The influence of emotional narrative content on the self-reference effect in memory

Highlights

• •Positive autobiographical memories aid the self-reference effect.
• •This is true for older adults with and without mild memory impairments.
• •This can serve as a memory intervention for aMCI patients and older adults.

Abstract

This study investigated how emotional language usage impacts self-referential effects in memory in healthy older adults and individuals with amnestic mild cognitive impairment (aMCI). To heighten self-focus, 37 healthy older adults and 22 aMCI participants narrated autobiographical memories and then encoded words using a self-referencing or a semantic strategy. We were interested in how narrating autobiographical memories impacted subsequent memory. We probed narrative language usage with the Linguistic Inquiry and Word Count text analysis program, testing the degree to which language from the narrated autobiographical memories contain emotional (positive and negative) words that predicted the self-reference effect across groups. Results indicated that higher levels of positive emotional language were related to larger self-reference effects in memory. In conclusion, narrating autobiographical memories using emotional language influenced the effectiveness of self-referencing as a memory strategy for both healthy older adults and aMCI participants.

Influence of emotional narrative content on the self-reference effect in memory

Changes to memory are one of the most noticeable aspects of cognitive change with aging (e.g., [1], [2], [3]), and are even more pronounced with cognitive decline associated with amnestic mild cognitive impairment (aMCI) or Alzheimer’s disease (AD). Strategies that can improve memory are of great interest, particularly when their benefits extend to people with memory disorders. In this study, we are interested in understanding whether there are specific qualities of memories that increase or decrease the effectiveness of the self-referencing strategy for older adults, including those with dysfunctional changes in memory. We investigate these qualities by focusing on the emotional aspects of language usage.

Self-referencing, relating information to the self, is an effective memory aid [4], [5], [6], [7] A typical self-reference paradigm involves reading a series of words or viewing pictures and responding to a question related to the self (e.g., “Does the word describe you?”), followed by a memory test to assess the effectiveness of the encoding strategy. Self-referenced items are typically compared to judgments in other conditions, such as deciding whether words describe someone else, belong to a particular category (e.g., an animal), or appear in capital letters [7].

Despite age-related memory declines, self-referencing is an effective memory strategy for healthy older adults [8], [9], [10], [11], [12], [13], [14]. Furthermore, older adults engage the same medial prefrontal cortex regions (mPFC) that support the self-reference effect in younger adults [10], [15], [16], [17], [18], indicating a rare case of convergence in neural activity across age groups.

When considering the potential memory benefits from self-referencing, it is important to understand what aspects of thinking about the self are likely to impact the magnitude of self-reference effects. A recent paradigm we developed is well-suited to examine this question. In contrast to typical self-referencing paradigms in which each trial requires a judgment about the self (or another condition) for a single stimulus, we asked participants to share autobiographical memories in order to induce a self-reference mode of thought [19]. The traditional self-reference memory tasks, such as those that involve making self-related decisions about individual items, invoke self-referencing at the item-level. However, recent paradigms have attempted to induce self-reference effects that can be induced more flexibly and sustained across several trials. For example, Jackson et al. [19] had participants write down autobiographical memories prior to the self-reference effect task. Invoking autobiographical memories requires individuals to retrieve specific events from the personal past [20] with a focus on the self [21], [22], [23], [24]. Similar to traditional self-reference memory tasks, retrieving autobiographical memories can result in a person thinking about themself, inducing a self-reference mode of thought [19], [25]. Prior work from our lab [19], [26] showed that autobiographical memories did not boost memory beyond the trial-by-trial induction of the self-reference effect for older adults and people with aMCI. In the current study, we are interested in examining the content of autobiographical memories to assess whether specific aspects of thinking about the self could potentially boost the self-reference effect.

