The Effects of Semantic Similarity and Emotion on False Memory

Production

Yingcheng He

Queen Mary School of Jiangxi Medical college, Nanchang University, Nanchang, 330031, China

Keywords: False Memory, Semantic Similarity, Emotional Valence.

Abstract: False memory is a long-term memory phenomenon when people report past events differently from the facts.

Scientists have been interested in the generation of false memory and numerous studies demonstrated that a

region in temporal pole called “semantic hub” of the brain is the origin of false memory induced by semantic

similarity among different concepts. According to this neural mechanism, auditorily-presented behavior task

and visually-presented retrieval task were conducted and the first hypothesis that semantic similarity promotes

false memory generation was confirmed. Additionally, this research aimed to compare the difference between

the influence caused by positive words and negative words on false memory. However, the result was not

consistent with the second hypothesis that negatively-valenced concepts induce more false memory than

positively-valenced ones. Therefore, the effect of emotion to memory distortion requires further investigation

while this paper may provide novel insight into the relationship between false memory and emotion.

1 INTRODUCTION

False memory is a long-term memory phenomenon

where one recalls something that did not happen or

recall it from a different way compared to the truth

(Henry, Roediger, Elizabeth 2009). There are

numerous factors influencing the generation of false

memory and one of the most significant is similarity

between new items and past events. In this condition,

people may incorrectly consider new items as old

things as they are very similar. Among various kinds

of similarity, in recent decades, scientists have

explored more about the relationship between

semantic similarity and false memory production.

Chadwick MJ et al. demonstrated that false memories

originate from a region known as the “semantic hub”

of the brain, which is a similarity-based neural code

in the temporal pole. Further, they found that the

degree of semantic similarity between different

events can be reflected by the extent of overlap

among neural representations of relevant concepts in

the neural code and this neural overlap predicts the

probability of false memory generation (Chadwick,

Anjum, Kumaran, Schacter, Spiers, Hassabis, 2016).

Another research indicates that in the recognition task

showing new words to participants, words of high-

similarity with old concepts had remarkably higher

likelihood to be improperly recalled as old compared

with low-similarity words, even after excluding out

the influence of lexical co-occurrence and associative

relatedness (Montefinese, Zannino, Ambrosini

2015). A 2016 study showes the neural mechanism of

false memory generation is both true and false

memories emerge. (Ye, Zhu, Zhuang, Lu, Chen, Xue

2016). In addition to the origin of false memory, the

effects of emotion on false memories are gaining

increased attention. This is usually reflected by the

impact of words’ valence on the strength of false

memories. For instance, it was reported that the

increase of false memories was more pronounced

when subjects were instructed to memorize and then

recognize negatively-valenced words compared to

remembering and recognizing positively-valenced

words (Brainerd, Holliday, Reyna, Yang, Toglia

2010). Furthermore, research revealed that words of

negative emotional valence tend to induce more false

memories due to its automatic neural processing

while positive ones often produce less false memories

because the related neural processing is more

controlled (Knott, Howe, Toffalini, Shah,

Humphreys 2018).

In this study, we conducted auditorily-presented

behavior task and visually-presented retrieval task

online and used words with distinctive emotional

valences and semantic similarity to investigate the

He, Y.

The Effects of Semantic Similarity and Emotion on False Memory Production.

DOI: 10.5220/0011290900003444

In Proceedings of the 2nd Conference on Artiﬁcial Intelligence and Healthcare (CAIH 2021), pages 225-229

ISBN: 978-989-758-594-4

225

false memory production and its related effects

caused by semantic similarity and emotion. We put

forwarded two hypotheses. One is that semantic

similarity can affect long-term memory and produce

false memories and the other is that concepts of

negative emotional valence produce more false

memory than that of positive ones.

2 METHODS AND MATERIALS

2.1 Collect Personal Information

We collected information of 30 subjects aged from 17

to 24, with the basic education level being high

school. Among the 30 subjects, 15 people are females

and 15 people are males. The information includes

gender, education background and age.

2.2 Behavioral Task: Verbal Memory

Task Presented Auditorily

Participants are native Chinese speakers and they

were instructed to memorize a set of 80 Chinese two-

character words (8 separate lists of 10 words each).

The words are broadcast in an audio file in sequence

and the time for each word is 2 seconds (includes the

time for broadcasting and the interval time). The total

time for memorizing is approximately 4 minutes.

These words were selected from the

Cornell/Cortland emotional list (Chang, Brainerd,

Toglia, Schmidt 2010) consisting of 32 DRM

(Deese–Roediger–McDermott) lists which are

divided into four groups of lists: 8 positively-

valenced/ high arousal lists and 8 positively-

valenced/ low arousal lists, 8 negatively-valenced/

high arousal lists and 8 negatively-valenced/ low

arousal lists. The mean arousal among all the high-

arousal lists is not significantly different and the

mean arousal among all the low-arousal lists is also

not. In our 8 lists, 4 of them contain words with

positive valence and 4 with negative valence.

