Neutral Word Lists Research Articles

A Bayesian Approach to Uncertainty in Word Embedding Bias Estimation

Abstract Multiple measures, such as WEAT or MAC, attempt to quantify the magnitude of bias present in word embeddings in terms of a single-number metric. However, such metrics and the related statistical significance calculations rely on treating pre-averaged data as individual data points and utilizing bootstrapping techniques with low sample sizes. We show that similar results can be easily obtained using such methods even if the data are generated by a null model lacking the intended bias. Consequently, we argue that this approach generates false confidence. To address this issue, we propose a Bayesian alternative: hierarchical Bayesian modeling, which enables a more uncertainty-sensitive inspection of bias in word embeddings at different levels of granularity. To showcase our method, we apply it to Religion, Gender, and Race word lists from the original research, together with our control neutral word lists. We deploy the method using Google, GloVe, and Reddit embeddings. Further, we utilize our approach to evaluate a debiasing technique applied to the Reddit word embedding. Our findings reveal a more complex landscape than suggested by the proponents of single-number metrics. The datasets and source code for the paper are publicly available.1

The effect of negative valence on false memory formation in the Deese-Roediger-McDermott paradigm: A preregistered meta-analysis and preregistered replication.

Participants in the Deese-Roediger-McDermott (DRM) paradigm learn lists of words (e.g., bed, tired) associated with a nonpresented lure (i.e., sleep). In subsequent memory tests, individuals tend to report the nonlearned lures, that is, exhibiting false memories. Priorly, the DRM task has been criticized for not capturing the aversive nature of (clinically and forensically relevant) real-life memories. To obtain a robust estimate of the influence of negative versus neutral word lists on the DRM effect, we conducted both a preregistered meta-analysis (krecall = 49, nrecall = 2,209, krecognition = 75, nrecognition = 3,008, kresponsebias = 31, nresponsebias = 1,128) and replication (nfinal = 278) predicting increased false memories for negative valence in recall and recognition. For recall, we found significant frequentist evidence in the meta-analysis for a reversed valence effect (d = -0.18, i.e., reduced false memories for negative content vs. neutral), whereas the replication displayed null results (d = 0.03). For recognition, both the meta-analysis (d = 0.23) and replication (d = 0.35) showed that negative valence (vs. neutral) increased false memories. However, this effect may be confounded by shifts in response tendencies as controlling for response bias nullified the valence effect in our meta-analysis (dmeta = 0.05), and we found evidence for differential response bias in our replication (dreplica = 0.39). Hence, the effect of valence on false memory reports in the DRM may not represent a systematic difference in emotional information but instead depend on how memory is tested, and be partly attributable to differential response tendencies. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

The impact of fabrication on recognition memory: An experimental study

We examined whether fabrication affects memory using a new paradigm combining the Deese/Roediger-McDermott paradigm (DRM) and the Amsterdam Short Term Memory task (ASTM). Participants were assigned to either a forced fabrication or honest condition, and encoded emotionally-negative and neutral wordlists by reading words out loud. The wordlists contained words that were associatively related to each other (e.g., symphony, sound, piano, radio, sing, orchestra), and converged to a non-presented critical word (e.g., music). Next, all participants had to indicate two words that were previously read out loud. However, participants in the forced fabrication condition had to fabricate that they read two additional words out loud, which were associated to the same DRM list. Participants in the forced fabrication condition formed false memories for their self-generated fabrications. Although fabrication did not increase spontaneous false memory rates for non-presented critical words, our results did highlight the perils of forcing people to fabricate.

False (or biased) memory: Emotion and working memory capacity effects in the DRM paradigm.

This study aimed to investigate the role of emotion and working memory capacity (WMC) on false memory by measuring memory sensitivity independently of response bias. We used the Deese-Roediger-McDermott (DRM) paradigm in which arousal levels were kept constant across positive, negative, and neutral word lists associated with unstudied critical lures. Participants' WMC was measured by the Operation Span Task. Although negative critical lures generated significantly more false recognition (i.e., false-alarm rates) compared to positive or neutral ones, Receiver operating characteristic (ROC) analysis based on signal detection theory showed that this effect could be ascribed to shifts in bias rather than actual memory sensitivity. Data revealed that the DRM effect is a robust illusion influenced by neither emotion nor WMC in terms of memory sensitivity. However, negative words led to a prominent increase in liberal bias to say "old" for both critical and noncritical lures. Furthermore, reaction time (RT) data suggested that mentally activated but actually unstudied critical lures were monitored as old words and that participants were faster to accept negative critical lures than positive or neutral ones, suggesting that the DRM illusion was clearly reflected on the RT data as well. These results were discussed emphasizing the role of negative emotion on response bias in recognition memory.

