Transfer of Perceptual Expertise: The Case of Simplified and...
Transcript of Transfer of Perceptual Expertise: The Case of Simplified and...
Cognitive Science (2016) 1–28Copyright © 2016 Cognitive Science Society, Inc. All rights reserved.ISSN: 0364-0213 print / 1551-6709 onlineDOI: 10.1111/cogs.12307
Transfer of Perceptual Expertise: The Case of Simplifiedand Traditional Chinese Character Recognition
Tianyin Liu,a Tin Yim Chuk,a Su-Ling Yeh,b Janet H. Hsiaoa
aDepartment of Psychology, University of Hong KongbDepartment of Psychology, National Taiwan University
Received 10 June 2014; received in revised form 23 July 2015; accepted 3 August 2015
Abstract
Expertise in Chinese character recognition is marked by reduced holistic processing (HP),
which depends mainly on writing rather than reading experience. Here we show that, while simpli-
fied and traditional Chinese readers demonstrated a similar level of HP when processing characters
shared between the simplified and traditional scripts, simplified Chinese readers were less holistic
than traditional Chinese readers in perceiving simplified characters; this effect depended mainly
on their writing rather than reading performance. However, the two groups did not differ in HP of
traditional characters, regardless of their difference in reading and writing performances. Our
image analysis showed high visual similarity between the two character types, with a larger vari-
ance among simplified characters; this may allow simplified Chinese readers to interpolate and
generalize their skills to traditional characters. Thus, transfer of perceptual expertise may be con-
strained by both the similarity in feature and the difference in exemplar variance between the cate-
gories.
Keywords: Holistic processing; Chinese character recognition; Reading; Writing
1. Introduction
Holistic processing (i.e., gluing features together into a Gestalt) has been found to be a
behavioral visual expertise marker for the recognition of faces (Tanaka & Farah, 1993)
and many other nonface objects, such as cars (Gauthier, Curran, Curby, & Collins, 2003),
fingerprints (Busey & Vanderkolk, 2005), and greebles (a kind of artificial stimuli; Gau-
thier & Tarr, 1997). In experts’ eyes, the parts of these stimuli are integrated and recog-
nized as a whole instead of various parts (Gauthier & Tanaka, 2002), and thus experts’
flexibility to access the information of individual parts is reduced (Maurer, Le Grand, &
Correspondence should be sent to Janet Hsiao, Department of Psychology, University of Hong Kong,
PokFulam Road, Hong Kong. E-mail: [email protected]
Mondloch, 2002). In the literature, holistic processing is commonly assessed through the
composite paradigm. Taking the composite face effect as an example, a composite face is
made by combining the top half of one face and the bottom half of another face together.
In the original composite face study done by Young, Hellawell, and Hay (1987), partici-
pants found it difficult to recognize the top of a celebrity’s face when it was aligned with
the bottom of another celebrity’s face. Later, Hole (1994) also demonstrated the compos-
ite face effect using unfamiliar faces. He asked participants to perform same/different
judgments about the top halves of two composite faces while ignoring the other face
halves, and the results showed that participants reported two identical top halves to look
different when they were aligned with two different bottom halves. Presumably, holistic
processing binds the two halves of the faces together and people get interference from
the irrelevant halves in the same/different judgment of the attended halves, and conse-
quently they perceive two identical top halves of faces as different when the top halves
are paired with different bottom halves. Thus, holistic processing assessed through the
composite paradigm can be understood as obligatory attention to all parts of an object,
which leads to failure of selective attention to parts (Richler, Tanaka, Brown, & Gauthier,
2008).
Chinese characters are the basic writing units in Chinese orthography. They consist of
strokes packed into a relatively constant square-shaped configuration, in contrast to words
in most alphabetic languages, which are linear combinations of letters. It has been sug-
gested that Chinese character recognition shares some similarities with face recognition
(McCleery et al., 2008). For example, both faces and Chinese characters have a homoge-
nous shape, are recognized at the individual level, and are learned in an upright orienta-
tion. Yet in contrast to face recognition, expertise in recognizing Chinese characters is
marked by reduced holistic (i.e., increased analytic) processing assessed using the com-
posite paradigm (Hsiao & Cottrell, 2009).1 This effect may be due to expert Chinese
readers’ orthographic knowledge about Chinese characters, and consequently they have a
better ability to selectively attend to different parts of a Chinese character than novices.
More specifically, Chinese characters are composed of strokes, which combine to form
over 200 different basic stroke patterns in Chinese orthography (Hsiao & Shillcock,
2006), and these basic stroke patterns are the smallest functional units in Chinese charac-
ter recognition (Chen, Allport, & Marshall, 1996). For expert Chinese readers, when rec-
ognizing Chinese characters, they may be more sensitive to the internal constituent
components/stroke patterns and have better ability to ignore some unimportant configural
information for recognition, such as exact distances between features (Ge, Wang,
McCleery, & Lee, 2006), as compared with novices (Ho, Ng, & Ng, 2003; Hsiao & Cot-
trell, 2009). Consequently, expert readers may process Chinese characters less holistically
than novices. This phenomenon suggests that holistic processing effects, as an expertise
marker in visual object recognition, depend on the characteristics of the stimuli and the
perceivers’ learning experience with the stimuli (Hsiao & Cottrell, 2009; see also Tso,
Au, & Hsiao, 2014). It is possible that holistic processing effects can also be influenced
by the perceivers’ prior experiences with similar object types. Nevertheless, it remains
unclear what factors may influence the transfer of holistic processing effects.
2 T. Liu et al. / Cognitive Science (2016)
There are currently two Chinese writing systems in use in Chinese speaking regions,
namely simplified and traditional Chinese. The simplification movement of Chinese char-
acters first emerged in the early 1900s, and the simplified script widely implemented
today was the version created during the writing reform initiated by the central govern-
ment of the People’s Republic of China in 1956 for easing the learning process in both
recognition (reading) and production (writing) (Zhao & Baldauf, 2011). Today the major-
ity of Chinese speaking regions, including Mainland China, Singapore, and Malaysia, use
the simplified script, while Hong Kong and Taiwan continue to use the traditional script.
