1. Introduction
Emotion enhances memory, particularly autobiographical memory (AM), which refers to a knowledge base of self-related and significant information that goes beyond the recall of personally experienced events to incorporate perspective, interpretation and emotion to make up an individual’s unique life story (Fivush, Reference Fivush2011; Williams et al., Reference Williams, Conway and Cohen2008). One major form of organization in the knowledge base is that autobiographical information is organized at varied levels of specificity, from vivid sensory, perceptual and emotional details of particular moments to high-level summaries of life periods (Barsalou, Reference Barsalou1988; Conway & Bekerian, Reference Conway and Bekerian1987; Conway & Pleydell-Pearce, Reference Conway and Pleydell-Pearce2000; Prebble et al., Reference Prebble, Addis and Tippett2013). This means that AM, organized in a form of hierarchical structure with different levels of specificity, consists of specific AM and general AM from bottom to top (Conway & Pleydell-Pearce, Reference Conway and Pleydell-Pearce2000; Williams et al., Reference Williams, Barnhofer, Crane, Hermans, Raes, Watkins and Dalgleish2007). Specific AM is a recollection of a past single event with episodic details that happened at a particular time and place and did not last longer than 1 day. In contrast, general AM refers to a general impression about past events without specific temporal and spatial details, including categorical memory (i.e., the memory refers to a class of generic events) or extended memory (i.e., the memory lasted longer than 1 day) (Sumner, Reference Sumner2012; Williams & Broadbent, Reference Williams and Broadbent1986; Williams & Scott, Reference Williams and Scott1988). Generally, humans are capable of retaining a wealth of AM details. Yet, the quantity and quality of the retrieval differ substantially across content, given that when retrieving personal past, not all memories come to mind with equal ease or with the same level of detail (Holland & Kensinger, Reference Holland and Kensinger2010). Events imbued with emotion seem to be remembered more vividly and durably than events lacking an emotional component (Buchanan, Reference Buchanan2007). Moreover, individuals may retrospect different emotional episodes with different levels of detail or specificity (Berntsen, Reference Berntsen2002; Talarico et al., Reference Talarico, Berntsen and Rubin2009). However, till now, little is known about the psychological mechanisms (i.e., emotional and cognitive processes as well as the thinking patterns) behind the construction of specific and general AM of different emotional events.
In the laboratory studies, one of the most common methods to measure AM is presenting participants with a list of cue words and asking them to recall a particular memory that corresponds to the cue words (see Holland & Kensinger, Reference Holland and Kensinger2010, for a review). These cue words can be emotional in nature (Buchanan, Reference Buchanan2007; Holland & Kensinger, Reference Holland and Kensinger2010). Previous studies examining self-narratives of past experiences have found that the words individuals use reveal the psychological states of the narrators (Pennebaker et al., Reference Pennebaker, Chung, Frazee, Lavergne and Beaver2014; Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010). For example, affective process words with both pleasant and unpleasant connotations (e.g., happy, sadness, hate, pride) are indicative of emotional processes in recalling past events, providing a global measurement of emotional reactions. Positively or negatively connoted words, as subcategories of affective process words, provide a window into individuals’ subjective feelings and show how they cope with such events (Pennebaker et al., Reference Pennebaker, Boyd, Jordan and Blackburn2015; Pennebaker et al., Reference Pennebaker, Mehl and Niederhoffer2003; Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010). The use of positive words in recollection of positive events exhibits narrators’ positive feelings, and that in recollection of negative events indicates a sound emotion regulation ability (Kahn et al., Reference Kahn, Tobin, Massey and Anderson2007; Robertson & Hopko, Reference Robertson and Hopko2009). The use of negative words in narrating positive memories is reflective of an attenuated positive affect or distorted emotion perception, while that in narrating negative memories indicates immersion in negative events (Gamber et al., Reference Gamber, Lane-Loney and Levine2013; Himmelstein et al., Reference Himmelstein, Barb, Finlayson and Young2018; Lee & Cohn, Reference Lee and Cohn2010). Cognitive process words include words that reveal thinking and causal reasoning (e.g., believe, think, reason, cause). They are indicative of the construction and rationalization processes during narration (Pennebaker, Reference Pennebaker2015; Pennebaker et al., Reference Pennebaker, Mehl and Niederhoffer2003). An increasing use of cognitive process words reflects an effort to elaborate on what happened and to integrate it with subjective