaoyu Luan&#x;$ Xiaoyu Luan^1†

Yayoi Kawasaki^2†

Qi Chen^3*†

Eriko Sugimori^2†

¹Graduate School of Human Sciences, Waseda University, Tokorozawa, Japan
²Faculty of Human Sciences, Waseda University, Tokorozawa, Japan
³Medical College, Yangzhou University, Yangzhou, China

We investigated the immediate and maintenance effects of mental-imagery-based mnemonic training on improving youths’ working memory, long-term memory, arithmetic and spatial abilities, and fluid intelligence. In Experiment 1, 26 Chinese participants (15 boys, 11 girls) aged 10–16 years were divided into an experimental group that received 8 days of mental-imagery-based mnemonic training and a no-contact control group. Participants completed pre-, post-, and three follow-up tests (3, 6, and 12 months after the pre-test). In Experiment 2, 54 Chinese children (28 boys, 26 girls), all 12 years old, were divided into experimental and control groups. Participants completed pre-, post-, and follow-up tests (three months after the pre-test). Results showed that the training significantly affected long-term memory-related task performance but no effects were observed on working memory, arithmetic or spatial ability, or fluid intelligence-related tasks. Moreover, the effect of the training on long-term memory lasted up to one year; the more frequently the training was used, the more effective it was. A content analysis of the feedback submitted by parents of participants in Experiment 2 three months after the training showed that the children used the strategy more for memorizing content such as Chinese and English, as well as for musical scores. Furthermore, there was also the possibility that the training improved abilities and academic performance such as concentration and math performance. Our results provide a basis for the further exploration of mental-imagery-based mnemonic training as a novel training modality.

Introduction

Cognitive training encompasses a series of training programs aimed at improving or maintaining one’s cognitive abilities (Simons et al., 2016), particularly general fluid intelligence (Gf; Tsubomi et al., 2019). In contrast to crystallized intelligence, which includes knowledge and experience, Gf is a collection of abilities that allows people to solve novel cognitive problems and adapt to new situations without relying on previous knowledge (Carpenter et al., 1990). Gf is highly predictive of academic and professional outcomes, including attention control, memory, problem solving, and reading (Jaeggi et al., 2008). Cognitive training research has been conducted for centuries (Katz et al., 2018), including on working memory (WM) training (Au et al., 2015), neuro-feedback training (Gruzelier, 2014), exercise training (Kramer and Colcombe, 2018), and music training (Schellenberg, 2011).

Existing cognitive training research tends to focus on two aspects of the commonality between trained capacities and Gf. For example, in WM training, research has explored whether WM capacity can be enhanced by WM training (McNamara and Scott, 2001; Klingberg, 2010) and whether WM training can improve cognitive abilities such as near and far transfer of Gf (Sala and Gobet, 2017b; Gathercole et al., 2019). Learning transfer occurs when skills acquired through particular behaviors of cognitive training are generalized to other fields (Sala and Gobet, 2017a). Near transfer refers to the transfer of skills among closely related tasks; for example, training on the n-back task leads to higher scores on WM tasks than on training tasks such as the digit span task. Far transfer refers to the transfer that occurs among tasks with a weak connection, such as higher-order cognitive tasks (e.g., mental rotation tasks) other than WM tasks (Sala and Gobet, 2017a). Numerous studies have demonstrated the attentional benefits of cognitive training (Chein and Morrison, 2010; Jaeggi et al., 2011; Karbach and Verhaeghen, 2014), although many have failed to find significant benefits (Harrison et al., 2013; Melby-Lervåg et al., 2016; Simons et al., 2016). Nevertheless, a growing body of literature exploring how cognitive training transfers to improve Gf or individuals’ daily lives shows that the possibility of cognitive training is receiving increasing attention.

Referring to Table 1, there are three limitations of current cognitive training in general. First, although studies have shown that cognitive training can produce some effects in terms of near transfer, such effects are strongly stimulus-type restricted, and thus are somewhat far from being applied to real-life situations (e.g., Lustig et al., 2009). Second, with regard to far transfer, as shown above, while some studies found that several cognitive trainings can produce far transfer, there is also a considerable amount of counter evidence; hence, their real effects are still under debate. Third, many training programs require the use of computers, equipment such as brainwave meters, musical instruments, or sports venues; these external necessities limit the possibility of long-term self-training in the daily life of the user. Furthermore, while some studies point to the need for long-term research (e.g., Morrison and Chein, 2011), most studies still focus only on the immediate effects of training after completion (e.g., Kuhn and Holling, 2014; Studer-Luethi et al., 2016), which presents a challenge to our understanding of the true role of these trainings in our lives.