Three basic emphases of the simplification process were (a) to simplify common radicals
of characters and apply in contexts where they occur (e.g., radical “言” was simplified as
“讠,” and applied to characters containing this radical, “討” to “讨,” “話” to “话,” etc.);
(b) to remove elements within characters (e.g., “愛” to “爱” and “髮” to “发”); and (c) to
combine some homophones and eliminate redundant characters (e.g., “範” and “范” were
combined into the latter, and the former was eliminated; McBride-Chang, Chow, Zhong,
Burgess, & Hayward, 2005). The simplification process did not apply to all characters;
among the most frequently used 3,500 characters, around 40% of them were simplified
and have approximately 22.5% fewer strokes than the traditional counterparts (Gao &
Kao, 2002); the other 60% remained the same, or in other words, are shared between the
two scripts. The contrast between simplified and traditional Chinese characters thus pro-
vides us a unique opportunity to examine factors that may influence transfer effects of
holistic processing across similar object types.
The effect of simplifying the script has been a subject of dispute over the last two dec-
ades. For instance, while simplified characters were designed to ease the learning process,
many researchers (e.g., Chen, 1999; Seybolt & Chiang, 1979) believe that pure reduction
in strokes or elimination of characters without standardization of principles was an artifi-
cial control of the natural direction of language evolution (Hodge and Louie, 1998), and
it may make the simplified characters harder to learn in the long term: On one hand,
reducing the number of strokes may make the characters easier to write for beginners as
they are visually less complex; on the other hand, because the simplification process did
not preserve all phonetic and semantic information of the characters, up to a certain point,
characters may become hard to learn due to violation of orthographic rules as readers’
lexicon size expands (Chen, 1999). Zhao and Baldauf (2011) also argued that the charac-
ters in the simplified script were “simplified individually, but complicated systematically”
(p. 171), indicating that fewer strokes did not necessarily make the characters easier to
learn and memorize given their inconsistency with the orthographic rules. In a nutshell,
simplicity does not guarantee efficiency.
Meanwhile, some recent studies suggest that learning simplified Chinese characters
may be beneficial. For example, McBride-Chang et al. (2005) found that visual skills of
children who learned simplified characters were significantly better than those of Hong
Kong children who learned the traditional script, especially visual discrimination skills.
They argued that because simplified Chinese characters have fewer visual features for dis-
crimination than their traditional counterparts, simplified Chinese readers need to develop
better visual discrimination skills in order to tell characters apart. Consistent with this
T. Liu et al. / Cognitive Science (2016) 3
finding, Peng, Minett, and Wang’s (2010) ERP data suggested that adult simplified char-
acter readers have developed a better ability to discriminate characters that are identical
in the simplified and traditional scripts (i.e., shared characters) from noncharacters that
were constructed by adding or removing one stroke from the characters, as compared
with traditional character readers. This result suggests that simplified character readers
may have developed better analytic visual discrimination skills than traditional character
readers. These findings thus suggest that reading simplified characters may require more
analytic (i.e., less holistic) processing than reading traditional characters.
Tso et al. (2014) recently showed how writing experience influences holistic process-
ing in Chinese character recognition (see also Tso, Au, & Hsiao, 2011, 2012, 2013). They
recruited proficient Chinese readers who were skilled in both reading and writing Chinese
characters (Writers), those who had proficient reading ability but limited writing experi-
ence in Chinese (Limited-Writers), and non-Chinese readers (Novices). They found that
compared with Novices, Limited-Writers perceived Chinese characters more holistically,
whereas Writers perceived Chinese characters less holistically. This result reflects an
inverted U-shape pattern of holistic processing with increased Chinese character exper-
tise: Perceptual experience alone increases holistic processing, whereas sensorimotor
experience reduces holistic processing. Thus, the reduced holistic processing effect
observed in expert Chinese readers may depend on writing rather than reading perfor-
mance. Although simplified Chinese readers may be able to read traditional characters
through their similarity with simplified characters or context information, they generally
do not know how to write them (and similarly for traditional Chinese readers to read and
write simplified characters). Thus, similar to Limited-Writers, they may perceive charac-
ters in their unfamiliar script more holistically, and this effect may depend on their writ-
ing rather than reading performance.
Here, we aim to examine whether native simplified Chinese readers process Chinese
characters more analytically (less holistically) than traditional Chinese readers due to the
differences in their scripts. In addition, we examine how simplified and traditional Chi-
nese readers process Chinese characters in the script they are not familiar with. We first
examine their perception of characters that are shared in the two scripts (i.e., shared char-
acters); according to the previous findings about simplified Chinese readers’ superior
visual discrimination skills to traditional Chinese readers (McBride-Chang et al., 2005;
Peng et al., 2010), simplified Chinese readers may exhibit stronger analytic processing
(weaker holistic processing) in processing shared characters. We then examine partici-
pants’ perception of Chinese characters in the simplified and the traditional forms. We
predict that simplified Chinese readers will perceive simplified characters less holistically
than traditional readers due to their expertise with the simplified script, and vice versa in
the perception of traditional characters; these holistic processing effects may be related to
their writing ability, that is, the ability to recall and write down the characters (Tso et al.,
2013). On the other hand, it is also possible that Chinese readers are able to transfer their
analytic processing skills to process characters in the script they are not familiar with due
to the similarity between the two Chinese scripts; in this case, the two groups may not dif-
fer in holistic processing in perceiving either the traditional or the simplified characters.
4 T. Liu et al. / Cognitive Science (2016)
This investigation thus enables us to examine transfer of analytic character processing in
readers with different character reading experiences in order to understand the factors that
may influence transfer effects of perceptual expertise.