Within each emotion set (positive or negative), there

are 2 lists contain words with high arousal and the

other 2 lists contain words with low arousal. We

provide English translations for presentation

purposes and recorded all the words in an audio file.

2.3 Distraction Task (Non-Mneumonic)

After encoding, participants were instructed to

complete a non-mneumonic task that involves

mathematical computations. The questions are 58-

25=? and 23+19=? Participants were given 30

seconds to write down the answers in the chat box.

Performance was measured using accuracy.

Following the distraction task, a retrieval period was

conducted.

2.4 Retrieval Task

During retrieval, participants were visually presented

with 120 words (40 old, 40 lures, 40 foils), each

presented one at a time for 2 second. The lures are

selected from Cornell/Cortland emotional list. For

each category of words (old, lure, foil), half were

positive and the other half were negative. Participants

were asked to respond in a table format (Table 1) to

determine whether they had seen the word during

encoding. If the answer is yes, they were supposed to

type Chinese character “是” in the table and Chinese

character “否” is corresponding to “No”.

Table 1: Table for subjects to write down answers.

The number Use the drop-down button to fill in "yes" or "no".

…

120

2.5 Data Analysis

We calculated the percent correct of studied words

(the number of “yes” for old/ the number of all the old

words tested (i.e. 40)) and that of critical lures (the

number of “no” for lures/ the number of all the lures

tested (i.e. 40)). Similarly, the percent correct of

negative/ positive studied words and that of negative/

positive lures were calculated. In addition, corrected

false memory is also computed by subtracting foils

judged as old from lures judged as old and corrected

true memory is computed by subtracting foils judged

as old from studied words judged as old. Further, we

calculated corrected false and true memory rates in

positive and negative sets.

(eg: positive corrected false memory rate =

percent correct of foils in positive set minus percent

correct of positive lures)

CAIH 2021 - Conference on Artiﬁcial Intelligence and Healthcare

226

(eg: positive corrected true memory rate = percent

correct of positive studied words plus percent correct

of positive lures minus one.)

3 RESULTS

3.1 The Effect of Semantic Similarity

on False Memory Generation

Compared with that of critical lures, the percent

correct of studied words is significantly higher

(p<0.05) (Figure 1A). Further, we compared the

percent correct of positive studied words and that of

positive lures and the result is consistent (p<0.05)

(Figure 1B). We also compared the percent correct

of negative studied words and that of negative lures.

Figure 1C shows that the percent correct of negative

studied words is significantly higher than that of

negative lures (p<0.05). The results above

demonstrate that semantic similarity promotes false

memory production.

Figure 1: Semantic similarity.

3.2 The Effect of Emotional Valence on

False Memory Production

We compared the corrected true memory rate and

false memory rate between positive words and

negative words. T test indicates that there is no

significant difference between the corrected false

memory rate of positive words and that of negative

words (p>0.05). Likewise, no significant distinction

is shown between corrected true memory rate of

positive words and negative words (p>0.05) but this

value (p=0.057) is close to the significance threshold

(Figure 2).

Figure 2: Positive and negative valences.

The Effects of Semantic Similarity and Emotion on False Memory Production

227

4 DISCUSSIONS

Memory is the ability of brain to encode, store and

retrieve information and is also a result of learning,

perception and attention. A true memory is important

for solving problems and recognizing items.

However, memory is not always reliable. False

memory, a recollection that is fake partially or in

whole but seems real in someone’s mind, often occur

in people’s life, attracting many psychologists and

neuroscientists. There are various studies

demonstrating that the generation of false memory is

related to neural overlap of neural representations of

related concepts. For instance, the similarity between

new items and past experiences increases the cortical

similarity, eventually leading to false memory (Wing,

Geib, Wang, Monge, Davis, Cabeza 2020).

Additionally, the same research revealed that the

false memory for a given concept’s lure tended to

reduce when high similarity in cortical region was

coexist with differentiated hippocampal patterns,

which suggests that hippocampus plays a crucial role

in discriminating similar concepts during encoding

and retrieval and its interaction with cortex

determines whether a similar lure will be incorrectly

recognized as old (Wing, Geib, Wang, Monge, Davis,

Cabeza 2020). Coane JH et al. found that false

identification of critical lures was significantly

greater after studying the word lists that have similar

characteristics with the critical lures, which indicates

that similarity at a taxonomic or categorical level

induces false memory (Coane, McBride, Termonen,

Cutting 2016). Our results show that semantic

similarity between different concepts increases false

memory production, which is demonstrated by the

higher percent correct of studied words than that of

critical lures. Specifically, the percent correct of

positive studied words is also higher than that of

positive critical lures and the same result is for

negative words. Therefore, our results are consistent

with previous studies.

