Divided Attention Can Enhance Memory Encoding: The Attentional Boost
Effect in Implicit Memory
Pietro Spataro
Sapienza University of Rome
Neil W. Mulligan
University of North Carolina at Chapel Hill
Clelia Rossi-Arnaud
Sapienza University of Rome
Distraction during encoding has long been known to disrupt later memory performance. Contrary to this
long-standing result, we show that detecting an infrequent target in a dual-task paradigm actually
improves memory encoding for a concurrently presented word, above and beyond the performance
reached in the full-attention condition. This absolute facilitation was obtained in 2 perceptual implicit
tasks (lexical decision and word fragment completion) but not in a conceptual implicit task (semantic
classification). In the case of recognition memory, the facilitation was relative, bringing accuracy in the
divided attention condition up to the level of accuracy in the full attention condition. The findings follow
from the hypothesis that the attentional boost effect reflects enhanced visual encoding of the study
stimulus consequent to the transient orienting response to the dual-task target.
Keywords: implicit memory, divided attention, attention and memory, repetition priming
The deleterious effects of distraction on memory encoding have
been amply documented from the earliest days of psychological
research (see Mulligan, 2008, for review). A common experimen-
tal technique uses the dual-task paradigm, in which memory en-
coding is carried out under full-attention (FA) or divided-attention
(DA) conditions (i.e., while simultaneously carrying out a second
task designed to compete for attentional resources). The results of
numberless studies make it abundantly clear that DA during en-
coding degrades later memory on tests such as recognition, free
recall, and cued recall (e.g., Craik et al., 1996; Mulligan, 1998,
2008).
Recently, Swallow and Jiang (2010) reported a surprising twist
on the usual adverse effects of DA. In these experiments, partic-
ipants studied a sequence of pictures, each with a small square
superimposed at the center. In the DA condition, participants were
instructed to remember all of the images and to monitor the color
of the square, pressing the space bar whenever they detected an
infrequent white square (targets) among frequent black squares
(distractors). In the FA condition, participants were told to ignore
the squares and to focus only on encoding the pictures. When
memory for the pictures was later tested in a four-choice recogni-
tion task, Swallow and Jiang (2010) found that in the DA condi-
tion, the images encoded together with the target squares (i.e.,
corresponding to the press response) were recognized significantly
better than were those encoded with the distractor squares (the
attentional boost effect). In the FA condition, in which the partic-
ipants made no response to the squares, no attentional boost effect
was found. What is importantly for present purposes ...
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
Divided Attention May Enhance Implicit Memory
1. Divided Attention Can Enhance Memory Encoding: The
Attentional Boost
Effect in Implicit Memory
Pietro Spataro
Sapienza University of Rome
Neil W. Mulligan
University of North Carolina at Chapel Hill
Clelia Rossi-Arnaud
Sapienza University of Rome
Distraction during encoding has long been known to disrupt
later memory performance. Contrary to this
long-standing result, we show that detecting an infrequent target
in a dual-task paradigm actually
improves memory encoding for a concurrently presented word,
above and beyond the performance
reached in the full-attention condition. This absolute facilitation
was obtained in 2 perceptual implicit
tasks (lexical decision and word fragment completion) but not
in a conceptual implicit task (semantic
classification). In the case of recognition memory, the
facilitation was relative, bringing accuracy in the
divided attention condition up to the level of accuracy in the
full attention condition. The findings follow
from the hypothesis that the attentional boost effect reflects
enhanced visual encoding of the study
stimulus consequent to the transient orienting response to the
dual-task target.
2. Keywords: implicit memory, divided attention, attention and
memory, repetition priming
The deleterious effects of distraction on memory encoding have
been amply documented from the earliest days of psychological
research (see Mulligan, 2008, for review). A common
experimen-
tal technique uses the dual-task paradigm, in which memory en-
coding is carried out under full-attention (FA) or divided-
attention
(DA) conditions (i.e., while simultaneously carrying out a
second
task designed to compete for attentional resources). The results
of
numberless studies make it abundantly clear that DA during en-
coding degrades later memory on tests such as recognition, free
recall, and cued recall (e.g., Craik et al., 1996; Mulligan, 1998,
2008).
Recently, Swallow and Jiang (2010) reported a surprising twist
on the usual adverse effects of DA. In these experiments, partic-
ipants studied a sequence of pictures, each with a small square
superimposed at the center. In the DA condition, participants
were
instructed to remember all of the images and to monitor the
color
of the square, pressing the space bar whenever they detected an
infrequent white square (targets) among frequent black squares
(distractors). In the FA condition, participants were told to
ignore
the squares and to focus only on encoding the pictures. When
memory for the pictures was later tested in a four-choice
recogni-
tion task, Swallow and Jiang (2010) found that in the DA condi-
tion, the images encoded together with the target squares (i.e.,
3. corresponding to the press response) were recognized
significantly
better than were those encoded with the distractor squares (the
attentional boost effect). In the FA condition, in which the
partic-
ipants made no response to the squares, no attentional boost
effect
was found. What is importantly for present purposes is that the
attentional boost effect was relative: For pictures accompanied
by
the more frequent black squares (the distractor trials), the DA
condition produced worse picture memory than did the FA
condi-
tion, a typical DA effect on memory encoding. For pictures ac-
companied by the target (white) squares, picture memory was
equal in the two attention conditions, which indicated the elimi-
nation of the DA effect. Thus, the attentional boost effect
reported
by Swallow and Jiang (2010) was a relative boost in memory
encoding, bringing the typically poor memory produced by DA
up
to the level of the FA condition.
In later studies, the authors ruled out a number of potential
accounts of the effect based on attentional cuing, reinforcement
learning, perceptual grouping, oddball processing and
distinctive-
ness. Rather, Swallow and Jiang (2011, 2012) concluded, the
attentional boost effect reflects enhanced visual encoding pro-
duced by the opening of an attentional gate consequent to the
transient orienting responses triggered by the detection of target
squares. According to the event segmentation theory (Zacks,
Speer, Swallow, Braver, & Reynolds, 2007), this gating mecha-
nism would be implemented in subcortical regions like the locus
coeruleus or the nucleus basalis and is involved in alerting the
observer to salient environmental changes. Once a modification
4. of
the stimulus’ properties occurs (e.g., the color of the target
squares
becomes red), the gating mechanism is activated, resulting in
increased attention to the perceptual properties of the
concurrently
presented images, as well as in the updating of their internal
representations. In other words, the attentional gate postulated
by
Swallow and Jiang (2010) would act as a filter to sensory input,
This article was published Online First January 28, 2013.
Pietro Spataro, Department of Psychology, Sapienza University
of
Rome, Rome, Italy; Neil W. Mulligan, Department of
Psychology, Uni-
versity of North Carolina at Chapel Hill; Clelia Rossi-Arnaud,
Department
of Psychology, Sapienza University of Rome.
Correspondence concerning this article should be addressed to
Neil W.
Mulligan, Department of Psychology, University of North
Carolina, Cha-
pel Hill, NC 27599-3270. E-mail: [email protected]
T
hi
s
do
cu
m
en
10. implicit memory than explicit memory (Mulligan & Brown,
2003;
Spataro, Cestari, & Rossi-Arnaud, 2011). Although implicit
mem-
ory is affected by DA under some conditions (e.g., depending
on
the type of implicit test, the type of dependent measure, or the
precise nature of the dual task; Mulligan & Peterson, 2008),
there
are numerous examples in which a dual task that impaired later
explicit memory left implicit memory unaffected (Mulligan,
Duke,
& Cooper, 2007; Parkin, Reid, & Russo, 1990). In general, re-
search on attention and implicit memory has found that dual
tasks
with infrequent response selection typically fail to affect
implicit
memory encoding despite disrupting explicit memory encoding
(Mulligan, 2003).
In the present study, we first reproduced the attentional boost
effect in a four-choice recognition test (explicit memory), using
words instead of pictures. Then we examined the effect in two
different perceptual implicit tasks (lexical decision and word-
fragment completion). We contrasted a DA condition, in which
participants read aloud a series of words and concurrently moni-
tored the color (red or green) of a small circle placed below
each
word, with a FA condition, in which participants read the words
but ignored the circles. In agreement with Swallow and Jiang
(2010), our approach was to consider the attentional boost
effect as
the combination of two related factors: a primary task
facilitation
as a consequence of increased attention and perceptual
processing
11. of the words due to target detection, plus a primary task
interfer-
ence as a consequence of the attentional demands associated
with
monitoring the circles to determine if a target is present. In the
case
of implicit memory, the mere monitoring required by the dual
task
is unlikely to reduce later implicit memory (Mulligan, 2003;
Mulligan et al., 2007). This implies that, for words encoded
with
green (distractor) circles, the FA and DA conditions should pro-
duce comparable amounts of priming. In contrast, for words en-
coded with red (target) circles, priming should be significantly
greater in the DA condition than in the FA condition, because
the
facilitating effect due to the detection of the infrequent red
circles
should overcome any small attentional interference produced by
the dual task. An alternative possibility stems from the
automatic-
ity hypothesis (Aloisi, McKone, & Heubeck, 2004; see Lozito &
Mulligan, 2010, for discussion), which states that implicit
memory
is the result of involuntary encoding processes. According to
this
account, any variation in attentional levels should not influence
implicit memory; thus, the attentional boost effect should not be
observed either in lexical decision or in the word-fragment com-
pletion task.
Experiment 1
Experiment 1 reproduced the attentional boost effect in a four-
choice recognition task, using words instead of pictures. In
addi-
12. tion, Swallow and Jiang (2010) repeated the presentation of the
to-be-remembered images 10 times, whereas in the present
exper-
iment, the study words were shown only once. Experiment 1
verifies that the attentional boost effect is observed with verbal
material and without an extended learning process.
Method
Participants. Thirty-six students from Sapienza University of
Rome participated (27 women, mean age � 24.6 years).
Materials and procedure. The critical items were 30 words,
seven to nine letters in length, from the LexVar database
(Barca,
Burani, & Arduino, 2002). They were divided into two lists of
15
words each (A and B), matched on several variables, including
length in letters (M � 7.80 vs. M � 7.73), written frequency (M
�
74.47 vs. M � 79.33), age of acquisition (M � 4.21 vs. M �
4.13),
familiarity (M � 6.01 vs. M � 6.13), imageability (M � 5.22
vs.