TABLE 1

Table 1. Benefits and limitations of existing cognitive trainings.

Considering the benefits and limitations of the various types of existing cognitive training, the purpose of this study is to propose a cognitive training that has no limitation of stimulus types in near transfer and can facilitate far transfer, improves users’ cognitive abilities more comprehensively, and does not require additional equipment or space, allowing users to easily and consistently apply it in their daily lives—mental-imagery-based mnemonic training (MIBMT), and validate its effectiveness. We position MIBMT as a method that can be “used” rather than “exercised,” that is, we hope that once users have mastered MIBMT, they do not need to dedicate a certain amount of time to daily training (e.g., core training, exercise training) to maintain its effects; rather, they can apply MIBMT directly to their daily studies or work, such as second-language learning.

Mental-Imagery-Based Mnemonic Training: Contents and Theoretical Basis

Mental-imagery-based mnemonic training is a generalized program designed to enhance WM and long-term memory (LTM) performance and improve yield transfer and Gf, using verbal and non-verbal mental-imagery mnemonic strategies in collaboration with mindfulness meditative practices. MIBMT consists of three elements: a short practice of mindfulness at the beginning of the training to improve participants’ attention, conversion of the target stimulus into mental imagery (MI) after the stimulus is presented auditorily/visually—the participant maintains it in their mind for the necessary encoding operations, and recollection of the target stimulus in reverse order.

Theoretical Basis

Mindfulness Practice

Mindfulness practice focuses on the present moment in a non-reactive way, and includes the seven attitudinal factors of “non-judging, patience, a beginner’s mind, trust, non-striving, acceptance, and letting go” (Kabat-Zinn and Hanh, 2009, p. 32). Studies have shown that experienced mindfulness users have better executive attentional performance (van den Hurk et al., 2010) and reduced mind wandering compared to those without mindfulness experience (Mrazek et al., 2012, 2013). Further, differences in attention can explain individual differences in WM capacity to some extent (Wiemers et al., 2019). Thus, mindfulness can extend WM capacity by limiting irrelevant information or expanding the amount of information that can be accommodated in WM (e.g., van Vugt and Jha, 2011).

Many studies have demonstrated the effectiveness of mindfulness training in adolescents (Quach et al., 2016), highlighting its effects during short-term training (Zeidan et al., 2010). Therefore, given the theoretical links among mindfulness practice, WM, attention, and mind wandering, using a short period of mindfulness practice initially in MIBMT may help participants focus their attention and reduce mind wandering, further improving training effectiveness.

Mental Imagery

Marschark et al. (1987) clearly showed that MI has a functional role in human cognition and that it contributes directly or indirectly to memory, list learning, language comprehension, and mind manipulation. Specifically, MI overlaps with many of the same mechanisms used in visual perception and with the central executive system in WM processing, such as information manipulation, and it plays an irreplaceable role in LTM. For example, concrete words are easier to recall than abstract words because MI can help form more characteristically cohesive encodings, and thus it provides additional retrieval avenues (Marschark et al., 1987). At the neural level, MI and WM share a common characterization function in the early visual cortex (Albers et al., 2013), and two-thirds of the brain’s regions that are activated during visual imagery processing coincide with the areas activated during visual perception (Kosslyn et al., 1997).

It is theoretically possible to improve the performance of WM and LTM by exploiting MI and intermediating the relationship between WM and LTM with Gf to further trigger a far transfer. Kosslyn (2005) argued that representations of LTM are inherently multimodal, linking visual, spatial, auditory, and tactile types of information together to build the structure of objects in memory. This coincides with a requirement of MIBMT—that participants determine multi-sensory characteristics of the target stimuli, such as how they look or feel, and create MI to enhance memorization.

In addition, according to the oral presentation effect, better results can be obtained when stimuli are presented for aural perception while using MI-based mnemonics, such as the method of loci (Cornoldi and de Beni, 1991; Moè and de Beni, 2005) and the cue-word method (de Beni and Moè, 2003). With this feature in mind, MIBMT presents stimuli aurally at the beginning of training. When users can use this strategy more consistently, the alternating presentation of audio and visual text helps them apply it to different real-life situations once the training sessions have concluded.

Recollection of the Target Stimulus in Reverse Order

One of the most important attributes of MIBMT is the third element that requires the user to maintain target stimuli in the form of MI and subsequently recall it in reverse order to enhance the processing level of encoding. Marschark et al. (1987) mentioned that abnormal sentences or jumbled prose, for example, allow attention to be focused on each individual word, while meaningful sentences are usually treated as a whole and result in errors generated by the top-down system during subsequent recalls (e.g., recalling “learning English is hard” as “it is hard to learn English”).