However, we did not find a significant difference

between the influence of positively-valenced words

and that of negatively-valenced words in false

memory production. Our results indicate that there is

no significant distinction between positive and

negative corrected false memory rates. Besides, no

significant difference was identified between positive

and negative correct true memory rates, though the p

value is close to the significance threshold. We think

that this result may be attributed to the distinctive

effects of emotion on false memory depending on

whether the emotion is in the content of items to be

memorized or in the participants’ moods reflecting

the context during encoding. A previous study found

that negatively-valenced concepts induce memory

distortion and cause false memory while negatively-

valenced moods keep memory from being distorted.

Moreover, enduring negative moods promotes false

memory production (Bookbinder, Brainerd, 2016).

Therefore, in our experiment, presenting words with

negative emotion may foment negative moods in

subjects and this effect may counteract the influence

of negative concepts in increasing false memory. As

a result, no significant difference can be found

between false or true memory rates of positive and

negative words.

There are several limitations in our research,

among which the most important is latency caused by

network and personal affairs as our experiment was

conducted online. For instance, after encoding,

subjects are supposed to take the distraction test

immediately but due to the network he or she may not

be able to see the math questions until several

minutes, which greatly influences the memory

encoding. Also, retrieval task may be interrupted by

trifles happening around the subject’s environment

(eg. someone knocks at the door and the subject needs

to stop watching the recognition video to open the

door). Another disadvantage is that semantic

similarity across lists may disturb our research

exploring the effect of semantic similarity of each

single list on false memory generation. Additionally,

how to properly control variables when investigating

the influence of emotion in false memory production

remains elusive.

Based on the limitations above, we came up with

several ideas to be realized in our future research.

First of all, conducting the experiment offline and in

this way, subjects tend to be more controlled, which

can decrease disturbances and enhance seriousness.

Second, we will re-select the words for encoding and

retrieval and reduce the semantic similarity across

lists to as small as possible. Third, as previous

research also suggested the effect of stimulus

modalities on false memory, we may alter the

encoding and retrieval modalities to explore the

difference between the influence of auditory and

visual modalities in memory distortion.

5 CONCLUSIONS

In summary, we conducted auditorily-presented

encoding task and visually-presented retrieval task

using words with different emotions and

demonstrated that semantic similarity of similar

concepts promotes false memory generation.

CAIH 2021 - Conference on Artiﬁcial Intelligence and Healthcare

228

However, the enhancement effect of negatively

emotional words on false memory production was not

verified but this paper offers new ideas about the

correlation between emotion and false memory

generation and further investigation is required.

REFERENCES

Bookbinder SH, Brainerd CJ. Emotion and false memory:

The context-content paradox. (2016) Psychol Bull.

142(12):1315-1351.

Brainerd CJ, Holliday RE, Reyna VF, Yang Y, Toglia MP.

(2010) Developmental reversals in false memory:

Effects of emotional valence and arousal. J Exp Child

Psychol. 107(2):137-54.

Chadwick MJ, Anjum RS, Kumaran D, Schacter DL, Spiers

HJ, Hassabis D. (2016) Semantic representations in the

temporal pole predict false memories. Proc Natl Acad

Sci U S A. 113(36):10180-5.

Chang M, Brainerd CJ, Toglia MP, Schmidt SR. (2010)

Norms for emotion-false memory lists. Behav Res

Methods. 53(1):96-112.

Coane JH, McBride DM, Termonen ML, Cutting JC.

(2016) Categorical and associative relations increase

false memory relative to purely associative relations.

Mem Cognit. 44(1):37-49.

Henry L. Roediger III, Elizabeth J. (2009). False memory.

http://www.scholarpedia.org/article/False_memory

Knott LM, Howe ML, Toffalini E, Shah D, Humphreys L.

(2018) The role of attention in immediate emotional

false memory enhancement. Emotion. 18(8):1063-

1077.

Montefinese M, Zannino GD, Ambrosini E. (2015)

Semantic similarity between old and new items

produces false alarms in recognition memory. Psychol

Res. 79(5):785-94.

Wing EA, Geib BR, Wang WC, Monge Z, Davis SW,

Cabeza R. (2020) Cortical overlap and cortical-

hippocampal interactions predict subsequent true and

false Memory. J Neurosci. 40(9):1920-1930.

Ye Z, Zhu B, Zhuang L, Lu Z, Chen C, Xue G. (2016)

Neural global pattern similarity underlies true and false

memories. J Neurosci. 36(25):6792-802.

The Effects of Semantic Similarity and Emotion on False Memory Production

229