M � 5.42), and concreteness (M � 5.88 vs. M � 6.08). The
mean
values of age of acquisition, familiarity, imageability, and con-
creteness were obtained from the LexVar database, as measured
on
7-point Likert-type scales, whereas the estimates of written fre-
quency were taken from the CoLFIS Vocabulary (which
includes
over 3 million occurrences; Laudanna, Thornton, Brown,
Burani,
& Marconi, 1995). An additional set of 120 words of medium
frequency (between 50 and 100 occurrences) were selected and
13. used as filler items during the study phase.
In the DA condition, participants were told to study (and read
aloud) each word while simultaneously monitoring the color of
a
small circle immediately below the word. Participants were in-
structed to press the space bar whenever they saw an infrequent
red
circle among more frequent green circles. A total of 150 words
were presented during the encoding phase, at a rate of 500 ms
per
word. The study procedure was modeled on Swallow and Jiang
(2010). On each trial, one word (Times New Roman, 44 points)
and one circle (red or green; 1 cm in diameter) appeared
simulta-
neously at the center of the screen for 100 ms, with a vertical
distance of 1 cm between them, after which only the word re-
mained visible for an additional 400 ms. There was no
interruption
between successive trials. The study list was constructed of 16
blocks of seven items (15 critical blocks preceded by one
practice
block). Critical words encoded with red circles (and associated
with a press response) were always located in the fourth
position.
Critical words encoded with green circles were placed either in
the
third position or in the fifth position; this choice followed from
the fact that in the study by Swallow and Jiang (2010), memory
for
the stimuli presented with distractor squares did not vary across
encoding positions. For simplicity, the critical words presented
on
target trials (i.e., with red circles) are referred to as target
words
and the critical words presented on distractor trials (i.e., with
14. green
circles) are referred to as distractor words. However, it should
be
noted that words on all trials were to be identified. The
designation
of target and distractor words refers to whether the word co-
occurred with a target or a distractor on the dual (circle
detection)
task.
In sum, only two critical words were presented per block,
yielding a total of 30 critical words across all blocks (15
encoded
T
hi
s
do
cu
m
en
t
is
co
py
ri
gh
te
d
18. d
is
no
t
to
be
di
ss
em
in
at
ed
br
oa
dl
y.
1224 SPATARO, MULLIGAN, AND ROSSI-ARNAUD
with red circles [target words] and 15 encoded with green
circles
[distractor words]). For half of the participants in each
condition,
eight distractor words were located in the third position, while
the
other seven were placed in the fifth position; for the remaining
19. participants, the pattern was reversed. All other words in a
block
were noncritical filler words and were accompanied by green
circles. One to five additional filler words, always presented
with
green circles, were randomly located between blocks to reduce
the
temporal regularity of the target trials. It should be noted that
the
block organization of the study list was not apparent to the par-
ticipant; all trials were presented as one continuous list. List A
and
List B words were counterbalanced across participants, so they
had
the same probability to be encoded with red and green circles.
The
FA condition was identical except that participants were told to
ignore the circles.
After a 5-min distractor task (arithmetic problems), a four-
choice recognition task was administered. There were a total of
30
quadruplets, one for each critical word. Each trial included a
fixation point for 500 ms, a test slide including four words that
remained visible until response, and a 1,500-ms pause. The four
words were arranged in different quadrants of the screen, with
the
studied word being located equally often in each quadrant. Two
foils were semantically unrelated to the studied word, while the
third foil was a semantic associate (e.g., for the studied word
hospital, the semantic associate was surgeon). For each studied
word, the corresponding foils were selected from the LexVar
database to be matched as closely as possible in terms of length
and written frequency (range: 20 –100 occurrences; M � 73.86).
Such a manipulation was used to mimic as closely as possible
the
20. original study by Swallow and Jiang (2010), where two of the
four
test images were similar scenes. Participants were required to
select the words studied during the encoding phase. For the pur-
pose of scoring, the choice of the semantic associates was
consid-
ered an error throughout.
Results and Discussion
During the encoding phase, participants correctly detected
98.2% of the red circles (mean response time � 419.9 ms). In
the
present experiment, as well as in the following ones, care was
taken to ensure that participants read aloud all the words, so
that
identification accuracy was always 100%. The proportions of
studied words correctly recognized (Figure 1) were submitted to
a
2 (word type: target words [encoded with red circles] vs.
distractor
words [encoded with green circles]) � 2 (attentional condition:
FA
vs. DA) mixed analysis of variance (ANOVA), in which word
type
was manipulated within subject and attentional condition was
manipulated between subjects. Results revealed (a) a significant
effect of word type, F(1, 34) � 9.41, mean square error (MSE)
�
0.024, p � .004, �2 � 0.22, indicating that target words were
recognized better than distractor words were (M � 0.64 vs. M �
0.53), and (b) a significant interaction, F(1, 34) � 6.55, MSE �
0.024, p � .015, �2 � 0.16. The effect of attentional condition
was
not significant, F(1, 34) � 1.12, p � .297. An analysis of simple
effects showed the typical attentional boost effect: In the DA
21. condition, accuracy was greater for target words than for
distractor
words, F(1, 34) � 15.82, p � .000, �2 � 0.32, whereas no
difference was found in the FA condition, F(1, 34) � 0.13, p �
.722). It is important to note that the attentional boost effect
was
relative, because it enhanced memory for target words in the
DA
condition to the same level of the FA condition, F(1, 34) �
1.24,
p � .272; in contrast, distractor words were recognized more
accurately in the FA condition than in the DA condition, F(1,
34) � 8.19,
p � .007, �2 � 0.19, replicating the usual negative effect of a
secondary task during encoding on explicit memory. The results
did not change when the performance for the target words was
conditionalized on prior success at the study-phase detection
task
(not surprisingly given the very high detection rate): indeed, the
critical interaction between word type and attentional condition
remained significant, F(1, 34) � 6.12, MSE � 0.023, p � .019,
�2 � 0.15. Furthermore, the findings are not explained by
differ-
ences in the likelihood of (incorrectly) selecting the semantic
associates, because a mixed ANOVA with the same factors as
above showed no significant effects or interaction on the
propor-
tion of semantic associates chosen, F(1, 34) � 2.23, p � .14.
In summary, the results of Experiment 1 replicated the relative
boost effect observed by Swallow and Jiang (2010), using
verbal
material (instead of pictures) and a single encoding presentation
(instead of multiple presentations). Moreover, our data extend
to
22. long-term memory the suggestion that the detection targets need
not overlap in space with the to-be-remembered items to
produce
a significant facilitation (Makovski, Swallow, & Jiang, 2011).
Experiment 2
In Experiment 2, we examined the attentional boost effect in the
lexical decision task (LDT), a perceptual implicit test
characterized
by a strong resilience to the negative consequences of DA (Mul-
ligan & Peterson, 2008; Newell, Cavenett, & Andrews, 2008;
Spataro, Mulligan, & Rossi-Arnaud, 2011).
Method
Experiment 2 used the methods of Experiment 1 with the
following modifications. A new pool of 36 participants took
part
(21 women, mean age � 24.2 years). A set of 45 critical words,
seven to nine letters in length (including the 30 used in
Experiment
1), were divided into three sublists of 15 words each. The prop-
erties of the additional 15 words were similar to those
illustrated
for Experiment 1 (M length in letters � 7.67; M written
frequency �
Figure 1. Experiment 1: Proportions of correct recognition, as a
function
of word type and attentional condition. DA � divided-attention;
FA �
full-attention. Bars represent standard errors.
T
hi
27. ed
br
oa
dl
y.
1225ATTENTIONAL BOOST EFFECT IN IMPLICIT
MEMORY
77.33; M age of acquisition � 4.23; M familiarity � 5.85; M
imageability � 5.30; M concreteness � 5.92; M neighborhood
size � 0.20). The three sublists were counterbalanced across
participants, so that all the critical words had the same
probability
to be presented in target trials (with red circles), in distractor
trials
(with green circles) or as unstudied words. Encoding was
inciden-
tal, because participants were required to read aloud all the
words
but were not told to remember them for a later memory task.
The test phase consisted of the LDT. Participants were
presented
with a total of 110 items, including the 45 critical words (30
studied and 15 unstudied) and 45 legal pronounceable
nonwords,
plus 20 practice items (10 words and 10 nonwords). Each trial
included three events: a fixation point for 500 ms, a string of
letters
until the participant’s response, and a pause of 1,500 ms. The
28. instructions were to decide if each item was a valid Italian word
by
pressing either S for word or N for nonword: Both speed and
accuracy were stressed. The order of presentation was
randomized
anew for each participant, and no mention was made about the
relationship with the study phase.
Results and Discussion
During the encoding phase, participants correctly detected
99.2% of the red circles (mean response time � 401.2 ms). Only
reaction times (RTs) for correct responses were analyzed and
RTs
more than 3 standard deviations from the participant’s mean
were
removed (Ziegler & Perry, 1998; less than 3% of the data).
Mean
RTs for words encoded with red circles, words encoded with
green
circles, and baseline (unstudied) words were 667.08, 701.81,
and
743.99 ms in the DA condition and 682.58, 679.40, and 720.25
ms
in the FA condition. Priming scores (the difference between RTs
for unstudied and studied words) are the dependent measure
(see
Figure 2). Preliminary t tests indicated that RTs for unstudied
words did not differ between the DA and FA conditions, t(34) �
– 0.59, p � .55, and that priming scores were significantly
greater
than zero for target words and for distractor words, t(17)s �
3.28, all
ps � 0.01. A mixed 2 (word type: target vs. distractor) � 2
(attentional condition: FA vs. DA) ANOVA, with word type as
the
29. within-subject variable and attentional condition as the
between-
subjects variable, found a significant effect of word type, F(1,
34) �
6.89, MSE � 644.85, p � .013, �2 � 0.17, indicating that
priming
scores were greater for target words (encoded with red circles)
than for distractor words (encoded with green circles; M � 57.2
ms
vs. M � 41.5 ms), and there was a significant interaction, F(1,
34) �
10.11, MSE � 644.85, p � .003, �2 � 0.23. The main effect of
attentional condition was nonsignificant, F(1, 34) � 1.60, p �
.213. Follow-up analyses showed an attentional boost effect in
the
DA condition, in which repetition priming was higher for target
than distractor words, F(1, 34) � 16.86, p � .000, �2 � 0.33,
but
not in the FA condition, F(1, 34) � 0.15, p � .698. It is
important
to note that the attentional boost was absolute: Priming scores
for
target words were significantly greater in the DA condition than
in
the FA condition, F(1, 34) � 4.73, p � .037, �2 � 0.12. In
contrast
with Experiment 1, no difference between the two conditions
was
obtained for distractor words, F(1, 34) � 0.01, p � .939,
confirm-
ing that a dual task with infrequent response rates did not
disrupt
implicit memory (Mulligan et al., 2007). Similar results were
obtained when the performance for the target words was condi-
tionalized on prior success at the study-phase detection task; in
30. particular, the crucial interaction between word type and atten-
tional condition remained significant, F(1, 34) � 11.30, MSE �
635.34, p � .002, �2 � 0.25. Error rates ranged from 3.4% to
4.9%
and did not vary across word type (target, distractor, and
unstud-
ied) or attentional condition, F(2, 68) � 0.39, p � .62.