Additionally, prior work suggests that a self-reference mode of thought also has the potential to increase false memories, potentially reflecting the fluent processing induced by the self, for both older adults and patients with aMCI [26]. However, the rich narrative information available in autobiographical memories could be used to investigate whether any specific aspects of those memories might influence self-reference effects. In the present study, we code aspects of the content of the autobiographical memories generated by participants and assess how those particular aspects impact the magnitude of self-referencing effects in memory. To do so, we use a dataset analyzed previously [26] to test the effects of retrieving autobiographical memories on a subsequent self-referential memory task, but which had not previously been used to probe the contribution of the content of the autobiographical memories retrieved in the task.

Emotion is a candidate process likely to be contained in autobiographical memories that could potentially impact self-reference effects. Emotion can enhance the subjective experience or vividness of emotional memories ([27], [28]; see [29] for a review). Emotionally-laden information has been shown to enhance memory [30], [31] such that memory for emotional information tends to be better than that for neutral information even among older adults, and these effects are comparable to those seen in younger adults [32], [33]. The affective state of an individual has also been found to influence the degree of self-focused attention, with studies showing increases in self-focused attention resulting from both negative affective states [34], [35], [36] and positive affective states [34], [37] (however see [35], [36] for instances when positive affective states have not influenced self-focused attention). In addition, emotion may contribute to self-referential processing due to shared mechanisms and neural regions in younger as well as older adults [38]. Further, emotional state can influence one’s ability to remember and learn new information, though the literature is mixed. There is some compelling support that positive emotional states improve memory and learning [39] but also some indications that negative emotional states improve learning and memory [40]. As positive and negative emotional states have both been shown to have beneficial effects on memory, we consider both types of emotional states and their relation to self-reference memory in our study.

We propose that the process of retrieving emotional autobiographical memories will activate the mechanisms that are shared between self-reference and emotion, leading to larger self-reference effects in memory. Thus, we propose to use the positive and negative emotional language content of retrieved autobiographical memories as measures of induced emotion, assessing whether higher levels of emotional language correlate with higher levels of self-reference memory on a subsequent task. However, as discussed later in the methods section, our participants were encouraged to narrate stories that were neutral or positive and not too negative. Therefore, it is important to note that while we assess the effects of both positive and negative emotion on the self-reference effect, the study design may best capture effects of positive emotion.

Self-referential memory and emotion are important to study in people with memory impairments because harnessing these processes may offer routes to enhance memory. People with amnestic mild cognitive impairment (aMCI) have disproportionately poorer memory performance than would be expected for their age. A diagnosis of aMCI is a risk factor for developing Alzheimer’s disease (AD), and is often related to abnormalities in the medial temporal regions, which are severely impacted as AD progresses. Beneficial effects of self-referencing have been seen to some extent among people with aMCI [41], [42], [43], although benefits do not always occur [43], [44]. Self-referencing may be a preserved process in aMCI, as the strategy relies on medial prefrontal regions [15], [16], [17], [18] that, unlike the medial temporal lobes, are still relatively well preserved in the early stages of AD. Some prior work indicates that self-referencing combined with emotionally-valenced information may be particularly beneficial for aMCI patients [41], [45], [46].

Although self-referencing may aid memory for people with aMCI, their autobiographical memory and quality of episodic memory retrieval can be impaired, relative to healthy older adults [47], [48], [49]. Although people with aMCI have shown similar patterns for retrieval of memories as their healthy counterparts in terms of accessibility, enhancement, and specificity in describing emotional memories compared to neutral memories, they nevertheless exhibit some impairments in retrieval of these memories compared to healthy controls, particularly emotional memories [28], [50]. For example, Philippi et al. [28] found that although healthy older adults showed an enhancement for emotional autobiographical memories, both in terms of the proportion of memories and the degree of specificity, aMCI participants showed an impairment in emotional autobiographical memories. Other research also suggests that aMCI patients may not exhibit the same memory benefit from emotional information as older adult controls [51], [52]. For instance, on a face recognition task Wang et al. [52] found that healthy older adults recognized positive and negative faces better than neutral faces, however aMCI participants failed to show this recognition benefit for emotionally valenced faces.