Furthermore, the textual anomalies produced by the act of recalling in reverse order improve participants’ grasp of the information related to each individual word/character, such as visual or spatial characteristics, and leads to a conscious calling into mind, or gaze, of the MI the participant holds in their mind. Reverse recall is likely to improve the accuracy of a person’s gaze, while a higher fidelity gaze can also improve recall accuracy (Laeng et al., 2014). Further, when people retrieve information from memory, they tend to anchor the information in the location where it appears during encoding (Scholz et al., 2015). Thus, an MI constructed with a high level of encoding and accuracy may help reproduce the exact position of the elements of the encoded stimulus during the retrieval period, thereby facilitating memory performance.

Research Objectives

Sala and Gobet (2017a) argued that since the probability of far-transfer occurrence is negatively related to the level of expertise in the discipline, theoretically, children are the ideal population to test the occurrence of far transfer. Therefore, we chose typically developing children as participants to explore the effects of MIBMT.

As mentioned above, the purpose of this study is to propose MIBMT, a cognitive training that aims to improve users’ cognitive abilities and can be easily and consistently used in daily life, and to experimentally test its practical effects. Therefore, considering the research on cognitive training to date, we examined the effects of MIBMT in the following three main aspects. First, we examined the scope of the training transfer. Specifically, digit span and non-word recall tasks were used to measure near transfer, and arithmetic and mental rotation tasks were used to measure far transfer. According to Shipstead et al. (2012), in many studies, near transfer to WM has been measured by simple digit span tasks, as improvement in WM capacity is often associated with short-term memory-related tasks. We also used a simple digit span task as a measure of WM storage capacity. Considering that one of the important characteristics of MIBMT is to focus on the extrinsic features of the stimulus instead of semantic processing, we chose a non-word as the retrieval object when selecting the LTM-related task (non-word recall task) to maximize the exploration of the impact of MIBMT on LTM. Regarding far transfer, WM capacity is associated with adolescents’ mathematical ability (Gathercole et al., 2004), and WM training is likely to improve children’s basic arithmetic skills (Sala and Gobet, 2017a). Thus, the arithmetic skills task has been used as one of the measures of the effects of far transfer in many studies (van der Molen et al., 2010). We also used the mental rotation task—a typical method to test spatial capabilities (Moreau, 2013).

Second, we assessed the maintenance of training benefits, hypothesizing that the benefits produced by MIBMT would be somewhat maintainable. Most studies have focused more on the immediate effects of cognitive training, and research on the effects of (even intermittent) maintenance training (Morrison and Chein, 2011) is scarce. Therefore, it is important to study the immediate and ongoing benefits of MIBMT.

Third, we explored the effect of individual differences on MIBMT effects, hypothesizing that the frequency of use of MIBMT by the experimental group after training would impact the effects of MIBMT. Since MIBMT is positioned as a mnemonic that can be applied to all aspects of life learning, the experimental group was encouraged to use this strategy actively in their daily lives after the training; the frequency of spontaneous use of MIBMT in users’ daily lives represents, to some extent, the level of recognition of MIBMT, the motivation to use it, and the level of proficiency in using it. Therefore, we assumed that the frequency of post-training MIBMT usage would also impact the training effect, and individuals who actively use it may have better maintenance results than their counterparts.

We conducted two experiments (Table 2). In Experiment 1, we evaluated the transfer effect, sustainability, and potential factors affecting the effectiveness of MIBMT. In Experiment 2, based on the results obtained in Experiment 1, the test contents were modified in four ways to further examine the effects and influencing factors of MIBMT. First, the digit span task was changed to a Stroop task. Second, a Raven’s Standard Progressive Matrices (SPM) task was added to measure users’ Gf. Third, we limited the experimental participants to sixth graders and changed the duration of the experiment from one year to three months. Finally, we conducted a contents analysis of the feedback submitted by parents of participants in Experiment 2 to investigate the impact of MIBMT on daily life.

TABLE 2

Table 2. Overview of the experiments.

Experiment 1

Ethical Approval

This study was approved by the Ethics Review Committee on Research with Human Participants of Waseda University (no. 2019-152).

