Electrophysiological Evidence of Human Memory …...associate recall were qualitatively different...

Electrophysiological Evidence of Human Memory Formation

by

Alice S. N. Kim

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy

Department of Psychology University of Toronto

© Copyright by Alice S. N. Kim 2013

ii

Electrophysiological Evidence of Human Memory Formation

Alice S. N. Kim

Doctor of Philosophy

Department of Psychology University of Toronto

2013

Abstract

This dissertation involves a series of experiments that investigated the neurocognitive processes

engaged in episodic association formation. Electroencephalograms were recorded from healthy,

young adults while they memorized visually presented words or word pairs. Event-related

potentials (ERPs) recorded during the study phase were examined based on participants’

subsequent memory performance. Study 1 demonstrated that association formation, but not item

encoding, was reflected by ERP positivity that occurred late during the ERP trial over the

parietal scalp region. Study 2 showed that encoding processes associated with subsequent paired-

associate recall were qualitatively different between two retention intervals (RIs). Sustained ERP

positivities were observed late during the ERP trial over the frontal and fronto-central scalp

regions for the long and short RIs, respectively. The fronto-central positivity was investigated

further in Study 3 using the distinction between primary memory (PM) and secondary memory

(SM). Items retrieved from SM have been absent from consciousness, whereas items in PM have

not. The observed fronto-central positivity was shown to reflect binding in a general sense (inter-

item and intra-item associations). Study 4 demonstrated that associative binding strength is

reflected by ERPs recorded late during the ERP trial over the right frontal and left centro-parietal

scalp regions. Study 4 also enabled an investigation on the “negative recency effect” – items

presented at the end of a series are remembered with the lowest probability on a delayed memory

iii

test, even though they are remembered best on an immediate memory test. The results suggest

that at least some of the factors that contribute to this effect occur during retrieval. Collectively,

the results demonstrate that the ERP methodology, combined with behavioural data, can be used

to provide an objective measure of depth of processing used to encode information, and in doing

so helps address one of the major criticisms of the levels-of-processing framework.

iv

Acknowledgments Looking back on my graduate career, I feel that I have learned many things. One of the most

important things I have come to appreciate is how blessed I have been in my life. From the very

beginning, as I know it, my parents devoted their lives to my siblings and me. They gave us

everything and they did everything so that we would be successful in our lives, and so that we

could each live the life that we wanted for ourselves. First and foremost, I thank my parents for

their ongoing support and for helping me get to this moment in my life. 제 연구와 그 모든

결과를 어머니, 아버지 두 분께 바칩니다.

Since I was a young elementary school student, I always wanted to do something great with my

life that would have a big and meaningful impact on the world. Without knowing anything about

the world, I had plans to have an important role in helping to save it. Fortunately for myself, I

have always enjoyed being a student (except in grade 9) and being involved in various extra-

curricular activities, and through my classes and activities I started to acquire a little bit of

knowledge about the world I was living in. Moreover, each chapter of my life seemed to get

better and better: elementary school was great, high school was even better, and my

undergraduate career at the University of Toronto was thrilling. The program that I was

registered in, Human Behavioural Biology, allowed me to take a wide breadth of courses. So I

did and I loved it. However, back then, I would never have thought that I could pursue so many

of my interests by following one path. Then, out of luck, I attended a guest lecture given by Dr.

Endel Tulving for a course that I was not enrolled in, and this experience not only changed my

opinion, it also changed the direction of my life.

Needless to say, Dr. Tulving is a very special person. For many years now, he has been one of

the most important people in my life. First he shared with me his passion for memory research,

and then he inspired me to devote my own career to studying memory. He taught me most of

everything I know about memory and the brain, but more importantly, he taught me how to think

like a scientist. Most importantly, he made me believe that I could do well in research, and that I,

with a lot of work and dedication, could one day become a great memory scientist. Per aspera ad

astra. His belief in me is what, for the most part, helps me to believe in myself as a scientist. I

v

thank Dr. Tulving for everything. Words cannot express my gratitude and how extremely

fortunate I feel to have him as a mentor.

As my graduate career progressed, I had the great fortune to work closely with another great

mentor, Dr. Claude Alain. In this case, too, I was in the right place at the right time, and luckily

for me, Dr. Alain thought I was the right PhD student to work with. Over the years, he has

become a very important figure in my life and he has had a tremendous impact on my training

and the way that I think about research. In addition to being the source for much of what I know

about electroencephalography, he continues to teach me how to conceptualize and think

differently about data and research, how to solve problems, and how to put things into

perspective. He has inspired much of the energy and passion I have for my research, and as my

advisor he has walked with me patiently through my thesis-writing journey. Without his

instruction and support, I could not be where I am today – for this, I could not be more thankful

to Dr. Alain.

As members of my Thesis Committee, Dr. Jennifer Ryan and Dr. Malcolm Binns, have both

gone above and beyond their designated roles. I thank them both for their ongoing instruction,

their contributions to my training and research, as well as their continued support. Their

mentorship throughout the years has helped me to strengthen my research skills, develop my own

research vision and has meaningfully improved my graduate experience. I could not be more

grateful for everything that Dr. Ryan and Dr. Binns have both done for me.

I am grateful to Drs. Ken Paller, Fergus Craik and Asaf Gilboa, my examiners, who provided

thoughtful and constructive feedback on my dissertation, which served to improve it, as well as

the corresponding papers that will be sent out for publication.

To date, my graduate career has been the most rewarding, but also the most challenging, period

of my life. Per aspera ad astra. I thank Fardid, my husband, for his continued support,

encouragement, advice, understanding and hilarity. He has been a source of strength and comfort

for me throughout this entire journey. I thank each of my family members (Susan, Grace,

Michael, Fahima, Firouz, Farbod), friends (particularly Toni, Jenny, Jeff, Claire, Rosanne and

Daniela) and the encouraging community at the Rotman Research Institute of Baycrest for their

ongoing support and for bringing balance to my life. Without all of you, colours would not be as

vibrant, food would not be as tasty, and science would not be as exciting or fulfilling.

vi

Table of Contents

Contents

Acknowledgments.......................................................................................................................... iv

Table of Contents........................................................................................................................... vi

List of Tables .................................................................................................................................. x

List of Figures ................................................................................................................................ xi

List of Acronyms ......................................................................................................................... xiii

Chapter 1 Introduction .................................................................................................................... 1

1 Methods of Investigation............................................................................................................ 2

2 Summary of the Main Behavioural Findings ............................................................................. 2

3 Event-related Potentials Studies on Encoding ........................................................................... 6

4 Outline and goals...................................................................................................................... 14

5 General Methods ...................................................................................................................... 15

5.1 General Hypotheses .......................................................................................................... 15

5.2 Participants........................................................................................................................ 16

5.3 EEG Recording and Analysis Procedures......................................................................... 16

Chapter 2 Differentiating ERP correlates of item encoding and episodic association formation (Study 1)................................................................................................................................... 18

6 Methods.................................................................................................................................... 19

6.1 Participants........................................................................................................................ 19

6.2 Material ............................................................................................................................. 20

6.3 Design ............................................................................................................................... 20

6.4 Procedure .......................................................................................................................... 20

6.5 Behavioural data analysis ................................................................................................. 21

6.6 Electrophysiological data analyses ................................................................................... 22

vii

6.7 Principal component analysis ........................................................................................... 22

7 Results ...................................................................................................................................... 23

7.1 Behavioural data ............................................................................................................... 23

7.2 Principal component analysis ........................................................................................... 23

8 Discussion ................................................................................................................................ 26

Chapter 3 An investigation of the ERP correlates of episodic association formation (Study 2) .. 28

9 Methods.................................................................................................................................... 31

9.1 Participants........................................................................................................................ 31

9.2 Design ............................................................................................................................... 31

9.3 Materials ........................................................................................................................... 31

9.4 Procedure .......................................................................................................................... 31


9.6 Electrophysiological data analysis.................................................................................... 33

9.7 Mean amplitude analysis................................................................................................... 34

10 Results ...................................................................................................................................... 36


10.2 ERP data............................................................................................................................ 36

11 Discussion ................................................................................................................................ 38

Chapter 4 An Investigation of the Neuroelectric Correlates of Primary Memory and Secondary Memory (Study 3) .................................................................................................. 42

12 Methods.................................................................................................................................... 45

12.1 Participants........................................................................................................................ 45

12.2 Materials ........................................................................................................................... 45

12.3 Design ............................................................................................................................... 45

12.4 Procedure .......................................................................................................................... 45


viii


12.7 Mean amplitude analysis................................................................................................... 47

13 Results ...................................................................................................................................... 47


13.2 Mean amplitude data......................................................................................................... 47

14 Discussion ................................................................................................................................ 48

Chapter 5 ERP Correlates of Associative Binding Strength and the Negative Recency Effect (Study 4)................................................................................................................................... 51

15 Methods.................................................................................................................................... 54

15.1 Participants........................................................................................................................ 54

15.2 Materials ........................................................................................................................... 54

15.3 Design ............................................................................................................................... 54

15.4 Procedure .......................................................................................................................... 55



15.7 Partial least squares analyses ............................................................................................ 58

15.8 Memory performance and ERP correlation analyses........................................................ 61

16 Results ...................................................................................................................................... 61


16.2 Mean amplitude results ..................................................................................................... 62

16.3 Partial least square results ................................................................................................. 62

16.4 Memory performance and ERP correlation results........................................................... 64

17 Discussion ................................................................................................................................ 64

Chapter 6 General Discussion....................................................................................................... 67

References..................................................................................................................................... 75

Appendices.................................................................................................................................... 89

x

List of Tables Table 1. Results of the repeated measures ANOVAs for the first principal component (PC1),

second principal component (PC2), and third principal component (PC3).

xi

List of Figures Figure 1. Summary of the experimental procedure used in Study 1.

Figure 2. The time course for the event-related potential trial to the paired-associates in Study 1

and Study 2.

Figure 3. Group level event-related potential (ERP) data recorded during encoding. (A) Grand

average ERP waveforms as a function of word order and subsequent memory performance; (B)

Word2-Word1 difference waves as a function of subsequent memory performance.

Figure 4. Results from the principal component analysis. (A) Topographical distributions of

electrode loadings from the rotated component matrix for the first principal component (PC1),

principal component 2 (PC2), and principal component 3 (PC3). The top of the figure

corresponds to the front of the head. (B) Plots of factor scores for PC1, PC2 and PC3. (C) Grand

average event-related potential data at representative electrode P2 for PC1, electrode FCz for

PC2, and electrode F8 for PC3.

Figure 5. Summary of the experimental procedure used in Study 2.

Figure 6. Group level event-related potential data recorded during encoding.

Figure 7. Event-related potential data for the ShortDelay condition. (A) The ShortDelay (R-N)

waveform recorded at representative electrode FC2. (B) Topographical distribution of the

ShortDelay(R-N) waveform between 2200 and 2800 ms.

Figure 8. Event-related potential data for the LongDelay condition. (A) The LongDelay(R-N)

waveform recorded at representative electrode F2. (B) Topographical distribution of the

LongDelay(R-N) waveform between 2200 and 2800 ms.

Figure 9. Mean amplitudes of the Dm effects observed for the ShortDelay and LongDelay

conditions during the time window of the LW as a function of electrode


Figure 11. The time course for the evoked potential trial to the study word presentation.

xii


Figure 13. Group level event-related potential data recorded during encoding from fronto-central

electrodes (FCz, FC2, Cz, C2).

Figure 14. Topographical distributions. (A) Topographical distribution of the primary memory

(PM) - not-recalled (NR) difference waveform between 1000 to 1600 ms after the presentation

onset of a word. (B) Topographical distribution of the secondary memory (SM) - NR difference

waveform between 1000 to 1600 ms after the presentation onset of a word.


Figure 16. The time course for the event-related potential trial to the paired-associates in Study 4.

Figure 17. Percentage of pairs recalled after both the Immediate and Delayed memory tests as a

function of serial position.


Figure 19. Results of the mean-centered partial least squares analysis. (A) The design scores and

corresponding scalp scores of the first latent variable, illustrating the effect that it reflected; (B)

Grand average event-related potential trial waveforms recorded at representative electrodes F6

and CP5 for the right fronto-central and the left centro-parietal scalp regions, respectively; (C)

The topographical distribution of the electrode saliences, corresponding to the 2200-2800 ms

time window highlighted in panel (B), indicating where on the scalp the effect picked up by the

latent variable was strongest.

xiii

List of Acronyms Dm: ERP differences at encoding correlated with subsequent memory performance

EEG: electroencephalography

ERP: event-related potential

ImmN_DelN: pairs that were not recalled after the short nor long retention intervals (Study 3)

ImmR_DelN: pairs that were recalled after the short retention interval but not after the long

retention interval (Study 3)

ImmR_DelR: pairs that were recalled after both the short and long retention intervals (Study 3)

ITRI: intra-trial retention interval

LongDelay: long retention interval (Study 2)

LongDelay-N: subsequently not-recalled pairs from the LongDelay condition (Study 2)

LongDelay-R: subsequently recalled pairs from the LongDelay condition (Study 2)

LV: latent variable

LW: late wave

MTL: medial temporal lobe

N: subsequently not-recalled pairs (Study 1)

NW1: first word of subsequently not-recalled pairs (Study 1)

NW2: second word of subsequently not-recalled pairs (Study 1)

N(W2-W1): Word2-Word1 difference waves for subsequently not-recalled pairs (Study 1)

Ospan: operation span (Study 3)

PC1: first principal component (Study 1)

xiv

PC2: second principal component (Study 1)

PC3: third principal component (Study 1)

PLS: Partial Least Squares

PM: primary memory

R: subsequently recalled pairs (Study 1)

RI: retention interval

RW1: first word of subsequently recalled pairs (Study 1)

RW2: second word of subsequently recalled pairs (Study 1)

R(W2-W1): Word2-Word1 difference waves for subsequently recalled pairs (Study 1)

ShortDelay: short retention interval (Study 2)

ShortDelay-N: subsequently not-recalled pairs from the ShortDelay condition (Study 2)

ShortDelay-R: subsequently recalled pairs from the ShortDelay condition (Study 2)

SM: secondary memory

WMC: working memory capacity

Word1: first word of a pair (Study 1)

Word2: second word of a pair (Study 1)

1

Chapter 1 Introduction

Memory, as an object of study, has been divided into different categories. Episodic memory is

conceptualized as a neurocognitive system that enables one to remember previously experienced

events (Tulving, 1972, 1983, 1985, 2005). Preceding the concept of episodic memory, Tulving

(1968) proposed that memory for lists of words that are studied in a laboratory setting can be

thought of more generally as memory for experienced events, as opposed to lexical units or items

in memory, which was the commonly held view at the time. This shift in perspective was critical

for the conceptualization of episodic memory. Episodic memory is associated with chronesthesia

– a form of consciousness that enables individuals to remember the personal past and imagine

the personal future or “mental time travel” (Nyberg, Kim, Habib, Levine, & Tulving, 2010;

Tulving, 2002). Other important types of memory, such as semantic memory and procedural

memory, do not share this feature of mental time travel that is unique to episodic memory.

Although episodic memory has been, and continues to be, the focus of a vast amount of memory

research (e.g., Dere, Easton, Nadel, & Huston, 2008) since its conception, the neurocognitive

mechanisms that underlie the encoding and binding of various components of an event is not yet

fully understood.

The present dissertation investigated the neurocognitive mechanisms that underlie the formation

of episodic (as opposed to semantic) associations between distinct items of information (e.g.,

English nouns) that constitute an event (e.g., the presentation of a pair of English words). The

association is a basic element of human memory, and refers to the binding of two memory traces

that effectively maps one onto the other (Murdock, 1974). Whereas semantic associations refer

to pre-experimental associations that are generally related to the meaning of the individual items

and developed over many experiences (e.g., TABLE – CHAIR, KING – QUEEN), episodic

associations enable one to remember the specific details of an experienced event (e.g., when was

the last time you had dinner with your best friend, where did you eat and what did you talk

about). In my thesis work, I used scalp-recorded event-related potentials (ERPs) to provide a

continuous measure of how healthy young adults processed visually presented pairs of words

during encoding. This made it possible to examine specific periods of processing that differed for

pairs that were successfully retrieved on a subsequent memory test and those that were not. In

2

the following sections of this chapter, I will briefly describe 1) methods used in the laboratory to

study episodic association formation; 2) the main behavioural findings; and 3) ERP studies on

encoding. Then I will conclude this chapter with an outline and specific goals of the present

dissertation and a description of the general methods that were used.

1 Methods of Investigation Encoding and retrieval of episodic associations are primarily studied in the laboratory using the

paired-associate learning method (Calkins, 1894). In this method, the experimenter explicitly

designates the two items that form each pair. The pairs may be composed of any type of discrete

unit (words, pictures, digits, nonsense syllables, etc). During the study periods, sets of paired-

associates are typically presented to the participant one at a time. During retrieval, participants

are presented with one item of a given pair and are asked to recall the item that it was presented

with during the study phase. Memory for associative information can also be tested using a

recognition task, which typically requires participants to discriminate between studied and novel

combinations of items that were presented during the study phase. Study and test periods can be

alternated (discrete-trials procedure) or intermixed (continuous-task procedure), and may be

separated by a distractor task. In studies that use a single-trial design, each pair is presented once

to the participant. In studies that use a multi-trial design, each pair is presented to the participant

many times.1

2 Summary of the Main Behavioural Findings Cognitive theorists have made a qualitative distinction between memory for associative

information and item information (Murdock, 1974). Item information corresponds to individual

items or events, such as the presentation of a single word. Associative information, on the other

hand, corresponds to links between items, such as the presentation of two words as a pair or

1 Another method for studying paired-associate learning is the anticipation method. In this method, participants are first presented with the cue for a given pair and asked to identify its associate. After the participant provides a response, both items of the pair are presented. This procedure is followed for each pair in a list, and participants are trained on a given list until they meet a predetermined criterion (e.g., two perfect cycles through the lists). Throughout the first cycle of a list, participants are typically instructed to anticipate or guess the correct response when they are presented with the cue for each of the pairs, since the pairings would not have been presented to them yet. Although this method is now used less commonly, it was the method of choice for the study of paired-associate learning up until the early 1970's (Murdock, 1974).

3

single unit. In support of this qualitative distinction, past studies have shown that amnesics,

patients with Alzheimer's disease and healthy older adults show more decline in memory for

associative information compared to item information (Yonelinas, 2002). Additionally,

compared to item information, associative information takes longer to retrieve, is enhanced for

high frequency words, and shows distinct receiver operating characteristic and forgetting curves.

Participants’ performance on episodic association tasks is strongly dependent on the use of

intentional encoding strategies. For pairs of words, in particular, memory for associations

depends largely on the formation of linguistic mediators, the use of imagery and other

elaborative encoding strategies (Bower, 1970; Montague, Adams, & Kiess, 1966; Paivio, 1969).

Furthermore, paired-associate learning is enhanced, sometimes greatly enhanced, when each pair

in a list belongs to a different conceptual category compared to when all pairs in the list belong

to the same conceptual category. This powerful effect of context was demonstrated, for example,

by Bower, Thompson-Schill, and Tulving (1994) in a study where participants were presented

with pairs composed of items from the same conceptual category, and intra-list semantic

similarity was manipulated. Memory for associations, like other forms of memory, is enhanced

when the meaning associated with the information is more fully processed, as proposed by the

levels-of-processing framework (Craik & Tulving, 1975; Craik & Lockhart, 1972). Additionally,

memory for associations is benefited by repeated testing and spaced (as opposed to massed)

practice. The “spacing effect” is a classic finding in memory research and it refers to the finding

that repetitive study of information becomes more effective for long-term retention as the

interval between repetitions increases (Cepeda, Pashler, Vul, Wixted, & Rohrer, 2006). The

facilitative role of repeated testing (retrieval) on long-term retention has long been known in the

literature (Gates, 1917; Roediger & Butler, 2011). Interestingly, recent research by Karpicke and

Roediger (2008; 2007) has produced a surprising finding: standard conditions of repeated study

are essentially useless for long-term retention after an item’s initial recall; instead, repeated

testing is a critical factor. However, additional work has revealed that retrieval is not always

beneficial for memory (Bridge & Paller, 2012; Roediger & Marsh, 2005), and that retrieval may

reinforce the memory trace that was retrieved, but not necessarily the information that was

originally encoded. For example, Buckner et al. (2001) demonstrated that participants’ memory

performance for lures on a recognition test and items that had been intentionally encoded were

comparable on an unexpected subsequent memory test.

4

Studies that involve repeated testing through immediate and end of session memory tests have

shown interesting patterns of forgetting using both single words and pairs of words: items

presented in the last serial position of a list are recalled best during immediate testing (recency

effect), but they are also recalled with the lowest probability in delayed memory tests (negative

recency effect; Craik, 1970). Madigan and McCabe (1971) showed an extreme case of recency

and negative recency using pairs of words, where paired-associate recall for pairs in the last

position of a list was perfect (at ceiling) on the immediate test, but these same pairs were never

recalled on a delayed paired-associate memory test. The negative recency effect was observed

for pairs that participants were cued to recall on both the immediate and delayed memory tests,

as well as for pairs that were only cued for recall on the delayed test. This finding suggests that

retrieval on the immediate test is not responsible for the negative recency effect. Other

explanations for the negative recency effect involve the concepts of primary memory (PM) and

secondary memory (SM; e.g., Craik, 1970; Tulving & Patterson, 1968), as discussed further in

Chapter 4 in relation to the study on associative binding strength and the negative recency effect.

James (1890) conceived of PM and SM as two distinct memory stores. According to James,

items (or events) in PM are those that have never left consciousness, whereas items (or events)

retrieved from SM have been absent from consciousness. Thus, items (or events) in PM refer to

the present time, whereas items (or events) retrieved from SM refer to the past. Moreover,

whereas PM is an accurate depiction of events that were just experienced, SM includes

distortions and missing pieces. Waugh and Norman (1965) proposed a model of PM and SM that

provides a basis for the operational distinction between the two memory stores. However, in

contrast to James who believed that events remain in PM for a fixed amount of time, Waugh and

Norman suggested that PM is composed of a certain number of events at any given moment.

According to their model, each verbal item (or event) that is attended to enters PM, and new

events displace old ones. Old items (or events) that are displaced from PM become lost and

inaccessible. However, items (or events) that are rehearsed remain in PM and may enter SM.

Thus, in this model, rehearsal is related to longer storage in memory. The model also assumes

that interference due to stimulus perception and response production has the same effect on an

item in PM. Thus, the likelihood of an item being retrieved from PM depends on the number of

new items perceived after it, as well as the number of items retrieved between its presentation

and attempted retrieval. This assumption is supported by studies showing that there are no

5

consistent differences in the effects of interference caused by stimulus perception and response

production in the retention of paired-associates (Tulving & Arbuckle, 1963).

Memory for paired-associates can be tested in the forward direction (where the item of a pair that

was presented first is used to cue retrieval of the second item) or in the backward direction

(where the item of a pair that was presented second is used to cue retrieval of the first item).

Consequently, the following question arises: after studying a pair A-B, can participants recall B

when given A as well as they can recall A given B? The vast majority of studies that have

examined forward and backward recall of paired-associates have shown that associative recall is

symmetrical (e.g., Kahana, 2002; Rizzuto & Kahana, 2000). Additionally, Kahana (2002)

examined the correlation between forward and backward recall at the level of individual pairs of

items and found that this correlation was close to perfect. The correlation between recall of pairs

tested in the same direction was also near perfect. In contrast, Kahana and Caplan (2002) showed

a forward advantage in study direction for triplets and serial lists, suggesting that there are

differences between associative processes that underlie association formation and memory for

series of three or more items. However, the same or similar processes underlying primacy effects

shown in list learning studies (Deese & Kaufman, 1957; Murdock, 1962) could also apply to

sequentially presented pairs of items, which would be consistent with explanations at the

neuronal level for primacy effects (Sikström, 2006; Tulving, 2007, 2010; Tulving & Rosenbaum,

2006). Also consistent with this notion, Caplan’s Isolation Principle (Caplan, 2005) posits that

paired-associate and serial learning are ends of a continuum rather than distinct types of

information. This principle proposes that pairs of consecutive items are relatively isolated from

other study items in paired-associate learning paradigms (e.g., the interval separating two items

of a pair is typically shorter than the interval separating consecutive pairs) but not in serial list

learning paradigms, resulting in differential interference that can account for the nearly perfect

correlation between forward and backward probes of pairs compared to the moderate correlation

for serial lists.

The studies that were conducted for the present dissertation were based on the assumption that

participants cannot begin to form associations between pairs of items until both items have been

presented. Participants may, however, begin preparing to make an association before both items

of a pair are presented. For example, in a situation where the words of a pair are presented

sequentially, participants may start preparing to form an association after the first word is

6

presented. According to the conceptual peg hypothesis (Paivio, 1963, 1965), the first word of a

pair may be used as a peg upon which the second word can be integrated to form an association.