Maltreatment increases spontaneous false memories but decreases suggestion-induced false memories in children.

We examined the creation of spontaneous and suggestion‐induced false memories in maltreated and non‐maltreated children. Maltreated and non‐maltreated children were involved in a Deese–Roediger–McDermott false memory paradigm where they studied and remembered negative and neutral word lists. Suggestion‐induced false memories were created using a misinformation procedure during which both maltreated and non‐maltreated children viewed a negative video (i.e., bank robbery) and later received suggestive misinformation concerning the event. Our results showed that maltreated children had higher levels of spontaneous negative false memories but lower levels of suggestion‐induced false memories as compared to non‐maltreated children. Collectively, our study demonstrates that maltreatment both increases and decreases susceptibility to memory illusions depending on the type of false memory being induced. Statement of contribution What is already known on this subject? Trauma affects memory.It is unclear how trauma affects false memory. What does this study add? This study focuses on two types of false memories.

Do emotional stimuli enhance or impede recall relative to neutral stimuli? An investigation of two “false memory” tasks

ABSTRACTMany eyewitness memory situations involve negative and distressing events; however, many studies investigating “false memory” phenomena use neutral stimuli only. The aim of the present study was to determine how both the Deese–Roediger–McDermott (DRM) procedure and the Misinformation Effect Paradigm tasks were related to each other using distressing and neutral stimuli. Participants completed the DRM (with negative and neutral word lists) and viewed a distressing or neutral film. Misinformation for the film was introduced and memory was assessed. Film accuracy and misinformation susceptibility were found to be greater for those who viewed the distressing film relative to the neutral film. Accuracy responses on both tasks were related, however, susceptibility to the DRM illusion and Misinformation Effect were not. The misinformation findings support the Paradoxical Negative Emotion (PNE) hypothesis that negative stimuli will lead to remembering more accurate details but also greater likelihood of memory distortion. However, the PNE hypothesis was not supported for the DRM results. The findings also suggest that the DRM and Misinformation tasks are not equivalent and may have differences in underlying mechanisms. Future research should focus on more ecologically valid methods of assessing false memory.

Maltreated and non-maltreated children’s true and false memories of neutral and emotional word lists in the Deese/Roediger–McDermott task

Maltreated (n=26) and non-maltreated (n=31) 7- to 12-year-old children were tested on the Deese/Roediger–McDermott (DRM) false memory task using emotional and neutral word lists. True recall was significantly better for non-maltreated than maltreated children regardless of list valence. The proportion of false recall for neutral lists was comparable regardless of maltreatment status. However, maltreated children showed a significantly higher false recall rate for the emotional lists than non-maltreated children. Together, these results provide new evidence that maltreated children could be more prone to false memory illusions for negatively valenced information than their non-maltreated counterparts.

On the roles of distinctiveness and semantic expectancies in episodic encoding of emotional words.

We examined the factors that contribute to enhanced recall for emotionally arousing words by analyzing behavioral performance, the P300 as an index of distinctiveness, and the N400 as an index of semantic expectancy violation in a modified Von Restorff paradigm. While their EEG was recorded, participants studied three list types (1) neutral words including one emotionally arousing isolate (either positive or negative), (2) arousing, negative words including one neutral isolate, or (3) arousing, positive words including one neutral isolate. Immediately after each list, free recall was tested. Negative, but not positive, words exhibited enhanced recall when presented as isolates in lists of neutral words and elicited a larger P300 for subsequently recalled than not-recalled words. This suggests that arousing, negative words stand out and that their distinctiveness contributes to their superior recall. Positive valence had an enhancing effect on recall only when the list contained mostly other positive words. Neutral isolates placed in either positive or negative lists elicited an N400, suggesting that semantic expectations developed in emotional word lists regardless of valence. However, semantic relatedness appeared to more strongly contribute to recall for positive than negative words. Our results suggest that distinctiveness and semantic relatedness contribute to episodic encoding of arousing words, but the impact of each factor depends on both a word's valence and its context.