Participants

From September 2019 to September 2020, 26 elementary- and middle-school children between 10 and 16 years old from Yangzhou, Jiangsu Province, China participated in this study. Eighteen participants (11 boys, 7 girls) were in the experimental group (mean age = 11.78 years; SD = 1.83), and eight participants (four boys, four girls) were in the control group (mean age = 11.63 years; SD = 1.92). To recruit participants, we advertised in SNS groups (WeChat) for parents of elementary and middle-school students. After asking participants if they had eight consecutive free days, we assigned those who had time to the experimental group and the rest to the control group. Participants in the control group were not informed of anything related to MIBMT. All participants were native Chinese speakers and had no previous events or illnesses that we believed would affect training effectiveness or test performance. All parents provided written consent for their children to participate in the study, and an oral assent was obtained from the children. No other exclusion criteria were used in the recruitment. Participants were free to choose whether to participate; moreover, they were informed that they could withdraw from the experiment at any time and that their data would be destroyed. Each participant was provided with an evaluation report after the experiment. Participation was voluntary. To ensure confidentiality and anonymity of the data, we used only the participant number during the experiment.

Materials

All tasks were originally designed by the primary investigator and ran on E-prime 2.0 (Psychology Software Tools; Washington, United States). The tasks of each test (pre-, post-, and follow-up) used different stimulus questions derived under the same rule/source, which means that all participants completed the same tasks in the five tests; however, the questions differed between tasks.

Near-Transfer Tasks

Digit Span Task

In this test, the screen presented successive integers from 0 to 9 in random order at a speed of one digit per 700 ms, with a dialog box appearing when the last digit of each span disappeared from the screen. The examinee used the keyboard to enter the recalled digit list into the dialog box before submitting it; no time limit was set for entry. The digit span started at five digits and increased gradually, and each span comprised two turns; when at least one turn represented a successful recall, the span would increase by one digit. No upper limit was set for the span, and the highest number attained by participants was 15 digits. The test ended when the participant could not recall the order for either turn in the same span. In each span, one point was counted if both inputs were correct, 0.5 points if only one was correct, and 0 points if both were incorrect. Four points were given at the beginning of this task as a basic score, which meant that even if a participant failed all the questions in this task, they would have four points. If a participant reached a larger span, one point was added for every passed sequence, even if only one of the spans was correct. For example, when a participant failed both times in an eight-digit span round and succeeded only once in the preceding seven-digit span round, their digit span score was 6.5, regardless of whether they succeeded once or twice in the six-digit span round or in preceding rounds (Table 3).

TABLE 3

Table 3. Example of scoring in the digit span task.

Non-word Recall Task

The non-word recall task was used to measure the participants’ LTM ability. Based on Ding’s (2009) method, we constructed 16 non-words: eight three-character non-words (e.g., 更元就) and eight four-character non-words (e.g., 退吴看跳). None of the characters exceeded the range of characters known to participants. Non-words were presented visually, with each word appearing on the screen for 5000 ms in sequence with a blank screen for 3000 ms between words. After all stimuli were shown, participants were asked to move on to other tasks (e.g., digit span), ensuring a 15-min interval between retrieval and stimulus presentation. Fifteen minutes after the end of the stimulus presentation, participants were asked to recall the stimuli in any order and write them down on paper.

Non-word recall task scores were calculated for the successful recall of each character rather than each whole non-word. Participants were not informed of the scoring method, as the memorization of overall textual content is a common occurrence in everyday life, and we did not want participants to attempt consciously to achieve high scores on their recall of individual characters but rather to focus on memorizing all non-words. In this way, we attempted to simulate the MIBMT mechanism in real-life scenarios. Every correct character written in the correct position was counted as one point; for example, if the stimulus was “更元就” and the participant recalled it as “○元○,” they were given one point. Thus, the scores for this task ranged from 0 to 56. The choice to score by character (including whether it was in the correct position) rather than by word was made to ensure that the task effectively reflected the encoding features of non-sense processing of stimuli when processing information with MIBMT. In simple terms, if the stimulus was non-meaningfully processed, then even if the stimulus was presented as a non-word, each character that made up the non-word should have been processed individually during encoding; therefore, the character was the minimum retrieval unit for the task.

Far-Transfer Tasks

Arithmetic Skills Task

Considering that participants included adolescents aged 10–16 years, to avoid both ceiling and floor effects, we referred to Huang’s (2018) method of constructing the task, using only addition and subtraction questions. We used six questions each on addition without regrouping (e.g., 33 + 62 = ?), addition with regrouping (e.g., 19 + 23 = ?), adding regrouping twice (e.g., 47 + 85 = ?), subtraction without regrouping (e.g., 97 – 42 = ?), subtraction with regrouping (e.g., 31 – 14 = ?), and subtraction with regrouping twice (e.g., 123 – 89 = ?) for 36 questions. No number greater than 1000 appeared during the calculation, and the numbers used were integers. After the question appeared on the screen, the participant used the keyboard to enter the answer in the dialog box and submit it; if no answer was submitted within 30 s, the system automatically displayed the next question, with a 1500 ms pause between questions. Participants did not receive feedback on the correctness of their answers during the task. The accuracy and response time (RT) were calculated separately.