2. Experiment 1
2.1. Methods
2.1.1. ParticipantsTwenty-four native readers of simplified Chinese from Mainland China and 24 native
readers of traditional Chinese from Hong Kong participated in the study. They were all
skilled writers in their own script: All Mainland China participants had passed the Chinese
test of National Entrance Examination to college, and all Hong Kong participants had
passed the Chinese test of Hong Kong Advanced Level Examination. They were all stu-
dents at University of Hong Kong; all simplified Chinese readers had resided in Hong
Kong for less than 1 year (average length of stay was 9.35 months) at the time they were
recruited. To ensure the two reader groups had limited exposure to the other script at the
time they were recruited, a questionnaire was designed to collect information about their
language background and exposure. Participants of both groups reported having no formal
education about the other script, and being exposed to the script they were less familiar
with for less than 10% of the text they encountered daily on average. Note that the official
written languages used in Hong Kong are English and traditional Chinese and the official
language for instruction at University of Hong Kong is English. All Hong Kong traditional
Chinese readers were born and raised in Hong Kong, and their experience with traditional
Chinese characters (calculated as years of education) was 15 years on average, signifi-
cantly more than Mainland China simplified Chinese readers’ exposure to traditional Chi-
nese since they arrived in Hong Kong (F(1, 46) = 1,194.95, p < .001). Note also that with
increasing interaction between Mainland China and Hong Kong, publications in simplified
Chinese characters are accessible in Hong Kong, and classes teaching simplified Chinese
characters have been included in some primary and high schools’ in recent years.
In addition, the two groups had similar education background (average years of educa-
tion, Mainland China = 15.79, SE = .38; Hong Kong = 15.29, SE = .42) and similar age
(Mainland China average = 22.71, SE = .70; Hong Kong average = 22.21, SE = .73). All
of them had normal or corrected-to-normal vision and were right-handed as measured by
the Edinburgh Handedness Inventory (Oldfield, 1971).
2.1.2. Holistic processingThe complete composite paradigm was used to examine holistic processing effects
(Gauthier & Bukach, 2007). The experiment procedure was adopted from Hsiao and Cot-
trell (2009). In each trial, participants were presented with a pair of Chinese characters
simultaneously and told to attend to only half of each character and judge whether they
were the same or different. There were in total four types of trials depending on congru-
T. Liu et al. / Cognitive Science (2016) 5
ency and response (see Fig. 1a): same in congruent trials, different in congruent trials,same in incongruent trials, and different in incongruent trials. In congruent trials, the
attended and irrelevant halves of the characters led to the same response (i.e., both were
the same or different); in incongruent trials, they led to different responses. The level of
holistic processing was assessed by the performance difference between the congruent
and incongruent trials.
Note that in contrast to the partial design where holistic processing is assessed by an
alignment effect (i.e., the performance difference between aligned and misaligned trials;
e.g., Young et al., 1987; Hole, 1994), in the complete design it is assessed by a congru-
ency effect (i.e., the performance difference between congruent and incongruent trials)
without a misalignment condition (Hsiao & Cottrell, 2009). Hsiao and Cottrell (2009)
showed that misalignment is able to reduce holistic processing observed in Chinese char-
acter processing, consistent with the literature on face and object recognition (e.g., Rich-
ler et al., 2008).
2.1.2.1. Materials: The materials consisted of 480 pairs of Chinese characters in Ming
font (the most common font used in print), divided equally into three script types: 160
pairs were unique simplified characters; 160 pairs were the corresponding traditional ver-
sion of the simplified characters, that is, having same meaning and pronunciation but dif-
fering in orthography; the remaining 160 pairs were characters shared between the two
scripts, that is, shared characters (see Table 1 for an illustration). The frequency of the
characters in the materials ranged from 2 per million to 250 per million based on single
character appearance in 80/90’s text (Ho & Kwan, 2001; the frequency of the characters
in the dataset ranged from 1 per million to 3,679 per million, with the median being 25
per million). Characters of different script types were matched in relative character fre-
quency, and the traditional characters were significantly more complex (complexity of
characters was measured by number of strokes throughout this paper) than the simplified
A B
Fig. 1. Illustration of stimulus pairs in the complete composite paradigm (a) and trial sequences (b). Stimulus
contrast was lowered to avoid ceiling effects. In (a), the attended components are the bottom halves, while
the top halves shaded in gray (not shown in the experiment) are irrelevant.
6 T. Liu et al. / Cognitive Science (2016)
ones (t(159) = 6.17, p < .01). In each script type, half of the characters had a top–bottom(TB) configuration, the other half were left–right (LR) structured (see Fig. 2, for exam-
ples), and the TB and LR characters were matched in complexity (F(1, 959) = 2.03,
p = .21) and frequency (F(1, 959) = .67, p = .50). We used both TB and LR characters
to counterbalance possible influence from character structure (Yeh & Li, 2002); the LR
structure is the most dominant structure in Chinese orthography, followed by the TB
structure (see, e.g., Hsiao & Shillcock, 2006).
The 80 character pairs in each script type and character configuration combination were
further divided into the four conditions in the complete composite paradigm, with 20 pairs
in each condition (shown in Fig. 1a). Each character was then divided into two compo-
nents, horizontally for TB and vertically for LR configured characters. In either character
configuration condition, the attended halves were the same across congruent and incongru-
ent trials (see Fig. 1a, for an illustration), and character frequency and visual complexity
were matched across congruent and incongruent trials (character frequency: t(239) = 0.82,
p = 0.51; visual complexity: t(239) = 1.2, p = 0.21). Summary of the descriptive proper-
ties of the stimuli used for the holistic processing task are presented in Table 2.
Fig. 2. Examples of top–bottom (TB) and left–right (LR) structured shared, simplified, and traditional coun-
terparts of simplified Chinese characters (the red line separates the top from the bottom, and the left from the
right components).