Positive Arousal Enhances the Consolidation of Item Memory

Emotional arousal induced after learning has been shown to modulate memory consolidation. However, it is unclear whether the effect of postlearning arousal can extend to different aspects of memory. This study examined the effect of postlearning positive arousal on both item memory and source memory. Participants learned a list of neutral words and took an immediate memory test. Then they watched a positive or a neutral videoclip and took delayed memory tests after either 25 minutes or 1 week had elapsed after the learning phase. In both delay conditions, positive arousal enhanced consolidation of item memory as measured by overall recognition. Furthermore, positive arousal enhanced consolidation of familiarity but not recollection. However, positive arousal appeared to have no effect on consolidation of source memory. These findings have implications for building theoretical models of the effect of emotional arousal on consolidation of episodic memory and for applying postlearning emotional arousal as a technique of memory intervention.

Age Alters the Effects of Emotional Valence on False Memory: Using the Simplified Conjoint Recognition Paradigm

Previous research has revealed robust differences between young and older adults' accurate emotional memory. Typically, negative emotion prompts memory performance in young adults, while positive emotion benefits memory performance in older adults. Relatively, much less efforts have been devoted to investigating how emotion divergently influences false memory in older adults compared to young adults. The main purpose of the present study is to explore, first, how age alters the effects of emotional valence on false memory; and second, what are the cognitive mechanisms for the interaction effect of age and emotional valence on false memory. The Fuzzy Trace Theory(FTT) uses verbatim memory trace and gist memory trace to interpret individuals' false memory, especially false memory under the framework of Deese-Roediger-Mc Dermott(DRM) paradigm. According to recent research, we proposed some flaws of traditional paradigms(such as the Remember-Know paradigm) in the area of false memory; and we also argued that it would be critical to include response bias in addition to verbatim and gist memory trace to investigate false memory. To fulfill this need, the simplified conjoint recognition(SCR) paradigm, combined with a statistical method of multinomial processing tree model, was used in the present study to investigate false memory and its associated cognitive mechanisms. A sample of 34 young adults(aged 23 ± 2 years) and 28 older adults(aged 68 ± 5 years) completed the SCR task. In the task, the Cornell/Cortland emotion word lists, along with neutral word lists adopted from previous research, were implemented as experimental materials. The results displayed a significant interaction effect of emotional valence and age on false memory(i.e. false alarm). Specifically, positive emotion decreased older adults' false memory, while negative emotion decreased young adults' false memory.The method of multinomial processing tree model was further employed to model, parameterize and inference the cognitive mechanisms related to the effect of emotional valence on false memory within young and older group. It was found that negative emotion contributed to a lower level of false memory in young adults by lowering their response bias toward negative words. As to older adults, positive emotion boosted retrieval of verbatim memory trace and hampered retrieval of gist memory trace, then led to less false memory for positive words. Our study uncovered interesting age-related differences in emotional false memory; it also confirmed the necessity of a simultaneous consideration of verbatim memory trace, gist memory trace and response bias when investigating false memory.

Brief report: Violent false memories and engagement in aggressive and delinquent behavior: An investigation in adolescents

The present study investigates the relationship between violent false memories and delinquent and aggressive behavior in a sample of adolescents. Two hundred eleven participants completed measures of aggressive and delinquent behavior and performed a modified version of the Deese-Roediger-McDermott (DRM) paradigm, a false memory task for lists of associated words. Participants were presented with a list of ambiguously violent words and three lists of neutral words. For each list a free recall task was performed. Violent false memories were significantly associated with delinquent behaviors in both genders, whereas a significant correlation with aggressive behaviors was found only in males. A multilevel multiple regression showed that the prediction of delinquent behaviors was improved by entering violent false memories into the model as a further predictor, whereas no effect was found for aggressive behaviors. These findings indicate a significant association of violent false memories with delinquent behavior in adolescents.