The differences in the quality of autobiographical memories, particularly emotional autobiographical memories, produced by healthy older adults and aMCI patients lead us to predict that emotional autobiographical memories may impact self-referencing in different ways for the two groups. If emotional autobiographical memories are impoverished for aMCI patients, then retrieving these memories may serve as a less effective induction of a self-reference mode for aMCI patients compared to healthy controls. In this case, there would be a stronger association between the emotional content in autobiographical memories and the magnitude of self-reference effects for older adults than those with aMCI. In contrast, it could be the case that the effects of emotion in autobiographical memory contribute similarly to self-referencing effects for both groups. To assess these different possibilities, we analyzed the emotional qualities of autobiographical memories and tested the relationship with the self-reference effect in healthy older adults and aMCI participants.

For the present study, we hypothesize that emotional language usage (both positive and negative) during autobiographical memory narrations would influence the magnitude of self-referential memory effects on a subsequent memory task (i.e., higher recognition hit rates for self-referenced words compared to control words). We also predicted the effects of emotional language content would be stronger among healthy older adults compared to aMCI participants, due to potentially less rich autobiographical memories [50], particularly those that are emotional [28], in aMCI participants.

Methods

Participants

Twenty-two participants (M_age = 79.33; female = 5) diagnosed with aMCI due to the AD pathology [53] by a neurologist at the VA Boston Memory Disorders Clinic, were recruited through the Center for Translational Cognitive Neuroscience, Veterans Affairs Boston Healthcare System. Thirty-seven healthy older adults (M_age = 75.86, female = 28) with no cognitive complaints were recruited through Brandeis University. The healthy controls also scored within the normal range on all cognitive tests (see neuropsychological test performance in Table 1). All participants were recruited as part of a prior study from our lab [26]. The sample size between the Zhang et al. [26] paper and the present study is different as Zhang et al. [26] used a matched sample.

Table 1.

Mean and SD for Age, Education and Neuropsychological Assessments by Group.

	Older Adults n = 37		aMCI n = 22		T-test
	M	SD	M	SD	p-value
Age	75.87	7.27	79.55	6.63	0.06
Years of Education	17.25	3.12	16.18	2.59	0.24
MMSE Score	28.73	1.31	27.23	1.59	<0.001
CERAD Delayed Recall	6.62	2.26	3.45	2.48	<0.001
CERAD Recognition	9.22	1.61	7.73	2.26	0.005
Geriatric Depression Scale (Short form)	1.47	2.43	2.82	2.53	0.05

Exclusion criteria included history of stroke or brain damage, and for healthy older adults, a score of less than 26 on the Mini-Mental Status Exam [54]. All participants were monetarily compensated for their participation.

Procedure

The data for this study was collected as part of an ongoing study in our lab [26]. The study was approved by the IRBs at Veterans Affairs Boston Healthcare System and Brandeis University. After providing informed consent, participants completed a demographics measure. This was followed by the experimental task during which subjects participated in two ostensibly unrelated tasks: an autobiographical memory interview and the common/self-descriptive task (described below).

Autobiographical Memory Interview: To induce a self-focused mode of thought, participants were asked to relate in detail an autobiographical memory from three different periods of their lives: childhood (before the age of 18), early adulthood (between ages 18–35), and within the last six months. They had five minutes for each period. Memories were elicited from each of these time periods in order to have more data, based on three independent memories, in order to test hypothesized influences of emotion on self-referential memory performance (but see [28]; and [55] for consideration of effects of time period). Participants were specifically asked to narrate memories that were neutral or at least not negative, in order to avoid the ethical concern of priming negative emotion that might be upsetting to participants. Participants were asked for permission to record their verbal responses.