Mental Rotation Task

This study used stimuli as described in Ganis and Kievit (2015). In a mental rotation task, the greater the angle of rotation, the longer the reaction time required (Estes, 2008). Therefore, we assigned the number of stimuli for each angle equally, wherein five groups were the same and five groups different at 0°, 50°, 100°, and 150°, resulting in 40 stimuli. Each set of stimuli was presented on the screen, and the participant had to use the keyboard (“O” for the same stimulus; “X” for a different stimulus) to make a judgment within 9000 ms; otherwise, the system would automatically display the next question. A blank screen was shown for 1500 ms between the questions. Participants did not receive feedback on whether their answers were correct during the task. The accuracy and RTs were calculated separately.

Questionnaire: Post-training Evaluation

For the participants in the experimental group who completed the entire test, the questionnaire was evaluated on a five-point Likert scale: used 0 times per week, used two times per week or less, used three to four times per week, used five to six times per week, and used daily, allowing users to rate their frequency of MIBMT use during the one-year post-training.

Procedure

The experimental group underwent eight days of MIBMT training (3 days of training, 3 days off, and 2 days of training), and the control group received no special training. All participants completed the pre-test on Day One of the study, the post-test on Day Nine, and three follow-up tests: 3 months, 6 months, and 1 year after the pre-test. Participants completed five tests in their computer classrooms. They were asked to complete the mental rotation and arithmetic skills tasks as quickly as possible while ensuring correct answers, whereas for the digit span and non-word recall tasks, participants were asked to recall items as accurately as possible without specific time limits. After all the five tests, which were spread across one year, were completed, the experimental group was asked to complete the Post-training Evaluation (PTE) questionnaire. For the procedure of Experiment 1, refer to Figure 1.

FIGURE 1

Figure 1. The procedure of experiment 1. MIBMT, mental-imagery-based mnemonic training; PTE, post-training evaluation.

The Experimental Group

Mental-imagery-based mnemonic training began after the completion of the pre-test, and training sessions were held for 5 h per day (2.5 h in the morning and 2.5 h in the afternoon) (Table 4). Research suggests that spreading the training period out yields better training effects (Penner et al., 2012), as does the use of longer sessions (Schwaighofer et al., 2015). Jaeggi et al. (2008) also proposed that when more (vs. less) time is spent training, the improvement in Gf is greater. Our MIBMT study, with a total duration of 25 h over 8 days, took this into consideration.

TABLE 4

Table 4. MIBMT training schedule.

The morning of Day One began with a half-hour explanation of MIBMT training. After the instructions were completed, a 20-min mindfulness practice was performed (only in the morning of each training day; no mindfulness practice was scheduled in the afternoon). In addition to MIBMT training instruction and mindfulness practice, time was spent on MI-based memory training. During the MI-based memory training, participants were asked to focus on the external characteristics of the target stimulus while encoding it and create the corresponding imagery personally. For example, when memorizing the number “6,” the participant should maintain the number “6” in mind, then define its color (e.g., red, yellow; participants who had difficulty building imagery on their own would be given a direct suggestion such as, “How about defining the image of 6 as yellow?” if necessary), tactile attributes (e.g., soft, rough), dimension (e.g., 2D, 3D), size (e.g., small, large), and other aspects to enhance the vividness of the imagery of “6.” MIBMT, unlike previous mnemonics, does not require the user to associate stimuli with other objects, such as animals or places, to promote recalling performance through associative means (e.g., association with spatial information, storytelling); rather, it creates individual imagery for each stimulus and reinforces their multi-sensory (but visually dominant) imagery traits.

The initial days consisted of the trainer’s reading of the stimulus to participants, with each set of stimuli consisting of seven meaningless characters (including kanji, numbers, letters, and shapes). The character contents are listed in Table 5. When reading each set of stimuli, the trainer read only one character at a time, with a pause between characters, ensuring that participants finished creating the relevant MI image before reading the next character. After participants had completed encoding each character in a set of stimuli using the MIBMT method, they were asked to repeat it in reverse order and then in the original order. After participants completed recalling the first set of stimuli, they encoded the next set of stimuli in the same way. However, during this training, participants were also required to recall the vertical order between groups of stimuli, for example, when asked about the third vertical column, participants had to recall the stimuli as they were presented from top to bottom (i.e., 看花的石 4 T) or from bottom to top (T 4 石的花看). In general, participants were required to grasp the location of each stimulus in Table 5 concerning the overall spatial relationship. The number of stimulation combinations presented depended on participants’ learning speed and capacity, and it might not have been possible for participants to recall all six stimulation groups on training day one.