According to this hypothesis, concrete nouns, compared to abstract nouns, should serve more

effectively as conceptual pegs because they are more conducive to imagery, which the

hypothesis regards as a mediator of recall. Furthermore, the conceptual peg hypothesis suggests

that as long as one word of a pair is highly imageable, a holistic association, as reflected by a

high correlation between forward and backward paired-associate recall, can be formed by

integrating the remaining word into the image generated for the peg. Pairs composed of two low

imageability words, however, cannot form a unified whole. Contrary to this notion, Madan and

colleagues (Madan, Glaholt, & Caplan, 2010) have shown that pairs composed of two low-

imageability nouns remained generally as holistic as pairs composed of two high-imageability

words and pairs composed of both high imageability and low-imageability words, even though

they did not include a high-imageability word that could be used as a conceptual peg.

3 Event-related Potentials Studies on Encoding Behavioural research has contributed several insights about how episodic associations are

formed. Supplementing behavioural work with neuroimaging techniques offers the added

advantage of providing measures of brain activity that occur during encoding, which provides

further insights into the neurocognitive processes that underlie encoding of information. For

example, Guo and colleagues (2005) showed that ERPs recorded during encoding that were

predictive of memory for associations between faces and names differed from ERPs that were

predictive of memory for the singly presented faces and names. Furthermore, ERPs that were

predictive of memory for associations were not equivalent to the sum of ERPs that were

predictive of memory for the constituent items (face, name). Other studies have produced similar

findings using functional magnetic resonance imaging (Small et al., 2001), providing further

support for a qualitative distinction between memory for associative information and item

information, as discussed above. A good deal of evidence about how the brain supports encoding

has been revealed through the use of the “subsequent memory paradigm” (Paller & Wagner,

2002; Rugg, Otten, & Henson, 2002; Wagner, Koutstaal, & Schacter, 1999). In this paradigm,

participants’ brain responses are recorded while they are presented with study items (encoding

phase). Afterwards, participants are tested on their memory for these study items (test phase).

The brain responses that were recorded for each item during the encoding phase are then sorted

7

and analyzed based on whether the given study item was subsequently retrieved on the memory

test. This type of analysis allows one to examine differences between encoding phase brain

responses recorded for subsequently recalled and subsequently not-recalled study items, as

measured by the employed memory test. The differences found in the brain responses are

thought to reflect how effectively a memory trace is formed.

The ERP technique is a useful tool for investigations on episodic association formation, and

encoding in general, for several reasons (Alain & Winkler, 2012; Paller, 2004; Rugg & Allan,

2000). Firstly, since ERPs allow neural activity to be measured with a temporal resolution on the

order of milliseconds, ERPs can be recorded easily with most perceptual and cognitive

paradigms without requiring much modification. Thus, neural correlates of a particular task can

be examined without introducing new variables that alter the perceptual or cognitive effects

under study. With respect to encoding, in particular, ERPs enable different operations of

encoding to be captured on the basis of their time course. Thus the ERPs elicited by subsequently

recalled pairs and subsequently not-recalled pairs can be compared for differences, and based on

when these differences occurred, it can be inferred which specific cognitive processes were

differentially engaged while participants were memorizing the pairs. Additionally, ERPs can be

used to examine whether encoding processes, or cognitive processes in general, engaged in

different experimental conditions are functionally dissociable. This is based on the premise that

qualitative differences between scalp distributions associated with different conditions reflect

neural processes that, at the least, do not-overlap completely. Moreover, ERP waveforms can be

produced “off-line” after the EEG traces recorded during the study phase have been organized

based on experimental conditions and participants’ later responses, allowing study phase neural

responses for subsequently retrieved and subsequently not-retrieved items to be contrasted.

ERPs are obtained by averaging multiple electroencephalography (EEG) traces that are time-

locked to a specific event. EEG refers to the practice of recording electrical brain activity. When

information is processed by the human brain the flow of electrical currents in the associated

neurons changes, which causes an electrical field that can be measured from the surface of the

scalp. ERP waveforms consist of components (also referred to as waves, and peaks and valleys)

that reflect synchronous activity from large ensembles of neurons in the brain (Alain & Winkler,

2012; Friedman & Johnson, 2000; Paller, 2004). Whereas some components are obligatory and

occur irrespective of the participants’ intentions (referred to as exogenous components), other

8

components are determined by a combination of internal psychological factors (e.g., expectation)

and properties of the external stimulus (referred to as endogenous components). ERP

components are commonly labeled in relation to their polarity and latency (measured in ms) in

relation to the onset of the event in question. It is also common for ERP components to be

labeled in terms of their polarity and ordinal position in the waveform. For example, ‘P1’ refers

to the first positive peak in an ERP waveform, whereas ‘N1’ refers to first negative peak and

‘P2’ refers to the second positive peak. Early sensory responses (e.g., P1, N1, P2, N2) are

exogenous components that are dependent on external factors. In contrast, the P3 is an

endogenous component that is dependent on internal factors.

In the literature, the P3 is also referred to as the ‘P300’, where the ‘P’ corresponds to the positive

polarity of the component, and the ‘300’ corresponds to the latency, in ms, of the component

peak as it was first reported (Sutton et al., 1965). However, the latency of the P300 has been

shown to vary widely, often exceeding 500 ms, since it is elicited following the completion of

stimulus evaluation (Karis, Fabiani & Donchin, 1984, Polich, 2007). The P300 or P3 is related to

target probability and is maximally recorded at the midline centro-parietal electrode locations; it

is usually largest over the parietal region of the scalp, slightly smaller over the central region of

the scalp, and very small over the frontal region of the scalp (Picton, 1992; Polich, 2007). The

original observation and examination of the P3 in the context of target evaluation and effects of

attention (Sutton et al., 1965) led to subsequent investigations on whether the P3 is related to

memory trace formation. These latter investigations employed the subsequent memory paradigm

to examine whether the P3 observed in encoding-phase ERP waveforms varied as a function of

participants’ subsequent memory performance. The general finding of these studies was that a

larger encoding-phase P3 was elicited by stimuli that were subsequently retrieved on a memory

test, compared to stimuli that were not subsequently retrieved. However, as described in more

detail below, the P3 only varied as a function of subsequent memory performance when shallow

encoding strategies were used, but not when deep encoding strategies were used (Karis, Fabiani

& Donchin, 1984; Fabiani, Karis & Donchin, 1990).

Past studies have used the subsequent memory paradigm to show that ERPs elicited by to-be-

remembered items, during both intentional (Fabiani, Karis, & Donchin, 1990a) and incidental

(Paller, Kutas, & Mayes, 1987) memory paradigms, can predict the accuracy of retrieval on a

subsequent memory test. The general finding is that ERPs elicited by those items that were

9

subsequently retrieved demonstrated a larger positive deflection compared to the ERPs elicited

by items that were subsequently forgotten. These ERP differences provide a measure of encoding

– a phenomenon that has been referred to in the literature using multiple terms, such as ‘Dm’ –

difference based on later memory performance (Paller et al., 1987), an ERP ‘memory effect’

(Friedman, 1990) and a ‘subsequent memory effect’ (Rugg, 1995). The Dm effects reported in

the ERP literature are composed of three components: the early and late positive components

(Johnson, 1995) and slow wave. The early component begins around 250 ms after stimulus

onset, and is largest over the frontal region of the scalp compared to the central and parietal scalp

regions. The late component begins around 450 ms after stimulus onset and has been

investigated much more extensively compared to the early component. Various parameters have

been shown to modulate the late Dm component, including the encoding strategy used (Paller et

al., 1987), whether the memory test was explicit versus implicit (Paller, 1990) and the type of

stimuli used (Sommer, Schweinberger, & Matt, 1991). The slow wave begins around 500 ms

after the stimulus onset and has been associated with elaborative, as opposed to rote, encoding

(Karis, Fabiani, & Donchin, 1984; Mangels, Picton, & Craik, 2001).

In the first study that reported encoding-related ERP data in relation to participants’ subsequent

memory performance, Sanquist and colleagues (1980) presented participants with pairs of words

to study, with each word of a pair presented sequentially. Participants’ task was to judge whether

the two words were the same or different based on one of three criteria: orthographic, phonemic,

or semantic attributes. Later, the participants were tested on recognition of words that were

presented second within a pair, but they were not tested on their memory for associations.

Participants’ performance on the recognition test for the second word of a pair was used as the

basis for the subsequent memory analysis. Semantic comparisons led to the highest percentage of

recognized words, followed by the phonemic and then the orthographic comparisons.

Subsequently recognized words elicited greater positivity in the late positive component and

slow wave compared to subsequently unrecognized words. For the semantic condition, the late

positive component difference peaked at about 500 ms after the onset of the first and second

words of a pair and was largest at midline parietal scalp region (i.e., Pz). The slow wave

difference started at about one second after the onset of the second word of a pair and appeared

larger in the more anterior, compared to posterior, regions of the scalp. A number of other

studies have found similar Dm effects in relation to recognition and recall of single words

10

(Donchin & Fabiani, 1991; Friedman & Johnson, 2000; Johnson, 1995; Rugg, 1995; Rugg &

Allan, 2000).

In one such study, the ERPs recorded during performance in a von Restorff paradigm were

examined as a function of subsequent recall performance (Karis et al., 1984). In a von Restorff

paradigm a series of items are presented, one of which (the “isolate”) deviates in some way from

the others. In this study participants were presented with a series of words and isolates were

presented in either a larger or smaller font compared to all of the other words. Participants were

instructed to memorize as many words as they could. Isolates were better recalled than non-

isolates presented in the same serial position of a list. Additionally, subsequently recalled words

elicited greater positivity in the encoding-phase ERP waveform compared to subsequently not-

recalled words, regardless of whether the word was an isolate. The positivity in the ERP

waveform peaked around 520 ms after the word onset and was largest at Pz, slightly smaller at

Cz, and very small at Fz (Karis et al., 1984). Based on the observed peak latency and scalp

distribution, Karis et al. interpreted this effect to be enhanced P300 response.

In a post hoc analysis of these data, the investigators also found a relation between the magnitude

of this ERP difference and the type of encoding strategy that the participants used. For

participants who used rote rehearsal strategies, the amplitude of the P300 elicited by

subsequently recalled words was larger than that of subsequently not-recalled words (Karis et al.,

1984). However, for participants who used deep encoding strategies, no such difference was

observed; instead, for these participants a frontal-positive slow wave, which appeared at 540 ms

and slowly increased over the remainder of the epoch, was observed to be more positive for

subsequently recalled, compared to subsequently not-recalled, words and more positive over the

frontal region compared to the central and parietal regions.

In a following study, Fabiani, Karis and Donchin (1986) hypothesized that the previously

observed P300 effect would emerge more consistently when elaborative strategies are not used.

To test this hypothesis, Fabiani and colleagues used an incidental memory paradigm to reduce

the use of rehearsal strategies. Participants were presented with a series of names and were

instructed to count either the number of female or male names. Participants were then

unexpectedly tested on free recall for the names that had been presented to them. Again

subsequently recalled items elicited greater positivity in what was interpreted to be the P300

11

compared to subsequently not-recalled items. This increased positivity had a peak latency of

about 550 ms and was largest at the mid-parietal region compared to the vertex and mid-frontal

regions.

Around the same time, Paller and colleagues (1987) also examined encoding-related ERPs

recorded during an incidental memory paradigm. For each word of a series, participants

performed either a semantic judgment that required a word’s meaning to be processed (e.g., is it

edible? Is it living?), or a non-semantic judgment that did not (e.g., Are there exactly two

vowels? Are the first and last letters in alphabetical order?). ERPs averaged over all tasks

showed greater late positive activity, as measured in the 400-800 ms latency range, for

subsequently recalled and recognized words, compared to words that were not subsequently

retrieved, at midline and lateral parietal electrodes. Paller et al. referred to this difference

between ERPs to subsequently retrieved and forgotten words as 'Dm' and defined it operationally

as “any ERP Difference based on later memory performance.” However, Paller and colleagues

found that Dm could not be accounted for solely in terms of changes in the P300 amplitude,

since the Dm was largest for the semantic tasks over the anterior scalp region. Similarly,

Friedman (1990) later reported a Dm scalp distribution for word stimuli that was significantly

different from that characteristic of the P300. The findings of Paller et al. (1987) are consistent

with those of Karis et al. (1984), who found that deep encoding strategies elicited a Dm effect

that was largest over the frontal region compared to central and parietal regions.

To examine further the relation between encoding strategies and Dm, Fabiani, Karis and

Donchin (1990) replicated their previous von Restorff study (Karis et al., 1984), this time

manipulating the encoding strategy that participants were instructed to use. For the rote

strategies, participants were instructed to repeat each word as it was presented, whereas for the

elaborative strategies participants were instructed to combine words in images, sentences, or

stories. The P3 elicited by subsequently recalled words was significantly larger than the P3

elicited by words that were not subsequently recalled when rote strategies were used, but not

when elaborative strategies were used. Replicating the results of Karis and colleagues (1984), the

amplitude of a frontal-positive slow wave was larger for subsequently recalled, compared to

subsequently not-recalled, words when elaborative strategies were used. On the basis of these

and similar findings, Donchin and Fabiani (1991) later proposed that the observed P3 and fontal-

positive slow wave Dm effects reflected different types of cognitive processes. They related the

12

P3 Dm effect to the ‘subjective distinctiveness’ of items, and the frontal-positive slow wave Dm

effect to the additional semantic processing of items.

Compared to the vast amount of work that has been done to investigate the electrophysiological

processes engaged in item encoding (Donchin & Fabiani, 1991; Friedman & Johnson, 2000;

Rugg & Allan, 2000), less work has been done to investigate the electrophysiological correlates

of association formation. There is, however, some evidence suggesting that a late occurring

positive slow wave may be reflective of association formation. In some studies, this sustained

positivity has a broad, fronto-central topography that is slightly biased to the right-hemisphere

and in other studies it has a frontal distribution with a slight right-hemisphere bias. For example,

in one study that used pairs of words to examine Dm effects, participants either created a

semantic association between the two words or made a semantic judgment separately for each

word during encoding (Weyerts, Tendolkar, Smid, & Heinze, 1997). A reliable Dm effect was

found only for pairs that were encoded by creating a semantic association: subsequently

recognized pairs elicited a more positive-going waveform, starting at about 200 ms to the end of

the epoch 1400 ms later, compared to pairs that were not subsequently recognized. This

positivity was maximal over the frontal scalp and larger over the right than left hemisphere.

However, since both words of a pair were presented at the same time during encoding, it is

unclear which aspects of the observed Dm effect corresponds to association formation as

opposed to the encoding of individual words. To help differentiate the EEG signal recorded for

the encoding of the first word from the EEG signal recorded for the second word and the

association formed between the two words, items of a pair should be presented one at a time.

Previous studies that investigated the neuroelectric correlates of association formation did not

focus on differentiating association formation from item encoding. For example, Kounios and

colleagues (2001) conducted an ERP study to determine whether the use of two different

associative strategies (compositional versus fused representations) would result in different

electrophysiological patterns. In this study, participants were presented with pairs of words, with

each word of a pair presented one at a time. Participants were later tested on their memory of the

order in which the words of a pair were presented. However, Kounios and colleagues did not

report the results of a subsequent memory analysis. Instead, for those pairs that were

subsequently remembered, the encoding-related ERP data were analyzed in relation to

participants’ response speed on the memory test to yield ERPs reflecting encoding activity that

13

resulted in more or less effective memory retrieval. The reasoning behind this was that those

pairs whose order were well encoded would be retrieved more quickly compared to pairs whose

order is more poorly encoded. Thus, the encoding data were analyzed to predict participants’

subsequent response speed for remembered pairs as opposed to participants’ subsequent memory

performance.

More recently, in a study by Caplan and colleagues (2009), participants were presented with

words to study, one word at a time. The words were either grouped into pairs or presented in

short lists composed of three words. Later, participants were tested on their memory of the pairs

using cued recall, where the target was probed with the word that has been shown before or after

it during the study phase. Participants’ memory performance was used as the basis for the

subsequent memory analysis. The study aimed to differentiate the neuroelectric correlates of

association formation and serial list learning. For the pairs, the encoding-related ERPs elicited by

the first and second words of a pair were averaged together, making it difficult to tease apart the

neuroelectric correlates of item encoding and association formation. However, the results of a

multivariate analysis on these data showed a latent variable reflecting a significant Dm effect for

pairs, but not lists, that was prominent over central electrodes close to the midline (e.g.,

electrodes C1 and C2) during the later portion of the epoch, starting from about 1200 ms post-

stimulus. The results also showed a second latent variable reflecting a significant Dm effect for

pairs, but not lists, that was prominent over posterior electrode sites and somewhat left-

lateralized. The timing of this Dm effect overlapped primarily with the time window of the slow

wave, but also showed some overlap with the early and late positive components. This finding is

consistent with one reported by Yovel and Paller (2004), which showed that positive ERPs

recorded late over the bilateral parietal scalp during encoding predicted subsequent recognition

of faces and the assigned occupation (contextual information), whereas ERPs recorded over the

right parietal scalp predicted subsequent recognition of faces without information about the

assigned occupation.

In a study by Kim and colleagues (2009) participants studied and encoded pairs of semantically

related words, with each word of a pair presented separately, one word at a time. The EEG data

were collected during the study phase under two conditions that varied in the degree of intra-list

semantic similarity. However, the ERPs collected for both of these conditions were combined for

the subsequent memory analysis to increase the overall number of observations of pairs that were

14

included in the analysis. Sustained fronto-central positivity, which occurred between 1-1.6 s after

the presentation onset of the second word, was associated with subsequent paired-associate

recall. This sustained positivity was referred to as the late wave (LW) and was thought to reflect

cognitive association formation. Additionally, subsequently recalled pairs, compared to

subsequently not-recalled pairs, demonstrated a larger positive deflection in the ERP waveform

around 555 ms after each word of a pair was presented. These positive deflections were thought

to reflect the encoding of each individual word. Guo and colleagues (2005) also found a late-

occurring sustained positivity that was larger for subsequently remembered pairs compared to

subsequently forgotten pairs, and this Dm effect was largest over a fronto-central electrode that

was located slightly right of the midline. More recently, Cheng and Rugg (2010) have shown a

positive-going ERP effect for a subsequent hit versus subsequent miss contrast between 1000 to

1400 ms post-stimulus for pairs of words. This contrast was significant over the frontal scalp

region and showed a right-hemisphere bias.

In light of previous work demonstrating interdependence between encoding and retrieval

processes and representations (Alvarez & Squire, 1994; Tulving & Thomson, 1973), ERP

patterns reflecting retrieval of associative information may also reveal clues about the

neuroelectric correlates of episodic association formation. The parietal old/new effect is of most

interest in this regard, because it is thought to reflect recollection (Mecklinger, 2006; Rugg &

Curran, 2007) and has been associated with recollection of associative information (Donaldson

& Rugg, 1998; Tendolkar, Doyle, & Rugg, 1997). The parietal old/new effect typically occurs

between 400 to 800 ms after stimulus onset and is largest over the left parietal scalp region.

4 Outline and goals Memory performance can be thought of as an outcome of processes that occur during different

stages of memory: encoding, consolidation and retrieval. The present thesis focused on encoding

with the goal of investigating specific neurocognitive processes engaged in episodic association

formation. Using the ERP methodology, combined with behavioural data, the present thesis

helps address one of the major criticisms of the levels-of-processing framework (Craik &

Tulving, 1975; Craik & Lockhart, 1972) – the lack of an objective index of depth of information

processing (Craik, 2002). Although past studies have investigated ERP patterns associated with

different levels of processing, as mentioned above and in Chapter 6, these past studies focused

15

on differences between ERPs elicited by deep encoding and shallow encoding processes, and did

not focus on ERPs elicited by different levels and types of deep encoding processes. The studies

included in the present thesis aimed to fill this gap by identifying specific cognitive processes

reflected by different Dm effects. In doing so, the present dissertation also helps address one of

the major concerns of ERP research on memory encoding – the observation that multiple ERP

components are involved in Dm effects, combined with an unclear understanding of the

circumstances under which each component makes its contribution (Rugg & Allan, 2000). Study

1 of the present thesis (Chapter 2) reports on a re-examination on the ERP data collected by Kim

et al. (2009). This re-examination was conducted to investigate ERP correlates of association

formation, but not item encoding, and thus laid the groundwork for the subsequent studies. Study

2 (Chapter 3) continued the investigation on specific processes engaged during episodic

association formation by focusing on ERP waveforms recorded after the time window

investigated in Study 1. Specific processes reflected by the results of Study 2 were further

investigated in Study 3 (Chapter 4) using the distinction between PM and SM. Study 4 (Chapter

5) investigated ERP correlates reflecting the strength of the association formed using an

experimental design that also allowed for an investigation of the neuroelectric processes engaged

in the negative recency effect, as described above. The overall findings of the present thesis are

discussed in the last chapter (Chapter 6) in relation to the extant literature.

5 General Methods

5.1 General Hypotheses The main hypotheses to be tested were as follows:

(1) Association formation is correlated with a specific pattern of neural activity that is revealed

by the ERP signature and distinguished from other memory-related ERP components.

(2) More specifically, during the time range of the parietal old/new effect (about 400 to 800 ms

after stimulus onset), differences in the ERPs recorded over the parietal scalp region during

encoding will be shown. These differences will be based on participants’ subsequent paired-

associate recall performance. This hypothesis is based on previous work demonstrating

interdependence between encoding and retrieval processes (e.g., Rugg, Johnson, Park, &

Uncapher, 2008; Tulving & Thomson, 1973) combined with the results of other studies that

16

suggest that the parietal old/new effect is thought to reflect recollection (Mecklinger, 2006; Rugg

& Curran, 2007), and more specifically recollection of associative information (Donaldson &

Rugg, 1998; Tendolkar et al., 1997).

(3) Additional encoding-related ERP differences that are based on participants’ subsequent

paired-associate recall performance will be shown after the time range of the parietal old/new

effect. These ERP differences will be shown over the anterior scalp region, as demonstrated in

past studies where participants were instructed to use elaborative encoding strategies (e.g.,

Fabiani, Karis, & Donchin, 1990).

5.2 Participants Healthy volunteers between the ages of 18 to 32 years were recruited from the participant pool at

the Rotman Research Institute of Baycrest. All participants were screened to ensure that (1) they

did not suffer from any disease or take any medications that might compromise brain function;

(2) they had normal or corrected-to-normal vision; and (3) were fluent English speakers. A new

group of participants was tested for each of the four studies. All participants provided written

informed consent, as approved by the Baycrest Research Ethics Board, prior to their participation

in any experiment. Participants who contributed fewer than 10 artifact-free trials for any given

condition were excluded from the analyses, as done in past studies (Johnson & Rugg, 2006;

Wilckens, Tremel, Wolk, & Wheeler, 2011; Woollams, Taylor, Karayanidis, & Henson, 2008).

Past studies, such as those reviewed in Section 3 of the present chapter (Event-related potential

studies of encoding), have shown that samples composed of a minimum of 10 participants are

sufficient to allow statistical tests to demonstrate the experimental effects and to support

generalization of the results. In line with past studies, the sample size for each of them main

analyses conducted for the studies of the present thesis consisted of 12 or more participants who

contributed a minimum of 10 artifact-free trials for each of the conditions of interest.

5.3 EEG Recording and Analysis Procedures The ERP experiments took place in a double-walled sound attenuated chamber. Stimuli were

presented visually on a computer screen 70 cm from the participant. EEG activity was recorded

continuously from a 64-channel EEG cap (Electro-Cap international, 10–10 system) using a

Neuroscan Synamps (El Paso, TX, USA). The neuroelectric activity was amplified (500 times),

17

filtered (.05–50 Hz), and sampled at 250 Hz. During EEG recording all electrodes were

referenced to Cz and subsequently converted to an average reference. Electrodes were located

over all scalp regions with four electrodes near the eyes and nine electrodes below the Fpz-T7-

Oz-T8 “equator.” All impedances were kept under 5 kOhms. Eye blinks, horizontal and vertical

eye movements were recorded immediately before and after the experiment to allow

compensation of ocular artifacts using source components (Picton et al., 2000).

The encoding-related ERPs were sorted and analyzed based on participants’ subsequent memory

performance. The ERPs were averaged separately for each electrode site and experimental

condition. The mean amplitudes and latencies of ERPs were measured after referencing voltages

to a pre-stimulus baseline. The ERP datasets were analyzed using univariate and multivariate

statistical techniques. Repeated measures analysis of variance (ANOVA) was used to examine

mean ERP amplitudes when the selection of electrodes and time window of analysis were based

on the results of previous work. Principal component analysis and partial least squares analysis,

which will be described in more detail in Chapter 5, were used to select electrodes and time

windows for statistical analysis based on the largest amount of variability that could be

accounted for in the data. These multivariate, data-driven approaches complemented the

hypothesis-driven univariate analyses that were conducted for the present dissertation. The

behavioural data were analyzed using tests of repeated measures ANOVA and paired t-tests. All

hypothesis tests were performed at an alpha level of 0.05.