Common/Self-Descriptive Task: In this computer-based task a series of 32 adjectives from Craik et al. [56] was presented in a random order for the encoding task and each word was paired with one of two instructional categories (common or self-descriptive), noted at the top of the screen. For half of the trials (prompted by the label “Common”), participants made a semantic decision, deciding whether or not the word presented is commonly used by the average person, someone other than themselves. For the other half of the trials (prompted by the label “Self-Descriptive”), participants decided whether or not the word describes them personally. Words were counterbalanced between conditions, across participants. Each adjective was presented for a period of 6 s during which the participants made a YES/NO response verbally, which the experimenter keyed into the computer. After this encoding phase, participants completed a verbal subtraction task (i.e., serial 6 subtraction) as a distractor for 1.5 min. They then completed a recognition task in which they were shown the 32 studied adjectives along with 32 new adjectives and were asked to decide whether each word was new (i.e., one that did not appear in the encoding stage) or old (i.e., one that appeared in the encoding stage). Participants responded by pressing a key on a 1–6 confidence scale, with 1–3 representing “old” judgments (1: most confident word was old), and 4–6 representing “new” judgments (6: most confident word was new). Data were collapsed into “new” and “old” to ignore confidence in the analyses (see [26] for further consideration of this factor).

Following these tasks, participants completed a series of neuropsychological tests (see Table 1), including the mini mental state exam (MMSE [54]), Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) word memory task [57], trail making, word fluency [58], Boston Naming Test [59], and the Geriatric Depression Scale (GDS; [60]). At the end of the study, participants were debriefed, and their questions were addressed.

Scoring of content

The recorded autobiographical memories were transcribed and then analyzed using the Linguistic Inquiry and Word Count (LIWC) automated text analysis program [61]. The program is specifically designed to analyze various emotional, cognitive and structural components of speech. After processing the narrative, it gives 90 output variables and a breakdown of all the aspects of speech that were used in the narrative text. The program’s raw output for each of the 90 variables is the total number of words belonging to that variable divided by the total number of words used in the text. For this study we selected positive emotional content, negative emotional content and first-person reference (i.e., the usage of “I” and “we”) for analysis. The results for first-person reference are discussed in detail in the supplementary materials.

Analytic Plan

We employed linear regression analyses to assess whether the usage of positive and negative emotional content (coded with LIWC) in autobiographical memories impact self-reference memory in healthy older adults and aMCI participants and how these relationships differ between the two groups. Data were analyzed using MATLAB Programming Software.

To investigate whether emotional components of narrative speech influence the magnitude of self-reference effect and memory, we tested a linear regression model in which the categorical predictor variable was group (i.e., controls and aMCI), the two language predictor variables were positive emotion and negative emotion usage during narrative content, coded from the autobiographical interviews, and the dependent variable was the self-reference effect. The self-reference effect was defined as the difference between the recognition hit rate for words encoded when making a “self” judgment and that for words encoded when making a “common” judgment. Because we were interested in seeing whether the two groups (i.e., controls and aMCIs) differed in how emotional language predicted accurate memory, an interaction model was set up for the linear regression model, testing the interaction between group and emotion language usage in predicting the self-reference effect. If the interaction model was not significant, then a simpler additive model was tested.

From the total sample, values that were ± 2.5 SD from the mean for both dependent variables and all predictor variables were considered outliers and removed before the final analysis. Thus, the final sample size for different variables ranged from 34 to 37 for controls and from 21 to 22 for aMCI participants (see Table 2). In addition, adding GDS scores as a covariate to the model did not alter which effects reached significance.

Table 2.

Mean and SD for Dependent and Predictor Variables by Group.