The results of Experiment 2 showed that an absolute attentional
boost effect is observed in the LDT. According to Swallow and
Jiang (2010), this happens because the attentional orienting re-
sponse following target detection facilitates the perceptual
encod-
ing of the concurrent word, overcoming any negligible
interference
caused by the dual task. Experiment 3 generalizes this finding
to a
second perceptual implicit test.
Experiment 3
Word-fragment completion (WFC) was used in Experiment 3
because it shares with the LDT two relevant properties: (a) It is
a
perceptually driven task (Roediger & McDermott, 1993) and (b)
it
is minimally affected by DA manipulations (Mulligan &
Hartman,
1996; Spataro, Mulligan, & Rossi-Arnaud, 2010).
Method
Experiment 3 used the same materials and procedures as Ex-
periment 2 with the following modifications. A new set of 36
participants took part (25 women, mean age � 24.6 years). The
test phase consisted of the WFC task. A word fragment was
31. constructed for each critical word by replacing three to five
letters
with blanks (the proportions of letters retained ranged from 0.43
to
0.57). The critical words were rotated across participants so that
they had the same probability to be presented in target trials
(with
red circles), in distractor trials (with green circles), or as
unstudied
words. On the WFC test, participants were presented with a
total
of 70 fragments for 4 s each, corresponding to the 45 critical
words
(30 studied and 15 unstudied), plus 25 filler words (i.e.,
additional
new words not presented during the encoding phase). Each trial
included three events: a fixation point for 500 ms, a fragment
for
4,000 ms, and a pause of 1,500 ms. The order of the fragments
was
randomized anew for each participant, with the constraint that
the
first 10 fragments were filler items that could not be completed
Figure 2. Experiment 2: Priming scores in the lexical decision
task, as a
function of word type and attentional condition. DA � divided-
attention;
FA � full-attention. Bars represent standard errors.
T
hi
s
do
36. oa
dl
y.
1226 SPATARO, MULLIGAN, AND ROSSI-ARNAUD
with a critical word. Participants tried to complete the
fragments
with the first word that came to mind. It should be noted that
the
time limit of 4 s is lower than that typically used in literature
(10
or 12 s; Mulligan, 1998; Roediger, Weldon, Stadler, & Riegler,
1992). This was done to minimize the use of explicit retrieval
strategies.
Results and Discussion
During the encoding phase, participants correctly detected
99.4% of the red circles (mean response time � 425.9 ms). The
mean proportions of fragments correctly completed for words
encoded with red circles, words encoded with green circles, and
baseline (unstudied) words were 0.32, 0.21, and 0.13 in the DA
condition and 0.26, 0.26, and 0.17 in the FA condition, respec-
tively. For the WFC test, priming is computed as the difference
between the proportion of correct completions for studied and
unstudied words (see Figure 3). Preliminary t tests indicated
that
the proportions of fragments completed with unstudied words
did
not differ between the DA and FA conditions, t(34) � –1.02, p
�
.31, and that priming scores were significantly greater than zero
37. for
target words and for distractor words, t(17)s � 2.25, all ps �
.05.
A mixed 2 (word type: target vs. distractor) � 2 (attentional
condition: FA vs. DA) ANOVA showed results identical to
those
of Experiment 2. Both the main effect of word type and the
interaction between word type and attentional condition were
significant, F(1, 34) � 4.22, MSE � 0.010, p � .048, �2 �
0.11,
and F(1, 34) � 4.82, MSE � 0.010, p � .035, �2 � 0.12, but
the
main effect of attentional condition was not, F(1, 34) � 0.96, p
�
.33. An analysis of simple effects showed that priming was
greater
for target than for distractor words in the DA condition, F(1,
34) � 9.02,
p � .005, �2 � 0.21, but not in the FA condition, F(1, 34) �
0.01,
p � .921. As in Experiment 2, an absolute boost effect was
obtained: For target words, the DA condition produced more
priming than the FA condition did, F(1, 34) � 4.67, p � .038,
�2 � 0.12; in contrast, for distractor words, the DA and FA
conditions produced equivalent priming, F(1, 34) � 0.04, p �
.854. The results did not change when the performance for the
target words was conditionalized on prior success at the
detection
task; in particular, the crucial interaction between word type
and
attentional condition remained significant, F(1, 34) � 4.52,
MSE
� 0.010, p � .041, �2 � 0.12.
The results of Experiment 3 replicated the findings of Experi-
38. ment 2 using a different implicit task (WFC). Overall, the
present
data suggest that DA can actually increase perceptual priming
for
the words presented with dual-task targets.
Experiment 4
If the attentional boost effect is due to enhanced perceptual
encoding (as proposed by Swallow & Jiang, 2010), then this
effect
should not extend to conceptually driven implicit tests that are
relatively insensitive to variation in perceptual processing. In
Ex-
periment 4, conceptual priming was assessed with a semantic
classification task (Brysbaert, Van Wijnendaele, & De Deyne,
2000). Such tasks are largely unaffected by dual-task manipula-
tions (Mulligan & Peterson, 2008), so no effect of DA is
expected
for the words encoded with green circles. The critical issue is
whether enhanced memory is found for words encoded with red
circles. The current analysis predicts no facilitation.
Method
A new set of 36 participants took part (22 women, mean age �
23.8 years). The critical words, the study phase, and distractor
phase were the same as those in Experiments 2 and 3. During
the
test phase, participants were presented with 100 stimuli, corre-
sponding to 45 critical words (30 studied and 15 unstudied), 45
Italian first names, and 10 practice items (five words and five
first
names). Each trial included a fixation point for 500 ms, a string
of
letters until the participant’s response, and a pause of 1,500 ms.
39. Participants responded by pressing the A key to classify the
stim-
ulus as a word and the L key to classify it as a first name. Both
speed and accuracy were stressed. The order of presentation was
randomized anew for each participant. As in previous
experiments,
no mention was made about the relationship with the study
phase.
Results and Discussion
During the encoding phase, participants correctly detected
98.4% of the red circles (mean response time � 419.1 ms). As
in
Experiment 2, only the RTs for correct responses were
analyzed,
and RTs more than 3 standard deviations from the participant’s
mean were removed (less than 5% of the data were discarded).
Mean RTs for words encoded with red circles, words encoded
with
green circles, and baseline (unstudied) words were 599.81,
599.54,
and 616.79 ms in the DA condition and 613.71, 607.10, and
627.82
ms in the FA condition, respectively. Priming scores were com-
puted as the difference between RTs for unstudied and studied
words (see Figure 4). Preliminary t tests confirmed that RTs for
unstudied words did not differ between the DA and FA
conditions,
t(34) � – 0.31, p � .75, and that priming scores were
significantly
greater than zero for target words and for distractor words,
ts(17) �
2.49, all ps � .05. In contrast to Experiments 2 and 3, a mixed 2
(word type: target vs. distractor) � 2 (attentional condition: di-
vided vs. full attention) ANOVA on priming scores showed no
40. main effects and no interaction between the two, F(1, 34) �
0.48,
p � .493. The same results were obtained when the performance
for target words was conditionalized on prior success at the de-
Figure 3. Experiment 3: Priming scores in the word fragment
completion
task, as a function of word type and attentional condition. DA �
divided-
attention; FA � full-attention. Bars represent standard errors.
T
hi
s
do
cu
m
en
t
is
co
py
ri
gh
te
d
by
th
44. no
t
to
be
di
ss
em
in
at
ed
br
oa
dl
y.
1227ATTENTIONAL BOOST EFFECT IN IMPLICIT
MEMORY
tection task, F(1, 34) � 0.59, p � .447. Error rates ranged from
1.8% to 4.6% and did not vary across word type (target,
distractor
and unstudied) or attentional condition, F(2, 68) � 1.98, p �
.15.
Experiment 4 showed no attentional boost effect in a semantic
classification task. Given that conceptual priming is relatively
sensitive to prior conceptual but not perceptual encoding, this
45. result is consistent with the hypothesis that the facilitation pro-
duced by the detection of dual-task targets affects processing of
perceptual but not semantic information. However, before
accept-
ing this conclusion, it is important to consider the statistical
power
of the critical comparison in this experiment. To this purpose,
we
computed the unbiased d effect sizes of the attentional boost
effect
in the DA conditions of Experiments 2 and 3 (Hedges & Olkin,
1985), yielding values of 0.71 and 0.81, respectively. Using the
software G�Power3 (Faul, Erdfelder, Lang, & Buchner, 2007),
we
found that the post hoc power to detect an advantage of target
words in the DA condition of Experiment 4 ranged from 0.97 to
0.99. The power to detect an effect less than one third of that
observed in Experiments 2 and 3 (d � 0.20, a medium-small
effect
size, according to Cohen, 1988) was 0.83. In addition, we also
performed a Bayesian analysis to test the likelihood that word
type
interacted with attentional condition, using a method illustrated
by
Masson (2011), which requires the transformation of the sum-
of-
squares values generated by the ANOVA. Within this approach,
the null hypothesis (interaction absent) and the alternative
hypoth-
esis (interaction present) are directly contrasted as competing
models; then, the Bayes information criterion value is estimated
and used to compute a Bayes factor and generate the posterior
probabilities for each hypothesis. For Experiment 4, this
analysis
indicated that the probability of the null model (interaction
absent),
46. given the data, p(H0|D), was 0.83. A similar value (0.81) was
obtained when analyzing the likelihood of the null hypothesis
for
the main effect of word type in the DA condition. Both results
provide positive evidence for the null hypothesis, following the
guidelines proposed by Raftery (1995). Thus, it appears that the
current experiment had substantial power to detect an
attentional
boost effect if one was to be found in the semantic classification
task.