18

Chapter 2 Differentiating ERP correlates of item encoding and episodic

association formation (Study 1)2 The aim of this first study was to identify patterns in the ERP waveforms that reflected episodic

association formation, but not the encoding of the individual items that the pair was composed

of. Although there are likely similarities between the ERP correlates of item encoding and

episodic association formation, the present study focused on ERP patterns that differentially

reflected episodic association formation and did not investigate the similarities. As reviewed in

the introduction, previous ERP studies have not differentiated the neuroelectric correlates of

episodic association formation and item encoding, as they either presented both items of a pair at

the same time and/or simply did not focus on this aspect (e.g., Caplan et al., 2009; Kounios et al.,

2001; Weyerts et al., 1997).

In the present study, the EEG data from an existing database (Kim et al., 2009) were re-analyzed

to investigate differences in encoding-phase ERPs elicited by the first and second words of a pair

(Word1 and Word2, respectively), where each word was presented one at a time. Whereas item

encoding was reflected in ERPs recorded for Word1 and Word2, the ERPs recorded for Word2

also reflected the association formed between the two words. Thus, the differences between the

ERP waveforms elicited by Word1 and Word2 were derived (Word2-Word1) to investigate

which aspects of the Word2 ERP waveform reflect association formation as opposed to item

encoding (encoding of Word2). To narrow in further on the ERP correlates of association

formation, the Word2-Word1 difference was derived separately for subsequently recalled and

subsequently not-recalled pairs [R(Word2-Word1) and N(Word2-Word1), respectively]. The

resulting difference waveforms were then contrasted, resulting in a second ERP difference,

which is referred to herein as the ‘double difference.’ The double difference waveform was

2 This study has been published in the journal Public Library of Science ONE [Kim, A.S.N., Binns, M.A. & Alain, C. (2012). Neuroelectric Evidence for Cognitive Association Formation: An Event-Related Potential Investigation. PLoS ONE, 7, e34856]. It was also presented as part of a conference talk [Kim, A.S.N. (February 12, 2012). Neuroelectric evidence for cognitive association formation: An event-related potential investigation. Talk presented at the Winter Conference on Neural Plasticity, 24th Annual Meeting, Frigate Bay, St. Kitts] and as a conference poster presentation [Kim, A.S.N., Binns, M.A., Alain, C., & Tulving, E. (November, 2010). Distinguishing association formation from item encoding using ERPs. Poster presented at the Society for Neuroscience Annual Meeting, San Diego, USA].

19

thought to reflect successful association formation: whereas the Word2-Word1 ERP difference

of both the subsequently recalled and subsequently not-recalled pairs likely reflected brain

responses related to the sequential presentation of the words of a pair (e.g., habituation), the

Word2-Word1 ERP difference of the subsequently recalled pairs also reflected brain responses

underlying association formation. Thus, ERP components reflecting association formation were

identified by contrasting the Word2-Word1 ERP differences between subsequently recalled and

subsequently not-recalled pairs [e.g., R(Word2-Word1)- N(Word2-Word1)].

The hypotheses of the present study were formed on the premise that any ERP component

reflecting association formation would occur primarily during or after the presentation of the

second item. Although participants may start preparing to make an association when they are

presented with the first item or even earlier, an association between two particular items can only

begin to form after the second item is presented. Therefore, the hypotheses of the present study

were as follows:

(1) ERPs recorded during the study phase will demonstrate a larger positive amplitude for Word2

compared to Word1. This Word2-Word1 ERP difference will be larger for subsequently recalled

pairs compared to subsequently not-recalled pairs.

(2) In light of previous work demonstrating interdependence between processes engaged in

encoding and retrieval (Rugg et al., 2008; Tulving & Thomson, 1973), combined with findings

that the parietal old/new effect reflects recollection of associative information (Mecklinger,

2006; Rugg & Curran, 2007), ERP differences reflecting episodic association formation will

occur over the parietal scalp region during the time range of the parietal old/new effect (about

400-800 ms post-stimulus).

6 Methods

6.1 Participants Fourteen healthy, young adults participated in this experiment. Data from two participants were

discarded: one of these participants had too few trials in one of the conditions to allow for ERP

analysis, and the other participant had large movement artifacts throughout the recordings. As a

20

result, ERP averages were obtained from 12 participants (6 female; mean age: 23 years, range:

19 to 32; first language: English).

6.2 Material Two hundred paired-associates were generated for this experiment, using nouns that were

selected without replacement from 20 conceptual categories. The following twenty conceptual

categories were used in the present experiment: 4-legged animals, articles of clothing, birds,

boats, chemical elements, cities, colours, countries, fields of study, fish, flowers, fruits, insects,

musical instruments, human body parts, professions, sports, trees, vegetables, weather

phenomena. The nouns in each pair belonged to the same conceptual category. These pairs were

used to compose two types of lists: Same lists, composed of pairs all belonging to the same

conceptual category; and Different lists, composed of pairs each belonging to a different

conceptual category. In total there were twenty lists: ten Same lists and ten Different lists. All

lists consisted of ten paired-associates. The pairs were counterbalanced across participants, so

that each pair was presented in each type of list.

6.3 Design Although the present study had two levels of intra-list semantic similarity (“Same” vs.

“Different”), for the purposes of the present study, ERPs from both the Same and Different

conditions were combined to increase the overall number of observations of pairs that were

subsequently recalled and pairs that were subsequently not-recalled. The ERP data collected

during the study phase served as the dependent variable.

6.4 Procedure The experimental procedure used in the present experiment is summarized in figure 1. Each

participant took part in one experimental session, which consisted of 20 study/test cycles. Each

study/test cycle consisted of three phases. During the first phase, participants were presented

with a list of 10 paired-associates to study. They were told that they would later be given a

paired-associate recall test during which they would be shown one of the two words

(bidirectional recall test). During the second phase participants solved simple arithmetic

equations, which served as a distractor task. During the third phase participants were tested on

paired-associate recall for the 10 pairs from the study phase. Each session began with a short

21

practice block to familiarize the participants with the experimental task. Each participant then

studied and recalled 20 lists, with short breaks after every fifth list.

The ERP trial corresponded to the presentation of a pair of words during encoding (Figure 2). At

the start of the ERP trial a 500 ms delay was followed by central ‘+’ which served as a warning

and lasted for 500 ms. Word1 was then presented for 1000 ms, followed by a blank screen for

200 ms. Word2 was then presented for 1000 ms. The inter-trial interval varied randomly between

1000 and 3000 ms.

After all 10 pairs had been presented, participants solved eight arithmetic problems of the form

A+B+C=? where A, B and C were randomly selected integers between 0 and 9. Each equation

was presented on the screen for 3750 ms, followed by a 250 ms blank screen. Within this 4000

ms period, participants were asked to respond aloud, as quickly and accurately as possible. After

giving their response, participants moved on to the next arithmetic problem without delay.

After participants finished solving the arithmetic equations, they were given a cued recall test for

all the pairs that had been presented in the study phase of that cycle. During recall, a central ‘+’

was presented for 200 ms, followed by a cue word for 7000 ms. Word1 and Word2 served

equally often as the cue. Participants responded vocally with the word they believed had been

paired with the presented cue word. The experimenter (A.S.N.K.) scored the responses in real-

time by referring to an answer key and pressing the “R” button of the keyboard for correct

responses and the “N” button for incorrect responses. Incorrect responses and the absence of any

response given by the participant within the allotted 7000 ms interval were classified as not-

recalled. The next cue word was presented once the experimenter had coded the participant’s

response into the computer or when the time limit of 7000 ms was reached.

6.5 Behavioural data analysis Paired-associate recall performance was measured as the percentage of correctly recalled pairs,

and was calculated separately for the Same and Different conditions. A paired t-test was used to

assess whether paired-associate recall performance differed between the Same and Different

conditions.

22

6.6 Electrophysiological data analyses For each participant, continuous EEG files were evaluated using BESA 5.1.8. The EEG data was

epoched into 4200 ms segments beginning 500 ms before the ‘+’ stimulus and lasting 1000 ms

after the offset of the second word in the pair. The encoding-related EEG data that were re-

examined in the present study were originally analyzed based on subsequent memory

performance and intra-list semantic similarity conditions (Same and Different), yielding the

following four categories of ERPs: (a) subsequently recalled pairs from the Same condition; (b)

subsequently non-recalled pairs from the Same condition; (c) subsequently recalled pairs from

the Different condition; (d) subsequently non-recalled pairs from the Different condition. For the

purposes of the present study, however, ERPs from the Same and Different conditions were

combined to increase the overall number of observations of pairs that were subsequently recalled

(R) and pairs that were not subsequently recalled (N). Next, R and N waveforms were broken

down into sections corresponding to the presentation of Word1 and Word2. The ERP data

recorded during the one second presentation of Word1 and the one second presentation of Word2

were both baseline corrected to each of the preceding 200 ms intervals. Then Word2-Word1

subtractions were derived separately for the R and N waveforms, resulting in two difference

waveforms: the Word2-Word1 difference for R pairs [R(Word2-Word1)] and for N pairs

[N(Word2-Word1)].

6.7 Principal component analysis A standard principal component analysis (PCA; Donchin & Heffley, 1978) was used to extract a

reduced number of components, which revealed the spatial distribution of electrodes that

displayed similar ERP patterns over time. The PCA was conducted with varimax rotation on the

R(Word2-Word1)-N(Word2-Word1) difference. The input to the PCA was the data matrix for

the 65 electrode site variables by 250 time point observations (250 Hz sampling rate and the

dataset covered a 1000 ms interval) averaged across 12 participants. For each component, the

corresponding factor scores were used to identify temporal addresses that showed a difference

between the R(Word2-Word1) and N(Word2-Word1) waveforms. Representative electrodes

were used to examine whether these differences could be attributable to amplitude differences

between the ERPs to Word1 and Word2 of R and N pairs, as opposed to time shifts.

Representative electrodes were selected based on large factor loadings and a central location

23

within the cluster of electrodes that demonstrated the largest factor loadings. A repeated

measures ANOVA was conducted to examine components of interest. Subsequent memory

performance (R vs. N), word order within the pair (Word1 vs. Word2), and electrode location

were included in the analysis as factors, and all interactions between these factors were also

examined. The ANOVA was conducted on the data recorded from six representative electrodes.

Unless specified as otherwise, significance testing was conducted over a +/- 25 ms range around

the peak identified in the principal component wave.

7 Results

7.1 Behavioural data Subsequent memory performance was calculated as a function of intra-list context condition. The

percentage of paired-associate recall was higher for the Different condition (M = 70.9, SE = 3.1),

which had low intra-list semantic similarity, compared to the Same condition (M = 42.4, SE =

3.4), which had high intra-list semantic similarity [t(11) = 9.70, p < .001].

7.2 Principal component analysis Differences between the R(Word2-Word1) and N(Word2-Word1) waveforms were largest over

the parietal region of the scalp. Waveforms recorded at multiple electrodes over the frontal,

central and parietal scalp regions are shown in figure 3a. Over the parietal region of the scalp, the

amplitude of the R(Word2-Word1) wave was more positive than the N(Word2-Word1) wave

starting at around 200 ms, and continuing through to the end of the 1000 ms period (figure 3b).

Figure 3a shows that this difference in amplitude between the R(Word2-Word1) and N(Word2-

Word1) waveforms was mainly attributable to differences in the ERP amplitude elicited by

Word1 and Word2 of the R pairs, whereas the amplitude of the ERPs to Word1 and Word2 of

the N pairs did not appear to differ.

The PCA on the R(Word2-Word1)-N(Word2-Word1) difference resulted in a six-component

solution when the eigenvalue threshold was set to one. The resulting six-component solution

accounted for 97% of the total variance in the dataset. Components five and six together

accounted for 6% of the total variance. These two components were discarded because there

were very few electrodes loading onto these components, and it was plausible that the

components were serving only to explain random noise detected by these electrodes. The fourth

24

component accounted for 8% of the total variance and was also discarded because the

corresponding waveform did not show a clear pattern. The topographic distributions of the first

three components were unchanged when the analysis was restricted to a three component

solution. The three-component solution accounted for 86% of the total variance: the first

principal component accounted for 55% of variance, the second principal component accounted

for 17% of variance, and the third principal component accounted for 14% of variance. The first

principal component (PC1) was of chief interest, because it provided evidence that was directly

relevant to the purpose of the present study, and is described in more detail below along with

principal component 2 (PC2) and principal component 3 (PC3).

The pattern of the factor loadings for PC1 (55% of variance) of the R(Word2-Word1)-N(Word2-

Word1) difference was most salient over the posterior scalp region, as shown by the

topographical distribution of the electrode loadings in figure 4a. The pattern of the PC1 factor

scores demonstrated a negative deflection at about 130 ms (N130), followed by a positive

deflection at about 460 ms (P460) and sustained positivity between 645–845 ms, as shown in

figure 4b. The grand average ERP waveforms recorded at representative electrode P2 (figure 4c)

showed that the N130 was due to a larger difference between the ERP amplitude to Word1 and

Word2 of N pairs compared to R pairs. In contrast, the P460 appeared to be due to an amplitude

difference between ERPs to Word1 and Word2 of R pairs, with no apparent difference between

the ERP amplitude to Word1 and Word2 of N pairs and ERPs to Word1 of R pairs. The data

recorded at representative electrode P2 also showed that the sustained positivity that was picked

up by PC1 between 645–845 ms was due to larger amplitude differences between the ERPs to

Word1 and Word2 of R pairs compared to N pairs.

A repeated-measures ANOVA was conducted on the N130, P460 and the sustained positivity

observed between 645-845 ms, which will hereafter be referred to as the positive slow wave, to

test whether the amplitude of these components demonstrated an interaction between subsequent

memory performance and word order. The ANOVA was conducted using the data recorded at

electrodes CPz, CP2, Pz, P2, P4, and P6. The full results of the ANOVA are listed in table 1.

Here, we report on the interaction between subsequent memory performance and word order, as

it is the result of interest for the purpose of the present study. The P460 demonstrated an

interaction between subsequent memory performance and word order [F(1, 11) = 8.97, p = .01,

ηp2 = .45]: the ERPs to Word2 were more positive than the ERPs to Word1 for the R pairs [t(71)

25

= -2.631, p = .01], but there was no such difference for the N pairs. For the positive slow wave,

the results of the repeated-measures ANOVA also showed an interaction between subsequent

memory performance and word order [F(1, 11) = 10.11, p = .009, ηp2 = .48]. For the

subsequently recalled pairs, the amplitude for Word2 was more positive than the amplitude for

Word1 [t(71) = -4.38, p < .001]. In contrast, the amplitudes for Word1 and Word2 did not differ

statistically for the subsequently not-recalled pairs. The N130 did not demonstrate an interaction

between subsequent memory performance and word order.

PC2. For PC2 (17% of variance) of the R(W2-W1)-N(W2-W1) difference, the topographical

distribution of the electrode loadings showed a broad positive focus over the fronto-central

region of the scalp, as shown in figure 4a. The pattern of the PC2 factor showed a positive

deflection that peaked around 540 ms (P540), followed by sustained positivity between 800-1000

ms (figure 4b). At around 540 ms, the data recorded at representative electrode FCz showed a

larger amplitude difference between the ERPs to W1 and W2 of N pairs compared to R pairs

(figure 4c). This was also the case for the pattern of sustained positivity that was observed

between 800-1000 ms in the factor scores. Both components were tested for significance using

data collected from electrodes FC1, FCz, FC2, F1, Fz, and F2. Neither of the two components

demonstrated a main effect of subsequent memory performance, or an interaction between

subsequent memory performance and word order. (The full results of the ANOVA are listed in

table 1.)

PC3. The waveform pattern that was observed for PC3 (14% of variance) of the R(W2-W1)-

N(W2-W1) difference was most salient over the right frontal region of the scalp, as shown by the

topographical distribution of the electrode loadings in figure 4a. The pattern of the PC3 factor

scores showed sustained positivity starting from about 500 ms to 750 ms, followed by sustained

negativity from about 800-1000 ms, as shown in figure 4b. When the grand average data

recorded at representative electrode F8 (figure 4c) was examined, this sustained negativity

appeared to be due to a larger amplitude difference between the ERPs to W1 and W2 of R pairs

compared to N pairs. However, the sustained positivity that was observed between 500-750 ms

did not appear to be due to amplitude differences between the ERPs to W1 and W2 of R and N

pairs. Consequently, only the sustained negative wave was tested for significance using data

collected at electrodes FP2, AF4, AF8, F6, F8, and FC6. The full results of the ANOVA are

listed in table 1. Here, only the results that are most relevant to the present study are highlighted.

26

The sustained negativity did not show a main effect of subsequent memory performance, but did

demonstrate an interaction between subsequent memory performance and word order [F(1, 11) =

8.16, p = .02, ηp2 = 0.43]. However, the ERPs to W1 were more positive than the ERPs to W2 for

both the R pairs [t(71) = 5.93, p < 0.001] and the N pairs [t(71) = 4.15, p < 0.001].

8 Discussion The present study extends previous investigations on cognitive association formation by

investigating ERP modulations that reflect episodic association formation, but not the encoding

of the individual items that the pair was composed of. The ERP modulations most relevant to the

purpose of the present study consist of those encoding-related ERP components that showed

significant amplitude differences between the first and second words of those pairs that were

subsequently recalled and no significant amplitude differences between the words of those pairs

that were not subsequently recalled. Two such ERP findings were observed in the present study:

the P460 and the positive slow wave, both of which were largest over the parietal scalp region

and discussed further below.

Interestingly, positive slow waves occurring over the parietal scalp region have been associated

with the completion of a task that is prompted by target detection (Ruchkin, Munson, & Sutton,

1982; Stuss & Picton, 1978). The slow wave is preceded by the P300 wave, which has been

related to stimulus evaluation (Kutas, McCarthy, & Donchin, 1977). To examine the generality

of the cognitive processes that parietal slow waves reflect, Garcia-Larrea and Cezanne-Bert

(1998) investigated whether parietal slow waves could be dissociated from the preparation or

execution of a motor response, updating of working memory, and response selection. To do so,

these investigators used a paradigm that consisted of two tasks. The first task required

participants to detect a target, which then prompted them to perform a second task that varied

between the experimental conditions. The results of the study suggest that parietal slow wave

positivities are related to the number of items retrieved from working memory, and can be

dissociated from processes related to motor response and response selection. In the context of the

present study, the observed positive slow wave may partially reflect the retrieval of the first and

second words of a pair from working memory. However, given that the ERP contrast examined

in the present study (the double difference) differentiated those pairs that were subsequently

recalled from those that were not, in addition to differentiating ERPs elicited by the first and

27

second words of a pair, the observed slow wave also likely reflects processes that are

supplementary to retrieval from working memory. These supplementary processes likely reflect

processes related to episodic association formation.

In a study by Caplan and colleagues (2009), where the neuroelectric correlates of association

formation and list learning were differentiated, the result of a multivariate analysis showed a

latent variable reflecting a significant subsequent memory effect for pairs but not lists. This

latent variable was prominent over posterior electrode sites and somewhat left-lateralized. The

timing of the latent variable overlapped primarily with the time window of the slow wave, but

also showed some overlap with the early and late positive components. Additionally, Yovel and

Paller (2004) found a Dm effect for associative information bilaterally over the parietal scalp.

The P460 and positive slow wave observed in the present study are consistent with the findings

of these past ERP studies and are thought to reflect association formation but not the encoding of

the individual items that make up a pair.

Based on the results of the present study alone, it is unclear whether the observed P460 and

positive slow wave correspond to the same or different cognitive processes. Additionally, it is

unclear whether the observed P460 and positive slow wave reflect memory processes that

differentiate episodic association formation from item encoding as opposed to different levels of

the same processes that underlie both item encoding and episodic association formation.

However, considering the results of past studies, the positive slow wave observed in the present

study likely reflects brain responses underlying the formation of associative bonds between the

first and second words of a pair. The observed P460, on the other hand, likely reflects brain

responses underlying the processing of the second word as the completion of the pair, which is

regarded as being necessary for association formation to occur, and may have lead to the positive

slow wave that followed. Study 2 was conducted to investigate further processes underlying

episodic association formation that occurred after the time window investigated in the present

study.

28

Chapter 3 An investigation of the ERP correlates of episodic association

formation (Study 2)3 The present study continued the investigation on episodic association formation. Whereas the

previous study focused on ERPs recorded during the presentation of both the first and second

words of a pair, the present study focused only on ERPs recorded during a later time window,

after the presentation of the second word. Previous work has shown that the pattern of ERPs

recorded during this latter time window varies as a function of subsequent paired-associate

recall. For example, in the study by Kim et al. (2009) participants’ ERPs were recorded as they

studied pairs of words with each word of a pair presented sequentially. A larger positive ERP

deflection was observed after the presentation of the second word for subsequently recalled pairs

compared to subsequently not-recalled pairs. This Dm effect was largest over the fronto-central

scalp region, with a slight right hemisphere bias, and was referred to as the LW, as mentioned in

the Introduction (Chapter 1). In light of the timing of the LW, combined with its amplitude

pattern, it was thought to reflect processes engaged in episodic association formation.

Other studies, however, have shown late-occurring Dm effects for paired-associates over the

frontal (e.g., Cheng & Rugg, 2010; Weyerts et al., 1997), as opposed to fronto-central, scalp

region. Differences in the DM scalp distribution observed for paired-associates may reflect

differences in the depth of encoding that was engaged during association formation or

differences in the depth of encoding required for successful retrieval on the subsequent memory

test. Whereas the former would be affected by factors including study instructions, the latter

would be affected by factors including the length of the retention interval (RI) and the selected

memory test. Deeper encoding may be reflected in the ERP signal by a Dm effect that is largest

over the frontal, as opposed to fronto-central, scalp region, with a slight right hemisphere bias.

For example, Weyerts and colleagues (1997) found a late-occurring Dm effect for pairs of words

that was largest over the right frontal scalp, with a larger positive slow wave elicited by

subsequently recalled pairs compared to subsequently not-recalled pairs. Interestingly, this Dm

3 This study was presented as part of a conference talk [Kim, A.S.N. (February 12, 2012). Neuroelectric evidence for cognitive association formation: An event-related potential investigation. Talk presented at the Winter Conference on Neural Plasticity, 24th Annual Meeting, Frigate Bay, St. Kitts].

29

effect was only found when an associative encoding task was used, and no Dm effects were

found when a non-associative encoding task was used. Similarly, Cheng and Rugg (2010) also

found a late-occurring frontal Dm effect for pairs of words, with a larger positive slow wave

elicited by pairs that were subsequently recognized compared to pairs that were missed (not

subsequently recognized). In this investigation, the study and test phases were separated by an

interval of about five minutes, during which participants performed a distractor task and rested.

During the test phase participants had to distinguish between old and recombined pairs, which

requires access to detailed information about the word pairings.

In contrast to the studies described above, in our previous study (Kim et al., 2009) we found a

Dm effect for pairs of words that occurred late and broadly over the fronto-central scalp.

Subsequently recalled pairs elicited a larger positive slow wave, which we referred to as the ‘late

wave’ or LW for short, compared to subsequently not-recalled pairs. In this study, participants

were not instructed to use a specific type of encoding strategy and they performed a self-paced

distractor task for a maximum of 32 seconds between the study and test phases. During the test

phase, participants were asked to recall the appropriate word pairings when they were given one

word of a pair as a cue. Thus, it could be that participants in the study by Kim et al. (2009) did

not have to encode the paired-associates to the same degree of depth, compared to participants in

the two studies described above, for successful retrieval on the subsequent memory test. For

example, Cheng and Rugg (2010) used a longer RI of 5 minutes compared to the maximum of 32

seconds used in our previous study (Kim et al., 2009), resulting in more opportunity for

interference and memory trace decay to occur. Thus, differences in the depth of processing

required for successful memory performance across different paradigms could explain why Dm

effects with different scalp topographies have been found for paired-associates.

It is also important to note that the topographic differences (frontal vs. fronto-central) described

above could be due to differences in the re-referencing procedures used as opposed to differences

in the encoding of information. For example, whereas Cheng and Rugg (2010) referenced their

data to the average of the mastoid electrodes, in our previous study (Kim et al., 2009) we used an

average-reference montage. The present study, as described further below, addresses this issue of

whether the topographic differences described above can be accounted for, at least in part, by

differences in how information is processed and/or encoded as opposed to being a consequence

of differing analysis procedures across studies.

30

The purpose of the present study was to investigate the effect of RI on Dm effects for paired-

associates that occur late during the ERP trial. Increasing the RI would require the associations

formed during the study phase to “survive” longer through time and through more interference

for successful paired-associate recall. The premise of the present study is that associations

formed during the study phase are more likely to survive through a longer RI when deeper

encoding processes are engaged. Based on this premise, we expected to find a prominent Dm

effect over the frontal scalp, with a slight right hemisphere bias, when a longer RI was used

compared to our previous study (Kim et al., 2009). Additionally, we expected to find a Dm effect

with a broad fronto-central scalp distribution, with a slight right hemisphere bias, when a RI

similar to the one used in our previous study (Kim et al., 2009) was employed. To investigate

these questions, two levels of RI were used in the present study, corresponding to the

“ShortDelay” and “LongDelay” conditions, as described below. In all other regards, the

experimental design used in the present study followed the one used in our previous study (Kim

et al., 2009). Qualitative differences between the scalp distributions of the Dm effects observed

for the ShortDelay and LongDelay conditions would support the notion that different encoding

processes were engaged across the two RIs used in the present study; moreover, such a finding

would suggest that ERPs recorded during encoding can be used as an objective index of depth of

processing.

The hypotheses for the present study were as follows:

(1) In the LongDelay condition, a prominent Dm effect will be observed over the frontal scalp

region, with a slight right hemisphere bias. This Dm effect will be sustained and occur late

during the time window of the LW. Larger, positive ERP amplitudes will be elicited by

subsequently recalled pairs compared to subsequently not-recalled pairs.

(2) In the ShortDelay condition, a Dm effect will be observed broadly over the fronto-central

scalp region corresponding to the LW observed by Kim et al. (2009). The amplitude of the LW

will be larger for subsequently recalled pairs compared to subsequently not-recalled pairs.

31

9 Methods

9.1 Participants ERP averages were obtained from sixteen healthy, young adults. The data from three participants

were discarded because there were too few trials in one or more of the conditions to allow for

ERP analysis. Consequently, the ERP averages were obtained from 13 participants (7 female;

mean age: 25 years, range: 20–30; first language: English).

9.2 Design Two levels of RI was used: the “ShortDelay” vs. “LongDelay” conditions, as outlined below in

the description of the procedure. The ERP data recorded during the study phase served as the

dependent variable.

9.3 Materials The pool of experimental words consisted of 384 English nouns of two syllables. This pool was

divided into 8 lists that were each composed of 24 pairs of words. In contrast to Study 1, the

words of a pair were not semantically related. Moreover, the word pool was not selected from

any particular groupings of semantically related words. None of the pairs appeared in more than

one list for a given participant, and the pairings never changed. The pairs were counterbalanced

across participants in the experiment so that each pair appeared equally frequently in both the

LongDelay and ShortDelay RI conditions.

9.4 Procedure The experimental procedure used in the present experiment is summarized in figure 5. Each

participant took part in one experimental session, which consisted of 8 study/test cycles. Each

study/test cycle consisted of three phases. During the first phase, participants studied a list of 24

pairs of words (encoding). The words of each pair were presented sequentially. Participants were

instructed to try to remember the pairs and were told that their memory for the pairs in the list

would be tested. The ERP trial (figure 1) corresponded to the presentation of a pair of words

during encoding. It started with a blank screen for 500 ms, followed by central fixation stimulus

‘+’ which served as a warning and lasted for 500 ms. The first word of a pair was then presented

32

for 1000 ms, followed by a blank screen for 200 ms, and the second word of a pair for 1000 ms.

The inter-trial interval varied randomly between 1000 to 3000 ms.

After all 24 pairs of a list were presented, participants completed the second phase of the

study/test cycle, where they solved eight arithmetic problems of the form A+B+C as a distractor

task. A, B and C were randomly selected digits from 0 to 9. Each equation was presented on the

screen for 3750 ms, followed by a 250 ms blank screen. Within this 4000 ms period, participants

were asked to respond aloud with the answer, as quickly and accurately as possible. After the

participant gave their response, the experimenter pressed a button allowing the participant to

move on to the next arithmetic problem.

During the third phase of a study/test cycle, which occurred immediately after participants

finished solving the arithmetic equations, participants were tested on paired-associate recall for

half of the pairs presented during the preceding study phase (12 of the 24 pairs). The first word

of a pair was always used as a cue to recall the word that it was paired with. During the test

phase, participants were presented with a central fixation stimulus (+) for 500 ms, followed by a

cue word for 7000 ms and then a black screen for 500 ms. Participants were asked to recall aloud

the word that had been presented immediately after the cue and the experimenter coded whether

the response was correctly recalled or not. Both an incorrect response and the absence of any

response within the time frame of the cue word presentation and the following blank screen

(7000 ms + 500 ms = 7500 ms) were classified as not-recalled. The presentation of the next cue

word was triggered when the experimenter made a button press to indicate the participant’s

response or when the time limit of 7500 ms was reached.

In the ShortDelay condition, the length of the RI was equivalent to duration of the second phase

of the study/test cycle, during which participants solved eight arithmetic equations. Since

participants had 4 s to solve each of the eight arithmetic equations, this second phase of the

study/test cycle lasted no longer than 32 seconds. At the end of the session, after participants had

completed the eight study/test cycles and the EEG cap was removed, participants were tested on

paired-associate recall for all of the pairs that they had not been tested on earlier during the

session (the 12 pairs from each of the eight study/test cycles that they were not tested on after the

ShortDelay). The interval of time separating the study phase of each list cycle and this latter final

memory test, which occurred at the end of the session after the EEG cap was removed,

33

constituted the RI for the LongDelay condition. The interval of time between the end of the last

study/text cycle and the beginning of the final memory test was about 40 minutes and no less

than 30 minutes. In the LongDelay condition, participants were tested on their memory for the

pairs of words according to the study/test cycle that it corresponded to: pairs that were presented

during the first study/test cycle were the first to be tested on, pairs presented during the second

study/test cycle were tested on next during the final memory test, and so forth. In the LongDelay

condition, participants were only tested on those pairs that they were not tested on in the

ShortDelay condition. It is important to note that pairs that were not successfully recalled in the

LongDelay condition could have potentially been recalled if they had been assigned to the

ShortDelay condition. However, to avoid any potential effects of repeated testing (Karpicke &

Roediger, 2008), a given pair was assigned to either the ShortDelay or LongDelay condition, but

not both.

Each session began with a practice block to familiarize the participants with the experimental

task. Each of the participants confirmed that they understood the task upon completion of the

practice block. Participants then completed the 8 study/test cycles, each of which corresponded

to one of the 8 lists of paired-associates. After every second study/test cycle, participants

received a short break (approximately 5 minutes).

9.5 Behavioural data analysis Paired-associate recall performance, measured as the percentage of correctly recalled pairs, was

calculated separately for the two RI conditions.

9.6 Electrophysiological data analysis For each participant, continuous EEG files were analyzed using BESA 5.1.8. The EEG data were

epoched into 4200 ms segments beginning 500 ms before the fixation stimulus (+) and lasting

1000 ms after the offset of the second word in the pair. The epoch was baseline corrected to the

500 ms segment preceding the fixation stimulus at the beginning of the epoch, when a blank

screen was presented.

The ERPs recorded during study were sorted based on whether the corresponding pair of words

was subsequently recalled or not during the test of paired-associate recall. This sorting of study

phase ERPs based on subsequent memory performance was done separately for the two RI

34

conditions (ShortDelay and LongDelay), yielding the following four categories of ERPs: (a)

subsequently recalled pairs from the ShortDelay condition (ShortDelay-R); (b) subsequently not-

recalled pairs from the ShortDelay condition (ShortDelay-N); (c) subsequently recalled pairs

from the LongDelay condition (LongDelay-R); (d) subsequently not-recalled pairs from the

LongDelay condition (LongDelay-N). Individual participant waveforms were created by

averaging the ERPs corresponding to these four categories. The mean trial counts going into the

grand-mean waveforms were 54 for ShortDelay-R (range: 41-68), 26 for ShortDelay-N (range:

11-42), 19 for LongDelay-R (range: 11-34) and 64 for LongDelay-N (range: 30-80).

9.7 Mean amplitude analysis To examine the ERP data of interest, the results of past studies on Dm effects for paired-

associates were used to guide the selection of electrodes for data analyses. Since past studies

have consistently shown Dm effects for paired-associates over the frontal scalp regions with a

slight right hemisphere bias (Cheng & Rugg, 2010; Weyerts et al., 1997), ERPs recorded from

frontal electrodes (F1, Fz, F2, F4) during the time window of the LW (between 1 to 1.6 seconds

after the presentation of the second item of a pair) were analyzed for the LongDelay condition. A

repeated-measures ANOVA was conducted on these data using subsequent memory performance

(R vs. N) and electrodes (F1, Fz, F2, F4) as factors. The results of Kim et al. (2009) were used to

guide further examination of the LW in the ShortDelay condition of the present study. Kim et al.

demonstrated a fronto-central distribution for the LW, with a slight right hemisphere bias,

between 1 to 1.6 seconds after the presentation of the second item of a pair. Consequently, in the

present study ERPs recorded at fronto-central electrodes Fz, F2, FCz, FC2, Cz and C2 during

this same time window were investigated. A repeated-measures ANOVA was conducted on

these data using subsequent memory performance (R vs. N) and electrodes (Fz, F2, FCz, FC2,

Cz and C2) as factors.

An additional analysis was conducted using the same electrodes for the data collected in the

LongDelay and ShortDelay conditions during the time window of the LW (between 1 to 1.6

seconds after the presentation of the second item of a pair). This was done to ensure that the

results of the two analyses described above were not due to the selection of different electrode

sets. For this additional analysis, a repeated-measures ANOVA was conducted using RI

(LongDelay vs. ShortDelay), subsequent memory performance (R vs. N), hemisphere (left vs.

35

right), location (anterior vs. posterior), and electrode as factors. The electrodes used in this

analysis corresponded to four quadrants: 1) a left anterior quadrant – electrodes F1, F3, FC1,

FC5; 2) a right anterior quadrant – electrodes F2, F4, FC2, FC6; 3) a left posterior quadrant –

electrodes CP1, CP5, P1, P3; and 4) a right posterior quadrant – electrodes CP2, CP6, P2, P4.

The Dm scalp distributions for the ShortDelay and LongDelay conditions were compared

statistically to provide further insight into whether the Dm effects observed for the two

conditions reflected different encoding processes. To evaluate any differences between the Dm

scalp distributions, differences between the ERPs elicited by subsequently recalled and

subsequently not-recalled pairs were first derived separately for the ShortDelay and LongDelay

conditions. This resulted in two difference waveforms: subsequently recalled minus subsequently

not-recalled pairs in the ShortDelay condition [ShortDelay(R-N)] and the LongDelay condition

[LongDelay(R-N)]. To compare scalp distributions, the EEG data were first normalized to

eliminate amplitude differences between the two conditions being compared – this allowed the

analysis to be focused on the pattern of voltage differences across electrode locations. The ERPs

were scaled using a normalization procedure similar to the one that was developed by McCarthy

and Wood (1985) and used by others (e.g., Crowley, Trinder, & Colrain, 2002) to compare scalp

distributions associated with different experimental conditions, different groups of participants

and different ERP components. In the present study,mean amplitude values corresponding to

each electrode location were assessed to identify the minimum and maximum values. This was

done separately for each condition at the level of individual participants. The minimum mean

amplitude value was then subtracted from the mean amplitude value of each electrode, and the

resulting differences were divided by the difference between the maximum and minimum values.

The normalized data were used to analyze differences between the scalp distributions of Dm

effects observed for the ShortDelay and LongDelay conditions [ShortDelay(R-N) and

LongDelay(R-N), respectively] along the anterior-posterior axis during the time window of the

LW (1 to 1.6 seconds after the presentation of the second item of a pair). The analysis was

conducted on ERPs recorded from four fronto-central electrodes along the anterior-posterior axis

that were located immediately to the right of the midline (electrodes F2, FC2, C2, CP2), using a

two [(ShortDelay(R-N) vs. LongDelay(R-N)] by 4 (electrode) repeated measures ANOVA

model. A significant interaction term between the two factors was considered to indicate scalp

distribution differences.

36

10 Results

10.1 Behavioural data Paired-associate recall performance was calculated as a function of RI condition (ShortDelay and

LongDelay). The percentage of paired-associate recall was higher for the ShortDelay condition

(M = 68.38, SE = 3.03) compared to LongDelay condition (M = 24.62, SE = 3.13, t(12) = 19.96,

p < .001.

10.2 ERP data An overview of the group level ERP data is shown in figure 6. As mentioned above, differences

between subsequently recalled and subsequently not-recalled pairs were derived separately for

the ShortDelay and LongDelay conditions, resulting in two difference waveforms:

ShortDelay(R-N) and LongDelay(R-N). The ShortDelay(R-N) waveform recorded at

representative electrode FC2, where this difference was largest, is shown in figure 7a. The

LongDelay(R-N) waveform was largest at electrode F2 and is shown in figure 8a. Figures 7b and

8b show the topographical distributions corresponding to the ShortDelay(R-N) and

LongDelay(R-N) difference waveforms, respectively, for the interval between 2200 and 2800

ms. The LongDelay(R-N) difference was largest over the frontal scalp region and highly focused

at electrode F2 between 2200 and 2800 ms, whereas the ShortDelay(R-N) difference was more

evenly distributed over fronto-central electrode sites.

For the ERP data collected in the ShortDelay condition, the results of a repeated-measures

ANOVA (factors of subsequent memory performance and electrode) that was conducted on the

fronto-central electrode sites (Fz, F2, FCz, FC2, Cz, C2) showed a main effect of subsequent

memory performance [F(1, 12) = 4.82, p = .05, ηp2 = .29]. The mean amplitude of the R pairs (M

= .94, SE = .27) was larger than and the N pairs (M = .15, SE = .28). The interaction between

subsequent memory performance and electrode site was not significant [F(5, 60) = .03, p = 1.00,

ηp2 = .002]. For the ERP data collected in the LongDelay condition, the results of a repeated-

measures ANOVA (factors of subsequent memory performance and electrode site) that was

conducted on the frontal electrode sites (F1, Fz, F2, F4) showed a main effect of subsequent

memory performance [F(1, 12) = 19.26, p = .001, ηp2 = .62]. The mean amplitude of the R pairs

(M = 1.76, SE = .25) was larger than and the N pairs (M = .35, SE = .24). There was also a

37

significant interaction between electrode site and subsequent memory performance [F(3, 36) =

3.47, p = .03, ηp2 = .22].

The additional repeated-measures ANOVA that was conducted on ERP data collected in the

LongDelay and ShortDelay conditions, using the same electrodes (F1, F3, FC1, FC5, F2, F4,

FC2, FC6, CP1, CP5, P1, P3, CP2, CP6, P2, P4), demonstrated a main effect of RI [F(1, 12) =

7.86, p = .02, ηp2 = .40], subsequent memory performance [F(1, 12) = 11.80, p = .005, ηp

2 = .50],

and borderline significance for hemisphere [F(1, 12) = 3.51, p = .09, ηp2 = .23]. Most

importantly, there was a significant three-way interaction between RI, subsequent memory

performance and hemisphere [F(1, 12) = 6.74, p = .02, ηp2 = .36]. However, the four-way

interaction between RI, subsequent memory performance, hemisphere and location (anterior vs.

posterior) was not significant [F(1, 12) = 1.03, p = .33, ηp2 = .08], suggesting that the data

collected in the ShortDelay and LongDelay conditions did not differ in terms of anterior vs.

posterior distribution. A follow-up ANOVA restricted to the data corresponding to subsequently

recalled pairs, and recorded from the electrodes located over the right anterior scalp region

(electrodes F2, F4, FC2, FC6), demonstrated an effect of RI [F(1, 12) = 10.65, p = .007, ηp2 =

.47]. The mean amplitude of pairs in the LongDelay condition (M = 1.26, SE = .20) was larger

than the mean amplitude of pairs in the ShortDelay condition (M = .25, SE = .20). However, the

results of a follow-up ANOVA restricted to the data corresponding to subsequently not-recalled

pairs, and recorded from electrodes located over the left anterior scalp region (electrodes F1, F3,

FC1, FC5) did not demonstrate a significant effect of RI [F(1, 12) = .97, p = .34, ηp2 = .08].

Together, the follow-up analyses suggest that the three-way interaction demonstrated between

RI, subsequent memory performance and hemisphere reflected differences in mean amplitude

between subsequently recalled pairs in the LongDelay and ShortDelay conditions over the right

anterior scalp region, which is consistent with the results of the two analyses described above.

The scalp distributions of the Dm effects observed for the ShortDelay and LongDelay conditions

[ShortDelay(R-N) and LongDelay(R-N), respectively] during the time window of the LW (1 to

1.6 seconds after the presentation of the second item of a pair) were compared using a repeated

measures ANOVA. The results of the ANOVA demonstrated a significant interaction between

the Dm effects of interest and electrode [F(3, 36) = 4.28, p = .01, ηp2 = .26], indicating that the

scalp distribution of Dm effects observed for the ShortDelay and LongDelay conditions differed

statistically. A significant quadratic trend [F(1, 12) = 8.54, p = .01, ηp2 = .42] was found for this

38

interaction. Figure 9 shows the normalized mean amplitudes of the Dm effects observed for the

ShortDelay and LongDealy conditions. This figure shows that the quadratic trend was driven by

the ShortDelay(R-N) difference, which increased in mean amplitude from the most anterior

electrode (F2) to the central electrode (C2) and then decreased in mean amplitude at the centro-

parietal electrode (CP2). Figure 9 shows that for the LongDelay(R-N) difference, the mean

amplitude was largest at the most anterior electrode (F2) and decreased as the electrodes reached

more posterior scalp locations.

11 Discussion The present study investigated Dm effects for paired-associates across two RIs. As mentioned

above, Kim et al. (2009) found a Dm effect for paired-associates that had a broad fronto-central

scalp distribution, whereas other studies have shown a late-occurring Dm effect for paired-

associates over the frontal scalp region (Cheng & Rugg, 2010; Weyerts et al., 1997). To

reconcile these two sets of findings, the present study examined whether a frontal Dm effect

would be found using a modified version of the experimental design used by Kim et al. (2009),

where a longer RI was used. Increasing the RI required the memory trace for a given pair to

survive a longer period of time and through more interference for successful paired-associate

recall. The premise of the present study was that a given item would be more likely to survive a

longer RI in cases where deeper, more elaborative encoding processes were engaged. Consistent

with past studies, the results of the present study showed a late-occurring Dm effect that was

prominent over the frontal scalp region, with a right hemisphere bias, for the longer RI.

However, for the shorter RI, which was similar to the one used by Kim et al. (2009), a late-

occurring Dm effect was found broadly over the fronto-central scalp, replicating the findings of

Kim and colleagues. Consistent with these findings, the analysis that was conducted on ERP data

collected in both RI conditions, using the same electrodes, demonstrated a larger positive mean

amplitude for subsequently recalled pairs in the longer RI condition, compared to the shorter RI

condition, over the right anterior scalp region.

The scalp distributions for the late-occurring Dm effects observed in the short and long RI

conditions were compared statistically to provide further insight into whether they differed. In

doing so, the ERP data were first scaled to eliminate amplitude differences between the two

conditions being compared so that the analysis would be focused on the pattern of voltage

39

differences across electrode locations. The results of this analysis showed that the Dm scalp

distributions differed significantly across RI conditions, suggesting that the neural mechanisms

underlying encoding were at least partially non-overlapping across the two RI conditions. It is

unclear whether this difference in Dm scalp distributions for the short and long RI conditions

reflects engagement of different memory systems and/or processes (Schacter & Tulving, 1994;

Tulving, 1999). Since pairs that were retrieved in the long RI condition would have also likely

been retrieved in the context of the short RI condition, it seems reasonable that the Dm scalp

distribution for the short RI would reflect memory systems and/or processes that were also

engaged in successful encoding in the long RI condition. However, additional memory systems

and/or processes, which were not engaged in the short RI condition, could have contributed to

the encoding of pairs in the long RI condition. Moreover, other factors could have contributed to

differences in performance, and the corresponding ERP data, between the two RI conditions. For

example, differences between the two RI conditions could have been due to the higher load of to-

be-recalled items for the paired-associate recall test in the longer RI condition, compared to the

paired-associate recall test that occurred at the end of each list in the shorter RI condition.

Additionally, intrusions of items in the longer RI condition could have strengthened or weakened

subsequent recall for those items. Furthermore, performance in the longer RI condition could

have been influenced by prior testing on items from the same study lists that were assigned to the

shorter RI condition; along the same lines as the retrieval-induced forgetting effect (Anderson,

Bjork, & Bjork, 1994) it could be that the testing of items in the shorter RI condition weakened

subsequent recall of items in the longer RI condition that were presented in the same list. Future

research will have to be conducted to investigate these possibilities further.

Based on the results of past studies (e.g. Cheng & Rugg, 2010; Weyerts et al., 1997), it seems

likely that the prominent frontal Dm scalp distribution that was observed during the long RI

condition reflected, at least to some extent, encoding processes that were deeper and more

elaborative compared to the processes engaged during the short RI condition, for which a LW

Dm effect was observed. It could be that the right frontal Dm effect observed in the present study

reflected additional elaborative encoding that served to strengthen the association formed, which

could in turn have been reflected by ERPs recorded over the fronto-central scalp. This was

investigated further in Study 4 (Chapter 5).

40

Mangels et al. (2001) showed a larger late-occurring, sustained ERP positivity over the right

frontal scalp for words that were subsequently recalled with a remember judgment compared to

words that were subsequently recognized with a familiarity judgment or missed altogether. Other

studies have shown similar findings for right frontal positivity (Friedman & Trott, 2000; Schott,

Richardson-Klavehn, Heinze, & Düzel, 2002), suggesting that this Dm effect reflected deep

encoding that lead to subsequent remembering. However, another possibility proposed by

Blanchet, Gagnon and Bastien (2007) is that late-occurring, sustained right frontal activity

reflects self-initiated semantic organization. This interpretation was based on the results of their

study where participants’ ERPs were recorded as they studied words in three conditions that

differed in the degree of semantic organization. In an “Unrelated” condition, the words did not

belong to any obvious semantic category. In the “Spontaneous” and “Guided” conditions,

however, the words belonged to one of four semantic categories. In the Spontaneous condition,

participants were not informed about this semantic organization. In the Guided condition,

participants were instructed to mentally group related words according to the category labels that

were provided to them. In their analyses of the data, Blanchet et al. (2007) examined the study-

phase ERPs elicited by subsequently recognized words only. The relevant finding was that words

from the Spontaneous condition elicited larger late-occurring, sustained positivity over the right

frontal scalp region compared to words from the Unrelated condition. This finding led Blanchet

et al. to suggest that the observed late positivity reflects self-initiated semantic organization.

However, it remains unclear what encoding strategies participants used in the Spontaneous

condition, and the degree to which they engaged in semantic organization. A finding that would

have strengthened the semantic organization interpretation would have been an additional

difference observed between the relevant ERP data collected for the Unrelated condition and the

Guided condition, where participants were instructed to perform semantic organization. It would

be interesting to know if different Dm effects would be elicited across the semantic organization

conditions.

As mentioned in the Introduction section and in the beginning of the present chapter, Weyerts

and colleagues (1997) recorded participants’ ERP activity as they encoded pairs of words using

an associative and non-associative task. For the associative task, participants had to judge

whether the words of a pair were semantically related (e.g., CELLAR – ROOF). For the non-

associative task, participants had to judge whether at least one of the words in the pair could be

41

associated with the colour white (e.g., LOAM – SWAN). Thus, both tasks required semantic

organization. However, only the associative task required participants to associate the words of a

pair with each other. Interestingly, a Dm effect was only found for those pairs that were encoded

using the associative task. This effect occurred late over the frontal scalp, with a right

hemisphere bias: subsequently recognized pairs elicited larger positive waveforms compared to

pairs that were not subsequently recognized. Consistent with the abovementioned results of

Blanchet et al. (2007), these results suggest that late-occurring right frontal positivity reflects

semantic organization to some degree and, more specifically, that Dm effects occurring late over

the right frontal scalp may reflect mechanisms engaged during associative binding that can be

promoted by semantic organization amongst items of a pair. As discussed further in the General

Discussion (Chapter 6), overall, the results of past studies suggest that late-occurring right frontal

activity reflect deep processing of information that supports later remembering. Additional

research will have to be conducted to examine in more detail the differences that underlie right

frontal Dm effects and LW Dm effects.

The LW Dm effect has been proposed to reflect the formation of episodic inter-item associations

(Kim et al., 2009), and the results of the present study are consistent with this proposal.

However, it is possible that the LW reflects mechanisms engaged during intra-item, as well as

inter-item, associative encoding. In addition to the binding that occurred between the two words

of a pair, in the present study binding could have also occurred between a given word and any

number of contextual features, including, for example, any internal thoughts, images, emotions

and memories that arose while studying the word. Thus, the LW may reflect binding, in a general

sense. Based on this interpretation, the LW Dm effect should also be observed for single items

that engaged elaborative encoding. The next study (Study 3) investigated whether encoding of

single words can elicit the LW Dm effect.

42

Chapter 4 An Investigation of the Neuroelectric Correlates of Primary

Memory and Secondary Memory (Study 3) According to James (1890), items in PM are those that have never left consciousness, whereas

items that have been absent from consciousness are retrieved from SM, which James also

referred to as “memory proper”. Since James, the conceptualization of PM and SM as distinct

memory stores has been pervasive in memory research (Craik, 1983). However, Tulving and

Patterson (1968) identified PM and SM with different types of retrieval mechanisms instead of

distinct memory stores. They also highlighted the importance of coding, which refers to the

encoding of additional information with each to-be-remembered item during its presentation.

They proposed that this additional information serves as sources of retrieval cues, which

provides access to information about the to-be-remembered item. Thus, more effective retrieval

cues lead to better access to the information in question. In line with this conceptualization of

PM and SM as different types of retrieval mechanisms and the importance of retrieval cues,

Craik (1983) has more recently suggested that the notion of information being transferred from

temporary storage (e.g., PM) to permanent storage (e.g., SM) is both misleading and

unnecessary. Instead, Craik suggested that memory should be conceptualized in terms of the

qualitative nature of the encoding processes and the compatibility between encoding and

retrieval operations, as outlined by the encoding specificity hypothesis (Tulving & Thomson,

1973).

The purpose of the present study was to investigate whether the amplitude of the LW increases

with the degree of elaborative encoding used for single words (or the degree to which the

processing of the words is enriched). The distinction between PM and SM items was used to

distinguish between items associated with elaborative coding (SM items) and items associated

with more shallow coding (PM items). This was based on past work that has shown that PM

items are recalled with lower probability on delayed memory tests compared to SM items (e.g.,

Craik, 1970).This differential accessibility of PM and SM items is thought to be due to

differences in the effectiveness of the types of retrieval cues that were encoded with the

respective items, where more elaborative coding resulted in more effective retrieval cues for SM

items and consequently better subsequent access compared to PM items.

43

In the present study, participants’ ERPs were recorded as they studied single words. To examine

the ERP correlates of encoding processes associated with PM and SM, subsequently recalled

words were decomposed further into components that corresponded to PM and SM using the

Tulving-Colotla classification method of PM and SM (Tulving & Colotla, 1970). Although items

that are part of PM and SM are not mutually exclusive, labelling specific items as ‘PM’ or ‘SM’

allowed for an initial investigation on neural mechanisms underlying coding of PM and SM

items. From this point onwards, ‘PM’ and ‘SM’ will be used to designate items that have been

labelled as belonging to PM and SM, respectively. According to the Tulving-Colotla (1970)

method, a recalled item is categorized as being part of SM if a critical number of other items, or

more, intervened between the presentation and recall of the item in question. It is important to

note that the intervening items can consist of subsequent study items, as well as participants’

responses. In the original paper, where the Tulving-Colotla method was first used (Tulving &

Colotla, 1970), a criterion of seven was used. The selection of this criterion was based on the

finding that when a criterion of seven was used, the average number of ‘PM’ items per list was

close to three, which was basically equivalent to the estimates of PM produced using other

measures of PM and SM (e.g., Tulving & Patterson, 1968). Following the original use of the

Tulving-Colotla method, the present investigation used a criterion of seven intervening items for

classification of SM items. The advantage offered by this method is that each recalled word can

be categorized specifically as being part of PM or SM. Others methods that also set out to tease

apart PM and SM, including the widely known method used by Waugh and Norman (1965), can

provide an average estimate of the proportion for PM and SM but cannot be used to identify

which specific items belong to either type of memory.

The Tulving-Colotla method (1970) has been used widely in clinical settings (Alexander, 2003;

Gibson, Gondoli, Flies, Dobrzenski, & Unsworth, 2010; Nilsson & Nilsson, 2009), as well as in

the context of experimental research on memory. For example, Unsworth and colleagues (2010)

extracted ‘PM’ and ‘SM’ components from participants’ performance on immediate free recall

using the Tulving-Colotla method. This was done to examine the extent to which the PM and SM

estimates would account for separate variance in participants’ working memory capacity (WMC)

using structural equation modeling. The results showed that the ‘PM’ and ‘SM’ factors each

accounted for unique variance in WMC and that the two factors were not correlated, suggesting

that both PM and SM contribute independently to WMC. More generally, these findings also

44

suggest that PM and SM are two independent sources of variance in immediate free recall.

Consistent with these findings, past behavioural studies have shown that ‘PM’ and ‘SM’ items

are affected by different experimental manipulations. For example, Tulving and Colotla (1970)

found that when trilingual speakers were presented with a series of words for free recall, the

number of languages incorporated into the list (i.e., unilingual, bilingual, trilingual lists) affected

‘SM’ but not ‘PM’. Other factors that have been shown to affect ‘SM’ but not ‘PM’ include age

and list length (Craik, 1968), as well as acoustic and semantic similarity (Craik & Levy, 1970).

In contrast ‘PM’, but not ‘SM’, has been shown to be grossly affected by a filled interval

preceding recall (Glanzer & Cunitz, 1966). Additionally, Craik (1970) has shown that ‘PM’, but

not ‘SM’, is affected by input mode (auditory versus visual presentation) and output mode

(spoken versus written).

To add to the abovementioned findings, neuroimaging can be used to investigate further any

neurocognitive processes that may be differentially engaged for PM and SM. The present study

investigated whether the amplitude of the LW increases with the degree of elaborative processes

engaged during encoding. Items that were subsequently categorized as part of SM (‘SM’ items)

were thought to engage more elaborative encoding compared to items that were subsequently

categorized as PM (‘PM’ items), and subsequently not-recalled items were thought to engage the

least amount of elaborative encoding. The following hypotheses were examined in the present

study:

(1) Items that are subsequently categorized as part of SM (‘SM’ items) will elicit the largest LW,

followed by items that are subsequently categorized as part of PM (‘PM’ items). Items that are

not subsequently recalled are expected to elicit the smallest LW amplitude.

(2) The scalp distribution of the Dm effect observed for ‘SM’ items during the period of the LW

will be largest over the fronto-central scalp region as observed in Study 2 for pairs of words,

demonstrating a LW Dm effect for single words when elaborative encoding processes are

engaged.

45

12 Methods

12.1 Participants ERP averages were obtained from 12 healthy, young adults (7 female; mean age: 23 years,

range: 18–30; first language: English).

12.2 Materials The pool of experimental words consisted of 288 English words (seven or eight letters long). The

words were divided into 16 lists: 8 long lists, composed of 24 words each; and 8 short lists,

composed of 12 words each. None of the words appeared in more than one list for a given

participant, and the words were counterbalanced across participants so that each word appeared

equally frequently in both the long and short lists. The order in which the words were presented

was also counterbalanced across participants.

12.3 Design Lists of 12 words and 24 words were used. The ERP data recorded during the study phase served

as the dependent variable.

12.4 Procedure The experimental procedure used in the present study is summarized in figure 10. Each

participant was tested in one experimental session. At the start of the session, participants were

given a short practice block so that they would become familiar with the experimental task.

Participants were instructed to pay careful attention to all of the words in each list and to recall as

many as they could, in any order that the words occurred to them. Each participant studied and

was tested on free recall for each of the 16 lists, with short breaks after every fourth list. The

length of the lists used alternated between 12 words and 24 words. The study words of each list

were presented one at a time on a computer screen. The ERP trial (figure 11), which

corresponded to the presentation of a study word, began with a central fixation cue (+) for 500

ms, followed by a study word for two seconds, and then a blank screen for 500-1000 ms.

At the end of each list, participants were allotted time for recall that was equivalent to two

seconds for each word presented in the list. Thus participants were given 48 seconds to recall the

46

long lists composed of 24 words, and 24 seconds to recall the short lists composed of 12 words.

Participants wrote out their responses on a sheet of paper provided by the experimenter after

each list.

12.5 Behavioural data analysis Recalled words were categorized as being part of PM or SM using the Tulving-Colotla

classification method of PM and SM (Tulving & Colotla, 1970). In this method, each recalled

word is classified as being part of PM or SM based on the length of its intra-trial retention

interval (ITRI). The length of ITRI for any given item is determined based on the number of

presentations and recall of other items occurring between the presentation and recall of the given

item. For example, if a participant is presented with items A, B, C, D, and E, and the participant

recalls items E, C, and D, in this sequence, the ITRIs for these three items are 0, 3, and 3,

respectively. Items that have an ITRI of seven or less are classified as being part of PM (‘PM’

items), whereas items with an ITRI of eight of more are classified as being part of SM (‘SM’

items). As explained above, this criterion of an ITRI of seven or less for ‘PM’ items was selected

based on the finding that when this criterion was used, the average number of ‘PM’ items per list

was basically equivalent to the estimates of PM produced using other measures of PM and SM.

The total number of items recalled is the sum of the number of ‘PM’ and ‘SM’ items. Memory

performance was measured as the mean number of recalled items classified as part of PM and

SM (‘PM’ and ‘SM’ items) per list.

12.6 Electrophysiological data analysis For each participant, continuous EEG files were analyzed using BESA 5.1.8. The EEG data was

epoched into 2500 ms segments beginning with a fixation stimulus (+) that lasted 500 ms,

followed by the word presentation that lasted 2000 ms. The epoch was baseline corrected to the

200 ms segment preceding the onset of the word presentation. The ERPs recorded during study

were sorted based on whether the corresponding words were classified as PM or SM items, or

not recalled, yielding three categories of ERPs corresponding to: (a) ‘PM’; (b) ‘SM’; (c) not

recalled (NR) items. The mean trial counts going into the grand-mean waveforms were 36 for

‘PM’ (range: 26-53), 70 for ‘SM’ (range: 45-131), and 170 for NR (range: 99-199).

47

12.7 Mean amplitude analysis To examine the ERP data of interest, the findings on the LW from Study 2 of the present thesis

and Kim et al. (2009) were used to guide the analyses of the present study. In the present study, a

repeated-measures ANOVA was conducted to examine ERPs recorded at fronto-central

electrodes (Fz, F2, FCz, FC2, Cz, and C2) between 1 and 1.6 seconds after the presentation onset

of a word. Subsequent memory categorization (NR, ‘PM’, ‘SM’) and electrode site (Fz, F2, FCz,

FC2, Cz, C2) were used as factors. A separate repeated measures ANOVA was conducted to

examine whether the LW would show a Dm effect for items that were subsequently categorized

as part of ‘SM’, using subsequent memory categorization (NR, ‘SM’) and electrode site (Fz, F2,

FCz, FC2, Cz, C2) as factors.

13 Results

13.1 Behavioural data The results of a repeated-measures ANOVA (factors of memory classification and list length)

showed a main effect of memory classification [F(1, 11) = 28.03, p <.001, ηp2 =.72] and list

length [F(1, 11) = 12.76, p =.004, ηp2 =.54], as well as an interaction between these two factors

[F(1, 11) = 11.06, p =.007, ηp2 =.50]. Follow-up analyses showed that the average number of

‘PM’ items recalled per list in the ShortList (M = 2.50, SE = .15) and Longs lists condition (M =

2.17, SE = .21) did not differ statistically; t(11) = 1.30, p = .22. However, the average number of

‘SM’ items recalled per list for the LongList condition (M = 5.92, SE = .71) was significantly

larger than the average number of ‘SM’ items recalled for the ShortList condition (M = 3.50, SE

= .26), t(11) = -3.62, p = .004.

13.2 Mean amplitude data Figure 12 displays an overview of the group level ERP data that were recorded at multiple

electrodes. Differences between ‘PM’ and ‘SM’ were largest over the posterior and central scalp

regions. Over the central scalp region, ‘SM’ items elicited larger positive ERPs compared to

‘PM’ items (figure 12) starting around 800 ms after stimulus onset. Figure 13 shows the grand

average waveforms recorded from fronto-central electrodes (Fz, F2, FCz, FC2, Cz, and C2) that

were used for statistical analyses and figure 14 shows the topographical distributions for the Dm

effects observed for the ‘SM’ and ‘PM’ items. The results of the repeated-measures ANOVA

48

that investigated ERPs recorded at fronto-central electrode sites (Fz, F2, FCz, FC2, Cz, and C2)

between 1 and 1.6 seconds after the presentation onset of a word showed a significant effect of

subsequent memory performance [F(2, 22) = 3.95, p =.03, ηp2 =.26]. There was also a

significant linear trend for subsequent memory performance [F(1, 11) = 7.52, p =.02, ηp2 =.41],

with ‘SM’ pairs demonstrating the largest mean amplitude (M = 1.62, SE = .18) followed by

‘PM’ pairs (M = 1.46, SE = .24) and then NR pairs (M = 1.00, SE = .15). There was also a main

effect of electrode site [F(5, 55) = 6.55, p < .001, ηp2 =.37] and a significant interaction between

memory and electrode site [F(10, 110) = 2.62, p =.007, ηp2 =.19].

14 Discussion The present study investigated whether the amplitude of the LW increases with the degree of

elaborative encoding engaged for single items. This was confirmed by the results of the present

study, where participants studied and encoded single words. The recalled words were broken

down into ‘PM’ and ‘SM’ components to distinguish between specific words that engaged

elaborative encoding (items that were categorized as being part of SM) and words that engaged

more shallow encoding (items that were categorized as being part of PM) during the study phase.

The Dm effect that was observed for ‘SM’ words showed a fronto-central scalp distribution, with

a slight right hemisphere bias, as observed in the short RI condition of the previous study (Study

2) of the present thesis, where the subsequent memory analysis was based on participants’

paired-associate recall performance. Thus, the results of the present study demonstrate that the

LW Dm effect can be elicited by single items and is not unique to the encoding of inter-item

associations. The results of the present study, combined with the results of Study 2, support the

notion that the LW reflects binding, in a general sense (both inter-item and intra-item

associations). It could be that the differences in the LW amplitude observed for different degrees

of elaborative encoding reflect the involvement of different memory stores or systems. Although

this interpretation is in line with two-store models of memory (e.g., Shiffrin & Atkinson, 1969),

it is difficult to reconcile with the results of Tulving and Patterson (1968) who, as mentioned

above, identified PM and SM with different types of retrieval mechanisms.

In the abovementioned study by Tulving and Patterson (1968), participants studied and recalled

four types of lists. One list type was composed of unrelated words (“C lists”), whereas the

remaining list types consisted of four highly related words (e.g., FATHER, MOTHER,

49

BROTHER, SISTER) as well as unrelated words. The four highly related words were presented

in a cluster at the end of “E lists” and in the middle of “M lists.” In contrast, in the “D lists”, the

highly related words were distributed throughout the list instead of being presented in a cluster. It

was expected that participants would “chunk” or “unitize” the strongly related words, especially

if they were presented in a cluster, resulting in a “functional unit.” The unrelated words were

expected to be more difficult to unitize compared to the related words, and thus result in smaller

functional units. The results of the study showed that recall performance was best for lists in

which the highly related words were presented in the middle of the list (M lists), followed by

when they were presented at the end of the list (E lists), and then by lists in which the related

words were distributed throughout the list instead of being clustered. Recall performance was

lowest for lists composed entirely of unrelated words.

To assess recall of functional units, participants were given credit for each of the unrelated words

that they recalled, as well as when they recalled at least one of the related words. For example, if

a participant recalled four unrelated words and four related words from a list, they received credit

for recall of five functional units. Tulving and Patterson (1968) pointed out that if one were to

assume that unitization occurs amongst unrelated words then this method overestimates the

number of functional units recalled. However, the investigators reasoned that this would affect

all list types equally for the most part and they were interested in comparing the number of

functional units for the different list types. D lists were not included in this analysis, because the

assumption about unitization of related words did not seem to apply to this list type: in E lists

and M lists, related words were always recalled as a group, however, the related words from D

lists were frequently recalled in a distributed manner. Interestingly, the same number of

functional units was recalled from M lists and C lists, but was lower for E lists. However, when

the recall of four related words was assumed to be approximately equivalent to the recall of three

functional units, the number of functional units recalled for the E lists and C lists became

roughly equivalent. Thus, Tulving and Patterson suggested that participants treated the related

words in E lists more like individual words as opposed to members of a functional unit.

If PM and SM are conceptualized as two different memory stores or systems, where items can

only be transferred into SM from PM, then the critical questions posed by Tulving and Patterson

(1968) are 1) how can one explain the high level of recall for lists in which the related words

were presented in the middle serial positions (M lists)? 2) How can one explain unitization of

50

these words in SM? If one were to assume that related words are unitized in PM, then it becomes

necessary to explain why recall was lower when the four highly related words were clustered at

the end of the list (E lists) compared to the middle of the list (M lists). One would also have to

explain why the number of functional units recalled for E lists was lower compared to C lists,

and why the recency effect was not extended over a larger number of terminal input positions in

E lists compared to C lists and M lists. Alternatively, if one were to assume that items are

unitized in SM but not PM, then it would be necessary to explain how related words are almost

always transferred into SM, and how they become unitized in SM. Due to these apparent

difficulties with a two store interpretation of PM and SM, Tulving and Patterson identified PM

and SM with different types of retrieval mechanisms. For these same reasons, it seems unlikely

that differences in the LW amplitude observed for different degrees of elaborative encoding can

be accounted for by different PM and SM memory stores or systems. However, it is possible and

likely that these differences can be accounted for by differing storage and retrieval mechanisms

associated with ‘PM’ and ‘SM’ items, where a greater degree of binding is thought to be

involved for ‘SM’ items compared to ‘PM’ items, as described above.

With the aim of further examining what specific processes the LW reflects, another interesting

question is whether the amplitude of the LW varies with the strength of an association formed.

The next study of the present thesis (Study 4) investigated whether the LW reflects the strength

of a memory trace formed using a manipulation of RI. Although a two levels of RI was used in

Study 2, it is more complicated to make a statement about the strength of the associations formed

based on the corresponding data because a given pair was either cued for recall after a short RI or

long RI. Thus, an unknown portion of pairs that were recalled after the short RI might have also

been recalled if they were cued for recall after the long RI. Conversely, an unknown portion of

pairs that were not recalled after the long RI might have been recalled if cued for recall after the

short RI. It is worth noting, however, that an estimate could be produced for these unknown

proportions with the assumption that words that were not-recalled after the short RI also would

not have been recalled if cued after the long RI. It would be interesting to investigate whether the

ERP waveforms corresponding to these estimates can be thought of as mixtures of collected

recordings. This may be the case if the processes engaged only differ quantitatively (or in terms

of amplitude), but not qualitatively.

51

Chapter 5 ERP Correlates of Associative Binding Strength and the Negative

Recency Effect (Study 4)4 The present study sought to examine whether the LW reflects associative binding strength. This

allowed for an additional investigation on the negative recency effect, which requires

participants to be tested on study items twice: first on an immediate memory test and then a

second time on a delayed memory test. As mentioned in the Introduction, Madigan and McCabe

(1971) showed the negative recency effect for pairs of words that were tested on during the

immediate test, as well as pairs that were not tested during the immediate test and only on the

delayed test. This finding suggests that the immediate test itself is not responsible for the

negative recency effect. One of the early explanations for the negative recency effect proposed

by Craik (1970) takes the perspective of a two-store model of memory and suggests that items in

the terminal list position are retrieved from PM but not effectively transferred into SM storage.

Consequently, these items in terminal list positions are recalled with a much lower probability

during a delayed recall test that requires retrieval from SM, compared to other items presented

earlier in the list. In contrast to items presented in terminal list positions, participants have the

opportunity to rehearse and process further items that were presented earlier in the list and,

moreover, incorporate newly presented items (from later list positions) into any ongoing

encoding processes. From a two-store model of memory perspective, these ongoing processes

could contribute to the effective transfer of items in earlier list positions from PM to SM. It

might be that items in terminal list positions are not effectively transferred into SM because

participants’ motivation to continue processing these items diminishes once these items have

been recalled on the immediate memory test, especially since participants are typically not

informed about the final memory test until it begins.

4 This study was presented as part of four conference talks: 1) Kim, A.S.N. (October 16, 2012). Associative binding strength: An ERP investigation. Talk presented at the Society for Neuroscience Annual Meeting (Nanosymposium Session 626: Human Long-Term Memory: Encoding and Sleep), New Orleans, USA; 2) Kim, A.S.N. (May 4, 2012). Is there a better way to learn? A cognitive and brain research perspective. Talk presented at the 1st Annual Brain Power Conference, Toronto, Ontario; 3) Kim, A.S.N. (April 14, 2012). Towards better practices of learning: Perspectives from cognitive and brain research. Talk presented at the 1st Annual Cross Discipline Graduate Student Association Conference – Modern Issues: The Role of the Researcher, Toronto, Ontario; 4) Kim, A.S.N. (November 23, 2011). What can we do to improve our memory? Evidence from cognitive and behavioural research. Talk presented at the 3rd Annual Applied Research Education Day, Toronto, Canada.

52

As mentioned in the previous chapter, Craik (1983) has more recently suggested that the notion

of transfer of information from temporary storage (PM) to permanent storage (SM) is both

misleading and unnecessary. Instead, Craik proposed that memory should be conceptualized in

terms of the qualitative nature of the encoding processes (Craik & Lockhart, 1972) and the

compatibility between encoding and retrieval operations, as outlined by the encoding specificity

hypothesis (Tulving & Thomson, 1973). This is in line with the conceptualization of PM and SM

with different types of retrieval mechanisms rather than different memory stores, and the

importance of coding (Tulving & Patterson, 1968). It could be that participants do not engage in

elaborative coding for items presented in terminal positions because they anticipate that these

items will remain in consciousness, and thus will remain easily accessible, if this information is

required on the following immediate memory test. However, this type of “shallow” coding

would result in fewer retrieval cues and decreased memory performance after the item in

question becomes absent from consciousness.

The present study examined whether encoding-related ERPs elicited by items in terminal list

positions differ from the ERPs elicited by items in other serial positions of the list. Any observed

differences would suggest that the negative recency effect is due, at least in part, to differential

coding of terminal and pre-terminal items of a list. Similar to Study 2, two levels of RI were used

(“Immediate” and “Delayed”, described further below). However, in contrast to Study 2,

participants in the present study were tested on their memory for pairs of words after both the

Immediate and Delayed RIs. Thus, the specific pairs that were recalled after the Immediate but

not Delayed RI could be identified, and likewise for those pairs that were recalled after both the

Immediate and Delayed RIs and pairs that were neither recalled after the Immediate or Delayed

RIs. This allowed for an investigation on the strength of the association formed during encoding.

For the analyses on association formation strength, the ERPs recorded in the present study were

sorted and analyzed based on three levels of associative strength: 1) pairs that were recalled after

both the Immediate and Delayed RIs, reflecting the strongest associations formed in the context

of the present study (ShortR_LongR); 2) pairs that were recalled after the Immediate RI but not

after the Delayed RI, representing an intermediate class of associative binding strength

(ShortR_LongN); and 3) pairs that were neither recalled after the Immediate RI nor after the

Delayed RI (ShortN_LongN).

53

As a secondary question, the present study also examined whether participants’ WMC is

correlated with (1) paired-associate recall performance or (2) any ERP components reflecting the

strength of the association formed. This examination addressed whether individual differences in

paired-associate recall performance could be linked to WMC, and was motivated by past studies

that have shown that WMC is closely correlated with a broad range of higher-order cognitive

capabilities, including reasoning and language comprehension (for a review see Engle & Kane,

2003). Since past studies have demonstrated that complex span tasks (e.g., operation span tasks)

are better predictors of higher-order cognitive ability compared to simple span tasks ( e.g.,

simple word or digit span tasks; Turner & Engle, 1989) a complex span task (the operation span

task) described in detail below, was used in the present study to measure participants’ WMC.

The hypotheses of the present study were as follows:

Strength of association formation:

(1) The LW, as an index of the strength of the association formed, will demonstrate a larger

positive amplitude as the strength of the formed association increases, such that the amplitude of

the LW will be largest for the strongest category of pairs (ShortR_LongR), followed by the

intermediate category of pairs (ShortR_LongN), and smallest for the weakest pairs

(ShortN_LongN).

(2) WMC, as measured by the operation span task described below, will be positively correlated

with participants’ memory performance and any ERP components reflecting the strength of the

association formed.

Negative recency, PM, and SM:

(3) The encoding-phase ERPs elicited by pairs in terminal list positions will differ (qualitatively

and/or quantitatively) from the encoding-phase ERPs elicited by pairs in earlier serial positions

of the list.

(4) As shown by Madigan and McCabe (1971), the negative recency effect will be found for

pairs that are tested on both during the immediate and delayed memory tests, as well as pairs that

are not tested on during the immediate test and only on the delayed test.

54

15 Methods

15.1 Participants Twenty-two healthy, young adults were tested in the present study. For the investigation on the

ERP correlates of associative binding strength, the data from eight participants were discarded

because they had too few ERP trials in one or more of the categories of pairs to allow for ERP

analysis. The data from one participant were discarded because their ERP data were more than

two standard deviations beyond the group mean. Consequently, for the investigation on

associative binding strength, the ERP averages were obtained from 13 participants (7 female;

mean age: 22 years, range: 18–29; first language: English) and the data from 9 participants were

excluded (7 female; mean age: 22 years, range: 18–26). For the analyses on the negative recency

effect, the ERP averages were obtained from 9 participants (5 female; mean age: 23 years, range:

19–29) and the data from 13 participants were excluded (9 female; mean age: 22, range: 18–26).

For this particular analysis, participants were excluded if they contributed less than nine ERP

trials (as opposed to the minimum criterion of 10 trials) for each serial position to increase the

number of participants included in the analysis. In total two participants contributed a minimum

of nine, as opposed to 10, ERP trials for one of the categories of pairs.

15.2 Materials The pool of experimental words consisted of 800 English nouns, which were divided into 80

lists. Each list was composed of 5 pairs of words. None of the pairs appeared in more than one

list for a given participant, and the same pairings were used for all participants.

15.3 Design Two levels of RI were used in the present study: “Immediate” and “Delayed”, as described

further below. For each participant, 80 pairs of words were tested on both after the Immediate

and Delayed memory tests, and another 80 pairs of words were tested on only during the

Delayed memory test. Both of these types of pairs were distributed equally across all serial

positions of a list. The ERPs recorded during the study phase served as the dependent variable.

55

15.4 Procedure The experimental procedure used in the present study is summarized in figure 15. Each

participant took part in one experimental session. At the beginning of each session, participants’

WMC was measured using a computer-administered version of the operation span (Ospan) task

(Turner & Engle, 1989; Unsworth & Engle, 2007; Unsworth, Heitz, Schrock, & Engle, 2005). To

complete this task, participants were required to try to remember a set of unrelated letters while

they solved a series of arithmetic equations. Participants were first presented with an arithmetic

equation and after they solved it, they were presented with a letter for 800 ms. Immediately after

the letter was presented participants were given another arithmetic equation to solve, followed by

the presentation of the next letter and so forth. Each of the five list lengths (three through seven

letters long) were used for three trials, with the order of list length varied randomly. During

recall, participants’ task was to identify letters from the current set in the same sequence they

were presented by clicking on the appropriate letters (see Unsworth et al., 2005 for more details).

Participants performed three sets of practice trials using a list length of two. The Ospan score

was the total number of recalled letters from the perfectly recalled sets and could range from 0 to

75. For example, if a participant recalled two letters in a set size of two, three letters in a set size

of three, and two letters in a set size of four, their Ospan score would be five (the sum of two,

three and zero). Participants’ Ospan scores were used to examine whether individuals differences

in paired-associate recall performance is correlated with WMC.

Following the Ospan task, participants performed the experimental tasks in 80 study/test cycles.

Each study/test cycle consisted of two phases. During the first phase, participants were presented

with a list of five pairs of words to study. The words of each pair were presented visually on a

computer monitor, one at a time. Participants were instructed to remember the pairs and were

told that they would be tested on their memory for the pairs. The evoked trial corresponded to the

presentation of a pair of words during encoding (figure 16). It started with a central fixation

stimulus ‘+’ that served as a warning and lasted for 500 ms. Then the first word of a pair was

presented for 1000 ms, followed by a blank screen for 200 ms, the second word of a pair for

1000 ms, and then a blank screen for 600 ms.

During the second phase of a study/test cycle, which occurred immediately after all five pairs of

a list were presented, participants were tested on paired-associate recall for one of the five pairs

56

that had been presented in the study phase of that cycle. This constituted the Immediate memory

test. Across all 80 study/test cycles, each probe position was tested 16 times in a randomized

fashion. During the Immediate memory test, participants were presented with a central fixation

stimulus (+) for 500 ms, followed by a cue word for 10 seconds and then a blank screen for 200

ms. The first word of a pair served as the cue for recall of the second word. Participants

responded vocally during the presentation of the cue and the experimenter coded whether the

response was correct by pushing a designated button on the keyboard. Both an incorrect response

and the absence of any response within the 10 second time frame of the cue word presentation

were classified as not-recalled.

After the last (80th) study/test cycle was completed, participants were given a final test of paired-

associate recall. This final test constituted the Delayed memory test, where participants were

tested a second time, in a randomized order, on all of the 80 pairs that they had been tested on

during the Immediate memory test. During the Delayed memory test, participants were also

tested on 80 additional pairs that were presented to them during the study phase, but on which

participants were not tested during the Immediate memory test. For this latter group of pairs,

each serial position was tested on an equal number of times accounting for the serial position that

was tested on during the immediate memory test. Participants were tested the same way during

the Delayed memory test as they were during the Immediate memory test, with a central fixation

stimulus (+) presented for 500 ms before the presentation of each cue, which lasted for 10

seconds and was followed by a blank screen for 200 ms. Participants provided their response

vocally during the presentation of the cue and both an incorrect response and the absence of any

response within the 10 second time frame of the cue word presentation were classified as not-

recalled.

Participants were given instructions on the experimental tasks and performed a practice block to

confirm that they understood what the tasks entailed before the study/test cycles started.

Participants took short breaks after every 10 study/test cycles, which corresponded to 10 lists that

were composed of 5 pairs each.

15.5 Behavioural data analysis Paired-associate recall performance, measured as the percentage of correctly recalled pairs, was

calculated separately for the Immediate and Delayed memory tests. Paired-associate recall

57

performance was also calculated based on whether pairs were recalled during both the Immediate

and Delayed memory tests. The small number of pairs that were recalled on the Delayed memory

test but not on the Immediate memory test, as described further in section 13.6, were excluded

from all analyses because these instances were so few in number and was not relevant to the

purposes of the present study. Paired-associate recall performance was also examined further as a

function of serial position to investigate the negative recency effect. Correlational analyses were

conducted to investigate the relation, if any, between participants’ WMC and paired-associate

recall performance as a function of associative binding strength (e.g., ImmR_DelR >

ImmR_DelN > ImmN_DelN). A positive finding would suggest that similar or overlapping

abilities support WMC and the strength of the association formed, and help clarify the specific

processes involved in association formation.

15.6 Electrophysiological data analysis For each participant, continuous EEG files were examined using BESA 5.3. The EEG data were

epoched into 3300 ms segments beginning with the fixation stimulus (+) and lasting 600 ms after

the offset of the second word in the pair. The epoch was baseline corrected to the 200 ms

segment preceding the presentation of the first word of a pair. The ERPs recorded during study

were sorted differently for the examination of (1) strength of association formation and (2) the

negative recency effect.

Strength of association formation. For this analysis, the encoding-related ERPs were sorted

based on whether the corresponding pair of words was subsequently recalled during the

Immediate and Delayed memory tests, yielding the following three categories of ERPs: (1) pairs

that were not subsequently recalled during the Immediate test nor the Delayed test

(ImmN_DelN); (2) pairs that were subsequently recalled during the Immediate test, but not

during the Delayed test (ImmR_DelN); (3) pairs that were subsequently recalled during the

Immediate test and the Delayed test (ImmR_DelR). Individual participant waveforms were

created by averaging the ERPs in these three categories. The mean trial counts going into the

grand-mean waveforms were 19 for ImmN_DelN (range: 10-33), 30 for ImmR_DelN (range: 17-

39) and 26 for ImmR_DelR (range: 13-42). Across all participants, a total of 14 trials were

excluded from the grand-mean waveform because they corresponded to pairs that were recalled

during the Delayed memory test but not during the Immediate memory test, which worked out to

58

a mean of 1 trial excluded per participant (range: 0-3). Based on the results of Study 2 and Kim

et al. (2009), a repeated measures ANOVA was conducted to examine the effect of the strength

of the association formed on the amplitude of the LW. This analysis was conducted on the ERPs

elicited by the three categories of pairs (ImmN_DelN, ImmR_DelN, ImmR_DelR) during the

time window of the LW (1 to 1.6 seconds after the presentation of Word2) that were recorded

over right fronto-central electrodes (F2, Fz, FCz, FC2, Cz, C2). Electrode location (F2, Fz, FCz,

FC2, Cz, C2) and pair category (ImmN_DelN, ImmR_DelN, ImmR_DelR) were used as factors.

The negative recency effect. For this analysis, only ERPs elicited by pairs that were subsequently

recalled on the Immediate memory test were considered. These ERPs were sorted based on the

corresponding pair's serial position in the list presented during the study phase, which resulted in

five categories of ERPs corresponding to: (1) subsequently recalled pairs presented in the first

serial position of a list (Pos1); (2) subsequently recalled pairs presented in the second serial

position of a list (Pos2); (3) subsequently recalled pairs presented in the third serial position of a

list (Pos3); (4) subsequently recalled pairs presented in the fourth serial position of a list (Pos4);

and (5) subsequently recalled pairs presented in the fifth (last) serial position of a list (Pos5).

Individual participant waveforms were created by averaging the ERPs in these five categories.

Given the number of participants excluded from the analyses because they contributed too few

trials to one or more categories of pairs, an exception was made to include two participants who

contributed nine trials (as opposed to the minimum of 10) for one category of pairs each. The

mean trial counts going into the grand-mean waveforms were 15 for Pos1 (range: 13-16), 13 for

Pos2 (range: 9-16), 14 for Pos3 (range: 11-16); 13 for Pos4 (range: 9-16), and 15 for Pos5

(range: 12-16).

15.7 Partial least squares analyses Partial least squares (PLS) is a multivariate technique that describes the relation between a set of

independent variables and a large set of dependent measures (McIntosh et al., 1996;

http://www.rotman-baycrest.on.ca/pls). This is done by computing the optimal least-squares fit to

the portion of a cross-block covariance matrix that is attributable to experimental manipulations

or behaviour (Wold, 1982). In this matrix, the sequence of time-points for each electrode is

ordered as columns, whereas the participants within each experimental condition are ordered as

rows. The PLS output includes latent variables (LVs) that describe patterns across experimental

59

conditions (referred to as design scores) expressed at each time point in terms of ERP amplitude.

The number of LVs extracted is equal to the number of experimental conditions, with the first

LV accounting for the most variance. A salience is then calculated for the LV at each electrode at

each time-point. The polarity and magnitude of the electrode salience denote the direction and

strength, respectively, of the identified differences among the experimental conditions shown in

the design scores. Scalp scores for each LV are obtained to measure how strongly each

individual participant contributes to the patterns depicted by the LV. These scores are the dot

product of the original data matrix and the saliences, resulting in a single value for each

participant.

Different versions of PLS have been used in the literature, and this technique has been applied

previously to analyze ERP data (Lobaugh, West, & McIntosh, 2001; Vallesi, McIntosh, & Stuss,

2011; Vallesi, Stuss, McIntosh, & Picton, 2009; Hay, Kane, West, & Alain, 2002). Mean-

centered PLS identifies the maximal experimental effects in the data set. This version of PLS is

referred to as mean-centered, because the means for each of the columns in the ERP amplitude

data matrix described above are subtracted from the grand mean to produce a mean-centered

deviation matrix. The LVs are then derived from singular value decomposition (SVD) of this

mean-centered deviation matrix. In the present study, a mean-centered PLS analysis was used to

identify where on the scalp the strongest experimental effects were expressed during the time

window of the LW (1 to 1.6 s after the onset of the second word of a pair), considering the three

categories of pairs: ImmN_DelN (weakest); ImmR_DelN (intermediate); ImmR_DelR

(strongest).

Statistical significance of the whole spatiotemporal pattern expressed by each LV was evaluated

by a permutation test using 500 permutations across the different experimental conditions

(Edgington, 1980; McIntosh, Bookstein, Haxby, & Grady, 1996). Permutations consist of

sampling without replacement to reassign the order of conditions for each subject. Mean-

centered PLS is recalculated for each new permuted sample, and the number of times the

permuted singular values exceeded the observed singular values in each LV is calculated and

expressed as a probability. A LV was considered significant at p < 0.05. A bootstrap test of 100

bootstrap samples was used to assess the stability of the saliences identified for each LV, on each

electrode and time-point. This was done to detect those portions of the ERP waveforms that

express robust experimental effects across subjects and to circumvent the effects of possible

60

outliers (Efron & Tibshirani, 1986). Bootstrap samples are produced by sampling with

replacement and keeping the assignment of experimental conditions to each subject fixed. Mean-

centered PLS is computed for each bootstrap sample. The ratio of the salience to its standard

error, estimated through the bootstrap procedure, approximately corresponds to a z-score.

Bootstrap ratios equal to or greater than 2.81 (roughly corresponding to a p level of 0.005) were

chosen as the cut-off for stable non-zero saliences. Thus, whereas permutation tests were used to

determine the significance of each LV, the stability of the saliences identified for each LV was

determined using bootstrap estimates of the salience standard errors.

Since both the bootstrap estimates and permutation tests involve resampling, it is important to

note that when a SVD is performed on a resampled matrix, reflection (a sign change in saliences)

and axis rotation (a change in the order of extracted LVs) can occur. These reflections and

rotations, however, are arbitrary and can be corrected using a Procrustes rotation of the

resampled SVD outcome to the original SVD outcome (for details on the Procrustes rotation,

please refer to McIntosh & Lobaugh, 2004).

Whereas PLS using SVD identifies the maximal experimental effects in the data set, an alternate

version of PLS can be used to constrain the analysis so the specified contrasts represent 100% of

the cross-block covariance. This allows the experimenter to test specific hypotheses about the

data, and is made possible by the relation between the singular values from the SVD to the total

sums-of-squares from the data matrix, where the singular values are derived from the sums-of-

squares from the data matrix and a SVD is not performed. Consequently, a Procrustes rotation is

not needed, which is why this particular version of PLS is referred to as “non-rotated task PLS”.

Similar to mean-centered PLS, in non-rotated PLS the significance of the specified contrast and

the corresponding electrode saliences are assessed through the permutation tests. The bootstrap

ratios are used to assess the time-points at which there is a stable difference between the

contrasted conditions.

To examine the negative recency effect a non-rotated analysis was conducted on the entire 3300

ms epoch considering all pairs that were recalled during the Immediate memory test (regardless

of their status after the Long RI) based on serial position in the list during presentation. To do so,

the following contrasts were submitted for analysis: A) ERPs elicited by subsequently recalled

pairs in the terminal position of a list vs. ERPs elicited by subsequently recalled pairs presented

61

in all pre-terminal positions (i.e., Pos1: 1; Pos2: 1; Pos3: 1; Pos4: 1; Pos5: -4); B) ERPs elicited

by subsequently recalled pairs in the terminal position of a list vs. ERPs elicited by subsequently

recalled pairs presented in the second, third and fourth serial positions (i.e., Pos1: 0; Pos2: 1;

Pos3: 1; Pos4: 1; Pos5: -3). The latter contrast did not include pairs presented in the first serial

position of a list to avoid variance that may correspond to the primacy effect.

15.8 Memory performance and ERP correlation analyses Based on the results of the PLS analysis, the encoding-related ERP data were examined further

in relation to participants’ Ospan scores, which were used as a measure of WMC as described

above. The purpose of this analysis was to investigate whether participants’ WMC is correlated

with any ERP components reflecting the strength of the association formed. If any specific ERP

components were found to correlate with WMC, this finding would help clarify the cognitive

processes that they reflect and speak to the specific relation, if any, between WMC and

associative binding strength. Correlational analyses were conducted on participants’ Opsan

scores and mean amplitude (corresponding to the time window of the LW) of electrode clusters

that showed strength of association formation effects.

16 Results

16.1 Behavioural data Strength of association formation. Memory performance was examined as the percentage of

pairs recalled, and was assessed separately for Immediate and Delayed memory tests. The

percentage of paired-associate recall, collapsed across serial positions, was higher for the

Immediate test (M = 72, SE = 3), compared to the Delayed test (M = 35, SE = 3), t(12) = -15.31,

p < .001, two tailed. There was no difference between the percentage of pairs that were recalled

on both the Immediate and Delayed memory tests (M = 33, SE = 3) and pairs that were recalled

on the Immediate test but not on the Delayed test (M = 39, SE = 2), t(12) = 1.36, p = .31, two

tailed. When participants’ Ospan scores (which served as a measure of WMC) were correlated

with the percentage of pairs in each of the three categories of pairs (Strongest = DelR_ImmR;

Intermediate = DelN_ImmR; Weakest = DelN_ImmN), none of the correlations were significant

[ImmR_DelR, r(11) = .07, p = .41, one tailed; ImmR_DelN, r(11) =.24, p = .22, one tailed;

ImmN_DelN, r(11) = -.30, p = .16, one tailed].

62

Negative recency effect. To investigate the negative recency effect, ImmR_DelR pairs were

examined as a function of serial position using a one-way repeated measures ANOVA. The

results show that the serial position during encoding had a significant effect on ImmR_DelR

pairs [F(4, 32) = 5.72, p = .001, ηp2 = .42]. There was a significant linear trend [F(1, 8) = 12.01,

p = .008, ηp2 = .60] and a significant quadratic trend [F(1, 8) = 8.56, p = .02, ηp

2 = .52], with a

pattern shown in figure 17 as a drop in memory performance for pairs that were presented in the

last or terminal serial position of a list during the study phase after being level for pairs in pre-

terminal serial positions. The results of the repeated measures ANOVA that was conducted on

pairs on which a participant was tested only during the Delayed, but not Immediate, memory test

showed a significant effect of serial position [F(4, 32) = 4.37, p = .006, ηp2 = .35]. There was also

a significant linear trend [F(1, 8) = 15.66, p = .004, ηp2 = .66], reflecting a decrease in memory

performance as a function of serial position, with memory performance being lowest for pairs

that were presented in the last serial position of a list. This latter finding suggests that repeated

recall cannot account for why items presented in terminal positions are recalled with the lowest

probability on delayed memory tests.

16.2 Mean amplitude results Figure 18 displays an overview of the group level ERP data that were recorded at multiple

electrodes. The results of the repeated measures ANOVA showed that there were no significant

effects of category pair [F(2, 24) = .93, p = .41] or electrode [F(5, 60) = 1.39, p = .24], and no

significant interaction between these two factors [F(10, 120) = .71, p = .72]. A data-driven

analysis procedure (mean-centered PLS) was used to identify what experimental effects were

most strongly expressed during the time window of the LW and where on the scalp theses effects

were expressed.

16.3 Partial least square results Strength of association formation. A mean-centered PLS analysis was conducted on the time

window of the LW (1 to 1.6 s after the onset of the second word of a pair), considering pairs that

belonged to the following categories: ImmN_DelN, ImmR_DelN, and ImmR_DelR. Only the

first of the three PLS latent variables (LVs) was significant (LV1, p < .04; LV2, p < .83; LV3, p

< .95). LV1 accounted for cross-block variance of about 63%. The design scores (figure 19a)

show that the ImmR_DelR pairs differed most strongly from the ImmN_DelN and ImmN_DelR

63

pairs. The mean scalp scores, which reflect the degree to which each participant expresses the

task–ERP relation indicated by the data, are also shown for each category of pairs in figure 19a,

with 95% confidence intervals plotted. The scalp scores for each category of pairs are considered

statistically reliable if the error bars do not cross 0. As shown in figure 19a, the errors bars for

the ImmN_DelN pairs crossed zero and did not contribute significantly to the overall pattern

picked up by this LV. The topographical distribution of the electrode saliences are shown in

figure 19c, and interpreted as follows: regions of the scalp showing positive saliences (shown in

red) indicate electrode locations where the ERP amplitudes were more negative for the

ImmR_DelR pairs compared to the ImmR_DelN pairs. Conversely, negative saliences (shown in

blue) indicated scalp regions where the ERP amplitudes were more positive for the ImmR_DelR

pairs compared to the ImmR_DelN pairs. The bootstrap re-sampling of the electrode saliences

indicate that effect picked up by this LV was stable over the right frontal and left centro-parietal

scalp regions throughout the analyzed time window (1 to 1.6 s after the onset of the second word

of a pair). Figure 19b shows the electrode saliences and grand average waveforms recorded from

two representative electrodes (F6 and CP5 for the right fronto-central and the left centro-parietal

scalp regions, respectively), with black markers above each plot to indicate time points when the

saliences were stable by bootstrap estimation. The ERPs recorded from these two scalp regions

were examined further in relation to WMC, as described in the section 16.3.

Negative recency effect. A non-rotated PLS analysis was conducted to examine two contrasts.

First, the ERPs elicited by subsequently recalled pairs in the terminal position of a list were

contrasted against ERPs elicited by subsequently recalled pairs presented in all pre-terminal

positions (i.e., Pos1: 1; Pos2: 1; Pos3: 1; Pos4: 1; Pos5: -4). Second, the ERPs elicited by

subsequently recalled pairs in the terminal position of a list were contrasted against ERPs elicited

by subsequently recalled pairs presented in the second, third and fourth serial positions (i.e.,

Pos1: 0; Pos2: 1; Pos3: 1; Pos4: 1; Pos5: -3). In the latter contrast, the ERPs elicited by pairs in

the first serial position of a list were not included to avoid the effects of any variance accounting

for the primacy effect in the results. These analyses allowed the specified contrast to represent

100% of the cross-block covariance. However, the results of both of these non-rotated analyses

were not significant (p < .35 and p < .34, respectively).

64

16.4 Memory performance and ERP correlation results Based on the results of the PLS analysis, participants’ Ospan scores, which served as a measure

of WMC, were correlated with the mean amplitude (corresponding to the time window of the

LW) of two clusters of electrodes: 1) a right frontal cluster, consisting of electrodes F6, F8, FC6;

and 2) a left centro-parietal cluster, consisting of electrodes C3, C5, CP5. For the centro-parietal

electrode cluster, there was a significant negative relation between participants’ Ospan scores

and the mean amplitude elicited by pairs in the intermediate strength category (ImmR_DelN)

[r(11) = -.63, p =.02, two-tailed]. This indicates that participants who demonstrated larger

positive waveforms over the left centro-parietal scalp regions for the intermediate pairs also

demonstrated lower WMC. Participants' Ospan scores were not significantly correlated with the

mean amplitude elicited by the strongest category of (ImmR_DelR) pairs, or with the mean

amplitude elicited by the weakest category of pairs (ImmN_DelN pairs; p > .05, two-tailed). For

the right frontal electrode cluster, no significant correlations were found between participants'

Ospan scores and mean amplitudes recorded for any category of pairs during the time range of

the LW (p > .05, two-tailed).

17 Discussion The use of repeated testing in the present study complemented the design of Study 2 of the

present thesis and allowed for two streams of investigation: one on the strength of association

formation, and another on the topic of the negative recency effect. The strongest results of the

present study, as demonstrated by the results of the mean-centered PLS analysis, showed a

pattern reflecting increasing strength of association formation during the time window of the LW

(1 to 1.6 s after the presentation of the second word of a pair) over the right frontal and left

centro-parietal scalp regions. This finding was mainly due to a contrast between pairs in the

intermediate strength (ImmR_DelN) and strongest (ImmR_DelR) categories, as the ERPs

elicited by pairs in the weakest category (ImmN_DelN) were not found to contribute

significantly to the pattern picked up by the LV.

Interestingly, there was a significant correlation between participants’ WMC and mean ERP

amplitudes recorded over the left centro-parietal region for pairs in the intermediate strength

(ImmR_DelN) category. Sustained ERP positivity over the left posterior scalp has been related to

the persistence of task-relevant information in working memory (Slagter, Kok, Mol, Talsma, &

65

Kenemans, 2005). Thus, the observed correlation could be due to the nature of the paired-

associate recall task: participants were presented with five pairs of words and then immediately

cued to recall one of the pairs, which likely tapped into working memory processes. Participants'

WMC was not significantly correlated with mean ERP amplitudes recorded for the other

categories of pairs (pairs corresponding to the weakest and strongest associative binding

strengths) from scalp regions shown by the PLS analysis to reflect associative binding strength

(right frontal electrodes and left centro-parietal electrodes). This is in line with the lack of

correlation found between participants' WMC and their paired-associate recall performance for

the three categories of pairs: pairs corresponding to the weakest, intermediate and strongest

associative binding strengths. This finding suggests that there is not a one to one overlap between

cognitive and neural processes supporting WMC and associative binding strength. As mentioned

above, this secondary analysis was motivated by past work that has demonstrated a close relation

between WMC and a broad range of higher-order cognitive abilities (for a review see Engle &

Kane, 2003). The results of the present study suggest that this is not the case for associative

binding strength. However, it is important to note that the data from 8 out of 22 participants were

excluded from the analyses because these participants had too few ERP trials in one or more of

the categories of pairs to allow for ERP analysis. Thus, it is possible that the analyzed data were

collected from a biased sample, which could be one reason why WMC was not significantly

correlated with associative binding strength or ERP correlates of associative binding strength.

Additionally, it could be that the ERPs elicited by pairs in the weakest category were not found

to contribute significantly to the pattern picked up by the PLS results reflecting associative

binding strength, because the weakest category of pairs was statistically underpowered,

consisting of an average of 19 trials per participant, whereas the intermediate strength and

strongest categories consisted of an average of 30 and 26 trials per participant, respectively.

The design of the present study also allowed for an investigation of the negative recency effect.

The results of the non-rotated PLS analyses, which allows the contrasts to be set by the

experimenter, was conducted to examine the negative recency effect. The results of these

analyses were not significant, which suggests that items in terminal and pre-terminal serial

positions of a list are not encoded differently. It is possible that participants continue to rehearse

previously presented items as each new item in the series are presented to them, which may have

masked differences in how items of a list are encoded as a function of serial position.

66

Additionally, it could be that at least some of the factors that contribute to the negative recency

effect occur during retrieval, including the effects of interference. In line with previous studies,

the effect of testing itself cannot account for the negative recency effect as pairs that were tested

on once, only after the Long RI, also showed that pairs that were presented in terminal positions

of a list were recalled with the lowest probability.

Study 1 of the present thesis demonstrated that association formation is reflected by positive

deflections in the ERP waveforms recorded over the parietal scalp region starting around 460 ms

after the second item of a pair is presented. Study 2 showed that around 1 to 1.6 seconds after the

presentation of the second item of a pair, association formation is correlated with positive

deflections in the ERP waveforms recorded over the frontal and fronto-central scalp regions,

depending on the RI. The results of Study 2 and Study 3 suggest that the LW may reflect binding

in a general sense (inter-item and intra-item associative encoding). The strongest effects of the

present study (Study 4) suggest that associative binding strength is reflected by ERPs recorded

late over the right frontal and left centro-parietal scalp regions.

67

Chapter 6 General Discussion

The present dissertation investigated the neuroelectric correlates of episodic association

formation. Use of the subsequent memory paradigm and different RIs provided the means to

examine ERPs associated with the encoding of pairs, as well as the strength of the association

formed. The results of the present dissertation identified three sets of Dm effects for episodic

association formation. First, over the parietal scalp region, a P460 and positive slow wave that

occurred between 645 to 845 ms both showed larger positive amplitudes for subsequently

recalled pairs compared to subsequently not-recalled pairs (Study 1). Second, a Dm effect was

also demonstrated by a late-occurring sustained positivity that occurred over the fronto-central

scalp, with a slight right hemisphere bias (the LW), between 1 to 1.6 s after the presentation

onset of the second word of a pair: a larger positive amplitude was observed for subsequently

recalled, compared to subsequently not-recalled, pairs (Study 2). This particular Dm effect was

also observed for single words (Study 3), suggesting that it is not unique to the encoding of

paired-associates. Lastly, when a longer RI was used, a prominent Dm effect for paired-

associates was observed over the frontal scalp, with a slight right hemisphere bias, during the

same time window as the LW (Study 2). Differences in the scalp distributions of the observed

right frontal and fronto-central Dm effects suggest that they reflect qualitatively different

encoding processes. The results of the final study (Study 4) suggest that associative binding

strength is reflected by ERPs recorded over the right frontal and left centro-parietal scalp

regions.

The present dissertation addresses one of the major criticisms of the levels-of-processing

framework (Craik & Tulving, 1975; Craik & Lockhart, 1972): the absence of an objective index

of depth of processing used to encode information (Craik, 2002). The results of the present

dissertation address this concern by providing ERP evidence that can help reveal the degree of

depth attributable to different types of information processing, such as associative binding and

semantic evaluation. In doing so, the results of the present dissertation demonstrate that ERPs

can be used in combination with behavioural data to provide an objective index of depth of

processing. The applications of this research have the potential to benefit individuals in a wide

range of settings where there is motivation to explicitly encode information that is encountered.

68

According to Craik (2002), in the context of encoding, depth refers to the “qualitative type of

information”, whereas elaboration refers to “the degree to which this information is enriched”.

Moreover, Craik (2002) proposed that both meaningfulness and elaboration must both be

measured by any objective index of depth of processing. The ERP technique shows promise as a

tool that can be used to help objectively index depth of processing, since ERPs can be used to

examine whether encoding processes engaged in different experimental conditions are

functionally dissociable (Rugg & Allan, 2000). This is based on the premise that qualitative

differences between scalp distributions associated with different conditions reflect neural

processes that, at the least, do not-overlap completely. In contrast, quantitative differences in the

ERPs (differences in ERP amplitude) associated with different conditions are thought to reflect

different levels of engagement of the same neural processes. Thus, qualitative differences

between scalp distributions can be used to help index different degrees of depth of processing,

whereas quantitative differences in the ERPs (differences in ERP amplitude) can be used to help

index the degree of elaboration of a specific type of processing.

Although Dm effects have been shown to differ based on the level of processing engaged during

the encoding task, all known studies have contrasted Dm effects for deep and shallow encoding

strategies and have not investigated Dm effects for different degrees of deep encoding.

Generally, past studies that investigated ERP correlates of item encoding showed frontal-positive

Dm effects when deep encoding strategies were used and posterior-positive Dm effects when

rote or shallow encoding strategies were used (for review see Donchin & Fabiani, 1991).

Furthermore, Van Petten and Senkfor (1996) observed a Dm effect for words that participants

responded affirmatively to on a semantic decision encoding task (e.g. a “yes” response to the

question “Is it living?”), but not for words that they gave negative judgments for, consistent with

the results of an earlier study by Paller and colleagues (1987). To explain their findings, Van

Petten and Senkfor referred to Craik and Tulving’s (1975) suggestion that a positive relation

between a study question and target, resulting in a “yes” response, would lead to more richly

elaborative encoding compared to a negative relation between a study question and target, which

would result in a “no” response.

In Study 2 of the present thesis, the scalp distributions for the observed Dm effects differed

significantly between the short and long RI conditions, suggesting that they reflected different

types of encoding processes or processes that were at least partially non-overlapping. Based on

69

the results of past studies, it seems likely that the prominent Dm effect observed over the frontal

scalp for the long RI condition reflected encoding processes that were deeper compared to the

encoding processes engaged during the short RI condition, for which a fronto-central Dm effect

was observed. Consistent with this notion, late-occurring right frontal ERP positivity has been

related to additional evaluation of specific perceptual attributes of an item in a specific context

(Ranganath & Paller, 2000), additional neural processing in response to unexpected information

(DeLong, Urbach, Groppe, & Kutas, 2011), self-initiated semantic organization (Blanchet,

Gagnon, & Bastien, 2007) and additional or deeper post-lexical processing (Fields & Kuperberg,

2012). In the context of encoding, Mangels, Picton and Craik (2001) showed a larger late-

occurring, sustained ERP positivity over the right frontal scalp for single words that were

subsequently recalled with a remember judgment compared to words that were subsequently

recognized with a familiarity judgment or missed altogether. Similar finding have been reported

by Friedman and Trott (2000) and Schott and colleagues (2002), suggesting that the observed

right frontal Dm effect reflected deep encoding processes that resulted in subsequent

remembering.

Late-occurring fronto-central ERP positivity, on the other hand, appears to be associated with

binding and memory encoding in general. For example, in a study that investigated the

neuroelectric correlates of association formation and serial list learning (Caplan et al., 2009), the

results of a multivariate analysis showed a latent variable reflecting a significant Dm effect for

pairs of words, but not word lists (which consisted of three words). This Dm effect was most

salient over central electrodes close to the midline (e.g., electrodes C1 and C2) during the later

portion of the epoch (starting from about 1200 ms post-stimulus). Guo, Voss and Paller (2005)

also found a late-occurring fronto-central Dm effect for associations formed between pictures of

faces and spoken names. Additionally, Gutchess, Ieuji and Federmeier (2007) found a late-

occurring fronto-central Dm effect for complex scenes in both young and elderly adults. It seems

likely that their participants made associations between different aspects of the complex scenes,

which improved their chances of recognizing the correct complex scenes on the subsequent

memory test. However, the results of Study 3 suggest that the observed fronto-central Dm effect

is not unique to inter-item associations, but instead also reflect intra-item associations, which is

also consistent with the findings of Gutchess et al. (2007) on complex scenes.

70

Given the design of the Studies 2 and 3, combined with the results of past studies, it seems likely

that the right frontal Dm effect observed in Study 2 for the long RI condition reflected deeper

encoding processes compared to the fronto-central Dm effect observed for the short RI condition.

More specifically, the right frontal Dm effect observed in Study 2 likely reflected evaluation of

the word pairs at a semantic level, invoking evaluation of specific word meanings and/or the

meanings of the specific word combinations being processed. On the other hand, the observed

fronto-central Dm effect likely reflected binding in a general sense, including inter-item and

intra-item associative binding. Study 2 is noteworthy because it provides ERP evidence that

helps delineate the relative ranking of depth attributable to two types of processing that are

commonly considered as deep encoding strategies: semantic processing that invokes evaluation

of associated meaning and associative binding.

Further to the primary goal of the present dissertation, based on past work demonstrating

interdependence between encoding and retrieval processes and representations (Tulving &

Thomson, 1973; Alvarez & Squire, 1994), investigations on the neuroelectric correlates of

episodic association formation can be guided by ERP data collected during retrieval of

associative information. The parietal old/new effect is thought to reflect recollection

(Mecklinger, 2006; Rugg & Curran, 2007), and more specifically, has been associated with

recollection of associative information (Donaldson & Rugg, 1998; Tendolkar et al., 1997).

Interestingly, the latency of the positive slow wave observed in Study 1 falls into the general

time range of the parietal old/new effect and its topographical distribution is similar to the

topographical distribution shown by Yu and Rugg (2010) for an ERP contrast that narrowed in

on recollection (‘recollected’ versus ‘confidently old’ judgments) between 500 to 800 ms after

stimulus onset. Similarly, the results of a study by Woodruff and colleagues (2006) also showed

a larger positive deflection in the ERPs elicited by recollected, compared to confidently

recognized, items between 500 to 800 ms over both the right and left hemispheres of the parietal

scalp region.

In Study 1, the Dm effect observed for episodic associations, but not item encoding, over the

parietal scalp region may also reflect semantic organization to some degree. In addition to

showing a right frontal effect for semantic organization, Blanchet et al. (2007) also showed a

larger late positive component over the parietal scalp for related words, regardless of whether the

participants were informed about the semantic relations, compared to unrelated words. Thus,

71

whereas the later-occurring right frontal ERP positivity, as mentioned above, was interpreted as

reflecting self-initiated semantic organizational processes, the late positive component observed

over the parietal scalp was interpreted to reflect voluntary associative processes involved in

binding related items. Additionally, in the study by Caplan et al. (2009), mentioned above, that

investigated ERP correlates of association formation and serial list learning, the results of the

multivariate analysis showed a latent variable reflecting a significant Dm effect for word pairs,

but not word lists, which was prominent over posterior electrode sites and somewhat left-

lateralized. The timing of this latter effect overlapped primarily with the time window of the

parietal slow wave. These findings are consistent with the notion that the positive slow wave Dm

effect observed over the parietal scalp in Study1 reflects associative binding. However, future

research will have to be conducted to investigate this possibility further.

The similarities between the observed positive slow wave and the parietal old/new effect, in

terms of latency and topographical distribution, suggest that similar cognitive and brain

processes are engaged during encoding and retrieval of content-specific information that allow

one to recollect a previously experienced event, and provides further support for the notion that

processes and representations that are active during encoding are reinstated during successful

retrieval. In the context of Study 1, participants may have performed study-phase retrieval while

they were encoding the pairs of words. The processes engaged during study-phase retrieval may

have then been reinstated during retrieval that occurred in the test phase, which would help

explain the correspondence observed between the encoding-related data of Study 1 and the

retrieval data of the studies discussed above (Woodruff et al., 2006; Yu & Rugg, 2010).

Alternatively, the correspondence observed between the encoding and retrieval data may reflect

encoding and retrieval of cues that provided access to the target or sought-after information.

The parietal old/new effect has been related to the retrieval of item and contextual information

and is thought to index the contribution of the medial temporal lobe (MTL) to episodic memory

retrieval (Rugg & Allan, 2000; Wilding & Rugg, 1996). However, the relation between the

parietal old/new effect and the hippocampus is likely to be indirect, as the activity of the

hippocampal formation contributes very modestly, at best, to scalp-recorded potentials (Rugg,

1995). Instead, it is more likely that the parietal old/new effect reflects cortico-hippocampal

interactions that are thought to underlie episodic memory retrieval (Rugg & Allan, 2000;

Wilding & Rugg, 1996). Interestingly, Düzel and colleagues (2001) examined ERPs recorded

72

from an amnesic patient with damage that appeared to be isolated to the hippocampus and did

not find a parietal old/new effect in the patient’s ERPs that were recorded during recognition.

They did, however, find an index of familiarity in the ERP data. These findings suggest that

recollection, compared to familiarity, is more dependent on the hippocampal formation and

further highlights the importance of this brain region to successful cortical reinstatement. In a

model that integrates the perspectives of cortical reinstatement with complementary cognitive

perspectives, including encoding specificity (Tulving & Thomson, 1973) and transfer-

appropriate processing (Morris, Bransford, & Franks, 1977), Rugg and colleagues (2008)

identified the hippocampus to be of central importance. In this model the hippocampus has the

role of encoding, storing and reinstating patterns of brain activity elicited by a stimulus event.

Interestingly, Ranganath and colleagues (2004) have shown that the activity of the hippocampus

measured during encoding, in addition to that of the posterior parahippocampal cortex, is

predictive of recollection-based memory performance. In contrast, the activity of the rhinal

cortex, as measured during encoding, is predictive of familiarity-based recognition.

Based on the evidence described above, the Dm effects observed over the parietal scalp region

(P460 and subsequent positive slow wave) in the present dissertation likely reflect encoding

processes that allow for subsequent recollection of an item and the context in which it occurred.

More specifically, the observed positive slow wave likely reflects brain responses underlying the

formation of associative bonds or binding between the first and second words of a pair. The

observed P460, on the other hand, likely reflects brain responses underlying the processing of the

second word as the completion of the pair, which is regarded as being necessary for association

formation to occur, and may have lead to the positive slow wave that followed.

In addition to investigating Dm effects reflecting episodic association formation, the present

dissertation included an investigation of the ERP correlates of associative binding strength

(Study 4). The strongest experimental effects of this investigation showed that associative

binding strength was correlated with sustained negativity in the ERP waveforms recorded over

the right frontal scalp. This effect was also observed over the left centro-parietal scalp region in

the opposite polarity. It is important to note that these ERP patterns occurred during the same

time window as the LW, which likely reflects associative encoding, and that these ERP patterns

appeared to be driven mainly by differences between the processing of pairs that were

subsequently recalled on the Immediate and Delayed memory tests compared to pairs that were

73

subsequently recalled on the Immediate, but not Delayed, memory test. The contrast between

these two categories of pairs can be viewed as isolating a subsequent memory effect for the

Delayed condition amongst those pairs that were recalled in the Immediate condition, thus

revealing clues about the types of processes associated with better memory performance after a

longer RI has passed.

A review of the literature suggests that the topography of negative slow waves varies with the

nature of the experimental task (for a review see Rösler, Heil, & Röder, 1997). Negative slow

waves recorded over the right frontal scalp region, in particular, have been related to time

perception (Monfort, Pouthas, & Ragot, 2000; Pouthas, Garnero, Ferrandez, & Renault, 2000)

and anticipatory attention for the offset of a stimulus (Pfeuty, Ragot, & Pouthas, 2003).

Interestingly, Pouthas et al. (2000) modeled the sources of right frontal negative slow wave

activity, which was related to processing the presentation duration of stimuli, and found that it

paralleled the activity of a right frontal dipole. This finding is consistent with patient data

showing that focal right frontal lesions, but not left frontal lesions, are associated with deficits in

time perception (Harrington, Haaland, & Knight, 1998). Thus, it could be that the frontal

negative slow wave that was observed in Study 4 reflected anticipatory attention and monitoring

of the amount of time remaining for encoding to occur, which could have in turn increased the

efficacy of encoding processes that were taking place. In contrast, late occurring, sustained ERP

positivity over the left centro-parietal scalp region has been related to semantic integration across

hierarchical levels during sentence comprehension (Zhou et al., 2010). Additionally, sustained

ERP positivity over the left posterior scalp has been related to need for additional processing of

word meaning (Liotti, Woldorff, Perez, & Mayberg, 2000) and the persistence of task-relevant

information in working memory (Slagter et al., 2005). Thus, the sustained positivity observed

over the left centro-parietal scalp region in Study 4 likely reflected evaluation of the study

material across different or deeper levels of meaning, leading to increased associative binding

strength. Based on the interpretation of the observed right frontal and left centro-parietal ERP

findings, encoding is likely to be more effective when one is cognizant of the amount of time

they have to process the to-be-remembered information and they do so in a manner that invokes

evaluation of meaning across different semantic levels, which it turn likely results in a larger

number of associated retrieval cues.

74

Collectively, the results of the present dissertation have the potential to benefit individuals in a

wide variety of setting where there is motivation to explicitly encode information, such as in

classroom and workplace settings. By providing neuroelectric evidence that speaks to the

hierarchy of different deep encoding strategies, the results of the present study demonstrate that

the ERP technique can be used in combination with behavioural data to provide an objective

index of depth of processing used to encode information. More specifically, the results of the

present dissertation suggest that processing information on the basis of its meaning, especially

across different semantic levels (e.g. by forming short stories or sentences using two words of a

pair), is more effective than the binding of contextual information, regardless of whether it is in a

compositional (e.g., imagining the two words of a pair merged together as one item) or non-

compositional manner, for later memory performance. The results of the present dissertation also

suggest that this type of processing that is centered on meaning is further enhanced by awareness

of the amount of time that is available to process the information, perhaps because this awareness

ensures that information processing is not disrupted at a critical or particularly vulnerable stage.

The results of the present dissertation show promise for applications directed towards individuals

who are not diagnosed with memory impairment, however, it is unclear whether and to what

extent the results of the present dissertation can be applied to benefit those individuals who

demonstrate memory impairment. In a context where explicit, conscious memory was made

irrelevant, Ryan and colleagues (2000) demonstrated that amnesics are impaired in their

declarative memory for relations among elements or items composing a scene or event using

eye-movement recordings. These results suggest that memory impairments demonstrated by

amnesics cannot solely be characterized as deficits related to explicit, conscious memory. Thus it

is unclear whether explicit instructions for use of specific encoding strategies will benefit

memory performance in this specific population. However, it could be that specific patient

populations will benefit from different types of information processing for encoding. Future

research will provide more insight on this topic. Additionally and of great interest, more recent

work has suggested that binding of information, or association formation, serves as the basis of

many cognitive operations above and beyond long-term memory, including perceptual

judgments (Olsen, Moses, Riggs, & Ryan, 2012). Consequently, it is possible that use of more

effective encoding processes can also indirectly benefit other domains of cognition. Future work

will illuminate this exciting avenue of research further.

75

References Alain, C., & Winkler, I. (2012). Recording event-related brain potentials: Application to study

auditory perception. In D. Poeppel, T. Overath, A. N. Popper, & R. R. Fay (Eds.), Human

Auditory Cortex (pp. 69–96). New York: Springer Science+Business Media, LLC.

Alexander, M. P. (2003). California Verbal Learning Test: Performance by patients with focal

frontal and non-frontal lesions. Brain, 126, 1493–1503. doi:10.1093/brain/awg128

Alvarez, P., & Squire, L. R. (1994). Memory consolidation and the medial temporal lobe: A

simple network model. Proceedings of the National Academy of Sciences of the United

States of America, 91, 7041–5. Retrieved from

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=44334&tool=pmcentrez&render

type=abstract

Anderson, M. C., Bjork, R. A., Bjork, E. L. (1994). Remembering can cause forgetting: Retrieval

dynamics in long-term memory. Journal of Experimental Psychology: Learning, Memory,

and Cognition, 20, 1063–1087. doi: 10.1037/0278-7393.20.5.1063

Blanchet, S., Gagnon, G., & Bastien, C. (2007). Event-related potential study of dynamic neural

mechanisms of semantic organizational strategies in verbal learning. Brain Research, 1170,

59–70. doi:10.1016/j.brainres.2007.07.024

Bower, G. H. (1970). Imagery as a relational organizer in associative learning. Journal of Verbal

Learning and Verbal Behavior, 9, 529–533. doi:10.1016/S0022-5371(70)80096-2

Bower, G. H., Thompson-Schill, S., & Tulving, E. (1994). Reducing retroactive interference: An

interference analysis. Journal of Experimental Psychology. Learning, Memory, and

Cognition, 20, 51–66.

Bridge, D. J., & Paller, K. A. (2012). Neural correlates of reactivation and retrieval-induced

distortion. The Journal of Neuroscience: The official journal of the Society for

Neuroscience, 32, 12144–12151. doi:10.1523/JNEUROSCI.1378-12.2012

76

Buckner, R. L., Wheeler, M. E., & Sheridan, M. A. (2001). Encoding processes during retrieval

tasks. Journal of Cognitive Neuroscience, 13, 406–415. Retrieved from

http://www.ncbi.nlm.nih.gov/pubmed/11371316

Calkins, M. W. (1894). Association: I. Psychological Review, I., 476– 483.

Caplan, J. B. (2005). Associative isolation: Unifying associative and list memory. Journal of

Mathematical Psychology, 49, 383–402. doi:10.1016/j.jmp.2005.06.004

Caplan, J. B., Glaholt, M. G., & McIntosh, A. R. (2009). EEG activity underlying successful

study of associative and order information. Journal of Cognitive Neuroscience, 21, 1346–

1364. doi:10.1162/jocn.2008.21167

Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in

verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132, 354–

380. doi:10.1037/0033-2909.132.3.354

Cheng, S., & Rugg, M. D. (2010). Event-related potential correlates of gist and verbatim

encoding. International Journal of Psychophysiology�: The official journal of the

International Organization of Psychophysiology, 77, 95–105.

doi:10.1016/j.ijpsycho.2010.04.010

Craik, F. I. M. (1983). On the transfer of information from temporary to permanent memory.

Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences,

302, 341–359.

Craik, F. I. M. (1968). Two components in free recall. Journal of Verbal Learning and Verbal

Behavior, 7, 996–1004.

Craik, F. I. M. (1970). The fate of primary memory items in free recall. Journal of Verbal

Learning and Verbal Behavior, 9, 143–148.

Craik, F. I. M. (2002). Levels of processing: Past, present . . . and future? Memory, 10, 305–318.

doi:10.1080/09658210244000135

77

Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic

memory. Journal of Experimental Psychology: General, 104, 268–294. doi:10.1037//0096-

3445.104.3.268

Craik, F. I. M., & Lockhart, R. (1972). Levels of Processing: A Framework for Memory

Research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684.

Crowley, K., Trinder, J., & Colrain, I. M. (2002). An examination of evoked K-complex

amplitude and frequency of occurrence in the elderly. Journal of Sleep Research, 11, 129–

140. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12028478

DeLong, K. A., Urbach, T. P., Groppe, D. M., & Kutas, M. (2011). Overlapping dual ERP

responses to low cloze probability sentence continuations. Psychophysiology, 48, 1203–

1207. doi:10.1111/j.1469-8986.2011.01199.x

Deese, J., & Kaufman, R. (1957). Serial effects in recall of unorganized and sequentially

organized verbal material. Journal of Experimental Psychology, 54, 180–187.

Dere, E., Easton, A., Nadel, L., & Huston, J. P. (Eds.). (2008). Handbook of episodic memory.

Amsterdam, The Netherlands: Elsevier.

Donaldson, D. I., & Rugg, M. D. (1998). Recognition memory for new associations:

Electrophysiological evidence for the role of recollection. Neuropsychologia, 36, 377–395.

Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/9699947

Donchin, E., & Fabiani, M. (1991). The use of event-related brain potentials in the study of

memory: Is P300 a measure of event distinctiveness? In J. Jennings & M. Coles (Eds.),

Handbook of cognitive psychophysiology: Central and autonomic system approaches (pp.

471–498). Chichester, NY: Wiley.

Donchin, E., & Heffley, E. (1978). Multivariate analysis of event-related potential data: A

tutorial review. In D. Otto (Ed.), Multidisciplinary perspectives in event-related brain

potential research (pp. 555–572). Washington, DC: U.S. Government Printing Office.

78

Düzel, E., Vargha-Khadem, F., Heinze, H. J., & Mishkin, M. (2001). Brain activity evidence for

recognition without recollection after early hippocampal damage, 98, 8101–8106. doi:

10.1073/pnas.131205798

Ebbinghaus, H. (1885/1964). Memory: A contribution to experimental psychology (H. A. Ruger

& C. E. Bussenius, Trans.). Mineola, NY: Dover Publications.

Engle, R. W., & Kane, M. J. (2003). Executive attention, working memory capacity, and a two-

factor theory of cognitive control. The Psychology of Learning and Motivation, 44, 145–

199.

Fabiani, M., Karis, D., & Donchin, E. (1986). P300 and recall in an incidental memory

paradigm. Psychophysiology, 23, 298–308. Retrieved from


Fabiani, M., Karis, D., & Donchin, E. (1990). Effects of mnemonic strategy manipulation in a

Von Restorff paradigm. Electroencephalography and Clinical Neurophysiology, 75, 22–35.


Fields, E. C., & Kuperberg, G. R. (2012). It’s all about you: An ERP study of emotion and self-

relevance in discourse. NeuroImage, 62, 562–74. doi:10.1016/j.neuroimage.2012.05.003

Friedman, D. (1990). ERPs during continuous recognition memory for words. Biological

Psychology, 30, 61–87.

Friedman, D., & Johnson, R. (2000). Event-related potential (ERP) studies of memory encoding

and retrieval: A selective review. Microscopy Research and Technique, 51, 6–28.

doi:10.1002/1097-0029(20001001)51:1<6::AID-JEMT2>3.0.CO;2-R

Friedman, D., & Trott, C. (2000). An event-related potential study of encoding in young and

older adults. Neuropsychologia, 38, 542–557. Retrieved from


Friedman, D. (1990). Cognitive event-related potential components during continuous

recognition memory for pictures. Psychophysiology, 27, 136–148.

79

Garcia-Larrea, L., & Cezanne-Bert, G. (1998). P3, positive slow wave and working memory

load: A study on the functional correlates of slow wave activity. Electroencephalography

and Clinical Neurophysiology: Evoked Potentials, 108, 260–273.

Gates, A. I. (1917). Recitation as a factor in memorizing. Archives of Psychology, 6, 1–104.

Gibson, B. S., Gondoli, D. M., Flies, A. C., Dobrzenski, B. A., & Unsworth, N. (2010).

Application of the dual-component model of working memory to ADHD. Child

Neuropsychology, 16, 60–79. doi:10.1080/09297040903146958

Guo, C., Voss, J. L., & Paller, K. A. (2005). Electrophysiological correlates of forming

memories for faces, names, and face-name associations. Cognitive Brain Research, 22,

153–164. doi:10.1016/j.cogbrainres.2004.08.009

Gutchess, A. H., Ieuji, Y., & Federmeier, K. D. (2007). Event-related potentials reveal age

differences in the encoding and recognition of scenes. Journal of Cognitive Neuroscience,

19, 1089–1103. doi:10.1162/jocn.2007.19.7.1089

Harrington, D. L., Haaland, K. Y., & Knight, R. T. (1998). Cortical networks underlying

mechanisms of time perception. The Journal of Neuroscience: The official journal of the

Society for Neuroscience, 18, 1085–1095. Retrieved from


James, W. (1890). The Principles of Psychology. New York, NY: Holt, Rinehart & Winston.

Johnson, J. D., & Rugg, M. D. (2006). Modulation of the electrophysiological correlates of

retrieval cue processing by the specificity of task demands. Brain Research, 1071, 153–164.

doi:10.1016/j.brainres.2005.11.093

Johnson, R. (1995). Event-related potential insights into the neurobiology of memory systems. In

F. Boller & J. Grafman (Eds.), Handbook of Neuropsychology (pp. 135–163). Amsterdam,

The Netherlands: Elsevier.

Kahana, M. J. (2002). Associative symmetry and memory theory. Memory & Cognition, 30,

823–840. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12450087

80

Kahana, M. J., & Caplan, J. B. (2002). Associative asymmetry in probed recall of serial lists.

Memory & Cognition, 30, 841–849. Retrieved from


Karis, D., Fabiani, M., & Donchin, E. (1984). “P300” and memory: Individual differences in the

von Restorff Effect. Cognitive Psychology, 16, 177–216.

Karpicke, J. D., & Roediger 3rd, H. L. (2008). The critical importance of retrieval for learning.

Science, 319, 966–968. doi:10.1126/science.1152408

Karpicke, J. D., & Roediger 3rd, H. L. (2007). Repeated retrieval during learning is the key to

long-term retention. Journal of Memory and Language, 57, 151–162.

doi:10.1016/j.jml.2006.09.004

Kim, A. S. N., Vallesi, A., Picton, T. W., & Tulving, E. (2009). Cognitive association formation

in episodic memory: Evidence from event-related potentials. Neuropsychologia, 47, 3162–

3173. doi:10.1016/j.neuropsychologia.2009.07.015

Kounios, J., Smith, R. W., Yang, W., Bachman, P., & D'Esposito, M. (2001). Cognitive

association formation in human memory revealed by spatiotemporal brain imaging, Neuron,

29, 297–306.

Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mental chronometry: The P300 as

a measure of stimulus evaluation time. Science, 197, 792–795.

Liotti, M., Woldorff, M. G., Perez, R., & Mayberg, H. S. (2000). An ERP study of the temporal

course of the Stroop color-word interference effect. Neuropsychologia, 38, 701–711.


Madan, C. R., Glaholt, M. G., & Caplan, J. B. (2010). The influence of item properties on

association-memory. Journal of Memory and Language, 63, 46–63.

doi:10.1016/j.jml.2010.03.001

Madigan, S. A., & McCabe, L. (1971). Perfect recall and total forgetting: A problem for models

of short-term memory. Journal of Verbal Learning and Verbal Behavior, 10, 101–106.

doi:10.1016/S0022-5371(71)80101-9

81

Mangels, J. A., Picton, T. W., & Craik, F. I. M. (2001). Attention and successful episodic

encoding: An event-related potential study. Cognitive Brain Research, 11, 77–95. Retrieved

from http://www.ncbi.nlm.nih.gov/pubmed/11240113

McCarthy, G., & Wood, C. C. (1985). Scalp distributions of event-related potentials: An

ambiguity associated with analysis of variance models, 62, 203–208.

Mecklinger, A. (2006). Electrophysiological measures of familiarity memory. Clinical EEG and

Neuroscience: The official journal of the EEG and Clinical Neuroscience Society, 37, 292–

299. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/17073167

Monfort, V., Pouthas, V., & Ragot, R. (2000). Role of frontal cortex in memory for duration: An

event-related potential study in humans. Neuroscience Letters, 286, 91–94. Retrieved from


Montague, W. E., Adams, J. A., & Kiess, H. O. (1966). Forgetting and natural language

mediation. Journal of Experimental Psychology, 72, 829–833. doi:10.1037/h0023877

Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer

appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16, 519–533.

doi:10.1016/S0022-5371(77)80016-9

Murdock, B. B., Jr. (1962). The serial position effect of free recall. Journal of Experimental

Psychology, 64, 482–488. doi:10.1037/h0045106

Murdock, B. B., Jr. (1974). Human memory: Theory and data. New York, NY: Lawrence

Erlbaum.

Nilsson, L.-G., & Nilsson, E. (2009). Overweight and cognition. Scandinavian Journal of

Psychology, 50, 660–7. doi:10.1111/j.1467-9450.2009.00777.x

Nyberg, L., Kim, A. S. N., Habib, R., Levine, B., & Tulving, E. (2010). Consciousness of

subjective time in the brain. Proceedings of the National Academy of Sciences of the United

States of America, 107, 22356–9. doi:10.1073/pnas.1016823108

82

Olsen, R. K., Moses, S. N., Riggs, L., & Ryan, J. D. (2012). The hippocampus supports multiple

cognitive processes through relational binding and comparison. Frontiers in Human

Neuroscience, 6, 146: 1–13. doi:10.3389/fnhum.2012.00146

Paivio, A. (1963). Learning of adjective-noun paired associates as a function of adjective-noun

word order and noun abstractness. Canadian Journal of Psychology, 17, 370–379.

Paivio, A. (1965). Abstractness, imagery, and meaningfulness in paired-associate learning.

Journal of Verbal Learning and Verbal Behavior, 4, 32–38.

Paivio, A. (1969). Mental imagery in associative learning and memory. Psychological Review,

76, 241–263.

Paller, K. A., Kutas, M., & Mayes, A. (1987). Neural correlates of encoding in an incidental

learning paradigm. Electroencephalography and Clinical Neurophysiology, 67, 360–371.


Paller, K. A. (1990). Recall and stem-completion priming have different electrophysiological

correlates and are modified differentially by directed forgetting. Journal of Experimental

Psychology, 16, 1021–1032.

Paller, K. A. (2004). Electrical signals of memory and of the awareness of remembering. Current

Directions in Psychological Science, 13, 49–55. doi:10.1111/j.0963-7214.2004.00273.x

Paller, K. A. & Wagner, A. D. (2002). Observing the transformation of experience into memory.

Trends in Cognitive Sciences, 6, 93–102. Retrieved from


Pfeuty, M., Ragot, R., & Pouthas, V. (2003). When time is up: CNV time course differentiates

the roles of the hemispheres in the discrimination of short tone durations. Experimental

Brain Research, 151, 372–379. doi:10.1007/s00221-003-1505-6

Picton, T. W., Bentin, S., Berg, P., Donchin, E., Hillyard, S. A., Johnson, R. Jr., … Taylor, M. J.

(2000). Guidelines for using human event-related potentials to study cognition: Recording

standards and publication criteria. Psychophysiology, 37, 127–152. Retrieved from


83

Pouthas, V., Garnero, L., Ferrandez, A. M., & Renault, B. (2000). ERPs and PET analysis of

time perception: Spatial and temporal brain mapping during visual discrimination tasks.

Human Brain Mapping, 10, 49–60. Retrieved from


Ranganath, C., & Paller, K. A. (2000). Neural correlates of memory retrieval and evaluation.

Cognitive Brain Research, 9, 209–222. Retrieved from


Ranganath, C., Yonelinas, A. P., Cohen, M. X., Dy, C. J., Tom, S. M., & D’Esposito, M. (2004).

Dissociable correlates of recollection and familiarity within the medial temporal lobes.

Neuropsychologia, 42, 2–13. doi:10.1016/j.neuropsychologia.2003.07.006

Rizzuto, D. S., & Kahana, M. J. (2000). Associative symmetry vs . independent associations.

Neurocomputing, 32–33, 973–978.

Roediger 3rd, H. L., & Butler, A. C. (2011). The critical role of retrieval practice in long-term

retention. Trends in Cognitive Sciences, 15, 20–7. doi:10.1016/j.tics.2010.09.003

Roediger 3rd, H. L., & Marsh, E. J. (2005). The positive and negative consequences of multiple-

choice testing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31,

1155–1159. doi:10.1037/0278-7393.31.5.1155

Ruchkin, D. S., Munson, R., & Sutton, S. (1982). P300 and slow wave in a message consisting of

two events. Psychophysiology, 19, 629–642. Retrieved from


Rugg, M. D. (1995). ERP studies of memory. In M. D. Rugg & M. G. H. Coles (Eds.),

Electrophysiology of Mind: Event-related Brain Potentials and Cognition (pp. 132–170).

Oxford, England: Oxford University Press.

Rugg, M. D., & Allan, K. (2000). Event-related potential studies of memory. In E. Tulving & F.

I. M. Craik (Eds.), The Oxford Handbook of Memory (pp. 521–537). New York, NY:

Oxford University Press.

84

Rugg, M. D., & Curran, T. (2007). Event-related potentials and recognition memory. Trends in

Cognitive Sciences, 11, 251–7. doi:10.1016/j.tics.2007.04.004

Rugg, M. D., Johnson, J. D., Park, H., & Uncapher, M. R. (2008). Encoding-retrieval overlap in

human episodic memory: A functional neuroimaging perspective. Progress in Brain

Research, 169, 339–352.

Rugg, M. D., Otten, L. J., & Henson, R. N. A. (2002). The neural basis of episodic memory:

Evidence from functional neuroimaging. Philosophical Transactions of the Royal Society of

London. Series B, Biological Sciences, 357, 1097–1110. doi:10.1098/rstb.2002.1102

Ryan, J. D., Althoff, R. R., Whitlow, S., & Cohen, N. J. (2000). Amnesia is a deficit in relational

memory. Psychological Science, 11, 454–461. Retrieved from


Rösler, F., Heil, M., & Röder, B. (1997). Slow negative brain potentials as reflections of specific

modular resources of cognition. Biological Psychology, 45, 109–141.

Sanquist, T. F., Rohrbaugh, J. W., Syndulko, K., & Lindsley, D. B. (1980). Electrocortical signs

of levels of processing: Perceptual analysis and recognition memory. Psychophysiology, 17,

568–576.

Schacter, D. L., & Tulving, E. (1994). What are the memory systems of 1994? In D. L. Schacter

& E. Tulving (Eds.), Memory Systems 1994 (pp. 1–38). Cambridge, MA: MIT Press.

Schott, B., Richardson-Klavehn, A., Heinze, H.-J., & Düzel, E. (2002). Perceptual priming

versus explicit memory: Dissociable neural correlates at encoding. Journal of Cognitive

Neuroscience, 14, 578–592. doi:10.1162/08989290260045828

Sikström, S. (2006). The isolation, primacy, and recency effects predicted by an adaptive

LTD/LTP threshold in postsynaptic cells. Cognitive Science, 30, 243–275.

Slagter, H. A, Kok, A., Mol, N., Talsma, D., & Kenemans, J. L. (2005). Generating spatial and

nonspatial attentional control: An ERP study. Psychophysiology, 42, 428–439.

doi:10.1111/j.1469-8986.2005.00304.x

85

Small, S. A., Nava, A. S., Perera, G. M., DeLaPaz, R., Mayeux, R., & Stern, Y. (2001). Circuit

mechanisms underlying memory encoding and retrieval in the long axis of the hippocampal

formation. Nature Neuroscience, 4, 442–449. doi:10.1038/86115

Sommer, W., Schweinberger, S., & Matt, J. (1991). Human brain potential correlates of face

encoding into memory. Electroencephalography and Clinical Neurophysiology, 79, 457–

463.

Stuss, D. T., & Picton, T. W. (1978). Neurophysiological correlates of human concept formation.

Behavioral Biology, 23, 135–162. Retrieved from


Tendolkar, I., Doyle, M. C., & Rugg, M. D. (1997). An event-related potential study of

retroactive interference in memory. Neuroreport, 8, 501–506.

Tulving, E. (1968). Theoretical issues in free recall. In T. Dixon & D. Horton (Eds.), Verbal

Behavior and General Behavior Theory (pp. 2–36). Englewood Cliffs, NJ: Prentice-Hall,

Inc.

Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.),

Organization of Memory (pp. 382–402). New York, NY: Academic Press.

Tulving, E. (1983). Elements of episodic memory. New York, NY: Oxford University Press.

Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26, 1–12.

Tulving, E. (1999). Study of memory: Processes and systems. In J. K. Foster & M. Jelicic (Eds.),

Memory: Systems, Process, or Function? (pp. 11–30). New York, NY: Oxford University

Press.

Tulving, E. (2002). Chronesthesia: Awareness of subjective time. In D. T. Stuss & R. C. Knight

(Eds.), Principles of Frontal Lobe Functions (pp. 311–325). New York, NY: Oxford

University Press.

86

Tulving, E. (2005). Episodic memory and autonoesis: Uniquely human? In H. S. Terrace & J.

Metcalfe (Eds.), The Missing Link in Cognition (pp. 4–56). New York, NY: Oxford

University Press.

Tulving, E. (2007). On the law of primacy. In M. A. Gluck, J. R. Anderson, & S. M. Kosslyn

(Eds.), Memory and Mind: A Festschrift for Gordon H. Bower (pp. 31–48). Hillsdale, NJ:

Lawrence Erlbaum.

Tulving, E. (2010). How do brains detect novelty? In L. Backman & L. Nyberg (Eds.), Memory,

Aging and the Brain: A Festschrift in Honour of Lars Goran-Nilsson (pp. 92–112). East

Sussex, England: Psychology Press.

Tulving, E., & Arbuckle, T. Y. (1963). Sources of intratrial interference in immediate recall of

paired associates. Journal of Verbal Learning and Verbal Behavior, 1, 321–334.

doi:10.1016/S0022-5371(63)80012-2

Tulving, E., & Colotla, V. A. (1970). Free recall of trilingual lists. Cognitive Psychology, 1, 86–

98.

Tulving, E., & Patterson, R. D. (1968). Functional units and retrieval processes in free recall.

Journal of Experimental Psychology, 77, 239–48. Retrieved from


Tulving, E., & Rosenbaum, R. S. (2006). What do explanations of the distinctiveness effect need

to explain? In R. Hunt & J. Worthen (Eds.), Distinctiveness and Memory (pp. 407–423).

New York, NY: Oxford University Press.

Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic

memory. Psychological Review, 80, 352–373.

Van Petten, C., & Senkfor, A. (1996). Memory for words and novel visual patterns: Repetition,

recognition, and encoding effects in the event-related brain potential. Psychophysiology, 33,

491–506.

87

Wagner, A. D., Koutstaal, W., & Schacter, D. L. (1999). When encoding yields remembering:

Insights from event-related neuroimaging. Philosophical Transactions of the Royal Society

of London. Series B, Biological sciences, 354, 1307–1324. doi:10.1098/rstb.1999.0481

Waugh, N. C., & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89–104.

Weyerts, H., Tendolkar, I., Smid, H., & Heinze, H.-J. (1997). ERPs to encoding and recognition

in two different inter-item. NeuroReport, 8, 1583–1588.

Wilckens, K. A., Tremel, J. J., Wolk, D. A., & Wheeler, M. E. (2011). Effects of task-set

adoption on ERP correlates of controlled and automatic recognition memory. NeuroImage,

55, 1384–92. doi:10.1016/j.neuroimage.2010.12.059

Wilding, E. L., & Rugg, M. D. (1996). An event-related potential study of recognition memory

with and without retrieval of source. Brain, 119, 889–905. Retrieved from


Woodruff, C. C., Hayama, H. R., & Rugg, M. D. (2006). Electrophysiological dissociation of the

neural correlates of recollection and familiarity. Brain Research, 1100, 125–135.

doi:10.1016/j.brainres.2006.05.019

Woollams, A. M., Taylor, J. R., Karayanidis, F., & Henson, R. N. (2008). Event-related

potentials associated with masked priming of test cues reveal multiple potential

contributions to recognition memory. Journal of Cognitive Neuroscience, 20, 1114–1129.

doi:10.1162/jocn.2008.20076

Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of 30 years of

research. Journal of Memory and Language, 46, 441–517. doi:10.1006/jmla.2002.2864

Yovel, G., & Paller, K. A. (2004). The neural basis of the butcher-on-the-bus phenomenon:

When a face seems familiar but is not remembered. NeuroImage, 21, 789–800.

doi:10.1016/j.neuroimage.2003.09.034

Yu, S. S., & Rugg, M. D. (2010). Dissociation of the electrophysiological correlates of

familiarity strength and item repetition. Brain Research, 1320, 74–84.

doi:10.1016/j.brainres.2009.12.071

88

Zhou, X., Jiang, X., Ye, Z., Zhang, Y., Lou, K., & Zhan, W. (2010). Semantic integration

processes at different levels of syntactic hierarchy during sentence comprehension: An ERP

study. Neuropsychologia, 48, 1551–1562. doi:10.1016/j.neuropsychologia.2010.02.001

89

Appendices Table 1

Results of the repeated measures ANOVAs for the first principal component (PC1), second principal component (PC2), and third principal component (PC3)

90

Figure 1. Summary of the experimental procedure used in Study 1. Each experimental session consisted of 20 study/test cycles. Each study/test cycle consisted of three phases. During the first phase (Study), participants were presented with a list of 10 pairs of words to study. During the second phase (Distractor Task), participants solved eight simple arithmetic equations. During the third phase (Memory Test), participants were tested on paired associate recall for the 10 pairs from the study phase.

91

Figure 2. The time course for the event-related potential trial to the paired associates in Study 1 and Study 2. TN is the beginning of the Nth trial and TN+1 the beginning of the subsequent trial. Word1 = the first word of a pair. Word2 = the second word of a pair.

92

Figure 3. Group level event-related potential (ERP) data recorded during encoding. (A) Grand average ERP waveforms as a function of word order and subsequent memory performance. The figure shows the ERPs elicited by the first word of subsequently not-recalled pairs (NW1), the second word of subsequently not-recalled pairs (NW2), the first word of subsequently recalled pairs (RW1), and the second word of subsequently recalled pairs (RW2); (B) Word2-Word1 difference waves as a function of subsequent memory performance. The figure shows Word2-Word1 difference waves for subsequently recalled pairs [R(W2-W1)] and subsequently not-recalled pairs [N(W2-W1)] at the group level.

93

Figure 4. Results from the principal component analysis. (A) Topographical distributions of electrode loadings from the rotated component matrix for the first principal component (PC1), principal component 2 (PC2), and principal component 3 (PC3). The top of the figure corresponds to the front of the head. (B) Plots of factor scores for PC1, PC2 and PC3. (C) Grand average event-related potential data at representative electrode P2 for PC1, electrode FCz for PC2, and electrode F8 for PC3. N_W1 = first word of subsequently not-recalled pairs; N_W2 = second word of subsequently not-recalled pairs; R_W1 = first word of subsequently recalled pairs; R_W2 = second word of subsequently recalled pairs.

94

Figure 5. Summary of the experimental procedure used in Study 2. Each experimental session consisted of 8 study/test cycles. Each study/test cycle consisted of three phases. During the first phase (Study), participants were presented with a list of 24 pairs of words to study. During the second phase (Distractor Task), participants solved eight simple arithmetic equations. During the third phase (Memory Test), participants were tested on paired associate recall for half of the pairs presented during the study phase (12 of the 24 pairs). After participants completed the 8 study/test cycles, they were given a final test of paired associate recall (End of Session Memory Test) for all of the pairs that they had not been tested on earlier during the session (the 12 untested pairs from each of the 8 study/test cycles).

95

Figure 6. Group level event-related potential data recorded during encoding. ShortR = subsequently recalled pairs from the ShortDelay condition; ShortN = subsequently not-recalled pairs from the ShortDelay condition; LongR = subsequently recalled pairs from the LongDelay condition; LongN = subsequently not-recalled pairs from the LongDelay condition.

96

Figure 7. Event-related potential data for the ShortDelay condition. (A) The ShortDelay(R-N) waveform recorded at representative electrode FC2. Short_R = subsequently recalled pairs from the ShortDelay condition; Short_N = subsequently not-recalled pairs from the ShortDelay condition; Difference Wave = ShortDelay(R-N). (B) Topographical distribution of the ShortDelay(R-N) waveform between 2200 and 2800 ms (highlighted in section panel A).

Figure 8. Event-related potential data for the LongDelay condition. (A) The LongDelay(R-N) waveform recorded at representative electrode F2. Long_R = subsequently recalled pairs from the LongDelay condition; Long_N = subsequently not-recalled pairs from the LongDelay condition; Difference Wave = LongDelay(R-N). (B) Topographical distribution of the LongDelay(R-N) waveform between 2200 and 2800 ms (highlighted in panel A).

97

0

0.2

0.4

0.6

0.8

1

F2 FC2 C2 CP2

Electrode

uV

Short

Long

Figure 9. Normalized mean amplitudes of the Dm effects observed for the ShortDelay and LongDelay conditions [ShortDelay(R-N) and LongDelay(R-N), respectively] during the time window of the LW (1 to 1.6 seconds after the presentation of the second item of a pair) as a function of electrode (F2, FC2, C2, CP2). Short = short retention interval condition; Long = long retention interval condition.

98

Figure 10. Summary of the experimental procedure used in Study 4. Each participant studied and was tested on free recall for each of 16 lists. The type of list used was alternated between the “ShortList” and “LongList” types.

99

Figure 11. The time course for the evoked potential trial to the study word presentation. TN is the beginning of the Nth trial and TN+1 the beginning of the subsequent trial.

100

Figure 12. Group level event-related potential data recorded during encoding. NR = not-recalled; PM = primary memory; SM = secondary memory.

101

Figure 13. Group level event-related potential data recorded during encoding from electrode sites used for statistical analysis. NR = not-recalled; PM = primary memory; SM = secondary memory.

102

Figure 14. Topographical distributions. (A) Topographical distribution of the primary memory (PM) - not-recalled (NR) difference waveform between 1000 to 1600 ms after the presentation onset of a word. (B) Topographical distribution of the secondary memory (SM) - NR difference waveform between 1000 to 1600 ms after the presentation onset of a word.

103

Figure 15. Summary of the experimental procedure used in Study 3. The experimental session began with a test of participants’ working memory (WM) capacity. Participants then performed the experimental tasks in 80 study/test cycles, which each consisted of two phases. During the first phase (Study), participants were presented with a list of five pairs of words to study. During the second phase (Test), participants were tested on paired associate recall for one of the five pairs that had been presented in the study phase. After the last (80th) study/test cycle was completed, participants were given a final test of paired associate recall (End of Session Memory Test).

104

Figure 16. The time course for the event-related potential trial to the paired associates in Study 3. TN is the beginning of the Nth trial and TN+1 the beginning of the subsequent trial. Word1 = the first word of a pair. Word2 = the second word of a pair.

105

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

1 2 3 4 5

Serial Position

Perc

enta

ge o

f pai

rs r

ecal

led

Figure 17. Percentage of pairs recalled after both the Immediate and Delayed memory tests as a function of serial position. Error bars = standard error.

106

Figure 18. Group level event-related potential data recorded during encoding. ImmN_DelN = pairs that were not subsequently recalled during the Immediate or Delayed memory tests; ImmR_DelN = pairs that were subsequently recalled during the Immediate memory test but not after the Delayed memory test; ImmR_DelR = pairs that were subsequently recalled during both the Immediate and Delayed memory tests.

107

Figure 19. Results of the mean-centered partial least squares analysis. (A) The design scores and corresponding scalp scores of the first latent variable, illustrating the effect that it reflected. ImmN_DelN = pairs that were not subsequently recalled after the Immediate nor the Delayed tests; ImmR_DelN = pairs that were subsequently recalled on the Immediate test, but not after the Delayed test; ImmR_DelR = pairs that were subsequently recalled on the Immediate and Delayed tests. (B) Grand average event-related potential trial waveforms recorded at representative electrodes F6 and CP5 for the right fronto-central and the left centro-parietal scalp regions, respectively. The highlighted 2200-2800 ms time window corresponds to the data that were submitted to the PLS analysis. Black markers at the top of each plot, within the highlighted time window, indicated when the saliences were stable by bootstrap estimation. (C) The topographical distribution of the electrode saliences, corresponding to the 2200-2800 ms time window highlighted in panel (B), indicating where on the scalp the effect picked up by the latent variable was strongest.

Electrophysiological Evidence of Human Memory …...associate recall were qualitatively different...

Documents

Transcript of Electrophysiological Evidence of Human Memory …...associate recall were qualitatively different...