TABLE 5

Table 5. Meaningless character combinations used on day one.

On Day Two, the stimulus was replaced by simple Chinese phrases (about 10 Chinese characters in length), and participants were asked to memorize and recall them in the same way. The difference was that the trainer did not stop after each character while reading the sentence but read the entire sentence at a slower pace and read each sentence one to three times depending on participants’ capacity. Immediately after the trainer finished reading the stimuli, participants repeated them in reverse order and then in the original order (e.g., auditory presentation: “I think English is very difficult”; participant recalling: “difficult very is English think I”). Memorization of the Chinese stimuli only required participants to recall backwards or forwards and did not require grasping the top-down spatial relationship between sentences. After participants became relatively proficient in this method, sentence length was gradually increased and visual stimuli were introduced (i.e., participants were asked to read simple sentences and construct MI images, and then repeat them in the same way). This training did not require participants to get every character correct, but rather to note whether the content of their backward and forward recall was consistent even in the case of repeating errors (e.g., missing the same word). If it was too inconsistent—for example, when there were more than five inconsistencies between forward and backward recall—participants were reminded of that inconsistency in subsequent training attempts on the same stimuli because the stark difference between the content of backward and forward recall might be because of the participants’ high dependence on semantic processing.

On the afternoon of training Day Three, English stimuli were presented in the same way as the Chinese stimuli. The stimuli did not contain Chinese characters or English words that were beyond the range of the visual or auditory understanding of participants. Stimuli came directly from the later, unlearned portions of participants’ own Chinese or English textbooks. At the end of Day Three, 15 min were taken to summarize the progress of the training, and participants were asked to apply this method for what they needed to remember in their daily lives over the next 3 days.

At the beginning of training Day Four, participants were asked to provide feedback about what aspects of their life and studies incorporated MIBMT during the off days, how well they thought they had applied it, and whether they had any questions. The rest of Days Four and Five involved Chinese and English auditory and visual stimulus material, and stimuli were alternately presented, requiring participants to memorize and recall them using MIBMT. The difficulty of the stimuli, determined by the number of Chinese characters or English words used in a span, gradually increased according to participants’ performance. Chinese characters are not equivalent to English words or letters but are more like a combination of both; depending on the type of Chinese character, its meaning is sometimes equivalent to a word in English and sometimes only equivalent to a letter. Therefore, we adjusted the number of Chinese characters to control stimulus difficulty; stimuli that included more Chinese characters were considered more difficult. By contrast, the difficulty of the English stimuli was controlled by the number of words.

At the end of Day Eight, participants were asked to continue using MIBMT as actively as possible in their daily lives, but there was no special supervision post-training.

The Control Group

Before the experiment began, the control group was informed that this study tracked the development of memory, arithmetic, and spatial abilities in children and adolescents over a one-year period. Controls were not provided any information about MIBMT. They were only asked to go to the school’s computer room during the prescribed period to complete five separate tests to the best of their ability. After the five tests were completed, the purpose of this experiment was explained again to the control group.

Data Analyses

Statistical analysis was performed with SPSS 25.0 (IBM; Armonk, NY, United States). We divided the experimental group into high-use (n = 10; five boys, five girls) and low-use (n = 8; six boys, two girls) groups according to the PTE score based on the median split technique (Rosen et al., 2003) and compared the changes in task performance of each group. The control group was used as the baseline. To reduce the possible underestimation of statistical power in between-groups comparisons in multiple time points, a 3 (high vs. low vs. control group) × 2 (pre-test and post-test; post-test and three-month follow-up test; post-test and 6-month follow-up test; post-test and 1-year follow-up test) analysis of variance (repeated measures ANOVA) was conducted to assess differences in the effect of the potential factor between the high, low, and control groups (Karbach et al., 2015). Furthermore, for significant interaction effects, we used the Ryan-Einot-Gabriel-Welsh F method as post-hoc analyses. Significance was set at p < 0.05, and p < 0.001 was considered highly significant. Additionally, we performed a post-hoc power analysis using G*Power (Faul et al., 2007) for 3 × 2 repeated measures ANOVA, and the effect size used in the calculation was partial eta-squared.

Single-task effect sizes were measured using Cohen’s d. The follow-up effect sizes were calculated in the same way as for the single-task effect size, using follow-up and pre-test results (Sala and Gobet, 2017a). Cohen’s d effect size was defined as small for d = 0.2, medium for d = 0.5, large for d = 0.8, and very large for d = 1.2 (Sawilowsky, 2009). That is, the larger the value, the greater the training effect.