A second important point to clarify is that the absence of
significant DA effects in Experiment 4 should not be taken as
evidence that the semantic classification task does not reflect
conceptual processing. Instead, this finding is consistent with
the
distinction drawn by Vaidya et al. (1997; see also Gabrieli et
al.,
1999) between competitive and noncompetitive access to
concep-
tual knowledge in memory. Competitive tests (e.g., category ex-
emplar generation and general knowledge) are those in which
the
retrieval cues are insufficient to guide the retrieval process to a
unique entry in semantic memory and thus initiate a competition
between alternative legitimate responses. Performance in this
type
of task is typically enhanced by conceptual elaboration
(Srinivas &
Roediger, 1990; Vaidya et al., 1997) and decreased by DA at
encoding (Light, Prull, & Kennison, 2000; Mulligan & Stone,
1999). In contrast, noncompetitive tests (e.g., category
verification
and semantic classification) are those in which the retrieval
cues
47. directly specify a unique solution. In such a circumstance, any
encoding task will result in full priming (provided that the
studied
stimuli are correctly identified): As a consequence,
noncompetitive
tests are relatively unaffected by conceptual elaboration and DA
at
encoding (Light et al., 2000; Mulligan & Peterson, 2008;
Vaidya
et al., 1997). However, these forms of priming are not mediated
by
perceptual processes, because it has been repeatedly
demonstrated
that they are insensitive to study–test modality changes (Light
et al.,
2000; Vaidya et al., 1997). Even more cogent evidence comes
from neuroimaging studies. Wig, Grafton, Demos, and Kelley
(2005), for instance, found that transcranial magnetic
stimulation
to the left frontal cortex (a cerebral region typically associated
with
semantic processing) disrupted repetition priming in a semantic
classification task, whereas stimulation to the middle and
inferior
occipital gyri (typically associated with visual processing) had
no
effect. Likewise, using functional magnetic resonance imaging,
Wagner, Koutstaal, Maril, Schacter, and Buckner (2000)
reported
significant signal reductions in the left inferior prefrontal cortex
when participants made semantic decisions about words
processed
in a semantic manner during the study phase but not when they
made decisions about words processed in a nonsemantic way.
All
of these findings unequivocally confirm that semantic
48. classifica-
tion tasks tap conceptual processes, even though they are unaf-
fected by DA at encoding.
General Discussion
In the present study, we document the surprising result that
dividing attention during encoding can actually enhance
memory.
Words that accompanied target stimuli from a dual task
produced
more perceptual priming (in WFC and LDT) than did words
encoded under full attention. This stands out as a highly
singular
result given that the overwhelmingly common result is that DA
impairs or at the very least does not enhance memory
performance
(Mulligan, 2008). The guiding hypothesis was that detection of
the
target enhances the perceptual encoding of a co-occurring
stimulus
(the study picture in Swallow & Jiang, 2010; the study word in
the
present experiments). Consistent with this, the attentional boost
effect was found on two perceptual priming tests, so categorized
because of their sensitivity to prior perceptual processing, but
not
on a conceptual priming test, which is relatively insensitive to
variation in prior perceptual processing.
More generally, dividing attention during encoding largely dis-
rupts elaborative processes (e.g., Craik et al., 1996; Mulligan,
2008). On the one hand, if a memory test is relatively
insensitive
Figure 4. Experiment 4: Priming scores in the semantic
49. classification
task, as a function of word type and attentional condition. DA �
divided-
attention; FA � full-attention. Bars represent standard errors.
T
hi
s
do
cu
m
en
t
is
co
py
ri
gh
te
d
by
th
e
A
m
er
53. di
ss
em
in
at
ed
br
oa
dl
y.
1228 SPATARO, MULLIGAN, AND ROSSI-ARNAUD
to variation in elaborative encoding, which is the case for
percep-
tual implicit tests, then the monitoring requirements of this dual
task will have minimal negative impact on later memory perfor-
mance. On the other hand, the enhanced perceptual encoding
brought about through target detection will produce an enhance-
ment that is not offset by the typical negative effects of the dual
task, with the consequent surprising result that a DA condition
ends up enhancing memory encoding as reflected by perceptual
priming. Recognition memory is often assumed to be sensitive
to
both semantic and elaborative processing on the one hand and
perceptual fluency processes on the other (Yonelinas, 2002).
This
renders recognition memory, according to this analysis,
sensitive
54. to both the negative effects of a dual task (the reductions in
semantic and elaborative processing) and the positive effects as-
sociated with enhanced perceptual analysis of study stimuli co-
occurring with targets. This is clearly the case in the picture
recognition tests used by Swallow and Jiang (2010, 2011). In
the
results of the present Experiment 1, the negative effect is
reflected
in worse memory for words encoded with green (distractor)
circles
in the DA condition compared with FA condition; however, the
positive effect is reflected in the attentional boost pattern, in
which
words encoded with red (target) circles are at an advantage
relative
to words encoded with green circles, and equivalent to words
presented in the FA condition.
It should be noted that a positive effect of DA on recognition
memory was reported by Voss, Baym, and Paller (2008). How-
ever, this beneficial effect was a retrieval rather than an
encoding
phenomenon and was due to a shifting in the type of retrieval
strategies used in the FA and DA conditions, with low-
confidence
guess responses (prevalent in the DA condition) being more ac-
curate than both recollection and familiarity responses
(prevalent
in the FA condition). In contrast, the positive effect produced
by
the attentional boost effect arises during encoding, due to
enhanced
perceptual processing of a stimulus (a study word or picture)
that
appears simultaneously with a dual-task target (Swallow &
Jiang,
55. 2010). Thus, the present findings provide evidence that
attending
to a second task may produce an absolute enhancement to the
perceptual processing of study words during the encoding
phase.
Another recent study (Lin, Pype, Murray, & Boynton, 2010) has
reported that memory for images presented at the same time as
dual-task targets was enhanced relative to images presented at
the
same time as distracters. Lin et al. (2010) had participants study
a
rapid serial visual presentation sequence of letters appearing
over
full-field urban and natural scenes. They were required to press
a
key when a target letter was presented and, immediately after
the
end of the sequence, to indicate whether a probe scene was
included in the sequence. Like in the attentional boost effect,
Lin
et al. (2010) found that memory performance was better for
scenes
co-occurring with target letters than for scenes co-occurring
with
distractor letters. Nonetheless, there are important differences
with
both the present findings and those illustrated by Swallow and
Jiang (2010, 2011) that prevent a direct comparison between the
two set of findings. Methodologically, the presentation rates of
the
scenes in the study by Lin et al. (2010) were much faster than
those
in our experiments (133 ms/item vs. 500 ms/item); participants
were tested after single sequences rather than after multiple se-
quences; and the test phase was performed immediately after
each
56. sequence, without breaks (thus tapping short-term memory
rather
than long-term memory). Even more important, Lin et al. (2010)
tested source memory, whereas in the presented experiments,
our
focus was on memory for the scenes themselves. This is a
crucial
difference, because source and item memory may rely on inde-
pendent brain systems (Davachi, 2006). Finally, Lin et al.
(2010)
found no evidence of dual-task interference, because source
mem-
ory for images presented together with targets in the DA
condition
were remembered better than the same images in the FA
condition.
In contrast, in the present Experiment 1 (explicit recognition),
as
well as in the studies by Swallow and Jiang (2010, 2011), the
performance for words presented at the same time of target
circles
did not differ between the FA and DA conditions. The latter
finding suggests that qualitatively different mechanisms may
un-
derlie the two phenomena.
It is unlikely that the absolute advantage observed in perceptual
implicit memory can be explained by alternative accounts of the
boost effect, such as attentional cuing, reinforcement learning,
distinctiveness, and oddball processing (see Swallow & Jiang,
2010, 2011, for extensive discussion). The attentional cuing hy-
pothesis states that participants might use the red circles as
selec-
tive cues to attend to the background images, thereby enhancing
their processing. Swallow and Jiang (2011, Experiment 1) ruled
57. out this explanation by showing that the attentional boost effect
was not obtained when the target squares appeared 100 ms
before
the to-be-remembered images (instead of being presented
concur-
rently with them). Similarly, the effect does not reflect the rein-
forcement of predictive information in memory, because
Swallow
and Jiang (2011, Experiment 2) demonstrated that there was no
significant advantage for images that preceded the presentation
of
the target squares by 100 ms (according to the reinforcement
hypothesis, any stimulus that consistently anticipates the reward
should be reinforced in memory). Given that the attentional
boost
effect shows the same characteristics in the present experiments
as
in those reported by Swallow and Jiang (2010, 2011; expect for
the
absolute nature of the target words’ advantage), the above evi-
dence implies that these hypotheses are unlikely explanations
for
our data. A third possibility is that the attentional boost effect
represents nothing more than the classical isolation effect in
mem-
ory (Hunt & McDaniel, 1993), whereby people tend to have
superior memory for an item when it is perceptually or semanti-
cally distinct from the other items in the list than when it is not.
As
outlined by Swallow and Jiang (2010), this hypothesis
incorrectly
predicts that the attentional boost effect should have occurred in
the FA condition, a result that was never observed in our
experi-
ments. Furthermore, all of the above accounts would expect a
significant attentional boost effect in Experiment 4 (with
58. semantic
classification), because the encoding phase was the same as
those
in Experiments 1–3: The fact that we found no advantage in that
condition supports an explanation based on the distinction
between
perceptual and conceptual processes.
Finally, the present results have consequences for theories of
implicit memory. In particular, our data are inconsistent with
the
claim that implicit memory is the result of involuntary,
automatic
processes (the automaticity hypothesis; Aloisi et al., 2004).
What
is more interesting is that they do not support the distinction
between implicit tests based on identification processes (i.e., in
which the retrieval cues have unique solutions) and production
processes (i.e., in which the retrieval cues induce a competition
between multiple solutions; Gabrieli et al., 1999), because the
T
hi
s
do
cu
m
en
t
is
co
63. attentional boost effect had a differential impact on three
implicit
tasks than can be all classified as identification tests.
In summary, four experiments illustrated the surprising finding
that, in contrast with the well-known interference hypothesis
(Craik et al., 1996), when participants are asked to study a list
of
words while simultaneously detecting infrequent targets in a
sec-
ond task, performance in the DA condition can be increased
over
and above the levels obtained in the FA condition. This
facilitation
occurred in two perceptual implicit tests but not in a semantic
classification task. Overall, the results support the hypothesis
that
the boost effect is caused by the opening of an attentional gate
after
the detection of infrequent targets, which increases the
perceptual
encoding of simultaneous stimuli (Swallow & Jiang, 2010,
2011).
References
Aloisi, B. A., McKone, E., & Heubeck, B. G. (2004). Implicit
and explicit
memory performance in children with attention
deficit/hyperactivity
disorder. British Journal of Developmental Psychology, 22,
275–292.
doi:10.1348/026151004323044618
Barca, L., Burani, C., & Arduino, L. S. (2002). Word naming
64. times and
psycholinguistic norms for Italian nouns. Behavior Research
Methods,
Instruments, & Computers, 34, 424 – 434.