	Older Adults			aMCI			t-test
	n	M	SD	n	M	SD	p-value
Memory Measures
Self-Reference Hit Rate	37	0.80	0.16	22	0.69	0.23	0.04*
Common Hit Rate	37	0.65	0.19	21	0.61	0.21	0.50
Self-Reference Effect (Self – Common) Hit Rate	37	0.15	0.17	21	0.11	0.26	0.50
False Alarm Rate	34	0.13	0.14	21	0.28	0.20	0.001*
Hit – False Alarm Rate: Self-Reference	37	0.68	0.22	22	0.42	0.32	<0.001*
Hit - False Alarm Rate: Common	37	0.53	0.24	21	0.33	0.30	0.007*
Emotion Usage in Narrations
Positive Emotion	37	2.76	0.62	22	2.32	0.74	0.02*
Negative Emotion	36	0.70	0.35	21	0.76	0.42	0.79

Note: The mean reported for the emotion variables is the mean of the proportion of the emotion variable (e.g., positive emotion; negative emotion) to the total number of words for each participant; * denotes p < .05.

Results

Descriptive statistics and independent samples t-tests for the demographics, neuropsychological assessment scores and predictor and criterion variables are presented in Table 1, Table 2. As shown in Table 2, older adults remembered significantly more items in the self-reference condition than aMCI patients, but the groups did not differ in memory in the common conditions or the overall self-reference effect (performance in the self condition compared to the common condition). In terms of language usage, a mixed ANOVA with emotion as a within-participant factor and group as a between-participants factor revealed a significant main effect of emotion. Positive language was used more than negative language, F(1, 55) = 307.10, p < .001, η_p² = 0.85, perhaps reflecting task instructions, but the groups did not differ overall in the amount of emotional language, F(1, 55) = 3.34, p = .07, η_p² = 0.06. There was a significant interaction between group and emotion, F(1, 55) = 6.37, p = .02, η_p² = 0.10. The results of follow-up t-tests to better understand the interaction, shown in Table 2, indicate that aMCI patients had lower usage of positive emotion language than healthy older adults, but that the groups did not differ in negative emotion language.

We first examined the scatterplots for the two language variables with the dependent variable of self-reference effect. Upon inspection, a non-linear relationship was noted between the self-reference effect and negative emotion. Negative emotion was then log transformed using the boxcox function in MATLAB and the linearity was tested. This transformation led to a reasonable linear relationship between the predictor and outcome variables, therefore the log transformation was used for the non-linear predictor variables (i.e., negative emotion). The relationship between self-reference effect and positive emotion usage was linear and therefore, raw data was used for this variable when examining its relationship with the self-reference effect.

The correlations of both predictor variables with the self-reference effect as the dependent measures are presented in Table 3. Of the two predictor variables, only positive emotion usage has a significant positive correlation with the self-reference effect (Table 3). In addition, the lack of a strong correlation between positive and negative emotion, as well as the evidence that each type of emotion has a different relationship to memory performance, supports the decision to separately analyze positive and negative emotion (rather than combining them into a measure of overall emotion usage).

Table 3.

Correlation between Self-Reference Effect Hit Rate and Emotion Language Usage during Autobiographical Memory Narration. Note: p-values provided in parentheses.

	Self-Reference Hit Rate	Positive Emotion Usage	Negative Emotion Usage (Log Transformed)
All Participants
Self-Reference Hit Rate	1	–	–
Positive Emotion Usage	0.31 (0.02)*	1	–
Negative Emotion Usage (Log Transformed)	−0.005 (0.97)	0.07 (0.60)	1
Older Adult Participants
Self-Reference Hit Rate	1	–	–
Positive Emotion Usage	0.28(0.10)	1	–
Negative Emotion Usage (Log Transformed)	0.10(0.57)	0.13(0.46)	1
aMCI Participants
Self-Reference Hit Rate	1	–	–
Positive Emotion Usage	0.33 (0.14)	1	–
Negative Emotion Usage (Log Transformed)	−0.11 (0.63)	0.05(0.82)	1