Results¹

The means and standard deviations for each task are listed in Table 6. Group-occasion interactions and main effects for each group and occasion, as well as the effect size for each task, are listed in Tables 7–9.

TABLE 6

Table 6. Means (standard deviations) of individual tasks in Experiment 1.

TABLE 7

Table 7. ANOVA analysis and effect sizes of near transfer tasks in Experiment 1.

Regarding the near-transfer tasks, Table 7 shows no intentional interaction between group and occasion in the digit span task either between the pre- and post-test, or between the post-test and the individual follow-up trials (3 months, 6 months, and 1 year later). A simple main effect of group, however, was observed in the individual ANOVA analyses. In the post-hoc analysis, intentional differences were consistently observed between the high-use and low-use groups, and between the high-use and control groups, while there were no intentional differences between the low-use and control groups (Pre-test – Post-test: High-use – Low-use: t = 3.259, p = 0.003, High-use – Control: t = 3.028, p = 0.006, Low-use – Control: t = 0.219, p = 0.828; Post-test – 3-month follow-up test: High-use – Low-use: t = 3.392, p = 0.003, High-use – Control: t = 2.332, p = 0.029, Low-use – Control: t = 1.006, p = .325; Post-test – 6-month follow-up test: High-use – Low-use: t = 3.782, p = 0.001, High-use – Control: t = 3.383, p = 0.003, Low-use – Control: t = 0.378, p = 0.709; Post-test – 1-year follow-up test: High-use – Low-use: t = 2.837, p = 0.009, High-use – Control: t = 2.440, p = 0.023, Low-use – Control: t = 0.376, p = 0.710). In addition, although a simple main effect of one occasion was also observed between the pre- and post-test, only the high-use group showed a moderate effect size of 0.59 for increase in performance, and the effect size kept on fluctuating in the post- and the three follow-up tests, indicating that the WM-related task performance was not stable. The fluctuation in effect size was also seen in the control group. At the same time, although the effect size of the low-use group kept on increasing, this increase did not lead to a meaningful difference being observed between the low-use group and the control group. These data may indicate that there was an inconsistency in the baseline between the high-use group and the other two groups from the beginning, and that MIBMT and its frequency of use did not have a significant effect on the WM ability of the users.

As for the non-word recall task, a significant group × occasion interaction was observed between the pre- and post-test (Table 7). Post-hoc analysis revealed that both the high- and low-use groups showed an intentional increase in performance in the post-test compared to the pre-test [High-use: F(1,23) = 13.060, p = 0.002; Low-use: F(1,23) = 8.140, p = 0.009], in contrast to the control group, which showed an intentional decrease [Control: F(1,23) = 7.476, p = 0.012]. There were no intentional differences between the groups’ performance in the pre-test [F(2,46) = 2.017, p = 0.146], whereas the post-test showed significant differences [F(2,46) = 9.062, p = 0.001] between the high- and low-use groups (t = 2.405, p = 0.020), and between the high-use and control groups (t = 4.326, p = 0.000), and a tendency for significant differences between the low-use and control groups (t = 1.823, p = 0.075). Between the post-test and the individual follow-up tests (3 months, 6 months, and 1 year later), the simple main effect of group remained, although the intentional group × occasion interaction was no longer observed. Post hoc analysis showed that the differences between the high- and low-use groups, and between the high-use and control groups were maintained (Post-test – 3-month follow-up test: High-use – Low-use: t = 2.824, p = 0.010, High-use – Control: t = 4.765, p = 0.000, Low-use – Control: t = 1.842, p = 0.078; Post-test – 6-month follow-up test: High-use – Low-use: t = 3.239, p = 0.004, High-use – Control: t = 4.996, p = 0.000, Low-use – Control: t = 1.667, p = 0.109; Post-test – One-year follow-up test: High-use - Low-use: t = 2.797, p = 0.010, High-use – Control: t = 4.654, p = 0.000, Low-use – Control: t = 1.761, p = 0.091). The effect size also showed that the gains in non-word recall tasks in the high- and low-use groups in the post-test compared to the pre-test were maintained in the subsequent follow-up tests. This suggests that the continuous use of MIBMT in daily life had a positive effect on improving LTM capacity, as users who tended to actively use MIBMT showed better LTM performance post-training.