Brysbaert, M., Van Wijnendaele, I., & De Deyne, S. (2000).
Age-of-
acquisition effects in semantic processing tasks. Acta
Psychologica, 104,
215–226. doi:10.1016/S0001-6918(00)00021-4
Cohen, J. (1988). Statistical power analysis for the behavioral
sciences.
Hillsdale, NJ: Erlbaum.
Craik, F. I. M., Govoni, R., Naveh-Benjamin, M., & Anderson,
N. D.
(1996). The effects of divided attention on encoding and
retrieval pro-
cesses in human memory. Journal of Experimental Psychology:
Gen-
eral, 125, 159 –180. doi:10.1037/0096-3445.125.2.159
Davachi, L. (2006). Item, context and relational episodic
encoding in
humans. Current Opinion in Neurobiology, 16, 693–700.
doi:10.1016/j
.conb.2006.10.012
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007).
G�Power 3: A
flexible statistical power analysis program for the social,
behavioral, and
biomedical sciences. Behavior Research Methods, 39, 175–191.
doi:
10.3758/BF03193146
65. Gabrieli, J. D., Vaidya, C. J., Stone, M., Francis, W. S.,
Thompson-Schill,
S. L., Fleischman, D. A., . . . Wilson, R. S. (1999). Convergent
behav-
ioral and neuropsychological evidence for a distinction between
identi-
fication and production forms of repetition priming. Journal of
Experi-
mental Psychology: General, 128, 479 – 498. doi:10.1037/0096-
3445
.128.4.479
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-
analysis.
Orlando, FL: Academic Press.
Hunt, R. R., & McDaniel, M. A. (1993). The enigma of
organization and
distinctiveness. Journal of Memory and Language, 32, 421–
445. doi:
10.1006/jmla.1993.1023
Laudanna, A., Thornton, A. M., Brown, G., Burani, C., &
Marconi, L.
(1995). Un corpus dell’italiano scritto contemporaneo dalla
parte del
ricevente. In S. Bolasco L. Lebart, & A. Salem (Eds.), III
Giornate
internazionali di Analisi Statistica dei Dati Testuali (Vol. 1, pp.
103–
109). Roma, Italy: Cisu.
Light, L. L., Prull, M. W., & Kennison, R. F. (2000). Divided
attention,
aging, and priming in exemplar generation and category
66. verification.
Memory & Cognition, 28, 856 – 872. doi:10.3758/BF03198421
Lin, J. Y., Pype, A. D., Murray, S. O., & Boynton, G. M.
(2010). Enhanced
memory for scenes presented at behaviorally relevant points in
time.
PLoS Biology, 8(3), Article e1000337. doi:10.1371/journal.pbio
.1000337
Lozito, J. P., & Mulligan, N. W. (2010). Exploring the role of
attention
during implicit memory retrieval. Journal of Memory and
Language, 63,
387–399. doi:10.1016/j.jml.2010.06.007
Makovski, T., Swallow, K. M., & Jiang, Y. V. (2011).
Attending to
unrelated targets boosts short-term memory for color arrays.
Neuropsy-
chologia, 49, 1498 –1505.
doi:10.1016/j.neuropsychologia.2010.11.029
Masson, M. E. J. (2011). A tutorial on a practical Bayesian
alternative to
null-hypothesis significance testing. Behavior Research
Methods, 43,
679 – 690. doi:10.3758/s13428-010-0049-5
Mulligan, N. W. (1998). The role of attention during encoding
in implicit
and explicit memory. Journal of Experimental Psychology:
Learning,
Memory, and Cognition, 24, 27– 47. doi:10.1037/0278-
7393.24.1.27
67. Mulligan, N. W. (2003). Memory: Implicit versus explicit. In L.
Nadel
(Ed.), Encyclopedia of cognitive science (pp. 1114 –1120).
London,
England: MacMillan.
Mulligan, N. W. (2008). Attention and memory. In H. L.
Roediger (Ed.),
Learning and memory: A comprehensive reference (pp. 7–22).
Oxford,
England: Elsevier. doi:10.1016/B978-012370509-9.00134-0
Mulligan, N. W., & Brown, A. S. (2003). Attention and implicit
memory.
In L. Jiménez (Ed.), Attention and implicit learning: Advances
in con-
sciousness research (pp. 297–334). Amsterdam, the
Netherlands: John
Benjamins.
Mulligan, N. W., Duke, M., & Cooper, A. W. (2007). The
effects of
divided attention on auditory priming. Memory & Cognition,
35, 1245–
1254. doi:10.3758/BF03193598
Mulligan, N. W., & Hartman, M. (1996). Divided attention and
indirect
memory tests. Memory & Cognition, 24, 453– 465. doi:10.3758/
BF03200934
Mulligan, N. W., & Peterson, D. (2008). Attention and implicit
memory in
the category verification and lexical decision tasks. Journal of
Experi-
mental Psychology: Learning, Memory, and Cognition, 34, 662–
68. 679.
doi:10.1037/0278-7393.34.3.662
Mulligan, N. W., & Stone, M. (1999). Attention and conceptual
priming:
Limits on the effects of divided attention in the category-
exemplar
production task. Journal of Memory and Language, 41, 253–
280. doi:
10.1006/jmla.1999.2648
Newell, B. R., Cavenett, T., & Andrews, S. (2008). On the
immunity of
perceptual implicit memory to manipulations of attention.
Memory &
Cognition, 36, 725–734. doi:10.3758/MC.36.4.725
Parkin, A. J., Reid, T. K., & Russo, R. (1990). On the
differential nature of
implicit and explicit memory. Memory & Cognition, 18, 507–
514.
doi:10.3758/BF03198483
Raftery, A. E. (1995). Bayesian model selection in social
research. In P. V.
Marsden (Ed.), Sociological methodology 1995 (pp. 111–196).
Cam-
bridge, England: Blackwell.
Roediger, H., & McDermott, K. (1993). Implicit memory in
normal human
subjects. In F. Boller & J. Grafman (Eds.), Handbook of
neuropsychol-
ogy (pp. 63–131). Amsterdam, the Netherlands: Elsevier, North-
Holland.
69. Roediger, H. L., III, Weldon, M., Stadler, M., & Riegler, G.
(1992). Direct
comparison of two implicit memory tests: Word fragment and
word stem
completion. Journal of Experimental Psychology: Learning,
Memory,
and Cognition, 18, 1251–1269. doi:10.1037/0278-
7393.18.6.1251
Spataro, P., Cestari, V., & Rossi-Arnaud, C. (2011). The
relationship
between divided attention and implicit memory: A meta-
analysis. Acta
Psychologica, 136, 329 –339. doi:10.1016/j.actpsy.2010.12.007
Spataro, P., Mulligan, N. W., & Rossi-Arnaud, C. (2010).
Effects of
divided attention in the word-fragment completion task with
unique and
multiple solutions. European Journal of Cognitive Psychology,
22, 18 –
45. doi:10.1080/09541440802685979
Spataro, P., Mulligan, N. W., & Rossi-Arnaud, C. (2011).
Attention and
implicit memory: The role of the activation of multiple
representations.
Experimental Psychology, 58, 110 –116. doi:10.1027/1618-
3169/
a000074
T
hi
s
do
75. http://dx.doi.org/10.1016/j.actpsy.2010.12.007
http://dx.doi.org/10.1080/09541440802685979
http://dx.doi.org/10.1027/1618-3169/a000074
http://dx.doi.org/10.1027/1618-3169/a000074
Srinivas, K., & Roediger, H. L., III. (1990). Classifying implicit
memory
tests: Category association and anagram solution. Journal of
Memory
and Language, 29, 389 – 412. doi:10.1016/0749-
596X(90)90063-6
Swallow, K. M., & Jiang, Y. V. (2010). The attentional boost
effect:
Transient increases in attention to one task enhance
performance in a
second task. Cognition, 115, 118 –132.
doi:10.1016/j.cognition.2009.12
.003
Swallow, K. M., & Jiang, Y. V. (2011). The role of timing in
the attentional
boost effect. Attention, Perception, & Psychophysics, 73, 389 –
404.
doi:10.3758/s13414-010-0045-y
Swallow, K. M., & Jiang, Y. V. (2012). Goal-relevant events
need not be
rare to boost memory for concurrent images. Attention,
Perception, &
Psychophysics, 74, 70 – 82. doi:10.3758/s13414-011-0227-2
Vaidya, C. J., Gabrieli, J. D. E., Keane, M. M., Monti, L. A.,
Gutierrez-
Rivas, H., & Zarella, M. M. (1997). Evidence for multiple
76. mechanisms
of conceptual priming on implicit memory tests. Journal of
Experimen-
tal Psychology: Learning, Memory, and Cognition, 23, 1324 –
1343.
doi:10.1037/0278-7393.23.6.1324
Voss, J. L., Baym, C. L., & Paller, K. A. (2008). Accurate
forced-choice
recognition without awareness of memory retrieval. Learning &
Mem-
ory, 15, 454 – 459. doi:10.1101/lm.971208
Wagner, A. D., Koutstaal, W., Maril, A., Schacter, D. L., &
Buckner, R. L.
(2000). Process-specific repetition priming in left inferior
prefrontal
cortex. Cerebral Cortex, 10, 1176 –1184.
doi:10.1093/cercor/10.12.1176
Wig, G. S., Grafton, S. T., Demos, K. E., & Kelley, W. M.
(2005). A causal
role for neural activity reductions during repetition priming.
Nature
Neuroscience, 8, 1228 –1233. doi:10.1038/nn1515
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
Zacks, J. M., Speer, N. K., Swallow, K. M., Braver, T. S., &
Reynolds,
J. R. (2007). Event perception: A mind/brain perspective.
Psychological
77. Bulletin, 133, 273–293. doi:10.1037/0033-2909.133.2.273
Ziegler, J. C., & Perry, C. (1998). No more problems in
Coltheart’s
neighborhood: Resolving neighborhood conflicts in the lexical
decision
task. Cognition, 68, B53–B62. doi:10.1016/S0010-
0277(98)00047-X
Received June 7, 2012
Revision received September 12, 2012
Accepted September 12, 2012 �
T
hi
s
do
cu
m
en
t
is
co
py
ri
gh
te
d
81. d
is
no
t
to
be
di
ss
em
in
at
ed
br
oa
dl
y.