Table 1
Examples of test stimuli of the complete composite paradigm
T. Liu et al. / Cognitive Science (2016) 7
Table
2
Summaryofthedescriptiveproperties
ofthestim
uliusedin
thecomplete
composite
paradigm
CharacterType
Shared
Sim
plified
Traditional
Congruency
Congruent
Incongruent
Congruent
Incongruent
Congruent
Incongruent
CorrectResponse
Sam
eDifferent
Sam
eDifferent
Sam
eDifferent
Sam
eDifferent
Sam
eDifferent
Sam
eDifferent
Number
ofpairs
TB
20
20
20
20
20
20
20
20
20
20
20
20
LR
20
20
20
20
20
20
20
20
20
20
20
20
Meancomplexity(number
ofstrokes)
TB
11.4
9.8
11
10.9
9.6
9.1
9.5
9.7
15.5
15.3
15.5
16.4
LR
9.5
8.9
9.7
9.0
9.1
8.4
8.6
8.6
14.2
14.1
14.6
15.2
Meanfrequency
(per
million)
TB
18
30
14
52
35
28
67
32
40
26
76
35
LR
47
36
22
20
20
28
29
29
20
26
25
25
8 T. Liu et al. / Cognitive Science (2016)
2.1.2.2. Design: The design had three within-subject variables: congruency (congruent
vs. incongruent), character configuration (TB vs. LR), and script type (simplified vs. tradi-
tional vs. shared); and a between-subject variable: group (simplified vs. traditional Chi-
nese readers). The dependent variable was discrimination sensitivity measured by A0,which is calculated by the following equation:
A0 ¼ :5þ signðH � FÞ ðH � FÞ2 þ jH � Fj4maxðH;FÞ � 4HF
" #
H and F present the hit rate and false alarm rate, respectively, and better performance is
indicated by a higher A0. Here we used A0 instead of D0 because A0 is a bias-free nonpara-
metric measure of sensitivity according to signal detection theory (Stanislaw & Todorov,
1999), and it can be calculated when the hit rate or the false-alarm rate is 1 or 0, which
was present in the data we collected.
2.1.2.3. Procedure: All characters were shown in low contrast (Michelson contrast = .2)
to avoid ceiling effects. In each trial, participants were presented with a central fixation
cross for 1,000 ms, followed by a symbol indicating which half of the character (top or
bottom for TB characters; left or right for LR characters) they should attend to. They
were then presented with a pair of characters above and below the initial fixation, respec-
tively, for 500 ms, followed by a mask with alternating black and white pixels (Fig. 1b).
Both characters were about 2.5 degrees of visual angle away from the center, and each
character was approximately 1.5 9 1.5 cm in size with a viewing distance of 50 cm,
occupying around 1.7 degrees of visual angle. Participants judged whether the attended
halves of the two characters were the same or different as quickly and accurately as pos-
sible with a serial response box; their accuracy was collected and no feedback was given.
There were six blocks of test; each block had 80 trials; characters with different configu-
rations or in different script types were presented in different blocks. The sequence of
blocks was counterbalanced across participants while the four different types of trials
were randomized in each block. A practice session with characters not used in the materi-
als was given before the test. Participants were given ample time to rest between blocks,
and they started a new block on their own when they were ready. On average, the holistic
processing task took 40–45 min to finish.
2.1.3. Reading and writing performanceTasks assessing participants’ reading and writing abilities were adopted from Tso et al.
(2011, 2013) to examine whether the differences in holistic processing, if any, can be
attributed to these factors. Participants’ reading ability was assessed by a character nam-
ing task, in which they named the characters in their mother tongue (i.e., Mandarin for
Mainland China participants and Cantonese for Hong Kong participants). Their writing
ability was assessed by a character copying and a word dictation task. Summary of the
descriptive properties of the stimuli used in examining reading and writing performances
T. Liu et al. / Cognitive Science (2016) 9
are presented in Table 3. All reading and writing assessment tasks were carried out after
the completion of the holistic processing task; breaks were given between tasks, and on
average it took 30–40 min to finish the reading and writing tasks (and hence less than
90 min to complete the entire study).
2.1.3.1. Character naming: The materials consisted of 120 Chinese characters, half with a
TB structure and the other half with a LR configuration. In either configuration, 20 charac-
ters were simplified characters, 20 were the corresponding traditional version of the simpli-
fied characters, and the remaining 20 were shared characters; these characters were not
used in the complete composite (holistic processing) task. All characters selected for the
naming task were of medium to high frequency (ranging from 28 per million to 1,316 per
million; Ho & Kwan, 2001) to mimic the day to day experience in speaking and reading in
Chinese; they were matched in relative frequency across the script types (F(2,117) = 1.69, p = .19). The traditional characters had significantly more strokes than the
simplified ones, (t(39) = 10.92, p < .01); in other words, they were more visually com-
plex. In each trial, after a 500 ms central fixation, participants were presented with a char-
acter occupying approximately 1.5 degrees of visual angle at the center of the screen, and
they were asked to read it out in front of a microphone. The onset of their pronunciation
was detected by a microphone attached to a serial response box. Their response time was
recorded as the duration between the onset of the character presentation and the onset of
the pronunciation. Upon their response, the screen turned blank and the experimenter
pressed buttons on a response box to record the accuracy and initiate the next trial.
2.1.3.2. Character copying: Participants copied 60 characters (20 shared, 20 simplified,
and 20 traditional) as quickly and as accurately as possible. The characters were
Table 3
Summary of the descriptive properties of the stimuli used in the character naming, copying, and dictation
tasks
Shared Simplified Traditional
TB LR TB LR TB LR
Character naming
Number 20 20 20 20 20 20
Mean complexity 9.9 7.7 7.6 7.5 13 14
Mean frequency (per million) 216 249 240 128 216 125
Character copying
Number 10 10 10 10 10 10
Mean complexity 14.7 19.1 14 9.2 20.7 19.6
Mean frequency (per million) 2.8 8.9 7.1 6 4.2 3
Word dictation
Number 10 10 10 10 10 10
Mean complexity (of target character) 9 9.4 8.7 9.1 17.9 19.8
Word frequency (per million) 273 385 167 197 189 211
Note TB, top–bottom; LR, left–right.