Self-reference effect: the effect of emotion usage and group

In this model, self-reference effect was regressed on the predictor variables positive emotion and log-transformed negative emotion usage. Additionally, the categorical variable group (i.e., controls and aMCI) was included as a predictor to test whether the effects of emotion usage on the self-reference effect differed across both groups. Thus, this model assesses whether emotion usage when relating autobiographical memories is associated with a larger or smaller self-reference effect on a subsequent (common/self-descriptive) task. We first tested the interactive model, which revealed no significant interactions between group and either positive emotion usage or the log-transformed negative emotion usage, further, the interaction between positive emotion and log-transformed negative emotion usage was marginal and did not reach significance (see Supplemental Table 5). Because the results from the interactive model were not significant, a simpler additive model was used (see Analytic Plan). In the additive model with all the three predictive variables (see Supplemental Table 4), only positive emotion usage was statistically significant. In this model, we found that negative emotion usage and group were not significant predictors of the self-reference effect, but positive emotion usage was a significant predictor. Therefore, using the principle of parsimony, we present the model results with only the significant predictor, i.e., positive emotion usage (see Table 4).

Table 4.

Predicting Self-Reference Effect (Self – Common Hit Rates) from Language Usage during Autobiographical Memory Narration. Coefficients are unstandardized.

Predictors	β	SE	t-value	p-value
Intercept	−0.10	0.10	−1.01	0.32
Positive Emotion Usage	0.09	0.04	2.47	0.02
F (1,57)		6.11		0.02
R2		0.10
Adjusted R2		0.08
Root Mean Squared Error		0.20

In this model, positive emotion usage was found to be a significant predictor of the self-reference effect. The estimates slope coefficient of positive emotion is 0.09 (t(57) = 2.47, p = .02) indicating that one unit difference (i.e., proportion of positive emotion words to total number of words) in positive emotion usage predicts a 0.09-unit difference in the self-reference effect. The Multiple R-squared is 0.10 indicating that 10% of the variation in the self-reference effect can be explained by positive emotion usage.

We also conducted additional exploratory analyses which we introduce briefly here and present in detail in the Supplementary Materials.

Exploratory analysis of the self-reference effect

The effect of first-person reference, emotion usage and group

Due to the potential for first-person reference language to reflect a self-focus, we also assessed how the first-person reference language (i.e., I, we) while narrating autobiographical memories affected the self-reference effect. This model assesses whether emotion and first-person reference usage when relating autobiographical memories are associated with a larger or smaller self-reference effect on a subsequent (common/self-descriptive) task. In this model, positive emotion usage continued to be a significant predictor of the self-reference effect. The contribution of first-person reference language was only marginally significant. See Supplementary Materials for full details of the models tested and results.

Exploratory analyses of the false alarm rate

The effect of emotion usage and group

Based on our prior findings that generating autobiographical memories can increase false alarm rates in analyses using these same samples of participants [26], we also conducted exploratory analyses to test the influence of emotional language usage on false memories. In this model, false alarm rate was regressed on the log transformed predictor variables positive and negative emotion usage. Additionally, group (i.e., controls and aMCI) was included as a categorical variable in the model to test how the effects of emotion usage on the false alarm rate differed across both groups. Thus, this model assesses what aspects of emotional language during the narration of autobiographical memories makes someone more or less prone to mistakenly endorse new information as previously studied, on a subsequent memory test. The interactive model was found to be significant and using the principle of parsimony, the non-significant variables were removed. In this model, we found a significant interaction between log-negative emotion and group (t(48) = −2.29, p = .03) and a marginal interaction between log-positive emotion and log-negative emotion usage (t(48) = −1.91, p = .06). The final model is presented in Supplementary Table 2 and is discussed in greater detail in the Supplementary Materials.

The effect of first-person reference, emotion usage and group

Exploratory analyses were conducted to assess the influence of an additional language variable of first-person reference usage during autobiographical memory narrations on the false alarm rate. This model assesses whether emotion and first-person reference usage when relating autobiographical memories are associated with greater or fewer false alarm rates on a subsequent (common/self-descriptive) task. In the interactive model, we found significant interactions between positive and negative emotion usage, positive emotion and first-person reference usage, and negative emotion usage and group. In the model with only the significant predictors, we found significant interactions between log positive emotion and log negative emotion usage (t(46) = −3.29, p = .002), log positive emotion and log first-person reference (t(46) = 3.11, p = .003) and log negative emotion usage and Group (t(46) = −3.33, p = .001). See Supplementary Materials for details.