1231ATTENTIONAL BOOST EFFECT IN IMPLICIT
MEMORY
http://dx.doi.org/10.1016/0749-596X%2890%2990063-6
http://dx.doi.org/10.1016/j.cognition.2009.12.003
http://dx.doi.org/10.1016/j.cognition.2009.12.003
http://dx.doi.org/10.3758/s13414-010-0045-y
http://dx.doi.org/10.3758/s13414-011-0227-2
http://dx.doi.org/10.1037/0278-7393.23.6.1324
http://dx.doi.org/10.1101/lm.971208
http://dx.doi.org/10.1093/cercor/10.12.1176
82. http://dx.doi.org/10.1038/nn1515
http://dx.doi.org/10.1006/jmla.2002.2864
http://dx.doi.org/10.1037/0033-2909.133.2.273
http://dx.doi.org/10.1016/S0010-0277%2898%2900047-
XDivided Attention Can Enhance Memory Encoding: The
Attentional Boost Effect in Implicit MemoryExperiment
1MethodParticipantsMaterials and procedureResults and
DiscussionExperiment 2MethodResults and
DiscussionExperiment 3MethodResults and
DiscussionExperiment 4MethodResults and DiscussionGeneral
DiscussionReferences
RESEARCH ARTICLE
Talking about Climate Change and Global
Warming
Maurice Lineman☯, Yuno Do☯, Ji Yoon Kim, Gea-Jae Joo*
College of Natural Sciences, Department of Biological
Sciences, Pusan National University, Busan, South
Korea
☯ These authors contributed equally to this work.
* [email protected]
Abstract
The increasing prevalence of social networks provides
researchers greater opportunities to
evaluate and assess changes in public opinion and public
sentiment towards issues of
social consequence. Using trend and sentiment analysis is one
method whereby research-
83. ers can identify changes in public perception that can be used to
enhance the development
of a social consciousness towards a specific public interest. The
following study assessed
Relative search volume (RSV) patterns for global warming
(GW) and Climate change (CC)
to determine public knowledge and awareness of these terms. In
conjunction with this, the
researchers looked at the sentiment connected to these terms in
social media networks. It
was found that there was a relationship between the awareness
of the information and the
amount of publicity generated around the terminology.
Furthermore, the primary driver for
the increase in awareness was an increase in publicity in either
a positive or a negative
light. Sentiment analysis further confirmed that the primary
emotive connections to the
words were derived from the original context in which the word
was framed. Thus having
awareness or knowledge of a topic is strongly related to its
public exposure in the media,
and the emotional context of this relationship is dependent on
the context in which the rela-
84. tionship was originally established. This has value in fields like
conservation, law enforce-
ment, or other fields where the practice can and often does have
two very strong emotive
responses based on the context of the problems being examined.
Introduction
Identifying trends in the population, used to be a long and
drawn out process utilizing surveys
and polls and then collating the data to determine what is
currently most popular with the pop-
ulation [1, 2]. This is true for everything that was of merit to
the political organizations present,
regarding any issue of political or public interest.
Recently, the use of the two terms ‘Climate Change’ and
‘Global Warming’ have become
very visible to the public and their understanding of what is
happening with respect to the cli-
mate [3]. The public response to all of the news and publicity
about climate has been a search
for understanding and comprehension, leading to support or
disbelief. The two terms while
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 1 / 12
a11111
OPEN ACCESS
Citation: Lineman M, Do Y, Kim JY, Joo G-J (2015)
Talking about Climate Change and Global Warming.
PLoS ONE 10(9): e0138996. doi:10.1371/journal.
86. use this ambiguity to their
favor in providing news to the public [4]. Within the news
releases, the expression ‘due to cli-
mate change’ has been used to explain phenomological
causality.
These two terms “global warming–(GW)” and “climate change–
(CC)” both play a role in
how the public at large views the natural world and the changes
occurring in it. They are used
interactively by the news agencies, without a thought towards
their actual meaning [3, 4].
Therefore, the public in trying to identify changes in the news
and their understanding of those
changes looks for the meaning of those terms online. The extent
of their knowledge can be
examined by assessing the use of the terms in online search
queries. Information searches
using the internet are increasing, and therefore can indicate
public or individual interest.
Internet search queries can be tracked using a variety of
analytic engines that are indepen-
dent of, or embedded into, the respective search engines (google
trend, naver analytics) and are
used to determine the popularity of a topic in terms of internet
searches [5]. The trend engines
will look for selected keywords from searches, keywords chosen
for their relevance to the field
or the query being performed.
The process of using social media to obtain information on
public opinion is a practice that
has been utilized with increasing frequency in modern research
for subjects ranging from poli-
tics [6, 7] to linguistics [8–10] complex systems [11, 12] to
87. environment [13]. This variety of
research belies the flexibility of the approach, the large
availability of data availability for min-
ing in order to formulate a response to public opinion regarding
the subject being assessed. In
modern society understanding how the public responds
regarding complex issues of societal
importance [12].
While the two causally connected terms GW and CC are used
interchangeably, they
describe entirely different physical phenomena [14]. These two
terms therefore can be used to
determine how people understand the parallel concepts,
especially if they are used as internet
search query terms in trend analysis. However, searching the
internet falls into two patterns,
searches for work or for personal interest, neither of which can
be determined from the trend
engines. The By following the searches, it is possible to
determine the range of public interest
in the two terms, based on the respective volumes of the search
queries. Previously in order to
mine public opinion on a subject, government agencies had to
revert to polling and surveys,
which while being effective did not cover a very large
component of the population [15–17].
Google trend data is one method of measuring popularity of a
subject within the population.
Individuals searching for a topic use search keywords to obtain
the desired information [5, 18].
These keywords are topic sensitive, and therefore indicate the
level of knowledge regarding the
searched topic. The two primary word phrases here “climate
change” and “global warming” are
88. unilateral terms that indicate a level of awareness about the
issue which is indicative of the indi-
viduals interest in that subject [5, 19, 20]. Google trend data
relates how often a term is
searched, that is the frequency of a search term can be
identified from the results of the Goo-
gle1 trend analysis. While frequency is not a direct measure of
popularity, it does indicate if a
search term is common or uncommon and the value of that term
to the public at large. The
relationship between frequency and popularity lies in the
volume of searches by a large number
of individuals over specific time duration. Therefore, by
identifying the number of searches
during a specific period, it is possible to come to a proximate
understanding of how popular or
common a term is for the general population [21]. However, the
use of trend data is more
appropriately used to identify awareness of an issue rather than
its popularity.
This brings us to sentiment analysis. Part of the connection
between the search and the pop-
ulations’ awareness of an issue can be measured using how they
refer to the subject in question.
This sentiment, is found in different forms of social media, or
social networking sites sites i.e.
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 2 / 12
twitter1, Facebook1, linked in1 and personal blogs [7, 22–24].
89. Thus, the original information,
which was found on the internet, becomes influenced by
personal attitudes and opinions [25]
and then redistributed throughout the internet, accessible to
anyone who has an internet con-
nection and the desire to search. This behavior affects the
information that now provides the
opportunity to assess public sentiment regarding the prevailing
attitudes regarding environ-
mental issues [26, 27]. To assess this we used Google1 and
Twitter1 data to understand public
concerns related to climate change and global warming. Google
trend was used to trace changes
in interest between the two phenomena. Tweets (comments
made on Twitter1) were analyzed
to identify negative or positive emotional responses.
Comparatively, twitter data is more indicative of how people
refer to topics of interest [28–
31], in a manner that is very linguistically restricted. As well,
twitter is used as a platform for
verbal expression of emotional responses. Due to the
restrictions on tweet size (each tweet can
only be 140 characters in length), it is necessary to be more
direct in dealing with topics of
interest to the tweeter. Therefore, the tweets are linguistically
more emotionally charged and
can be used to define a level of emotional response by the
tweeter.
The choice of target words for the tweets and for the Google
trend searches were the specific
topic phrases [32, 33]. These were chosen because of the
descriptive nature of the phrases. Scien-
tific literature is very specific in its use and therefore has very
definitive meanings. The appropri-
90. ation of these words by the population as a method for
describing their response to the variation
in the environment provides the basis for the choice as target
words for the study. The classifica-
tion of the words as being positive versus negative lies in the
direction provided by Frank Lutz.
This politicization of a scientific word as a means of directing
public awareness, means the pre-
scription of one phrase (climate change) as being more positive
than the other (global warming).
Global warming is defined as the long-term trend of increasing
average global temperatures;
alternatively, climate change is defined as a change in global or
regional climate patterns, in
particular a change apparent from the mid to late 20th century
onwards and attributed to the
increased levels of atmospheric carbon dioxide arising from the
use of fossil fuels. Therefore,
the search keywords were chosen based on their scientific value
and their public visibility.
What is important about the choice of these search terms is that
due to their scientific use, they
describe a distinctly identifiable state. The more specific these
words are, the less risk of the
algorithm misinterpreting the keyword and thus having the
results misinterpreted [34–36].
The purpose of the following study was to identify trends within
search parameters for two
specific sets of trend queries. The second purpose of the study
was to identify how the public
responds emotionally to those same queries. Finally, the
purpose of the study was to determine
if the two had any connections.
91. Methods
Data Collection
Public awareness of the terms climate change and global
warming was identified using Google
Trends (google.com/trends) and public databases of Google
queries [37]. To specify the exact
searches we used the two terms ‘climate change’ and ‘global
warming’ as query phrases. Queries
were normalized using relative search volume (RSV) to the
period with the highest proportion
of searches going to the focal terms (i.e. RSV = 100 is the
period with the highest proportion
for queries within a category and RSV = 50 when 50% of that is
the highest search proportion).
Two assumptions were necessary for this study. The first is, of
the two terms, climate change
and global warming, that which draws more search results is
considered more interesting to
the general population. The second assumption is that changes
in keyword search patterns are
indicators of the use of different forms of terminology used by
the public. To analyze
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 3 / 12
sentiments related to climate change and global warming, tweets
containing acronyms for cli-
mate change and global warming were collected from Twitter
API for the period from October
12 to December 12, 2013. A total of 21,182 and 26,462 tweets
92. referencing the terms climate
change and global warming were collected respectively. When
duplicated tweets were identi-
fied, they were removed from the analysis. The remaining
tweets totaled 8,465 (climate change)
and 8,263 (global warming) were compiled for the sentiment
analysis.
Data Analysis
In Twitter1 comments are emotionally loaded, due to their
textually shortened nature. Senti-
ment analysis, which is in effect opinion mining, is how
opinions in texts are assessed, along
with how they are expressed in terms of positive, neutral or
negative content [36]. Nasukawa
and Yi [10]state that sentiment analysis identifies statements of
sentiment and classifies those
statements based on their polarity and strength along with their
relationship to the topic.