Regarding the arithmetic skills task, according to Table 8, both accuracy and RT of the task showed no significant effect of experimental group × occasion interaction. In the pre-test, the post-hoc analysis revealed that the high- and low-use groups showed an intentional between-group difference in RT (t = 3.347, p = 0.003). There was also an intentional tendency that appeared between the high-use and control groups (t = 1.933, p = 0.066), while there were no intentional differences between the low-use and control groups (t = 1.341, p = 0.193). Referring to the effect sizes of the groups and Table 6, it seems that there is an inconsistency in the baseline of RT for arithmetic skills between the high-use group and the other two groups probably from pre-test onwards (Post-test – 3-month follow-up test: High-use – Low-use: t = 2.984, p = 0.007, High-use – Control: t = 2.098, p = 0.047, Low-use – Control: t = 0.841, p = 0.409; Post-test – 6-month follow-up test: High-use – Low-use: t = 2.726, p = 0.012, High-use – Control: t = 1.148, p = 0.263, Low-use – Control: t = 1.497, p = 0.148). No evidence was found that MIBMT and its frequency of use improved users’ performance in arithmetic skills.

TABLE 8

Table 8. ANOVA analysis and effect sizes of far transfer arithmetic skills task in Experiment 1.

Concerning the mental rotation task, Table 9 shows that neither the accuracy of the task nor RT showed significant effects for group and occasion. In combination with the effect sizes of individual groups and Table 6, this suggests that MIBMT had no effect on users’ performance in the mental rotation task.

TABLE 9

Table 9. ANOVA analysis and effect sizes of far transfer mental rotation task in Experiment 1.

Discussion

We found that MIBMT improved participants’ LTM task performance and had a maintenance effect for at least 1 year. Further, the active use of MIBMT in daily life may reflect users’ acceptance of the training, which may in turn affect their motivation during the training. The results showed that participants who could actively use MIBMT later performed better on post-tests related to LTM capabilities, with a further long-term maintenance effect. There was no evidence that MIBMT improved the performance of WM-related short-term tasks. The difference in the performance of the groups on WM and LTM tasks may be because MIBMT helped improve the efficiency and quality of encoding while dealing with LTM-related tasks rather than improving task performance by increasing the storage capacity of WM (Carretti et al., 2007). Considering extant WM models, these models have two aspects in common: the limited capacity of WM and cognitive control capacity (Baddeley, 2003; Unsworth and Engle, 2007; Schwaighofer et al., 2015). While the digit span task focused only on the limited capacity of WM, to further explore the relationship between MIBMT, WM, and LTM, it was necessary to use a Stroop task in Experiment 2 that could measure participants’ capacity for cognitive control (e.g., Pugin et al., 2014; Karbach et al., 2015).

In the case of far transfer, the results suggest that MIBMT had no effect on users’ arithmetic and spatial abilities. The main goal of cognitive training is to improve or maintain the Gf of users (Tsubomi et al., 2019), and Gf includes the arithmetic, spatial, and reasoning abilities (Jaeggi et al., 2008). Morrison and Chein (2011) noted that the SPM task developed by Raven (1936) as a non-verbal Gf test is an appropriate tool for assessing the far-transfer effects of cognitive training. Therefore, to further clarify the effectiveness of MIBMT as a cognitive training, it was necessary to use the SPM task in Experiment 2.

Experiment 1 examined the effectiveness of MIBMT in LTM and the possibility of maintaining this effect. The frequency of daily use may be a factor that influences the effectiveness of MIBMT. Further, Experiment 1 had four limitations: first, for WM capacity, Experiment 1 tested only the storage capacity of the participants and not other associated capacities, such as the distribution or control of attention in the WM system; second, we did not utilize some mainstream testing methods such as the SPM task for exploring Gf (reasoning ability) improvement in far transfer. Third, inconsistencies in the baseline between groups emerged in the WM and arithmetic tasks, which may be a limitation of the too small sample size. Fourth, although the goal of MIBMT is to allow users to easily and consistently apply it in their daily lives, the daily use of MIBMT by participants was measured using only a single Likert scale. To further investigate the usefulness of MIBMT, more surveys of participants’ use status are necessary.

Experiment 2

Objectives

Based on the results and limitations of Experiment 1, four modifications were made in Experiment 2: first, the digit span task was changed to a Stroop task to fully measure WM function; second, an SPM task was added to measure users’ Gf; third, the experimental participants were limited to sixth graders, and the experimental period was changed to 3 months to maximize sample size. Fourth, 3 months after the training, parents of all participants in the experimental group were asked to submit feedback on their children’s use status of MIBMT, academic performance, and daily life status during the three months (since the participants were elementary school students, we chose their parents as the reporter to ensure the objectivity and accuracy of the feedback). We conducted a contents analysis to investigate the impact of MIBMT on daily life. Experiment 2 is a revalidation of the results of Experiment 1 and an extension of the study in terms of both quantitative and qualitative analyses to further clarify the effectiveness and sustainability of MIBMT, as well as the impact of the frequency of daily use of MIBMT.