Discussion

This study analyzed language usage during autobiographical memories to assess whether emotional aspects of the narrative content predicted the self-reference effect. We first discuss the findings regarding emotional language usage in the autobiographical memory narrations for healthy older adults and aMCI participants, then examine how emotional content influences self-reference effects, and finally consider the lack of group differences.

In terms of emotional language used when sharing autobiographical memories, both groups produced significantly more positive than negative emotion language usage, consistent with prior work demonstrating preferential processing of positive than negative emotional information with age [62], [63]. Moreover, older adult controls used more positive emotional language than aMCI patients, though the groups did not differ on negative emotional language. It is important to note that the reduced usage of negative, compared to positive, emotion could reflect task instructions that encouraged participants to relate neutral or positive autobiographical memories. Although it does not appear as if the use of negative language was at floor (see Table 2; scores ranged from 0.14 to 1.69 across participants) and there is no indication that the groups differed in their approach to the task or failed to follow instructions, we interpret findings regarding negative emotion with caution. In light of our task instructions, we focus on the effects of positive emotion on self-reference effects.

In line with our predictions, emotional language usage during autobiographical memories appears to influence the self-reference effect. Specifically, we found that greater usage of positive emotional language was associated with better subsequent memory for self-referenced information in both healthy older adults and people with aMCI. These findings go beyond our previous studies attempting to use autobiographical memories to induce a sustained self-reference effect [19], [26]. In those studies, using autobiographical memories alone was not sufficient to increase the magnitude of the self-reference effect. The present results, however, suggest that positive emotion may be a critical factor in the potential for autobiographical memory to enhance self-reference effects. More generally, the effects of emotion converge with prior work showing that positive emotional states enhance memory and learning [39]. Despite the greater use of positive emotional language in older control participants compared to aMCI patients, the mechanisms through which emotion influences self-referential memory may be intact in aMCI. Contrary to the literature in support of negative emotional states enhancing learning and memory [40], our results did not find this effect for negative emotion language usage; we do not discuss this effect further given the potential limitations of the design.

Contrary to our prediction that autobiographical memories would differentially influence self-referencing in older adults and people with aMCI, there were no group differences in the way autobiographical memory retrieval influenced the self-reference effect for these two populations. Despite the typical declines in memory seen in people with aMCI, they still show the same benefits from self-referencing following the retrieval of positive autobiographical moods as their healthy counterparts. These findings further support the use of self-referencing strategies to enhance memory in both healthy older adults and people with aMCI. Although prior work has demonstrated some differences between healthy older adults and people with aMCI in autobiographical memory narrations [28], [50], these group differences may emerge when considering the richness or amount of detail, or emerge when more negative emotional content is elicited.

Our work extends the literature on the mnemonic benefits of positive emotion to self-referential memory, demonstrating that retrieving positive autobiographical memories prior to self-referencing is a useful technique in aiding self-related memory for both healthy older adults and people with aMCI. This finding converges with prior studies in showing that relating positive information to the self during encoding aids memory, although in these studies the effects were more prominent for people with aMCI; older adult controls were not as sensitive to the emotional content of the information [41], [45], [46]. Other studies have examined whether the valence of words to be encoded influenced the self-reference effect in healthy older adults and people with mild memory impairments [41], [43], or whether the valence of statements modulated the self-reference effect [41]. Whereas these studies focused on the valence of the content of information to be remembered, the present study considers the emotional content of the autobiographical memories retrieved by individuals, apart from the information to be remembered. Converging with prior work, positive emotion can enhance the effectiveness of self-referencing. These behavioral findings converge with some neuroimaging evidence indicating larger effects for positive than negative information, with positive self-referenced information activating right dorsomedial prefrontal cortex more than negative words [64].