Sentiment analysis was conducted using Semantria1 software
(www.semantria.com), which
is available as an MS Excel spreadsheet application plugin. The
plugin is broken into parts of
speech (POS), the algorithm within the plugin then identifies
sentiment-laden phrases and
then scores them from -10 to 10 on a logarithmic scale, and
finally the scores for each POS are
tabulated to identify the final score for each phrase. The tweets
are then via statistical inferences
tagged with a numerical value from -2 to 2 and given a polarity,
which is classified as positive,
neutral or negative [36]. Semantria1, the program utilized for
this study, has been used since
2011 to perform sentiment analyses [7, 22].
93. For the analysis, an identity column was added to the dataset to
enable analysis of individual
tweets with respect to sentiment. A basic sentiment analysis was
conducted on the dataset
using the Semantria1 plugin. The plugin uses a cloud based
corpus of words tagged with senti-
mental connotations to analyze the dataset. Through statistical
inference, each tweet is tagged
with a sentiment value from -2 to +2 and a polarity of (i)
negative, (ii) neutral, or (iii) positive.
Positive nature increases with increasing positive sentiment.
The nature of the language POS
assignation is dependent upon the algorithmic classification
parameters defined by the Seman-
tria1 program. Determining polarity for each POS is achieved
using the relationship between
the words as well as the words themselves. By assigning
negative values to specific negative
phrases, it limits the use of non-specific negation processes in
language; however, the program
has been trained to assess non-specific linguistic negations in
context.
A tweet term frequency dictionary was computed using the N-
gram method from the cor-
pus of climate change and global warming [38]. We used a
combination of unigrams and
bigrams, which has been reported to be effective [39]. Before
using the N-gram method, typo-
logical symbols were removed using the open source code editor
(i.e. Notepad) or Microsoft
Words’ “Replace” function.
Differences in RSV’s for the terms global warming and climate
change for the investigation
period were identified using a paired t-test. Pettitt and Mann-
94. Kendall tests were used to iden-
tify changes in distribution, averages and the presence of trends
within the weekly RSV’s. The
Pettitt and MK tests, which assume a stepwise shift in the mean
(a break point) and are sensi-
tive to breaks in the middle of a time series, were applied to test
for homogeneity in the data
[40]. Temporal trends within the time series were analyzed with
Spearman’s non-parametric
correlation analysis. A paired t-test and Spearman’s non-
parametric correlation analysis were
conducted using SPSS software (version 17.0 SPSS In corp.
Chicago IL) and Pettitt and MK
tests were conducted using XLSTAT (version 7.0).
To determine the accuracy and reliability of the Sentiment
analysis, a Pearson’s chi-square
analysis was performed. This test identifies the difference ratio
for each emotional response
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 4 / 12
http://www.semantria.com
group, and then compares them to determine reliance and
probability of interactions between
the variables, in this case the terms global warming and climate
change.
Results
According to Google trend (Fig 1) from 2004–2014, people
searched for the term global warm-
95. ing (n = 8,464; mean ± S.D = 25.33 ± 2.05) more frequently
than climate change (n = 8,283;
mean ± S.D. = 7.97±0.74). Although the Intergovernmental
Panel on Climate Change (IPCC)
published its Fourth Assessment Report in 2007 and was
awarded the Nobel Prize, interest in
the term global warming as used in internet searches has
decreased significantly since 2010
(K = 51493, t = 2010-May-23, P<0.001). Further the change in
RSV also been indicative of the
decreased pattern (Kendall’s tau = -0.336, S = -44563,
P<0.001). The use of the term “climate
change” has risen marginally since 2006 (K = 38681, t = 2006-
Oct-08, P<0.001), as indicated
by a slight increase (Kendall’s tau = -0.07, S = 9068, P<0.001).
These findings show that the dif-
ference in usage of the two terms climate change and global
warming has recently been
reduced.
The sentiment analysis of tweets (Fig 2) shows that people felt
more negative about the term
global warming (sentiment index = -0.21±0.34) than climate
change (-0.068±0.36). Global
warming tweets reflecting negative sentiments via descriptions
such as, “bad, fail, crazy, afraid
and catastrophe,” represented 52.1% of the total number of
tweets. As an example, the tweet,
“Supposed to snow here in the a.m.! OMG. So sick of already,
but Saturday says 57 WTF!” had
the lowest score at -1.8. Another observation was that 40.7% of
tweets, including “agree, recom-
mend, rescue, hope, and contribute,” were regarded as neutral.
While 7.2% of tweets conveyed
positive messages such as, “good, accept, interesting, and
truth.” One positive global warming
96. tweet, read, “So if we didn’t have global warming, would all
this rain be snow!”. The results
from the Pearson’s chi-square analysis showed that the
relationship between the variables was
significant (Pearson’s chi-square –763.98, d.f. = 2, P<0.001).
Negative climate change tweets
represented 33.1% of the total while neutral tweets totaled
49.8%, while positive climate change
tweets totaled 17.1%.
Understandably, global warming and climate change are the
terms used most frequently to
describe each phenomenon, respectively, as revealed by the N-
gram analysis (Table 1). When
people tweeted about global warming, they repeatedly used
associated such as, “ice, snow, Arc-
tic, and sea.” In contrast, tweets referring to climate change
commonly used, “report, IPCC,
world, science, environment, and scientist.” People seem to
think that climate change as a phe-
nomenon is revealed by scientific investigation.
Discussion
Internet searches are one way of understanding the popularity of
an idea or meme within the
public at large. Within that frame of reference, the public looks
at these two terms global warm-
ing and climate change and their awareness of the roles of the
two phenomena [41]. From 2004
to 2008, the search volumes for the term global warming far
exceeded the term climate change.
The range for the term global warming in Relative search
volumes (RSV) was more than double
that of climate change in this period (Fig 1). From 2008 on the
RSV’s began to steadily decrease
until in 2014 when the RSV’s for the term global warming were
97. nearly identical to those for the
term climate change. From 2008 there was an increase in the
RSVs for CC until 2010 at which
point the RSVs also began to decline for the term climate
change. The decline in the term cli-
mate change for the most part paralleled that of the term global
warming from 2010 on to the
present.
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 5 / 12
While we are seeing the increases and decreases in RSVs for
both the terms global warming
and climate change, the most notable changes occur when the
gap between the terms was the
greatest, from 2008 through to 2010. During this period, there
was a very large gap found
between the RSVs for the terms global warming and climate
change; however, searches for the
term climate change was increasing while searches for the tem
global warming were decreasing.
The counter movement of the RSV’s for the two terms shows
that there is a trend happening
with respect to term recognition. At this point, there was an
increase in the use of the CC term
while there was a corresponding decrease in the use of the GW
term. The change in the use of
the term could have been due to changes in the publicity of the
respective terms, since at this
point, the CC term was being used more visibly in the media,
and therefore the CC term was
98. showing up in headlines and the press, resulting in a larger
number of searches for the CC
term. Correspondingly, the decrease in the use of the GW term
is likely due to the changes in
how the term was perceived by the public. The public press
determines how a term is used,
since they are the body that consistently utilizes a term
throughout its visible life. The two
terms, regardless of how they differ in meaning, are used with
purpose in a scientific context,
yet the public at large lacks this definition and therefore has no
knowledge of the variations in
the terms themselves [42]. Therefore, when searching for a
term, the public may very well,
choose the search term that they are more comfortable with,
resulting in a search bias, since
they do not know the scientific use of the term.
The increase in the use of the CC term, could be a direct result
of the release of the fourth
assessment report for the IPCC in 2007 [43]. The publicity
related to the release of this docu-
ment, which was preceded by the release of the Al Gore
produced documentary “An Inconve-
nient Truth”, both of which were followed by the selection by
the Nobel committee of Al Gore
Fig 1. Change in relative search volume (RSV) for “global
warming” and “climate change” as search terms (2007–2013);
dash line represents the
mean for the RSV for each period.
doi:10.1371/journal.pone.0138996.g001
Climate Change and Global Warming
99. PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 6 / 12
and the IPCC scientists for the Nobel Prize in 2007 [43]. These
three acts individually may not
have created the increased media presence of the CC term;
however, at the time the three
events pushed the CC term and increased its exposure to the
public which further drove the
public to push for positive environmental change at the political
level [44, 45]. This could very
well have resulted in the increases in RSV’s for the CC term.
This point is more likely to depict
accurately the situation, since in 2010 the use of the two terms
decline at almost the same rate,
with nearly the same patterns.
Thus with respect to trend analysis, what is interesting is that
RSVs are paralleling the press
for specific environmental events that have predetermined value
according to the press. The
press in increasing the visibility of the term may drive the
increases in the RSV’s for that term.
Prior to 2007, the press was using the GW term indiscriminately
whenever issues affecting the
Fig 2. Distribution of positive, neutral, and negative sentiments
in tweets about global warming and climate change.
doi:10.1371/journal.pone.0138996.g002
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
100. 2015 7 / 12
global climate arose; however, after the movie, the report and
then the Nobel prize the terminol-
ogy used by the press switched and the CC term became the
word du jour. This increased the vis-
ibility of the word to the public, thereby it may be that
increasing public awareness of the word,
but not necessarily its import, is the source for the increases in
RSV’s between 2008 and 2010.
The decline in the RSV’s then is a product of the lack of
publicity about the issue. As the
terms become more familiar, there would be less necessity to
drive the term publicly into the
spotlight; however, occasionally events/situations arise that
refocus the issue creating a resur-
gence in the terms even though they have reached their peak
visibility between 2008 and 2010.
Since these terms have such an impact on the daily lives of the
public via local regional
national and global weather it is understandable that they have
an emotional component to
them [46]. Every country has its jokes about the weather, where
they come up with cliché’s
about the weather (i.e. if you don’t like the weather wait
10minutes) that often show their dis-
cord and disjunction with natural climatological patterns [47].