Emotional true and false memories in children with callous-unemotional traits

Several studies have found that children with callous-unemotional (CU) traits have a deficit in processing emotionally negative material. The present study examined whether this deficit also affects emotional memory. Twenty-two children with low CU traits and 24 children with high CU traits between 8 and 12 years of age were selected from a community sample and presented with neutral and negative emotional words, using the Deese–Roediger–McDermott paradigm. On true recall, there was no difference between the groups. Both groups had higher true recall rates for the neutral word lists than for the negative lists. However, on false recall, although there were no group differences for neutral word lists, the high CU group recalled significantly fewer critical lures on the negative word lists than the low CU group. Furthermore, the high CU group had significantly less false recall on the negative word lists compared to the neutral word lists, while the low CU group showed no difference in false recall between the word lists. These results indicate that children with high CU traits have no deficiencies in true memory performance, yet are less susceptible to developing false memories concerning emotionally negative material.

Recalling unpresented hostile words: False memories predictors of traditional and cyberbullying

This study investigated the relationship between hostile false memories (violent and verbal/aggressive) and engagement in traditional and cyberbullying, controlling for their co-occurrence. Two hundred eleven adolescents completed measures of traditional and cyberbullying and performed a modified version of the “DRM paradigm”, a false memory task for lists of associated words. Five lists were used: one of ambiguously violent words, oneof insults and three lists of neutral words used as controls. For each list a free recall task was performed. A path analysis showed that both violent false memories for ambiguously hostile words and verbal/aggressive false memories for insults were positively associated with cyberbullying and, in males, also with traditional bullying. These data indicate a contribution of hostile memory distortions to bullying behaviours in adolescents. Findings are discussed according to the general aggression model.

A gender difference in the false recall of negative words: Women DRM more than men

Gender differences in susceptibility to associative memory illusions in the Deese/Roediger–McDermott paradigm were investigated using negative and neutral word lists. Women (n=50) and men (n=50) studied 20 lists of 12 words that were associates of a non-presented critical lure. Ten lists were associates of negatively valenced lures (e.g., cry, evil) and ten were associates of neutral lures (e.g., chair, slow). When asked to recall the words after each list, women falsely recalled more negative lures than men, but there was no gender difference in the false recall of neutral lures. These findings suggest that women reflect on associations within negative lists to a greater degree than men and are thereby more likely to generate the negative critical lures.

Emotional working memory capacity in posttraumatic stress disorder (PTSD)

Participants with a lifetime history of posttraumatic stress disorder (PTSD) and trauma-exposed controls with no PTSD history completed an emotional working memory capacity (eWMC) task. The task required them to remember lists of neutral words over short intervals while simultaneously processing sentences describing dysfunctional trauma-related thoughts (relative to neutral control sentences). The task was designed to operationalise an everyday cognitive challenge for those with mental health problems such as PTSD; namely, the ability to carry out simple, routine tasks with emotionally benign material, while at the same time tackling emotional laden intrusive thoughts and feelings. eWMC performance, indexed as the ability to remember the word lists in the context of trauma sentences, relative to neutral sentences, was poorer overall in the PTSD group compared with controls, suggestive of a particular difficulty employing working memory in emotion-related contexts in those with a history of PTSD. The possible implications for developing affective working memory training as an adjunctive treatment for PTSD are explored.

Attentional bias in excessive massively multiplayer online role-playing gamers using a modified Stroop task

There is considerable dispute regarding the nature of excessive or problematic Internet-related behaviour and whether it constitutes a clinical addiction. Classification of excessive gaming is hindered by a lack of experimental research investigating behavioural responses from gamers and comparing these patterns to those found in established addictions. We investigated whether an attentional bias for gaming-related words existed for addicted Massively Multiplayer Online Role-Playing Gamers (MMORPGers) identified using the Addiction–Engagement Questionnaire. Forty frequent MMORPGers (15 female) and 19 non-MMORPGers (eight female) completed a computerised modified Stroop task comprised of game-related, negative and neutral word lists, Addiction–Engagement Questionnaire, Depression, Anxiety and Stress Scale 21, gaming-related variables. The results indicated that addicted MMORPGers had significantly longer reaction times to negative and MMORPG words compared to neutral words, whereas highly engaged and non-MMORPG participants showed no such bias. The presence of an attentional bias in addicted MMORPGers is comparable with research investigating this behavioural response in established addictions.