feelings to recognize a coherent story about past events (Boals & Klein, Reference Boals and Klein2005; Crespo & Fernandez-Lansac, Reference Crespo and Fernandez-Lansac2016). For example, the use of cognitive process words in recalling negative events has been shown to positively associate with psychological well-being since systematically analyzing the causes and perception of these events might facilitate the adaptive processes for unpleasant experiences (D’Andrea et al., Reference D’Andrea, Chiu, Casas and Deldin2012; Patwardhan & Kulkarni, Reference Patwardhan and Kulkarni2023). Thinking styles, which are embodied using a series of function words (e.g., I, all, even, never, but), indicate how people are thinking (Pennebaker et al., Reference Pennebaker, Chung, Frazee, Lavergne and Beaver2014). Generally, frequent use of articles and prepositions reflects analytic thinking such that individuals are thinking in a logical, hierarchical and formal way. In contrast, a low proportion of articles and prepositions and a concurrent high rate of pronouns, auxiliary verbs, conjunctions, adverbs and negations reflect a narrative style in which people tell stories and speak in the here and now (Pennebaker, Reference Pennebaker2015; Pennebaker et al., Reference Pennebaker, Chung, Frazee, Lavergne and Beaver2014). Previous studies have shown that overgeneral memory is responsive to the manipulation of the extent of analytic thinking, with analytical thinking contributing to the maintenance of overgeneral memory (Watkins & Teasdale, Reference Watkins and Teasdale2001). The use of analytical thinking words after traumatic events suggests that individuals try to make sense of the negative information through a formal and logical way, posing a huge impact on the meanings, causations and consequences of these events (Monzani et al., Reference Monzani, Vergani, Pizzoli, Marton and Pravettoni2021).
Taken together, the linguistic analysis provides an effective way to measure the content of memory and indicate the psychology behind it. Therefore, the present study analyzed the texts of AM narratives to address how the psychological mechanisms of memory would differ between specific and general AM and how the differences would vary across different emotions. Based on the literature, we focused on emotional and cognitive processes and thinking styles underpinning emotion-dependent psychological mechanisms of specific and general AM. We analyzed narratives of AM categorized as specific and general under cues of discrete emotions, instead of valence, given that different influences of emotion on cognition may go beyond valence and occur between distinct emotion types within the same valence (DeSteno et al., Reference DeSteno, Petty, Rucker and Wegener2000; Zhang et al., Reference Zhang, Hu, Lai, Hu and Gao2021). While extensive evidence has suggested six basic emotion categories (happiness, sadness, disgust, anger, fear and surprise), these emotions are not discrete categories fully independent of each other. Happiness and sadness feature their own uniqueness, while disgust and anger, as well as fear and surprise, share substantial overlaps (Wang et al., Reference Wang, Zhu and Chen2019). Therefore, some studies have proposed that basic emotions may include four categories (happiness, sadness, surprise/fear and disgust/anger) rather than six (Jack et al., Reference Jack, Garrod and Schyns2014; Jack et al., Reference Jack, Sun and Delis2016). Considering that previous studies on autobiographical memory often recruit happiness, sadness, fear and anger (Mills & D’Mello, Reference Mills and D’Mello2014; Wang et al., Reference Wang, Hutt, Kulkofsky, McDermott and Wei2006), we chose them as emotional cues, with the neutral cue as the baseline.
To the best of our knowledge, no previous research has compared the linguistic markers of AM of different specificity across different emotional events. Therefore, the hypotheses in our study remained relatively general. Given that specific AM are often recalled with great feelings of vividness but general AM are often recalled without rich sensory and perceptual details (Holland & Kensinger, Reference Holland and Kensinger2010), we expected higher use of affective process words in specific AM as compared to general AM. We also expected higher use of analytical thinking words in general AM relative to specific AM due to the important role of analytical thinking in the maintenance of overgeneral memory (Watkins & Teasdale, Reference Watkins and Teasdale2001). Moreover, we anticipated that specific versus general memory differences in using emotionally valenced words, cognitive process words and thinking style words would vary across AM of different emotional events since previous findings have suggested that the use of such words in narrations of different emotional experiences has different psychological implications (D’Andrea et al., Reference D’Andrea, Chiu, Casas and Deldin2012; Gamber et al., Reference Gamber, Lane-Loney and Levine2013; Monzani et al., Reference Monzani, Vergani, Pizzoli, Marton and Pravettoni2021; Robertson & Hopko, Reference Robertson and Hopko2009).
2. Methods
2.1. Participants
The present study aimed to obtain 210 AM reports of each emotion (105 in either of the specificity conditions). The target sample was determined using G*Power 3.1 (Faul et al., Reference Faul, Erdfelder, Lang and Buchner2007) based on a minimum power of 0.95 at an α level of 0.05 as well as a medium effect size (0.25) reported in previous studies examining word use of emotional autobiographical narratives (Tani et al., Reference Tani, Peterson and Smorti2016). Considering that every participant was asked to report six memories following each emotional cue, the required number of data could be collected in 35 participants. However, participants would report overgeneral or very specific AM under a particular emotion, given previous evidence showing that individuals may retrospect different emotional episodes with different levels of specificity (Berntsen, Reference Berntsen2002; Talarico et al., Reference Talarico, Berntsen and Rubin2009). Therefore, we enlarged the sample size to ensure that we could collect at least 210 general and specific AM of each emotion. Finally, the study included 122 healthy participants (60 males; age: 22.3 ± 2.09) recruited by local advertisement. To control for the potential confounding effects of handedness on AM recall (Parker et al., Reference Parker, Parkin and Dagnall2017), only right-handers were included. Given that there is a strong positive association between depression and overgeneral memory (Sumner et al., Reference Sumner, Griffith and Mineka2010), participants with a score of the second version of the Beck Depression Inventory (BDI-II) (Beck et al., Reference Beck, Steer and Brown1996) greater than 13 were precluded from participation. The mean BDI-II score in participants was 4.47 ± 3.71. All participants gave written informed consent in accordance with the latest version of the Declaration of Helsinki and the study was approved by the Ethics Committee of the University of Electronic Science and Technology of China.
2.2. Task and procedure
The autobiographical memory task (Williams & Broadbent, Reference Williams and Broadbent1986) was computerized and administered using Eprime 2.0. Before the task, participants were provided both written and verbal instructions explaining that they were to be shown on a computer screen five emotion-label words (i.e., happy, sad, angry, fearful and neutral) and then verbally recall an event that happened to them in the past in accordance with the given emotion. The instructions emphasized that the event should be personally experienced and the report should be given in detail (e.g., where and when the event occurred, what happened, who was present). In each of the five emotional contexts, participants elaborated six self-relevant memories, which constituted five (emotional) blocks of six (memory) trials. The permutation of the emotional blocks was randomized across participants. Each memory trial lasted 30 s, a time limit widely used in previous studies assessing the retrieval specificity of autobiographical memory under different cues (Hermans et al., Reference Hermans, Van den Broeck, Belis, Raes, Pieters and Eelen2004; Kleim & Ehlers, Reference Kleim and Ehlers2008). When time was up, a fixation was shown on the screen indicating the end of the report. Upon the presentation of the next cue, participants started to recall another corresponding memory. Speech was recorded by a computer with a Lenovo UM10C microphone.
Data preprocessing
The speech recording files collected from each participant under each of the five emotional contexts were transcribed into texts using Xunfei’s speech recognition platform (https://www.iflyrec.com). After transcribed texts were manually checked word by word, AM was classified into two categories (specific and general) by two trained reviewers who were unaware of the experiment design. The reviewers were instructed to code memories of an event that happened at a particular time and place within a time span less than 1 day as specific AM but summaries or classifications of repeated events (i.e., categorical memory), or recollection of an event that lasted longer than 1 day (i.e., extended memory) as general AM (Prebble et al., Reference Prebble, Addis and Tippett2013; Sumner, Reference Sumner2012; Williams & Broadbent, Reference Williams and Broadbent1986; Williams & Scott, Reference Williams and Scott1988). When participants provided a verbal response derived from general knowledge rather than a memory (i.e., semantic associates), the response was coded as general AM given that semantic associates represent a type of overgeneral memory positioned at a level even higher than categorical memory in the hierarchy structure of memory system (Conway & Pleydell-Pearce, Reference Conway and Pleydell-Pearce2000; Debeer et al., Reference Debeer, Hermans and Raes2009; Raes et al., Reference Raes, Hermans, Williams and Eelen2007). The reviewers independently classified all transcripts, and Cohen’s kappa coefficient was calculated (k = 0.937, p < 0.001), showing perfect agreement between their classifications (Conger, Reference Conger2017). Disagreements were discussed and coded until a consensus was reached between them (Glachet & El Haj, Reference Glachet and El Haj2019). Also, all the reported memories (3660 trials in total) were judged by the two reviewers on whether or not the memories were self-relevant and whether or not the emotion of the memories conformed to the emotion category required in the task. All the memory trials were evaluated as self-relevant except for one happy trial, one sad trial, six angry trials (from 5 participants) and 17 neutral trials (from 15 participants). All the emotional memories retrieved met the requirement of the emotion category, whereas one happy narration, one angry narration, three fearful narrations (from 3 participants) and 10 neutral narrations (from 10 participants) were not consistent with the required emotion category and thus excluded from the statistical analysis. Finally, a total of 3620 trials (happy: specific 510, general 220; sad: specific 464, general 267; angry: specific 426, general 299; fearful: specific 480, general 249; neutral: specific 198, general 507) were included in the following analysis.
Regarding linguistic features, the Simplified Chinese version of Linguistic Inquire and Word Count (SCLIWC) dictionary, an exclusive Simplified Chinese text analysis dictionary established according to the English LIWC dictionary and the Traditional Chinese version of LIWC dictionary to bridge the gap between LIWC software and Simplified Chinese (Gao et al., Reference Gao, Hao, Li, Gao and Zhu2013), was imported into the Natural Language Processing and Information Retrieval (NLPIR) Chinese lexical analysis system (http://ictclas.nlpir.org) for the segmentation of transcribed texts. Afterward, TextMind, a Chinese language psychological analysis program optimized for processing Simplified Chinese and developed based on the SCLIWC to analyze preferences of different word categories in text, was used to generate word frequencies (Gao et al., Reference Gao, Hao, Li, Gao and Zhu2013). Most of the input words were recognized by TextMind (general AM texts: M = 90.13%; specific AM texts: M = 90%) and the dictionary cover rate did not show significant differences between AM categories (Z = −0.117, p = 0.907, Mann–Whitney U test, two-tailed).
TextMind could map texts in 80 word categories, reflecting different aspects of the psychology and each category is computed as a ratio with respect to the total word count (Gao et al., Reference Gao, Hao, Li, Gao and Zhu2013). Given that we focused on emotional and cognitive processes and thinking styles reflected by AM texts, 14 LIWC word categories were used and classified into three groups (emotional process, cognitive process and thinking style). The emotional process group (e.g., love, cry, worry) consisted of affective process words (Pennebaker, Reference Pennebaker2015). Considering that previous research has shown that the content of memory and the associated psychology behind it differ in words with differently valenced connotations (Gamber et al., Reference Gamber, Lane-Loney and Levine2013; Himmelstein et al., Reference Himmelstein, Barb, Finlayson and Young2018; Kahn et al., Reference Kahn, Tobin, Massey and Anderson2007; Robertson & Hopko, Reference Robertson and Hopko2009), we further compared the use of positive and negative emotion words, respectively, between specific and general AM following the analysis of the overall emotional word use. The cognitive process group (e.g., because, should, maybe) consisted of insight, causation, discrepancy, tentativeness and certainty words (Moore et al., Reference Moore, Yen and Powers2021). Previous research has suggested that thinking styles are embodied using a series of function words, in which a high proportion of articles and prepositions signal analytic thinking, a relatively logical and precise thinking style and a high proportion of pronouns, auxiliary verbs, conjunctions, adverbs and negations indicate narrative thinking, a relatively dynamic and flexible thinking style (Pennebaker, Reference Pennebaker2015; Pennebaker et al., Reference Pennebaker, Chung, Frazee, Lavergne and Beaver2014). Therefore, the thinking style group (e.g., until, usually, very) here was divided into two subgroups representing analytic and narrative thinking styles. It is notable that due to the unique feature of Chinese, there is no article category; the analytic thinking group only included prepositions. The narrative thinking group included pronouns, auxiliary verbs, conjunctions, adverbs and negations.
2.3. Data analysis
Before performing the analysis in IBM SPSS version 22, the ratios of all word categories in each word group were added up to a total ratio within each emotion and specificity condition (see Supplementary Table S1 for the details). Therefore, six variables, namely the ratios of affective process, positive emotion, negative emotion, cognitive process, analytical thinking and narrative thinking words, were obtained for either specific or general AM in each of the five emotions. Then, the data of all the conditions were tested for normality and homogeneity of variance assumptions using the Kolmogorov–Smirnov test and the Levene test, respectively, which showed that the data did not follow the normal distribution (ps <0.001) and did not verify the assumption for equal variances (ps < 0.05). Thus, we turned to a nonparametric test to compare linguistic features between specific and general AM under each emotion condition. We used the Mann–Whitney U test, which does not rely on distributional assumptions and allows for unequal variances, and it has been shown to have a robust statistical power with disproportioned sample sizes (Karch, Reference Karch2021; Treister et al., Reference Treister, Nielsen, Stubhaug, Farrar, Pud, Sawilowsky and Oaklander2015; Zimmerman, Reference Zimmerman1987). False discovery rate (FDR) was used for multiple comparison correction to control for the type I error. The threshold of p < 0.05 (FDR corrected) was considered to be significant.
3. Results
Mann–Whitney U test results revealed that significantly fewer affective process words were used in specific AM than in general AM of sad (Z = −2.93, p = 0.01), angry (Z = −4.74, p < 0.001) and fearful (Z = −3.85, p < 0.001) events (Figure 1). However, the differences in using such words were not significant between AM of different specificity when retrieving happy (Z = -1.3, p = 0.26) and neutral (Z = −0.76, p = 0.54) events. Furthermore, specific AM was provided with fewer negative words than general AM in sad (Z = −2.82, p = 0.02), angry (Z = −3.46, p < 0.001) and fearful (Z = −4.67, p < 0.001) narrations. In terms of the use of positive emotion words, however, no significant differences were identified between the two categories of AM under any emotional conditions (all ps > 0.1).
Figure 1. Emotion-dependent differences in word use between specific and general autobiographical memory. Black horizontal bars represent the median and the first and third quartiles are identified by the bottom and top of the bold vertical bars, respectively. The bottom and top of the thin vertical lines represent the lower and upper adjacent values, respectively. * p < 0.05, ** p < 0.01.
Regarding cognitive process words, general AM relative to specific AM featured a higher proportion of such words when happy (Z = −2.66, p = 0.02), sad (Z = −4.55, p < 0.001), angry (Z = −3.27, p < 0.001) and fearful (Z = −4.48, p < 0.001) experiences were recalled. There were no significant differences in using cognitive words for describing past neutral events (Z = −1.65, p = 0.44).
Regarding the use of words reflecting formal thinking styles, specific AM was reported with fewer such words than general AM under happy cues (Z = −2.33, p = 0.03), while in recollection of fearful events, specific AM was reported with more such words than general AM (Z = −3.28, p < 0.001). Retrievals under sad (Z = −1.68, p = 0.29), angry (Z = −0.24, p = 0.96) and neutral (Z = −0.69, p = 0.36) cues did not show any significant differences between AM categories in using formal thinking words. Regarding the use of informal thinking words, general relative to specific AM under happy cues featured more such words (Z = −2.58, p = 0.02). However, the use of informal thinking words did not show any significant differences under any other types of emotion (all ps > 0.1).
4. Discussion
To reveal the emotion-dependent psychological mechanisms of specific and general AM, we conducted a linguistic analysis of the recordings collected from an autobiographical memory task in happy, sad, angry, fearful and neutral contexts.
In the retrieval of events imbued with sadness, anger and fear, a higher proportion of emotional process words were used in general AM than in specific AM. Emotional words are cues reflecting emotional responses to experienced events (Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010), and more frequent use of emotional words indicates enhanced emotional reactivity, which has been shown to associate with decreased emotion regulation efficiency (Robertson & Hopko, Reference Robertson and Hopko2009, Reference Robertson and Hopko2013). Therefore, our results suggest that recalling negative experiences at a general level may lead to decreased emotion regulation abilities and increased emotional reactions. This interpretation is supported by the finding in Philippot et al. (Reference Philippot, Schaefer and Herbette2003) that priming a general (versus specific) mode of negative AM reconstruction results in more intense emotion in a subsequent task in which participants mentally re-lived an associated emotional experience.