Furthermore, some sectors of
society (farmers) have a direct relationship with the climate and
their means of living; bad
Table 1. Tweet Terms-frequency Dictionary for Global
101. Warming and Climate Change.
Rank Global warming Climate change
Words Frequency Words Frequency
1 Global warming 821 Climate change 802
2 Climate change 507 Global warming 267
3 Ice 177 Ow 177
4 Years 158 Report 167
5 Snow 143 IPCC 143
6 Arctic 136 World 136
7 Scientist 124 Science 119
8 Sea 119 Environment 105
9 Cause 114 Scientist 101
10 Ow 109 Help 100
11 Time 101 Action 97
12 Show 97 Impacts 85
13 Report 94 Arctic 82
14 Science 91 Time 79
15 Data 88 Australia 77
102. 16 World 85 Study 75
17 Earth 82 Caused 72
18 Environment 75 Talk 70
19 Coverage 70 Human 68
20 Percent 68 Need 65
21 Human 67 People 63
22 Study 65 Deniers 60
23 Satellite 63 Huff 58
24 IPCC 60 Risk 57
25 EPA 56 Fight 56
26 Expert 54 Years 54
27 Stop 53 Make 53
28 Fight 52 Politics 52
29 Million 51 Nations 51
30 People 50 Carbon 49
doi:10.1371/journal.pone.0138996.t001
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 8 / 12
103. weather is equal to bad harvests, which means less money. To
understand how society repre-
sents this love hate relationship with the weather, the twitter
analysis was performed. Twitter, a
data restricted social network system, has a limited character
count to relay information about
any topic the sender chooses to relate. These tweets can be used
to assess the sentiment of the
sender towards a certain topic. As stated previously, the
sentiment is defined by the language of
the tweet within the twitter system. Sentiment analysis showed
that the two terms differed
greatly. Based on the predefined algorithm for the sentiment
analysis, certain language compo-
nents carried a positive sentiment, while others carried a
negative sentiment. Tweets about GW
and CC were subdivided based on their positive, neutral and
negative connotations within the
tweet network. These emotions regardless of their character still
play a role in how humans
interacts with surroundings including other humans [48, 49] As
seen in Fig 2 the different
terms had similar distributions, although with different ranges
in the values. Global warming
showed a much smaller positive tweet value than did climate
change. Correspondent to this the
respective percentage of positive sentiments for CC was more
than double that of GW. Com-
paratively, the neutral percentiles were more similar for each
term with a small difference.
However, the negative sentiments for the two terms again
showed a greater disparity, with neg-
ative statements about GW nearly double those of climate
104. change.
These differences show that there is a perceptive difference in
how the public relates to the
two terms Global Warming and Climate Change [50, 51].
Climate change is shown in a more
positive light than global warming simply based on the tweets
produced by the public. The dif-
ference in how people perceive climate change and global
warming is possibly due to the press,
personal understanding of the terms, or level of education.
While this in itself is indefinable,
since by nature tweets are linguistically restrictive, the thing to
take from it is that there is a
measurable difference in how individuals respond to
climatological changes that they are
experiencing daily. These changes have a describable effect on
how the population is respond-
ing to the publicity surrounding the two terms to the point
where it can be used to manipulate
governmental policy [52].
Sentiment analysis is a tool that can be used to determine how
the population feels about a
topic; however, the nature of the algorithm makes it hard to
effectively determine how this is
being assessed. For the current study, the sentiment analysis
showed that there was a greater
negative association with the term global warming than with the
term climate change. This dif-
ference, which while being an expression of individual like or
dislike at the time the tweet was
created, denotes that the two terms were either not understood
in their true form, or that indi-
viduals may have a greater familiarity with one term over the
other, which may be due to a lon-
105. ger exposure to the term (GW) or the negative press associated
with the term (GW).
Conclusions
Trend analysis identified that the public is aware of the
terminology used to describe climato-
logical variation. The terminology showed changes in use over
time with global warming start-
ing as the more well-known term, and then its use decreased
over time. At the same time, the
more definitive term climate change had less exposure early on;
however, with the increase of
press exposure, the public became increasingly aware of the
term and its more accurate defini-
tion. This increase appeared to be correspondent with the
increasing publicity around three
very powerful press exposure events (a documentary, a
scientific report release and a Nobel
Prize). The more the term was used the more people came to use
it, this included searches on
the internet.
Comparatively sentiment analysis showed that the two terms
had differential expressions in
the population. With climate change being seen in a more
positive frame than global warming.
Climate Change and Global Warming
PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 9 / 12
The use of sentiment analysis as a tool to evaluate how the
population is responding to a feature
106. is an important tool. However, it is a tool that measures, it does
not define.
Social network systems and internet searches are effective tools
in identifying changes in
both public awareness and public perception of an issue.
However, in and of itself, these are
bell ringers they can be used to determine the importance of an
issue, but not the rationale
behind the why it is important. This is an important fact to
remember when using analytical
tools that evaluate social network systems and their use by the
public.
Acknowledgments
This study was financially supported by the 2015 Post-Doc.
Development Program of Pusan
National University
Author Contributions
Conceived and designed the experiments: YD GJJ. Performed
the experiments: ML YD. Ana-
lyzed the data: ML YD. Contributed reagents/materials/analysis
tools: JK YD. Wrote the paper:
ML YD GJJ.
References
1. Motteux NMG. Evaluating people-environment relationships:
developing appropriate research method-
ologies for sustainable management and 2003.
2. Castle NG, Brown J, Hepner KA, Hays RD. Review of the
literature on survey instruments used to col-
lect data on hospital patients' perceptions of care. Health
Services Research. 2005; 40(6p2):1996–
107. 2017.
3. Weingart P, Engels A, Pansegrau P. Risks of communication:
discourses on climate change in science,
politics, and the mass media. Public Understanding of Science.
2000; 9(3):261–83.
4. McCright AM, Dunlap RE. Challenging global warming as a
social problem: An analysis of the conser-
vative movement's counter-claims. Social Problems. 2000;
47(4):499–522.
5. Ripberger JT. Capturing curiosity: Using internet search
trends to measure public attentiveness. Policy
Studies Journal. 2011; 39(2):239–59.
6. Chadwick A. The Political information cycle in a hybrid news
system: the British Prime Minister and the
"Bullygate" Affair. The International Journal of Press/Politics.
2011; 16(1):3–29.
7. Abeywardena IS. Public opinion on OER and MOCC: a
sentiment analysis of twitter data. 2014.
8. Choi Y, Cardie C, editors. Learning with compositional
semantics as structural inference for subsenten-
tial sentiment analysis. Proceedings of the Conference on
Empirical Methods in Natural Language Pro-
cessing; 2008: Association for Computational Linguistics.
9. Wiegand M, Balahur A, Roth B, Klakow D, Montoyo A,
editors. A survey on the role of negation in senti-
ment analysis. Proceedings of the workshop on negation and
speculation in natural language process-
ing; 2010: Association for Computational Linguistics.
108. 10. Nasukawa T, Yi J, editors. Sentiment analysis: Capturing
favorability using natural language process-
ing. Proceedings of the 2nd international conference on
Knowledge capture; 2003: ACM.
11. Eagle N, Pentland A. Reality mining: sensing complex
social systems. Personal and Ubiquitous Com-
puting. 2006; 10(4):255–68.
12. Bossomaier T. Complexity: Metrics and modules. Advances
in Complex Systems. 2003; 6(3):313–29.
13. De Longueville B, Smith RS, Luraschi G, editors. Omg,
from here, i can see the flames!: a use case of
mining location based social networks to acquire spatio-
temporal data on forest fires. Proceedings of
the 2009 international workshop on location based social
networks; 2009: ACM.
14. Whitmarsh L, O'Neill S. Green identity, green living? The
role of pro-environmental self-identity in deter-
mining consistency across diverse pro-environmental
behaviours. Journal of Environmental Psychol-
ogy 2010; 30:305–14.
15. Roberts ML, Hite PA, Bradley CF. Understanding attitudes
toward progressive taxation. Public Opin Q.
1994; 58(2):165–90.
16. Rigg R, Findo S, Wechselberger M, Gorman ML, Sillero-
Zubiri C, et al. Mitigating carnivore-livestock
conflict in Europe: lessons from Slovakia. Oryx. 2011;
45(2):272–80.
Climate Change and Global Warming
109. PLOS ONE | DOI:10.1371/journal.pone.0138996 September 29,
2015 10 / 12
17. Miksza P, Roeder M, Biggs D. Surveying Colorado Band
Directors' Opinions of Skills and Characteris-
tics Important to Successful Music Teaching. Journal of
Research in Music Education. 2010; 57
(4):364–81.
18. Choi H, Varian H. Predicting the present with google trends.
Economic Record 2012; 88:2–9.
19. Granka L. Inferring the public agenda from implicit query
data. Understanding the User-Logging and
Interpreting User Interactions in Information Search and
Retrieval (UIIR-2009). 2009: 28.
20. Do Y, Kim JY, Lineman M, Kim DK, Joo GJ. Using internet
search behavior to assess public awareness
of protected wetlands. Conservation Biology. 2015; 29(1):271–
9. doi: 10.1111/cobi.12419 PMID:
25412014
21. Kim JY, Do Y, Im R-Y, Kim G-Y, Joo G-J. Use of large
web-based data to identify public interest and
trends related to endangered species. Biodiversity and
Conservation. 2014; 23(12):2961–84.
22. Aston N, Liddle J, Hu W. Twitter sentiment in data streams
with perceptron. Journal of Computer and
Communications. 2014; 2014(2):11–6.
23. Wang X, Wei F, Liu X, Zhou M, Zhang M, editors. Topic
sentiment analysis in twitter: a graph-based
110. hashtag sentiment classification approach. Proceedings of the
20th ACM international conference on
Information and knowledge management; 2011: ACM.
24. Kouloumpis E, Wilson T, Moore J. Twitter sentiment
analysis: The good the bad and the omg! ICWSM.
2011; 11:538–41.
25. Mobley C, Vagias WM, DeWard SL. Exploring additional
determinants of environmentally responsible
behavior: The influence of environmental literature and
environmental attitudes. Environment and
Behavior. 2009.
26. Bollen J, Pepe A, Mao H. Modeling public mood and
emotion: Twitter sentiment and socio-economic
phenomena. arXiv preprint arXiv:09111583. 2009.
27. Zhao D, Rosson MB, editors. How and why people Twitter:
the role that micro-blogging plays in informal
communication at work. Proceedings of the ACM 2009
international conference on Supporting group
work; 2009: ACM.
28. Signorini A, Segre AM, Polgreen PM. The use of Twitter to
track levels of disease activity and public
concern in the US during the influenza A H1N1 pandemic. Plos
One. 2011; 6(5):e19467. doi: 10.1371/
journal.pone.0019467 PMID: 21573238
29. Eysenbach G. Can tweets predict citations? Metrics of social
impact based on Twitter and correlation
with traditional metrics of scientific impact. Journal of Medical
Internet Research. 2011; 13(4).
30. Phelan O, McCarthy K, Smyth B, editors. Using twitter to