Psychosocial stress enhances non-drug-related positive memory retrieval in male abstinent heroin addicts

Stress exposure in addicted individuals is known to provoke drug craving, presumably through a memory-like process, but less is known about the effects of stress on non-drug-related affective memory retrieval per se in such individuals, which is likely to provide important insights into therapy for relapse. In present study, we explored the effect of stress on retrieval of neutral and emotionally valenced (positive and negative) words in abstinent heroin addicts. In present study, 28 male inpatient abstinent heroin addicts and 20 sex-, age-, education- and economic status-matched healthy control participants were assessed for 24 h delayed recall of valenced and neutral word lists on two occasions 4 weeks apart—once in a nonstress control condition, once after exposure to the Trier Social Stress Test in a counterbalanced design. In addition, attention, working memory, blood pressure, heart rate and salivary cortisol were assessed. We found acute stress at the time of word list recall enhanced retrieval of positively valenced words, but no effect on negative and neutral word retrieval in abstinent heroin addicts was observed. No changes were detected for attention and working memory. The stressor induced a significant increase in salivary free cortisol, blood pressure and heart rate. Stress can enhance non-drug-related positive memory in abstinent heroin addicts. Our findings will provide richer information in understanding dysregulation of their emotional memory processing under stress and hopefully provide insight into designing improved treatments for drug addiction.

False Memories of Emotional and Neutral Words

This study used the Deese-Roediger-McDermott paradigm [34] to investigate the direction and the extent to which emotional valence in semantic word lists influences the formation of false memories (FM). The experimental paradigm consisted of 1) a study phase (learning of neutral and negative lists of words semantically associated to a non-presented critical lure (CL), 2) a free recall phase, and 3) a recognition phase. Participants had to indicate whether the displayed item was "new" (new item or non-studied CL) or “old” (studied list item). CL associated with negative word lists elicited significantly more FM than CL associated with neutral word lists. This finding is in contrast to previous work showing that emotional words elicit fewer FM than neutral words. The results of our study also suggest that valence is capable of influencing emotional memory in terms of encoding and retrieval processes.

False memory in aging: Effects of emotional valence on word recognition accuracy.

Memory is susceptible to distortions. Valence and increasing age are variables known to affect memory accuracy and may increase false alarm production. Interaction between these variables and their impact on false memory was investigated in 36 young (18-28 years) and 36 older (61-83 years) healthy adults. At study, participants viewed lists of neutral words orthographically related to negative, neutral, or positive critical lures (not presented). Memory for these words was subsequently tested with a remember-know procedure. At test, items included the words seen at study and their associated critical lures, as well as sets of orthographically related neutral words not seen at study and their associated unstudied lures. Positive valence was shown to have two opposite effects on older adults' discrimination of the lures: It improved correct rejection of unstudied lures but increased false memory for critical lures (i.e., lures associated with words studied previously). Thus, increased salience triggered by positive valence may disrupt memory accuracy in older adults when discriminating among similar events. These findings likely reflect a source memory deficit due to decreased efficiency in cognitive control processes with aging.

Recall performance, plasma cortisol and plasma norepinephrine in normal human subjects

This study investigated hypothalamic–pituitary–adrenal (HPA)-axis and sympathetic nervous system (SNS) correlates of recall performance in normal human subjects. Twenty-two normal human subjects were given one memory task: short-term recall of unrelated non-organizable lists of neutral words, in immediate recall conditions. Two types of memory were individualized: measures reflecting effortful processing and measures reflecting automatic processing, which were related to 3 daytime plasma cortisol (CORT) and plasma NE values, and assessed after venipuncture. It was hypothesized that plasma CORT is positively related and plasma norepinephrine (NE) is negatively related to effortful processing. Pearson correlation was computed and regression analysis was performed. Positive correlation appeared between plasma CORT values and negative correlation appeared between plasma NE values and measures reflecting effortful processing. However, stepwise multiple regression analysis showed that only morning plasma CORT values are functionally positively and afternoon plasma NE values are functionally negatively related to effortful processing. This suggests that morning HPA-axis activities enhance and afternoon SNS activities inhibit effortful processing.