Given that the emotional process word category includes words with both pleasant and unpleasant connotations (Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010) and the valence of an experience influences how likely AM is to be remembered (Williams et al., Reference Williams, Barnhofer, Crane, Hermans, Raes, Watkins and Dalgleish2007), the findings of emotional words in narratives were further clarified regarding the use of positive and negative emotion words.
Negative words were used more often in general AM than in specific AM when events imbued with sadness, anger and fear were retrieved. Previous research has shown that negatively emotion-laden words are used frequently in describing negative experiences and the increased use of these words is associated with the enhancement of perceived negative emotions (Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010). High frequencies of negative words indicate an immersion in negative events, leading to an impaired adaptive ability to cope with such events (Gamber et al., Reference Gamber, Lane-Loney and Levine2013; Lee & Cohn, Reference Lee and Cohn2010). Our findings on the use of negative words therefore suggest that maladaptive adjustment to experiences imbued with sadness, anger and fear results in recalling negative events in a general way, confirming our previous interpretation on the use of emotional word use. Indeed, previous studies have suggested that people troubled by their negative experiences are likely to develop a general retrieval style to avoid short-term distress, whereas this avoidance strategy comes at a long-term cost with maladaptive emotion regulation (Sumner, Reference Sumner2012; Williams et al., Reference Williams, Barnhofer, Crane, Hermans, Raes, Watkins and Dalgleish2007).
Regarding the use of positive words, there were no differences between AM of different specificity under any emotional conditions. Positive words are often reported in positive narratives and are cues of perceived positive emotions (Tausczik & Pennebaker, Reference Tausczik and Pennebaker2010). Positive emotions broaden the scope of attention and cognition (Fredrickson, Reference Fredrickson2001). Happiness has been shown to activate a global information processing strategy, such that when happy, people attend to a broad range of information in the environment (Gasper & Clore, Reference Gasper and Clore2002). This information processing strategy, as it turns out, affects memory as well, such that people recall a wide variety of memory details and use less effortful heuristics to reconstruct experiences in narrating happy events (Levine & Pizarro, Reference Levine and Pizarro2004). In the current study, the absence of general and specific AM differences in positive word use in the happy context indicates that the global information processing strategy makes no differences in the perceived positive emotion when retrieving happy events of different specificity. In recollection of aversive events (i.e., sad, angry and fearful), specific and general AM also did not show any significant differences in using positive emotion words. Finding positive meaning in memory of negative events is considered as mood incongruent recall, which is associated with a mood repair effect, lessening negative mood states (Brockmeyer et al., Reference Brockmeyer, Kulessa, Hautzinger, Bents and Backenstrass2015). Such recall provides an adaptive way to reframe negative events, leading to the re-emergence of positivity at future recollections, in turn alleviating the experienced negative emotions (Speer et al., Reference Speer, Ibrahim, Schiller and Delgado2021). Our findings suggest that reinterpreting negative experiences in a positive way bears no impact on the perceived positive emotion between recollections of different specificity.
Taken together, the general AM advantages in overall use of emotional process words in the sad, angry and fearful contexts seem to be contributed by the greater use of negative words. Avoidance strategy triggered by aversive events exerts an adverse effect on the adaptive ability to cope with negative emotion, leading to diminished negative emotion regulation efficiency.
Cognitive process words were used more frequently in general AM than in specific AM in all emotional (but not neutral) narrations. Cognitive words reflect the ruminative processes of assimilating what happened to recognize a coherent story (Crespo & Fernandez-Lansac, Reference Crespo and Fernandez-Lansac2016; D’Andrea et al., Reference D’Andrea, Chiu, Casas and Deldin2012; Patwardhan & Kulkarni, Reference Patwardhan and Kulkarni2023). Previous evidence has shown that the construction of specific episodic memory features minimal inter-episode interference, while generalized episodic memory features overlapping information across multiple episodes (Kumaran & McClelland, Reference Kumaran and McClelland2012). Therefore, the higher frequency of cognitive words reported in happy general AM indicates that integrating happy event-specific details into a holistic impression, as compared to recalling a single happy event, involves increased cognitive processes. On the other hand, higher frequencies of cognitive words reported in sad, angry and fearful general AM suggest an adaptive integration of negative memory under avoidance strategy comes at a cognitive cost since general AM with negative valence depends on integrating both threatening and unpleasant feelings, typically difficult to accept (Weinstein et al., Reference Weinstein, Deci and Ryan2011). However, when retrieving neutral events, the use of cognitive words did not between AM categories, possibly because most of the neutral memories in the present study were general conceptual knowledge divested of personal experience involvement (e.g., I had two cups. One was put in the lab and the other one was put in the dormitory. I used to carry the one in the dormitory). Given that emotion increases self-focused attention during recalling (Mukadam et al., Reference Mukadam, Zhang, Liu, Budson and Gutchess2021), our finding suggests that although low emotion load decreases personal involvement, it makes no difference in cognitive complexity between AM categories.
We found general and specific AM differences in discrete emotional contexts regarding the use of both analytic and narrative thinking words. When past events imbued with fear were recalled, specific AM featured a higher proportion of analytic thinking words than general AM. Fear has been shown to motivate people to selectively encode and retrieve information associated with threats and means of avoiding them since such emotion alerts them to potential dangers that are similar to previously experienced circumstances (Lench & Levine, Reference Lench and Levine2005; Levine & Pizarro, Reference Levine and Pizarro2004). Our finding suggests that the potential threat posed by a single fearful event relative to that in a general fearful impression contributes to promoting a more precise and rigorous thinking style. However, in recollection of happy events, specific AM contained a lower proportion of both analytic and narrative thinking words than general AM. As previously discussed, when people are happy, they are likely to attend to global and heuristic characteristics of information rather than its local details (Gasper & Clore, Reference Gasper and Clore2002). On the other hand, healthy individuals generally experience positive events often, and thus, there may be a number of related events activated, interfering with the accurate retention details of any single positive experience and increasing the occurrence of false retrieval details (Holland & Kensinger, Reference Holland and Kensinger2010; Storbeck & Clore, Reference Storbeck and Clore2005). Therefore, our results suggest that when people retrieve happy events in a general way, their recollections may be less interfered with by concurrent activation of retention details of the same valence, as general retrieval implicates a more dynamic and flexible thinking style.
5. Conclusion
Manipulating the type of emotion engaged in AM, the present study elucidates the emotion-dependent psychological mechanisms underlying top-down and bottom-up retrieval processes. In sum, top-down retrieval of emotion-laden events incurs a cognitive cost. This retrieval style for sad, angry and fearful experiences involves an additional cost of reduced negative emotion regulation efficiency, while it contributes to avoiding re-experiencing painful feelings. For recollections imbued with happiness, although top-down retrieval demands more cognitive resources, it offers the benefits of a rigorous and flexible thinking style in return. On the other hand, bottom-up retrieval of accurate details for fearful experiences is driven by the alert for potential risks. However, to better understand the nature of the findings here, future research should measure participants’ mood states to rule out potential influences of mood on emotional memory recall (Greenberg & Meiran, Reference Greenberg and Meiran2014). Also, future research should have more control over individual differences in traits, given their potential effects on emotional AM recall. For example, previous studies have shown that people with high levels of alexithymia experience decreased intensity of emotion during emotional memory retrieval (Mantani et al., Reference Mantani, Okamoto, Shirao, Okada and Yamawaki2005) and those with high trait anxiety exhibit increased arousal of negative emotion during narration, particularly when recalling negative events (Herrera et al., Reference Herrera, Montorio and Cabrera2017). In addition, the current study ensured a gender balance among participants without primarily aiming to examine gender effects, since many studies have not found gender differences in emotional AM. However, there is evidence for such differences (see Grysman & Hudson, Reference Grysman and Hudson2013, for a review). Therefore, exploring potential interactions between emotion and gender in future AM research would be valuable.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/langcog.2025.10039.
Data availability statement
Raw data used for analysis in this study are available at OSF, https://osf.io/sguze/overview.
Funding statement
This work was supported by the National Natural Science Foundation of China under grant number 31800961; the China Postdoctoral Science Foundation under grant number 2021 T140094 and the Fundamental Research Funds for the Central Universities in China (ZYGX2024ZSZ001).
Competing interests
The authors report there are no competing interests to declare.
Open access