10 T. Liu et al. / Cognitive Science (2016)
randomly selected from those used in the character naming task; half of them had a TB
structure, whereas the other half had a LR configuration. All characters were of low to
medium frequency (ranging from 2 per million to 26 per million) and were matched
in relative frequency across script types (F(2, 57) = 2.05, p = .14). High-frequency
characters were not used in the copying task to reduce the chance of ceiling effects.
The traditional characters had significantly more strokes than their simplified counterparts
(t(19) = 8.26, p < .01). In each trial, after a 500 ms central fixation, participants were
shown a stimulus at the center of the screen, occupying around 1.5 degrees of visual
angle, and were asked to copy it as quickly and as accurately as possible. After they fin-
ished copying, they pressed a button on a serial response box to signal completion and
the screen turned blank. Then the experimenter pressed buttons on a serial response box
to record the accuracy and to initiate the next trial. The 60 stimuli were presented in a
random order in one block.
2.1.3.3. Word dictation task: Forty characters (20 shared and 20 traditional/simplified)
were selected from the character naming task. Each character was concatenated with a
second character to compose a two-character word, and these words were used here.
Words instead of characters were used to avoid ambiguity, because a single Chinese char-
acter has around 11 homophones on average (Tan & Perfetti, 1998). All words were of
medium to high frequency (from 162 per million to 1,037 per million) to mimic the daily
experience of writing in Chinese (Taiwan Ministry of Education, 1997) and were matched
in relative word frequency across different script types (F(2, 57) = 2.77, p = .11). Partici-
pants listened to the words presented in a female voice in their native language, that is,
Cantonese for Hong Kong participants and Mandarin for Mainland China participants.
The audio recordings of the words were presented by a computer in a random order. Par-
ticipants wrote down each word in their own script first and then in the other script, even
if they thought the characters were the same in the two scripts. If they did not know how
to write a character, they indicated it by putting a cross on the space. In each trial, after
the words “get ready” were presented on the screen for 500 ms, participants were pre-
sented with a stimulus; they then pressed a button on a serial response box to indicate
whether they knew how to write it or not. After writing the word in both scripts, they
pressed a button to indicate completion and start of the next trial. Their accuracy of writ-
ing the first character of each word was assessed (to match the character naming task).
2.2. Results
In the A0 measure, repeated-measures ANOVA revealed main effects of congruency (F(1,46) = 68.89, p < .01, g2
p = .61) and character configuration (F(1, 46) = 26.08, p < .01,
g2p = .37). Participants did better in congruent (M = .98, SE < .01) than incongruent tri-
als (M = .94, SE = .01), and in processing LR (M = .97, SE = .02) than TB (M = .95,
SE = .02) characters. There was also a significant interaction between configuration and
congruency (F(1, 46) = 35.28, p < .01, g2p = .44). Participants were more holistic in
processing TB characters than LR characters; yet this effect did not interact with group
T. Liu et al. / Cognitive Science (2016) 11
(F(1, 46) = .02, p = .89). In addition, there was a marginal three-way interaction between
script, congruency, and group (F(2, 92) = 2.94, p = .058, g2p = .06), suggesting that the
interaction between congruency and group might be different across the three script types.
Meanwhile, there was no main effect of group (F(1, 46) = 2.64, p = .11, g2p = .03). To
investigate how the two groups differed in processing different script types, we first com-
pared their behavior in processing shared characters, and then contrasted their perfor-
mance in perceiving simplified and traditional characters.
2.2.1. Shared charactersA main effect of congruency was found (F(1, 46) = 52.34, p < .01, g2
p = .53), but
there was no interaction between congruency and group (F(1, 46) = 2.05, p = .16,
g2p = .10), nor main effect of group (F(1, 46) = 2.01, p = .16, g2
p = .04; Fig. 3a), and
both groups demonstrated a similar level of holistic processing. This result suggests that
the two groups do not differ in holistic processing of shared Chinese characters because
they are both experts in reading shared characters.2
2.2.2. Simplified versus traditional charactersHere we contrasted the processing of simplified and traditional characters using
repeated-measures ANOVA with script type (simplified vs. traditional character), congru-
ency, character configuration, and group as the independent variables. The results
revealed main effects of congruency (F(1, 46) = 59.08, p < .01, g2p = .57) and character
configuration (F(1, 46) = 26.53, p < .01, g2p = .37). There was a marginal interaction
between congruency and group (F(1, 46) = 3.92, p = .054, g2p = .08): Simplified Chinese
readers tended to perceive the characters less holistically overall. There was also a three-
way interaction between script type, congruency, and group (F(1, 46) = 4.93, p < .05,
g2p = .10), suggesting the interaction between congruency and group was different
between the two scripts. No main effect of group was found (F(1, 46) = 2.19, p = .15,
g2p = .05). We then examined their performance in processing the two scripts separately.
A B C
Fig. 3. Performance of Mainland China and Hong Kong Chinese readers in the complete composite paradigm
with (a) shared Chinese characters, (b) simplified Chinese characters, and (c) traditional Chinese characters.
12 T. Liu et al. / Cognitive Science (2016)
In processing simplified characters, as predicted, there was a main effect of congruency
(F(1, 46) = 36.35, p < .01, g2p = .44; Fig. 3b) and the interaction between congruency
and group (F(1, 46) = 6.51, p < .05, g2p = .12): Simplified Chinese readers processed
simplified characters less holistically than traditional Chinese readers, possibly due to
their expertise with simplified characters.