In addition, autobiographical memories may represent a robust intervention able to improve both mood and memory in older adults, even for those with some degree of memory impairment [65]. The present results suggest that there may be benefits to combining both an autobiographical memory intervention focused on positive memories with a self-referencing strategy, potentially allowing for more flexible and self-guided interventions based on reminiscing and applicable in a variety of naturalistic settings. Whereas much of the prior research relies on trial-by-trial manipulations of self-referencing, perhaps retrieving positive autobiographical information has the potential to more flexibly improve memory performance. Perhaps autobiographical memories support both robust encoding and retrieval of memories. For instance, when attempting to remember a sequence of errands, one could retrieve past memories related to the tasks (e.g., remembering the last time I saw my nephew, for whom I must mail a birthday card) to help encode and retrieve the action items, or when grocery shopping, one could retrieve a positive autobiographical memory about a previous time they cooked the same meal for which they are purchasing ingredients, and use that to aid retrieval of the items to be bought. Additional research would be needed to determine the situations in which such an approach would be effective and how to best implement the strategy. If successful, the approach investigated in this study has implications for supporting memory performance of both healthy older adults and aMCI patients.

The pattern of findings also speaks to the different systems involved in memory, and the ways in which those are impacted by aMCI. Although medial temporal regions that are critical for long-term memory are affected early in AD, including for those with aMCI (e.g., [53]), other brain systems continue contributing to memory until later stages of the disease, allowing aMCI patients to benefit from positive emotion and self-referencing much like older adult controls. Further, recent literature has argued for shared mechanisms between self-referential processing and emotion [38]. It is possible that some shared mechanisms between these two processes, such as those governed by mPFC, are relatively preserved in aMCI and early stages of AD. As this study shows that retrieval of positive emotional autobiographical memories can enhance the effectiveness of self-referencing in memory, this could be evidence of shared mechanisms between self and positive emotion.

There are some limitations to the present study. Based on the design, it is not possible to determine whether the increased self-reference effect is due to a positive mood induction alone or is a function of retrieving positive autobiographical memories. Indeed, retrieving autobiographical memories is often used as a mood induction technique [66], [67]. Although it is unclear whether positive mood enhances the self-reference effect in memory [68], it is possible that mood alone accounts for the effects reported in this study. To fully evaluate the source of effects, future studies would need to directly assess mood or compare conditions that separately manipulate mood and the emotional content of the autobiographical memories. We are less inclined to think that the effects reflect mood induction alone, as positive and negative emotion were not correlated with each other and did not impact the self-reference effect in consistent ways, perhaps arguing against the possibility that positivity and negativity operated together to create an overall mood manipulation. However, it is important to note that participants were not specifically asked to narrate emotional memories, but rather were instructed to narrate memories that were neutral and not negative. Thus, the instructions may have limited the range of positive and negative emotional language used by participants. Relatedly, this leads to another shortcoming of the study in that this manipulation may have evoked incidental rather than more controlled emotional states. These features of the study could restrict the results from generalizing to more naturalistic emotional situations.

The present results could be extended by comparing performance to younger adults and to people with more advanced stage of memory decline due to AD, in order to clarify effects of aging and memory impairment1. As is, results indicate the usefulness of studying emotional language usage to reflect different aspects of being in a self-reference mode. Our results reveal that the novel combination of positive emotion induction through autobiographical memory retrieval and self-referencing may offer an approach to maximally enhance memory performance in older adults with and without memory impairment.

CRediT authorship contribution statement

Nishaat Mukadam: Conceptualization, Methodology, Formal analysis, Visualization, Writing - original draft, Writing - review & editing. Wanbing Zhang: Investigation. Xiaodong Liu: Methodology, Writing - review & editing. Andrew E. Budson: Resources, Writing - review & editing, Funding acquisition. Angela Gutchess: Methodology, Resources, Writing - review & editing, Supervision, Funding acquisition.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: