2013. Vol.4, No.9, 695-703
Published Online September 2013 in SciRes (
Copyright © 2013 SciRes. 695
A Novelty-Induced Change in Episodic (NICE) Context
Account of Primacy Effects in Free Recall
Eddy J. Davelaar
Department of Psychological Sciences, Birkbeck, University of London, London, UK
Received June 1st, 2013; revised July 2nd, 2013; accepted July 29th, 2013
Copyright © 2013 Eddy J. Davelaar. This is an open access article distributed under the Creative Commons At-
tribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
Formal cognitive models of episodic memory assume that during encoding list items become associated
with a changing context representation. However, this representation is recency-biased and thus can not
account for primacy effects under conditions that prevent rehearsal. In this paper, it is hypothesized that
one source underlying primacy effects is the detection of novelty. In three experiments, it is shown how
novelty at the perceptual and semantic level can explain the full serial position function of first recall
probabilities, including primacy effects. It is proposed that an item becomes distinctive due to increase in
the change within a distributed episodic context representation, induced by novelty detection. The theory
makes three assumptions. First, items become associated with a distributed context representation. Second,
the context representation changes with item presentation. Third, the rate of contextual change is related
to the perceptual and conceptual difference computed between the presented item and the previous item
(or items in the buffer). This theory captures primacy effects in first recall probabilities without recourse
to a rehearsal process and provides a mechanistic account of distinctiveness.
Keywords: Distinctiveness; Novelty-Detection; Episodic Context; Distributed Context; Contextual
One of the most robust results in cognitive psychology is the
U-shaped serial position curve obtained in the immediate free
recall task (Murdock, 1962). In the free recall task, participants
are presented with a sequence of words and are instructed to
recall as many words as possible in any order. Words presented
at the beginning or end of the sequence are remembered better
than words presented in the middle of the sequence. These
phenomena are referred to as primacy and recency effects, re-
spectively. There exists a number of theories that can account
for these results and have advanced our understanding of the
mechanisms of human memory.
Primacy effects in list memory have been explained in a va-
riety of ways. Many dual-store and dual-trace models assume
that the first item enters an empty short-term memory (STM)
buffer from which it is displaced after the buffer is filled to
capacity, leading to that item residing in STM longer than sub-
sequent items. As episodic trace strength is a function of the
duration that items reside in the buffer, the episodic trace for
the first item will be stronger than subsequent items (Atkinson
& Shiffrin, 1968; Davelaar et al., 2005; Raaijmakers & Shiffrin,
1980). Other encoding models explain primacy effects in terms
of more opportunity for rehearsing the first items (Murdock &
Metcalfe, 1984; Tan & Ward, 2001; Ward, 2002). There are
two types of rehearsal explanations. In one, increased rehearsal
opportunities for the first few items lead to increased opportu-
nities for encoding and hence to a stronger trace. This re-
hearsal-enhanced encoding is assumed in early work on re-
hearsal (Atkinson & Shiffrin, 1968; Rundus, 1971; Rundus &
Atkinson, 1971; Brodie & Murdock, 1977). In the second type
of rehearsal explanation, increased rehearsal opportunities for
the first few items lead to increased probabilities that the items
are still in the rehearsal cue at the end of the list presentation.
This rehearsal-enhanced accessibility of the items is assumed in
the more recent work on rehearsal (Tan & Ward, 2001) in
which no short-term buffer is presumed. This denial of a short-
term buffer requires that such models liken the rehearsal proc-
ess that occurs during list presentation to mini-retrievals (Ward,
2002; see also Laming, 2006). Retrieval-based explanations of
primacy effects appeal to the notion of the first items being
more distinctive (Murdock, 1960; Neath, 1993). This notion has
proven useful to account for the finding that after retrieval of
end-of-list items, participants continue retrieving begin-of-list
items, as if the first item contains a tag that is accessible at
retrieval (Davelaar et al., 2005; Murdock & Metcalfe, 1984).
Although the explanations in terms of buffer-enhanced encod-
ing, rehearsal-enhanced encoding, rehearsal-enhanced accessi-
bility, and distinctiveness all capture the basic pattern, they are
not mutually exclusive.
When procedures are in place that minimizes the use of re-
hearsal, primacy is still found. For example, Richardson and
Baddeley (1976) found primacy effects in immediate free recall
under articulatory suppression. Primacy effects in immediate
free recall are also found in an incidental task (Baddeley &
Hitch, 1977) and when participants are required to make a se-
mantic judgment for each word (Howard & Kahana, 1999). In a
procedure called the continuous-distractor paradigm, each item
is preceded and followed by an interval of distractor activity.
Even though this procedure abolishes the opportunity to re-
hearse, the items during the distractor activity and equates the
duration that each item resides in the buffer, primacy effects are
still found (e.g., Bjork & Whitten, 1974; Neath, 1993; Thapar
& Greene, 1993). These results suggest that not all primacy
gradients found in serial position functions are a consequence
of a rehearsal mechanism or a buffer. However, this is not to
say that all primacy gradients are due to distinctiveness-based
retrieval mechanisms.
Evidence supporting the view that primacy effects may have
an origin during encoding comes from neuroscientific investi-
gations. Electrophysiological studies have revealed that en-
hanced gamma band activity is correlated with enhanced mem-
ory performance for earlier list positions (Sederberg et al.,
2006). In the neuroscientific literature these results have been
discussed with reference to the computation of novelty (De-
bener, Herrman, Kranczioch, Gembris, & Engel, 2003): the
higher the novelty of an item, the larger the neural response.
The neural response for the first item would then be the result
of a larger difference between the first item and the fixation
stimulus (that is typically presented before the first item) than
between the second and the first item. Importantly, it has been
shown that increased activation in the same brain areas that
compute novelty also predicts recall performance (Kirchoff et
al., 2000).
Whereas the neuroscientific evidence points to the me-
dial-temporal cortex as the site for novelty signals that are re-
lated to increased memory, neurocomputational models of
memory have implicated the medial-temporal cortex as a con-
textual system (e.g., Howard et al., 2005; Norman & O’Reilly,
2003). This triple conjunction of novelty, contextual change
and enhanced memory in the same brain area raises the ques-
tion whether a novelty signal is directly related to contextual
change, which in turn is related to increased memory perform-
ance. In what follows next, I will put forward the hypothesis
that novelty is one source for change in episodic context and
that contextual change leads to enhanced memory for the first
item after the change. The hypothesis is critically different from
its nearest neighbors. That is, the novelty signal itself does not
lead to an enhanced memory trace or causes an increase in di-
mensional distinctiveness at retrieval. Instead, the contextual
change at encoding leads to the first item after a change to be
relatively less different from the context during retrieval than
expected based on temporal distance alone. As above, the pro-
posed hypothesis is compatible with the other explanations of
primacy effects and the question which view is superior is not
taken up here.
Together, the behavioral data on primacy effects supports the
existence of a mechanism that favors the first item in a list and
the neuroscientific data points towards a novelty-signal that is
computed on-line. To directly test the idea that novelty contrib-
utes to primacy effects, a series of experiments was conducted
addressing the following questions. First, can primacy effects
be enhanced by making the first item in a list more novel?
Second, can novelty during list presentation create a midlist
primacy effect? To preview, the data confirms the role of nov-
elty in generating primacy effects. In the next section, a for-
malization of the novelty-induced primacy effect is presented in
what will be called the novelty-induced change in episodic
context model or NICE context model for short. This is then
followed by three experiments and a general discussion.
Distinctiveness through Novelty-Driven Change
in Episodic Context
Within the memory literature, the term distinctiveness has
been used as an explanation for such well-known empirical
findings as the primacy effect in serial position functions
(Murdock, 1960; Neath, 1993) and the isolation effect (Brown,
Neath, & Chater, 2007). However, some researchers have ar-
gued that “Distinctiveness itself cannot be used as an explana-
tion for effects of distinctiveness” (Hunt & Lamb, 2001: p.
1359). Instead perceived novelty leads items to become en-
coded in episodic contexts that have just undergone a relatively
large change and are therefore seen at retrieval as being more
distinctive. In other words, distinctiveness is not a process, but
an end-product. The model presented here adds a further nu-
ance in that local distinctiveness and encoding-retrieval match
are tightly connected.
In this paper, a novelty-based mechanism is proposed that
produces primacy effects. The model is a changing-context
model in which contextual change is driven by novelty. The
model is named the Novelty-Induced Change in Episodic
(NICE) context model and has three critical assumptions, two
of which were employed successfully in previous chang-
ing-context models (i.e., Glenberg et al., 1983; Howard & Ka-
hana, 1999; Mensink & Raaijmakers, 1988). First, context is a
distributed representation of active and inactive elements. Sec-
ond, the context representation changes slowly over time by
making active elements inactive (with probability ) and inac-
tive elements active (with probability ). Third, the probability
of activation is a positive function of the novelty of a stimulus.
In short, the model is a Markov system with novelty-dependent
transition probabilities. Each of these assumptions will be ad-
dressed in turn.
Distributed Context
In following several investigators (e.g., Burgess & Hitch,
1999; Dennis & Humphreys, 2001; Glenberg et al., 1983;
Howard & Kahana, 2002; Mensink & Raaijmakers, 1988;
Norman & O’Reilly, 2003) context is a distributed representa-
tion of active and inactive elements. The anatomical location of
this context representation may be debated, but in some com-
putational models the entorhinal cortex in the medial-temporal
lobe is assumed to be the location (Howard et al., 2005; Nor-
man & O’Reilly, 2003). It is known that patients with damage
to the medial-temporal lobe suffer from long-term memory loss
(Scoville & Milner, 1957; Baddeley & Warrington, 1970), be-
ing unable to recall items that were presented early in the list.
Assuming that the elements correspond to (groups of) neurons
in the medial-temporal lobe allows an investigation whether the
model is able to address the finding that activation in the me-
dial-temporal lobe reflects not only the amount of novelty de-
tected, but also is predictive of subsequent memory.
Contextual C h an ge
During presentation of a stimulus, active contextual elements
may be de-activated, while currently inactive elements may be
activated. This process is probabilistic with being the prob-
ability that an active element becomes inactive and with be-
ing the probability that an inactive element becomes active. In
following Mensink and Raaijmakers (1988; see also Estes,
1955; Murdock, 1972), the total number of active elements A(t)
Copyright © 2013 SciRes.
at time t is given by:
 
AtA eNe
which asymptotes at N/( + ) (the maximum number of ac-
tive elements). A(0) is the number of active elements at time =
In the models of Mensink and Raaijmakers (1988) and How-
ard and Kahana (2002), the transition probabilities were as-
sumed to remain constant. However, this assumption makes the
system’s dynamics independent of the nature of the items. This
counters the evidence reviewed above showing systematic
variations in brain activation based on the type of material and
more specifically the novelty of an item. Here, it is assumed
that and are variable (see also Murdock, 1972) and are
positively correlated with the novelty of an item. Novelty, , is
measured as the total number of attended features in a (distrib-
uted) stimulus representation that become activated. Stimulus
representations contain semantic features of the to-be-remem-
bered word and the features of the background picture on which
the word is presented, but only those that are attended to by the
participant. The probabilities and for a given could then
be formulated as = F(0, ) and = G(0,). Here, 0 and 0
are the minimal probabilities that are present when a stimulus is
immediately repeated. The functional form of equations F()
and G() are not known, but suffice is to say that > 0 for
several of the data sets looked at. Future work could address the
precise equations in detail. These equations show that the dis-
tance between the values of and becomes smaller with in-
crease in novelty and implies that the asymptotic level of the
total number of active elements, N/( + ), would then be-
come larger (with N/2 as its maximum). This is important, as it
suggests that with these three assumptions in place, the number
of active elements that will be encoded with the stimulus is
larger for a novel than for a similar stimulus. In other words,
similar stimuli are encoded in similar contexts. Novel stimuli
are encoded in dissimilar contexts (increased contextual change)
with more contextual elements.
Memory Encoding a nd Retrie va l
During encoding, active stimulus features and active contex-
tual elements become associated. This matrix of connections
represents the episodic memory of the events and is used in the
retrieval of those events. Memory performance is assumed to be
dependent on the similarity between the context at encoding
and the context at test (Tulving, 1983; see Nairne, 2002 for a
challenge). This is equivalent to the number of elements that
are active at the time of test and were associated with a particu-
lar stimulus. The details of a retrieval process are less important
than the actual consequence that novelty-driven contextual
change has on the retrieval probability (for detailed models see
Howard & Kahana, 2002; Mensink & Raaijmakers, 1988).
However, several general aspects of this contextual retrieval are
relevant. First, the similarity and therefore the probability of
retrieval decreases with increasing distance between the en-
coded context and the test context, leading to a recency effect
(see Howard & Kahana, 2002). Second, due to the transition
probabilities being novelty-dependent, a novel stimulus will be
associated with more contextual elements and the context rep-
resentation will differ more strongly from the context represen-
tation associated with its predecessor. This would lead to a
higher likelihood of retrieving novel stimuli.
Model Simulation
Figure 1 shows a demonstration of the NICE context model.
Vectors of 100,000 binary elements were used in the simulation
with values = 0.8 and = 0.01. A list of 10 items is simulated
in which no novelty signal is provided at the beginning of the
list, novelty is provided at the beginning, and where a novelty
signal is also provided during presentation of the item in posi-
tion 6. Novelty is implemented as a momentary increase in =
0.08. The striking observation is that the NICE context model is
able to produce a primacy effect due to change in episodic con-
text as well as showing an enhancement in memory perform-
ance for the novel item in position 6. In other words, the model
gave rise to primacy via a process of novelty-induced contex-
tual change.
To understand the reason why the model produces this strik-
ing pattern, the dissimilarity among the context vectors that
were encoded with the ten items and the context vector prior to
the first item (labeled “B”) and after the last item (labeled “R”)
are plotted in a multidimensional scaling (MDS) solution in
Figure 2. The model was rerun 1,000 times for each of the
three lists and subjected to MDS analyses, forcing the retrieval
vector, R, in the upper right quadrant.
In the simulation without novelty for the first item, the MDS
solution forms a horseshoe in two dimensions with B and R at
either side. The MDS solution using three dimensions were not
significantly worse, but revealed that along the third dimension
the B and R are far removed. When the presentation of the first
item coincides with increased novelty, the horseshoe pattern
changes such that the vector for the first item moves away from
the B vector and gets closer to the R vector. In other words, the
first item becomes locally more distinctive from the start of the
list and more similar to the retrieval context. When the item in
the sixth position coincides with increased novelty, the same
happens in that the sixth vector gets closer to the R vector.
Moreover, two solutions exist with different orientations for the
sequence involving the first five items, which implies that the
Figure 1.
Simulated serial position functions showing lack of primacy
effects without novelty and primacy effects when novelty is
Copyright © 2013 SciRes. 697
Figure 2.
Multidimensional scaling solutions for the three simulations. The solu-
tions are arranged to have the retrieval context vector, R, in the up-
per-left quadrant. Each panel includes the context vectors for all ten
items and the context vector prior to the first item, B. The lower panels
show MDS solutions (without B) of the simulation with novelty in
positions 1 and 6. The left and right panels present 840 and 160 solu-
tions, respectively.
novelty during presentation of the sixth item created two sub-
lists. This is further strengthened by the fact that the distance
between context vectors 5 and 6 is larger in both solutions
compared to the unitary lists. This provides the computational
argument that the enhanced memory for the first item after a
change is contextually similar to a primacy effect.
Experimental Study
In applying the model to data, the question arises what type
of data is suitable. The model only addresses contextual change
and therefore can only be tested with data on first recall prob-
abilities. To date no published report addressed first recall
probabilities in relation to novelty-induced change. Three free
recall experiments were conducted to provide such data. In
Experiment 1, the relative novelty of the very first item is ma-
nipulated to test the assertion that novelty-induced change in
episodic context contributes to primacy effects in the absence
of rehearsal. In Experiment 2, the background context is ma-
nipulated while participants memorized words. In Experiment 3,
a critical item is embedded in a list of category exemplars from
a different semantic category.
Experiment 1
In Experiment 1, participants were presented with a modified
delayed free recall task in order to explore the mechanisms
underpinning the primacy effect. The novelty of the first item
was manipulated by preceding it with a variable number of
nonword items consisting of random consonants. In addition,
we looked at whether rehearsal interacts with novelty. If the
primacy effect due to novelty interacts with that induced by
rehearsal then it is assumed that both are manifestations of a
single process. Thus lack of an interaction can be interpreted as
a validation of a separate process that contributes to primacy
The rationale for the particular manipulation of novelty is as
follows. The calculation of novelty implies that a current item
is different from the items that preceded it. This difference
might be at several levels, such as perceptual (different colors)
and conceptual (different semantics). As we used unrelated
word in the experiment, each word is equally different concep-
tually and perceptually from its predecessor, except the very
first item. This item is preceded by the fixation prompt that tells
the participant that the presentation of the list will begin. Thus,
the very first item coincides with additional perceptual change.
To equate the differences at the perceptual level, letters are
used instead of a fixation cross, such as a plus sign. In addition,
it is expected that the computation of novelty is larger after a
series of similar items. Thus, in the sequence “+ A B”, the word
“B” is less novel than in the sequence “+ A A A A B”. As we
are interested in the resulting primacy effect that is due to en-
coding processes and not due to dynamics unfolding during the
retrieval phase, we looked at the first recall of a trial. This is the
purest measure of primacy effect, uncontaminated with re-
trieval-related strategies. We expected that there will be more
primacy than recency when there are more nonwords preceding
the very first item.
Materials and Methods
Forty-eight participants (22 females) took part in this expe-
riment. Half were allocated to the no-rehearsal condition and
the other half to the rehearsal condition.
The experiment conformed to a 5 × 10 × 2 mixed factorial
design crossing the between-subjects factor rehearsal (“no
rehearsal” group, n = 24, “rehearsal” group, n = 24) with two
within-factors: nonwords (number of nonwords 0 - 4) and word
serial position.
A total of 230 common words and 45 nonwords were used as
stimuli. The words were selected such that none had any nega-
tive connotations (which could lead to enhanced novelty) and
all had a word frequency count between 100 - 1000 in a million.
Words within each list did not have any semantic or phono-
logical relation. Letter repetition on the same letter-position
was avoided. Nonwords consisted of strings of 3 to 5 conso-
nants, matching the range of word lengths. Example nonwords
include: kpnsk, mjn, klbn. Each of the 23 lists consisted of ten
words and 0 to 4 nonword items, creating five conditions with 4
lists in each condition (and 3 practice lists).
Participants were tested individually and received instruc-
tions on the computer screen and verbally by the experimenter.
All stimuli were presented on the computer screen. Participants
in the rehearsal condition received instructions to rehearse the
words from the list during list presentation. After instructions
about the delayed free recall task, participants practiced two
10-word lists, one without nonwords (nonword - 0 condition)
and one with three nonwords (nonword - 3 condition). The
sequence of ten words and variable number of nonwords was
presented at a rate of one item per second in the middle of the
screen, each item masking the previous one. After the words
were presented, a prompt appeared indicating the start of the
Copyright © 2013 SciRes.
distractor task. The distractor task consisted of nine mathe-
matical problems, such as “7 3 = 3” and participants indicated
the accuracy of the equation by pressing the K- or S-key on the
keyboard when it was correct or incorrect, respectively. The
maximum duration for each mathematical problem was 2.5 s
and participants were advised to respond within 2 seconds to
each of the problems as accurately as possible.
Following this distractor activity, three question marks ap-
peared, prompting the participant to start recalling as many
words from the list as possible in any order. Although only the
very first response was of interest, all answers were record by
the experimenter to encourage the participant to memorize all
the words during list presentation. After the two practice trials,
21 trials were presented of which the first one was a “warming
up list” with two nonwords and was thus excluded from the
analyses. The test lasted approximately 25 minutes.
Results and Discussion
Figure 3 presents the serial position curves of the first recall
probabilities for the no-rehearsal and the rehearsal group. From
the figure it is clear that the distractor task abolished the re-
cency effect in the no-rehearsal group. However, a recency
effect is present in the rehearsal group, which is due the fact
that participants are able to rehearse items during a difficult
distractor task if requested to do so. As the first recall prob-
abilities are normalized (the values sum to one), the rehearsal
group necessarily have lower levels of primacy.
As our focus is on the primacy effect, we used the PR-index
Figure 3.
Serial position curves for both groups as a function of the
number of nonwords seen prior to the very first item.
(Davelaar et al., 2005) to get a relative estimate of the primacy/
recency gradient. The PR-index is larger than 0.5 when there is
more recency than primacy and smaller than 0.5 when there is
more primacy than recency. Therefore, the PR-index summa-
rizes the entire the serial position curve in a single value. This
value varies as a function of the number of nonwords presented
as shown in Figure 4. The data in Figure 4 were subjected to a
2 (rehearsal group) × 5 (number of nonwords) mixed ANOVA.
There was a main effect of number of nonwords [F(4, 184) =
2.573, MSe = .017, p < .05], showing more primacy with in-
creasing number of nonwords in both groups. The main effect
of group and the interaction between group and number of
nonwords were not significant (p > .12).
Participants in both groups showed tendency to recall more
primacy words from lists preceded by the maximum four non-
word items. This finding supports the hypothesis that the pri-
macy effect is enhanced when the very first item coincides with
a larger level of novelty. In addition, instead of enhanced pri-
macy effects in the rehearsal group, the group shows recency
effects. This finding shows that the rehearsal process retrieves
only the more recent items. Typically, the immediate free recall
task is used in which the most recent items can be retrieved
from short-term memory and thus rehearsal can not contribute
to the performance. However, in the delayed free recall task, all
items have been displaced from short-term memory. Any re-
hearsal process that can unfold will not take the very first item,
but instead the most recent item, as expected from a re-
cency-based account of retrieval (Tan & Ward, 2000; Ward,
2002). Finally, the lack of an interaction between novelty and
rehearsal confirms that the two are operating on different parts
of the list.
Experiment 2
Experiment 1 confirmed the view that novelty of the very
first item contributes to the primacy effect in free recall. This
novelty was induced by presenting a number of items that differ
from the very first item and were similar to each other. In prin-
ciple, it should be possible to have a primacy effect present in a
single list of items in which the first half of the list differs from
the second half of the list. In our conception, the presentation of
the very first item of the second half will coincide with a high
level of novelty and should thus be better remembered. In Ex-
Figure 4.
The primacy/recency index computed over the first recall
probabilities. The smaller the index, the more primacy
there is in the serial position curve.
Copyright © 2013 SciRes. 699
periment 2, we varied the background on which words are pre-
sented. The rationale for this choice was that the background is
unrelated and irrelevant to the task, but forms part of the epi-
sodic context in which the word is encoded. When the same
context is presented during retrieval, memory for items encoded
in that context should be better. This allows investigating
whether the match in context at encoding and retrieval interacts
with the effect of novelty. If this is the true, novelty and epi-
sodic context are related.
Materials and Methods
Forty-one participants (35 females) took part in this experi-
The current experiment conformed to a 5 × 10 within-subject
design crossing the factors background (5 levels) and word
serial position.
A total of 150 common words were used as stimuli. The
words were selected such that none had any negative connota-
tions and all had a word frequency count between 100 - 1000 in
a million. Words within each list did not have any semantic or
phonological relation. Letter repetition on the same letter-posi-
tion was avoided. Each of the 15 lists consisted of ten words.
Background pictures were landscapes of forests, beaches, cities,
and mountain ranges.
Participants were tested individually and received instruc-
tions on the computer screen and verbally by the experimenter.
Each word was presented for one second on the computer
screen on top of a picture that filled the entire screen. Partici-
pants were told to ignore the background picture and focus on
the words to be remembered. After the final word, participants
recalled as many words as possible in any order. There was no
time limit imposed for recall.
The background picture varied according to 5 conditions.
Condition 1 (AA/A): the background picture was the same
for all 10 words and was also present during the retrieval.
Condition 2 (AA/B): the background picture was the same
for all 10 words, but a different picture was presented during
Condition 3 (AB/A): words 1-5 were presented on top of a
different picture than words 6-10 and during the retrieval the
picture that was presented during the first half was re-presented.
Condition 4 (AB/B): same as condition 3, but during re-
trieval the picture that was presented during the second half was
Condition 5 (AB/C): same as condition 3, but during re-
trieval the picture was a totally different one.
The lists were counterbalanced across participants so that the
different lists were presented in each of the different con-
text-conditions. There were 3 lists per condition and no picture
was presented in more than one trial.
Results and Discussion
Figure 5 shows the first recall probabilities as a function of
Figure 5.
Top panel: Serial position curves for the conditions in
Experiment 2. The first five items were presented with
the same background picture, denoted by “A”. The sec-
ond five items were presented with the same background
picture, but could be either the same or different than the
first list half. During retrieval, a picture stayed on the
screen which was either the same or different than the
picture presented during encoding. Compare with the
simulation results in Figure 1. Bottom panel: the raw
scores for all experimental conditions.
context condition.
A deviation score was calculated between conditions 2 - 5
and condition 1 (not shown), which serves as a baseline condi-
tion, for the words in positions 1 and 6. A 4 (condition) × 2
(list-half) ANOVA conducted on the deviation scores revealed
a marginal effect of condition [F(3, 120) = 2.50, MSe = 0.026,
p = .063] and a significant contrast effect on the interaction
[F(1, 40) = 6.38, MSe = 0.022, p < .05]. This was further un-
packed using t-tests. Comparing conditions AA/B and AB/C
revealed a significant effect of a contextual change despite the
lack of an encoding-retrieval match [t(40) = 2.23, p < .05].
When the retrieval context matched either list-half, the list po-
sition matching the context was enhanced (position 1 in AB/A
and position 6 in AB/B), although this did not reach signifi-
cance. The AB/B [t(40) = 3.11, p < .01] and AB/C [t(40) = 2.23,
p < .05] conditions showed enhanced recall for position 6. Thus,
both change in episodic context during list presentation and
contextual encoding-retrieval match contribute to first recall.
Finally, comparing the AB/B and AB/C conditions revealed no
statistical difference, suggesting that the change-induced en-
Copyright © 2013 SciRes.
hancement and the encoding-retrieval match are not independ-
Experiment 2 reveals that changing an irrelevant background
picture is sufficient to produce enhanced recall performance for
the first item after the change. In analogy, this constitutes a
mini-primacy effect. As shown in the simulations in Figures 1
and 2. This is consistent with the view that novelty as induced
by change in episodic context contributes to primacy effects.
The data in Experiment 2 is too sparse to run correlational
analyses relating the size of the primacy effect with the size of
the novelty effect. This is taken up in Experiment 3.
Experiment 3
Experiments 1 and 2 provide the first evidence of primacy
effects being driven by novelty. However, in Experiment 1, the
novelty was induced at the item level, whereas in Experiment 2
this was induced at the level of episodic context. It is therefore
possible that the mini-primacy effect was driven by the change
in context and not necessarily by item-level novelty. The third
and final experiment employs a delayed free recall task of se-
mantically related items in which one item in the middle of the
list is semantically unrelated to the rest. The novelty for that
item is then based on item-level analysis. This unrelated item,
the isolate, is positioned either at the beginning or in the middle
of the list. If item-level novelty (and not just context-level nov-
elty) contributes to primacy effects, then normal primacy ef-
fects should correlate with the increased memory for the isolate
when presented in the middle list position.
Materials and Methods
Seventeen participants (11 females) took part in this experi-
The current experiment conformed to a 2 × 12 within-subject
design crossing the factors isolate position (position 1 or posi-
tion 6) and word serial position.
A total of 240 common words were used as stimuli. The
words were selected such that none had any negative connota-
tions and all had a word frequency count between 100 - 1000 in
a million. Words within each list were semantically related by
virtue of being exemplars for a single category. Each of the 20
lists consisted of twelve words. A word that was unrelated to
the words in the list was presented either on position 1 (b-list)
or position 6 (m-list). There were 10 lists per condition and the
location of the isolate was counterbalanced across participants.
Participants were tested individually and received instruct-
tions on the computer screen and verbally by the experimenter.
Each word was presented for one second on the computer
screen. After the final word, a prompt appeared indicating the
start of the distractor task. The distractor task consisted of nine
mathematical problems, such as “7 – 3 = 3” and participants
indicated the accuracy of the equation by pressing the K- or
S-key on the keyboard when it was correct or incorrect, respec-
tively. The maximum duration for each mathematical problem
was 2.5 s and participants were advised to respond within 2
seconds to each of the problems as accurately as possible. Fol-
lowing this distractor activity, three question marks appeared,
prompting the participant to start recalling as many words from
the list as possible in any order. There was no time limit im-
posed for recall.
Results and Discussion
Figure 6 presents the first recall serial position curves. As
can be seen, the isolate is reported more often that the words in
the same position, but related to the other words in the list. This
constitutes the basic isolation effect also known as the Von
Restorff effect. Our focus is on whether the size of the isolation
effect on position 6 (m-list minus b-list) correlates with the
normal primacy effect (position 1 in the m-list) and with the
additional benefit when the isolate is in position 1 (b-list minus
m-list). To provide a baseline, the correlational analyses were
repeated for all other serial positions (2 - 5, 7 - 12). The corre-
lation between the recall difference in position 6 correlated
significantly with the primacy effect [r(17) = .63, p < .01] and
with the addition to the primacy effect [r(17) = .91, p < .001].
The only other position that showed significance was the final
list position with the normal primacy effect [r(17) = .51, p
< .05]. This pattern makes sense given that for the recall, the
end-of-list context is used as a memory cue and is directly re-
sponsible for the first recalls. The last item is encoded in an
episodic context that is similar to the end-of-list context. How-
ever, the final list item does not coincide with enhanced novelty
and therefore it does not correlate with the novelty-induced
enhancement on the primacy effect or even with the novel item
The results support the view that item-level novelty drives
better first recall performance of the items and that normal pri-
macy effects is partly driven and is enhanced by item-level
novelty. As an aside, the lack of correlations between the pri-
macy effects and item number 7 suggests that the semantic
novelty of item 7 is not as great as item number 6. This sug-
gests that novelty is calculated against all items in short-term
memory, which is consistent with the results of Experiment 1
where the primacy gradient is gradually enhanced when placing
more nonwords in short-term memory. However, Experiment 2
clearly showed a close connection between novelty-induced
memory enhancement and change in episodic context. The
overall pattern of results can be accommodated in the NICE
Figure 6.
Serial position curves of the data in Experiment 3. The
isolate was presented either in position 1 or in position 6.
Copyright © 2013 SciRes. 701
context model to which we now turn.
General Discussion
The three experiments provide evidence for the view that
item-level novelty and context-level novelty enhance memory
performance and contribute to primacy effects in free recall. As
expected (Glenberg et al., 1983; Howard & Kahana, 1999;
Murdock, 1972), the NICE context model produces recency
effects. However, the important difference is that the probabili-
ties of activating and de-activating contextual elements are
made conditional on the novelty value between the previously
and the newly presented items. As the first item is different
from the information preceding it, its presentation coincides
with an enhanced level of novelty, which in turn speeds up the
change in episodic context. This additional change makes that
first item more locally distinctive and more accessible during
Although the begin- and end-points have been used in sev-
eral models in memory retrieval (Brown, Preece, & Hulme,
2000; Burgess & Hitch, 1999; Davelaar et al., 2005; Henson,
1998; Metcalfe & Murdock, 1981; Neath, 1993; Shiffrin &
Cook, 1977), none of these models provide a mechanistic ex-
planation for the higher distinctiveness of the beginning and
end of a list. The NICE context model provides such an expla-
One of the assumptions is that novelty is measured as the
difference in representation between the current and the previ-
ous item (plus background). This is consistent with single-store
models that do not postulate a limited-capacity buffer. When a
buffer is postulated, however, the novelty-signal could be
computed from the current item and the buffer items. Such a
change to the model does not invalidate the central thesis of this
paper (i.e., Novelty can Induce Change in Episodic context) and
may even provide a novel way to address the need for a lim-
ited-capacity buffer in understanding free recall performance.
In addition, it is assumed that stimulus representations con-
tain semantic features of the to-be-remembered word and the
features of the background picture on which the word is pre-
sented. Although this captures the data with a minimum set of
assumptions, this operationalization of stimulus representations
comes from the view that the background environment is also
input and differs from the internal context representation. De-
spite the regular usage of context in computational models of
memory, no clear understanding has been reached about what
context is (and is not) (see for a recent review, Klein, Shiffrin &
Criss, 2007).
Regarding novelty detection, it has been proposed that detec-
tion of novelty is related to increase in the release of acetylcho-
line, a neurotransmitter that originates from the basal forebrain
and modulates several brain areas, including structures in the
medial-temporal lobe. The mechanism by which the brain de-
tects novelty is yet unclear, but there are several ways in which
a novelty signal can be obtained in a neurocomputational model.
For example, in ART models (Grossberg, 2012) an interplay
between bottom-up input and top-down feedback leads to a
novelty-signal when the top-down feedback does not resonate
with the bottom-up input. In those models, the resulting nov-
elty-signal is used to allow a new higher-level representation to
become activated and get associated with the novel bottom-up
input. A non-interactive (bottom-up and top-down) architecture
is also possible when using units that adapt over time. This
adaptation (aka habituation, neural fatigue, synaptic depression)
leads to the total activation to be an indicator of perceived
stimulus novelty (Davelaar et al., 2011).
The NICE context model provides a solution to the question
of how a context model can retrieve the first item of a preced-
ing list. Jang and Huber (2008) replicated the so-called list-
before-last experiments by Shiffrin (1970), showing that par-
ticipants are able to skip the just encoded list and retrieve items
from the preceding list. This provides a major challenge to all
context models that are recency-biased. One solution is to as-
sume that a hierarchy of contextual elements exists, such that
different levels have faster or slower transition probabilities (cf.
Davelaar & Usher, 2002; Glenberg et al., 1983). The modeler
can then choose those elements that fluctuate in the same
rhythm as the presentation of the trials. Another approach is to
assume that each first item is more distinctive than the other
items and thus can provide an anchor point during retrieval.
The NICE context model produces memory traces for the first
items of all lists that are locally distinctive. Thus retrieving the
first item of the preceding list is easier than retrieving the mid-
dle item of the preceding lists, despite that the middle items are
temporally more recent. It also naturally accounts for the ob-
servation that after retrieving items from the end of the list, the
next item to be retrieved is most likely to be the very first list
item (Laming, 1999).
Concluding Remarks
The current approach is not meant as an alternative to exist-
ing models, but instead introduces a mechanism by which other
context models can capture primacy effects. For example, in
TCM (Howard & Kahana, 2002), where contextual change is
item-dependent, a similarity structure between the retrieved
contexts of list items could be introduced. As the retrieved con-
text vector of the first item differs more from that of the fixa-
tion stimulus than that of the second item, the change in the
ongoing context will be larger for the first than the second item.
It remains to be seen whether this modification to TCM would
produce the desired results. Nevertheless, the basic concept will
be the same, i.e., a source of contextual change is the novelty of
an item at perceptual and conceptual levels.
Author Note
Preliminary versions of this work were presented at the 3rd
Context and Episodic Memory Symposium in 2005 and the 1st
Computational Cognitive Neuroscience Meeting in 2005. The
author would like to thank Marta Sibilska for running Experi-
ment 1.
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed
system and its control processes. In K. W. Spence, & J. T. Spence
(Eds.), The psychology of learning and motivation: Advances in re-
search and theory (Vol. 2, pp. 89-195). New York: Academic Press.
Baddeley, A. D., & Hitch, G. J. (1977). Recency re-examined. In S.
Dornic (Ed.), Attention and performance (Vol. 6). Hillsdale, NJ:
Lawrence Erlbaum Associates Inc.
Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and the distinc-
tion between long- and short-term memory. Journal of Verbal Learn-
ing and Verbal Behavior, 9, 176-189.
Copyright © 2013 SciRes.
Copyright © 2013 SciRes. 703
Bjork, R. A., & Whitten, W. B. (1974). Recency-sensitive retrieval
processes in long-term free recall. Cognitive Psychology, 6, 173-189.
Brodie, D. A., & Murdock, B. B. (1977). Effects of presentation time
on nominal and functional serial position curves of free recall. Jour-
nal of Verbal Learning and Verbal Behavior, 16, 185-200.
Brown, G. D. A., Neath, I., & Chater, N. (2007). A temporal ratio
model of memory. Psychological Re v i e w , 114, 539-576.
Brown, G. D. A., Preece, T., & Hulme, C. (2000). Oscillator-based
memory for serial order. Psychological Review, 107, 127-181.
Burgess, N., & Hitch, G. J. (1999). Memory for serial order: A network
model of the phonological loop and its timing. Psychological Review,
106, 551-581. doi:10.1037/0033-295X.106.3.551
Craik, F. I. M., & Watkins, M. J. (1973). The role of rehearsal in short
term memory. Journal of Verbal Learning and Verbal Behavior, 12,
599-607. doi:10.1016/S0022-5371(73)80039-8
Davelaar, E. J., & Usher, M. (2002). An activation-based theory of
immediate item memory. In J. A. Bullinaria, & W. Lowe (Eds.),
Connectionist models of cognition and perception: Proceedings of
the seventh neural computation and psychology workshop. Singapore
City: World Scientific.
Davelaar, E. J., Goshen-Gottstein, Y., Ashkenazi, A., Haarmann, H. J.,
& Usher, M. (2005). The demise of short-term memory revisited:
Empirical and computational investigations of recency effects. Psy-
chological Review, 112, 3-42. doi:10.1037/0033-295X.112.1.3
Davelaar, E. J., Tian, X., Weidemann, C. T., & Huber, D. E. (2011). A
habituation account of change detection in same/different judgments.
Cognitive, Affective, & Behavioral Neuroscience, 11, 608-626.
Debener, S., Herrman, C. S., Kranczioch, C., Gembris, D., & Engel, A.
K. (2003). Top-down attentional processing enhances auditory evoked
gamma band activity. Neuroreport, 14, 683-686.
Estes, W. K. (1955). Statistical theory of distributional phenomena in
learning. Psychological Review, 62, 369-377. doi:10.1037/h0046888
Glenberg, A. M., Bradley, M. M., Kraus, T. A., & Renzaglia, G. J.
(1983). Studies of long-term recency effect: Support for a contextu-
ally guided retrieval hypothesis. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 9, 231-255.
Grossberg, S. (2012). Adaptive resonance theory: How a brain learns to
consciously attend, learn, and recognize a changing world. Neural
Networks, 37, 1-47. doi:10.1016/j.neunet.2012.09.017
Henson, R. H. A. (1998). Short-term memory for serial order: The start-
end model. Co g n it i v e Psychology, 36, 73-137.
Howard, M. W., & Kahana, M. J. (1999). Contextual variability and
serial position effects in free recall. Journal of Experimental Psy-
chology: Learning, Me m o ry and Cognition, 25, 923-941.
Howard, M. W., & Kahana, M. J. (2002). A distributed representation
of temporal context. Journal of Mathematical Psychology, 46, 269-
299. doi:10.1006/jmps.2001.1388
Howard, M. W., Fotedar, M. S., Datey, A. V., & Hasselmo, M. E.
(2005). The temporal context model in spatial navigation and rela-
tional learning: Toward a common explanation of medial temporal
lobe function across domains. Psychological Review, 112, 75-116.
Hunt, R. R., & Lamb, C. A. (2001). What causes the isolation effect?
Journal of Experimental Psychology: Learning, Memory, & Cogni-
tion, 27, 1359-1366. doi:10.1037/0278-7393.27.6.1359
Jang, Y., & Huber, D. E. (2008). Context retrieval and context change
in free recall: Recalling from long-term memory drives list isolation.
Journal of Experimental Psychology: Learning Memory and Cogni-
tion, 34, 112-127. doi:10.1037/0278-7393.34.1.112
Kirchoff, B. A., Wagner, A. D., Maril, A., & Stern, C. E. (2000). Pre-
frontal-temporal circuitry for episodic encoding and subsequent
memory. Journal of Neuroscience, 20, 6173-6180.
Klein, K. A., Shiffrin, R. M., & Criss, A. H. (2007). Putting context in
context. In J. S. Nairne (Ed.), The foundations of remembering: Es-
says in honor of Henry L. Roediger III. New York: Psychology
Laming, D. (1999). Testing the idea of distinct storage mechanisms in
memory. International Journal of Psychology, 34, 419-426.
Laming, D. (2006). Predicting free recall. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 32, 1146-1163.
Mensink, G.-J., & Raaijmakers, J. G. W. (1988). A model of interfere-
ence and forgetting. Psychological Review, 95, 434-455.
Metcalfe, J., & Murdock, B. B. (1981). An encoding and retrieval
model of single trial free recall. Journal of Verbal Learning and
Verbal Behavior, 20, 161-189. doi:10.1016/S0022-5371(81)90365-0
Murdock, B. B. (1960). The distinctiveness of stimuli. Psychological
Review, 67, 16-31. doi:10.1037/h0042382
Murdock, B. B. (1962). The serial position effect of free recall. Journal
of Verbal Learning and Verbal Behavior, 64, 482-488.
Murdock, B. B. (1972). Short-term memory. In G. H. Bower & J. T.
Spence (Eds.), The psychology of learning and motivation (Vol. 5).
New York: Academic Press.
Nairne, J. S. (2002). The myth of the encoding-retrieval match. Memory,
10, 389-395. doi:10.1080/09658210244000216
Neath, I. (1993). Distinctiveness and serial position effects in recogni-
tion. Memory & Cognition, 21, 689-698. doi:10.3758/BF03197199
Norman, K. A. & O’Reilly, R. C. (2003). Modeling hippocampal and
neocortical contributions to recognition memory: A complementary
learning systems Approach. Psychological Review, 110, 611-646.
Raaijmakers, J. G. W., & Shiffrin, R. M. (1980). SAM: A theory of
probabilistic search of associative memory. In G. Bower (Ed.), The
psychology of learning and motivation (Vol. 14). New York: Aca-
demic Press.
Richardson, J. T. E., & Baddeley, A. D. (1975). The effect of articula-
tory suppression in free recall. Journal of Verbal Learning and Ver-
bal Behavior, 14, 623-629. doi:10.1016/S0022-5371(75)80049-1
Rundus, D. (1971). An analysis of rehearsal processes in free recall.
Journal of Experimental Ps ych olog y, 89, 63-77.
Rundus, D., & Atkinson, R. C. (1970). Rehearsal procedures in free
recall: A procedure for direct observation. Journal of Verbal Learn-
ing and Verbal Behavior, 9, 99-105.
Scoville, W. B., & Milner, B. (1957). Loss of recent memory after
bilateral hippocampal lesions. Journal of Neurology, Neurosurgery,
and Psychiatry, 20, 11-21. doi:10.1136/jnnp.20.1.11
Sederberg, P. B., Gauthier, L. V., Terushkin, V., Miller, J. F., Barna-
than, J. A., & Kahana, M. J. (2006). Oscillatory correlates of the pri-
macy effect in episodic memory. NeuroImage, 32, 1422-1431.
Shiffrin, R. (1970). Forgetting: Trace erosion or retrieval failure? Sci-
ence, 168, 1601-1603. doi:10.1126/science.168.3939.1601
Shiffrin, R. M., & Cook, J. R. (1978). Short-term forgetting of item and
order information. Journal of Verbal Learning and Verbal Behavior,
17, 189-218. doi:10.1016/S0022-5371(78)90146-9
Tan, L., & Ward, G. (2000). A recency-based account of primacy ef-
fects in free recall. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 26, 1589-1625.
Tulving, E. (1983). Elements of episodic memory. New York: Oxford
University Press.
Ward, G. (2002). A recency-based account of the list length effect in
free recall. Memory & Cognition, 30, 885-892.