To examine whether the difference in holistic processing of simplified characters was
dependent on their reading and writing performance measures3 (Table 4), separate ANCO-
VAs were conducted with reading and writing task performances as covariates. We found
that the interaction between congruency and group was still significant when putting their
simplified character naming accuracy or RT as covariate (reading accuracy: F(1,46) = 4.76, p < .05, g2
p = .10; reading RT: F(1, 46) = 4.60, p < .05, g2p = .10). After
controlling for simplified character copying accuracy, the interaction between congruency
and group became marginal (F(1, 46) = 4.08, p = .05, g2p = .08). Only when we put sim-
plified character dictation accuracy as a covariate did the interaction become not signifi-
cant (F(1, 46) = .42, p = .52, g2p = .01). These results are consistent with Tso et al.’s
(2013) finding that the reduced holistic processing effect in expert Chinese character pro-
cessing may depend more on writing rather than reading or copying performance.
In contrast to the results with simplified Chinese characters, in processing traditional
characters, there was only a main effect of congruency (F(1, 46) = 50.48, p < .01,
g2p = .53), but no interaction between group and congruency (F(1, 46) = 0.20, p = .66;
Fig. 3c). This showed that the two groups processed traditional characters with a similar
level of holistic processing, regardless of their performance difference in reading and
writing traditional characters (Table 4). This result suggests that, while traditional Chi-
nese readers processed simplified characters more holistically than simplified Chinese
readers, possibly due to the lack of experience in reading and especially in writing sim-
plified characters, simplified Chinese readers were able to transfer their analytic process-
ing (reduced holistic processing) skills to the processing of traditional characters, even
Table 4
Mainland China and Hong Kong participants’ reading and writing performance in shared, simplified, and tra-
ditional Chinese characters
Task Script Type
Mainland China (Simplified
Chinese Readers)
Hong Kong (Traditional
Chinese Readers)
Accuracy RT (ms) Accuracy RT(ms)
Naming Shared 1.00 (0.00) 284.85 (16.97) 0.99 (0.00) 329.38 (12.30)
Simplified 0.99 (0.00) 280.74 (16.50) 0.97 (0.01) 354.10 (11.48)
Traditional 0.97 (0.01) 329.24 (14.91) 0.99 (0.00) 328.40 (13.20)
Copying Shared 0.96 (0.00) 0.96 (0.00)
Simplified 0.98 (0.00) 0.94 (0.00)
Traditional 0.65 (0.02) 0.94 (0.00)
Dictation Shared 0.99 (0.00) 0.98 (0.00)
Simplified 0.99 (0.00) 0.66 (0.05)
Traditional 0.21 (0.04) 0.99 (0.00)
T. Liu et al. / Cognitive Science (2016) 13
though they particularly had difficulty in writing traditional characters, as reflected in
their copying and dictation performances.
3. Experiment 2
In Experiment 1, we compared holistic processing of Chinese characters in simplified
Chinese readers from Mainland China and traditional Chinese readers from Hong Kong.
Nevertheless, there are several noticeable differences in Chinese teaching methods
adopted in Mainland China and Hong Kong. For example, in Hong Kong, children learn
to read and write Chinese mainly through rote repetition and memorization (Cheung &
Ng, 2003), whereas in Mainland China, children are taught explicitly about character
components and rules of combination. As a result, simplified Chinese readers in the
Mainland may be generally more sensitive to the internal constituent components of char-
acters than traditional Chinese readers in Hong Kong (McBride-Chang et al., 2005). In
addition, Cantonese is the spoken language in Hong Kong, and Mandarin is spoken by
Mainland Chinese; the two spoken languages differ in various ways. For example, Can-
tonese has 19 onsets (initial consonants) and nine tones (six working), while Mandarin
has 21 onsets and four tones. Thus, their character reading performance may differ due to
the differences in the acoustic properties between the two spoken languages. In addition,
while Cantonese is the spoken language, the written Chinese is still based on Mandarin
words and grammar, which gives rise to the disparity between the oral and written lan-
guage in Hong Kong (Cheung & Ng, 2003). To rule out the influence from the teaching
methods and spoken languages in accounting for the current results, in Experiment 2 we
recruited traditional Chinese readers in Taiwan, where children are also taught about
character components explicitly and speak Mandarin, and compared their character pro-
cessing with the simplified Chinese readers’ behavior in Experiment 1.
3.1. Methods
Twenty-four native traditional Chinese readers from Taiwan participated in the study,
and their performance was compared with the Mainland Chinese participants recruited in
Experiment 1. All Taiwanese participants had passed the Chinese subtest of General
Scholastic Ability Test, which insured their proficiency level in traditional Chinese, and
they were recruited from National Taiwan University. The same questionnaire used in
Experiment 1 was adopted for participants to report their language background and expo-
sure; none of the participants had formal education about the other Chinese script, and
they reported being exposed to the other script for less than 10% of the text they encoun-
tered daily on average. Note that with increasing communication between Taiwan and
Mainland China, books published by Mainland China publishers in simplified Chinese
can be found in bookstores in Taiwan. In addition, the development and flourishing of
the Internet and social media platforms, such as Weibo (Mainland China’s equivalence to
twitter), have attracted users from Mainland China and Taiwan, and both simplified and
14 T. Liu et al. / Cognitive Science (2016)
traditional Chinese characters are supported by these platforms, exposing users to both
scripts. In our experiment, simplified Chinese readers’ length of stay in Hong Kong was
significantly shorter than Taiwan participants’ years of education (F(1, 46) = 1204.13,
p < .001). Both Mainland China and Taiwan participant groups had a similar education
background (average years of education, Mainland China = 15.79, SE = .38; Tai-
wan = 15.79, SE = .35) and similar age (Mainland China = 22.71, SE = .70; Tai-
wan = 22.04, SE = .66). All of them had normal or corrected-to-normal vision and were
right-handed as measured by the Edinburgh Handedness Inventory (Oldfield, 1971). The
materials, design, and procedure used in Experiment 2 were identical to those used in
Experiment 1.
3.2. Results
ANOVA was used to analyze the data. There were three within-subject variables: congru-
ency, character configuration, and script type; and a between-subject variable: group. In
A0, repeated-measures ANOVA revealed main effects of congruency (F(1, 46) = 106.54,
p < .01, g2p = .70) and character configuration (F(1, 46) = 36.48, p < .01, g2
p = .45).
Participants did better in congruent (M = .97, SE = .01) than incongruent trials
(M = .93, SE = .01), and in processing LR (M = .96, SE = .01) than TB (M = .94,
SE = .01) characters. Similar to Experiment 1, there was a significant interaction between
configuration and congruency (F(1, 46) = 53.25, p < .01, g2p = .54), and it did not inter-
act with group (F(1, 46) = 1.53, p = .22, g2p = .03). There was also a significant three-
way interaction between script, congruency, and group (F(2, 92) = 3.74, p < .05,
g2p = .08), suggesting that the interaction between congruency and group was different
across the three script types. Similar to the results in Experiment 1, no significant main
effect of group was found, although it was marginal (F(1, 46) = 3.16, p = .09, g2p = .06).
To investigate how the two groups differed in processing different script types, we
first compared their behavior in processing shared characters, and then contrasted
their performance in processing simplified versus traditional characters as performed in
Experiment 1.
3.2.1. Shared charactersA main effect of congruency was found (F(1, 46) = 50.67, p < .01, g2
p = .52), but
there was no interaction between congruency and group (F(1, 46) = .82, p = .37,
g2p = .02), nor main effect of group (F(1, 46) = .78, p = .38, g2
p = .02; Fig. 4a). The
results echoed the findings in Experiment 1: Both groups demonstrated a similar level of
holistic processing because they are both experts in reading shared characters (Table 5).4
3.2.2. Simplified versus traditional charactersRepeated-measures ANOVA revealed main effects of congruency (F(1, 46) = 75.50,
p < .01, g2p = .63) and character configuration (F(1, 46) = 27.64, p < .01, g2
p = .38),
but no significant main effect of group, although it was marginal (F(1, 46) = 3.42,
p = .08, g2p = .08). There was also a significant interaction between congruency and
T. Liu et al. / Cognitive Science (2016) 15
group (F(1, 46) = 4.63, p < .05, g2p = .10): Mainland China simplified Chinese readers
tended to perceive both characters less holistically overall; and a three-way interaction
between script type, congruency, and group (F(1, 46) = 7.25, p < .05, g2p = .14), suggest-
ing that the interaction between congruency and group was different between the two
scripts. Below we examined their performance in processing the two scripts separately.
In processing simplified characters, as predicted, there was a main effect of congruency
(F(1, 46) = 45.55, p < .01, g2p = .50; Fig. 4b), and the interaction between congruency
and group was significant (F(1, 46) = 8.95, p < .01, g2p = .16): Simplified Chinese read-
ers processed simplified characters less holistically than traditional Chinese readers, possi-
bly due to their expertise with simplified characters.
Again the two groups differed in some reading and writing performance measures in
processing simplified characters5 (Table 5); to examine whether the difference in holistic
processing of simplified characters was dependent on these measures, we put them as
A B C
Fig. 4. Performance of Mainland China and Taiwan Chinese readers in the complete composite paradigm
with (a) shared Chinese characters, (b) simplified Chinese characters, and (c) traditional Chinese characters.
Table 5
Mainland China and Taiwan participants’ reading and writing performance in shared, simplified, and tradi-
tional Chinese characters
Task Script Type
Mainland China (Simplified
Chinese Readers)
Taiwan (Traditional Chinese
Readers)
Accuracy RT(ms) Accuracy RT(ms)
Naming Shared 1.00 (0.00) 284.85 (16.97) 0.99 (0.00) 358.95 (9.86)
Simplified 0.99 (0.00) 280.74 (16.50) 0.93 (0.01) 406.10 (10.78)
Traditional 0.97 (0.01) 329.24 (14.91) 0.98 (0.00) 373.10 (9.00)
Copying Shared 0.96 (0.00) 0.98 (0.01)
Simplified 0.98 (0.00) 0.89 (0.02)
Traditional 0.65 (0.02) 0.98 (0.01)
Dictation Shared 0.99 (0.00) 1.00 (0.01)
Simplified 0.99 (0.00) 0.32 (0.03)
Traditional 0.21 (0.04) 0.99 (0.00)
16 T. Liu et al. / Cognitive Science (2016)
covariates in separate ANCOVA tests. We found that the interaction between congruency
and group was still significant when putting either their simplified character naming RT
(F(1, 46) = 5.07, p < .05, g2p = .10) or simplified character copying accuracy (F(1,
46) = 4.60, p < .05, g2p = .10) as covariates. The interaction between congruency and
group became not significant after controlling for their simplified character naming accu-
racy (F(1, 46) = 2.01, p = .11, g2p = .05), suggesting reading accuracy of simplified Chi-
nese characters might play a role in explaining the difference in holistic processing of
simplified Chinese characters between Mainland China and Taiwan participants. This phe-
nomenon was in contrast to the results of Hong Kong participants, in which the interac-
tion between congruency and group remained significant after controlling for their
simplified character naming accuracy. This difference between Hong Kong and Taiwan
participants may be due to a larger difference in reading accuracy of simplified Chinese
characters between Mainland China and Taiwan participants (.99 vs. .93) than between
Mainland China and Hong Kong participants (.99 vs. .97). In addition, after controlling
for their simplified character dictation accuracy, the interaction effect became not signifi-
cant (F(1, 46) = 1.91, p = .23, g2p = .07). This result was consistent with what we found
in Mainland China and Hong Kong participants, and it is also consistent with Tso et al.’s
(2013) finding that the reduced holistic processing effect in expert Chinese character
processing may depend on writing (dictating) performance.
In contrast, in processing traditional characters, there was only a main effect of con-
gruency (F(1, 46) = 50.92, p < .01, g2p = .53), but no interaction between group and con-
gruency (F(1, 46) = 0.10, p = .99; Fig. 4c). This suggests that the two groups processed
traditional characters with a similar level of holistic processing, regardless of their perfor-
mance difference in reading and writing traditional characters. These findings were con-
sistent with what we found between Mainland China and Hong Kong participants,
suggesting that simplified Chinese readers were more capable of transferring their analytic
processing skill to the processing of the characters in the script unfamiliar to them.
4. Discussion
4.1. Behavioral findings
Experiments 1 and 2 examined whether simplified and traditional Chinese readers pro-
cessed Chinese characters differently in terms of holistic processing, and whether they
were able to transfer their analytic (reduced holistic) processing skills to the script unfa-
miliar to them. The results showed that, first, when processing characters that are identi-
cal in the two scripts (shared characters), simplified Chinese readers and traditional
Chinese readers did not differ in holistic processing, most likely because they were both
experts in reading shared characters. In other words, simplified Chinese readers do not
seem to process shared characters more analytically than traditional Chinese readers due
to their experience in reading simplified Chinese characters, as assessed through the com-
plete composite paradigm. This result is in contrast to Peng et al.’s (2010) finding, which
T. Liu et al. / Cognitive Science (2016) 17
suggests that simplified Chinese readers may be more analytic in processing shared char-
acters than traditional Chinese readers. More specifically, Peng et al. (2010) showed that
in simplified Chinese readers, shared characters elicited larger ERP P300 amplitude than
noncharacters that differed from the shared characters by one stroke; in contrast, such
effect was not observed in traditional Chinese readers. Perhaps the difference in analytic
processing of shared characters between simplified and traditional Chinese readers is at
the stroke level, and thus the composite paradigm, which examines selective attention to
parts/components, is not sensitive to such difference.
Second, we found that when processing simplified characters, simplified Chinese read-
ers from Mainland China were less holistic than traditional Chinese readers from Hong
Kong and Taiwan, and the difference seemed to be more dependent on their word dicta-
tion performance than naming or copying performances. This finding is consistent with
Tso et al.’s (2011, 2013) studies, which showed a close relationship between writing
experience and reduced holistic processing in Chinese character recognition. More specifi-
cally, they showed that compared with novices, proficient Chinese readers who had lim-
ited writing experience showed increased holistic character processing, whereas proficient
Chinese readers who were also proficient in writing characters showed reduced holistic
processing. Thus, writing, or more specifically dictation, the ability to recall and write
Chinese characters from memory, seems to enhance analytic (reduce holistic) character
processing. This finding suggests a close relationship between perceptual and sensorimo-
tor processing in visual word recognition (see, e.g., James & Atwood, 2009).
Third, although simplified Chinese readers performed worse in writing (dictating) tradi-
tional characters than traditional Chinese readers, their performance in holistic processing
of traditional characters did not differ from traditional Chinese readers. This effect sug-
gests that writing experience might not be the only attribution to reduced holistic process-
ing in Chinese characters. The data reveal that simplified Chinese readers are able to
transfer the analytic processing skills acquired in learning to read and write simplified
characters to the processing of traditional characters, indicating that holistic processing
can be influenced by transfer effect in addition to writing experiences. However, the
transfer is asymmetrical; while simplified Chinese readers can transfer the analytic pro-
cessing skills to the processing of traditional Chinese characters, traditional Chinese read-
ers could not do so in perceiving simplified Chinese characters. This finding is consistent
with the results of McBride-Chang et al.’s (2005) study, which showed that children who
learned to read simplified Chinese characters had better visual skills in matching line
drawings or geometric forms (including Visual Closure, Visual Discrimination, and
Visual Spatial Relationships subtests from Gardner, 1996) than those who learned to read
traditional Chinese characters. Taken together, these findings seem to suggest that simpli-
fied Chinese readers are more able to transfer their analytic processing skills acquired
from reading and writing simplified Chinese characters to unfamiliar/novel stimuli, such
as traditional Chinese characters, or line drawings/geometric forms. Thus, transfer of per-
ceptual expertise effects between two categories is not always symmetrical; it may
depend on the perceptual representation developed as the result of one’s learning experi-
ence of a category. We explore this possibility below.
18 T. Liu et al. / Cognitive Science (2016)
4.2. Visual similarity analysis of the characters in the two scripts
Since writing experience could not account for the asymmetrical transfer effect, this
asymmetric transfer effect may be related to the differences in visual forms between sim-
plified and traditional characters. As discussed in the Introduction, Chen (1999) pointed
out that there are fewer strokes in simplified characters than their traditional counterparts,
and this could make them visually more distinct from each other as compared with tradi-
tional Chinese characters. Zhao and Baldauf (2011) also suggested that simplification of
some of the original traditional script with variations in rules might increase the variance
among the simplified script.
To examine whether the visual form representation of simplified characters indeed has
a larger variance than that of traditional characters, we conducted a visual similarity anal-
ysis on a set of most frequently used characters (Ho & Kwan, 2001). From the set, we
identified 1,492 simplified characters and 3,131 shared characters; then we included the
corresponding 1,492 traditional characters into the set. The set therefore contained the
most frequently used simplified, traditional, and shared characters, forming a representa-
tive Chinese character space; in total there were 6,115 characters. We generated an image
for each character in Ming font (the most common font in print); each image was
40 9 40 pixels in size. To obtain an estimated mental representation of the characters at
an early perceptual stage, we used overlapping 2D Gabor filters to filter the images,
because Gabor filter responses were found to be a good approximation of the neural
responses in the human primary visual cortex V1 (Sanger, 1989). At each pixel there
were Gabor filters of five frequency scales and eight orientations; the five scales corre-
sponded to 2–32 (21–25) cycles per character (see, e.g., Hsiao & Lam, 2013; Hsiao,
Shieh, & Cottrell, 2008). After obtaining the Gabor representation of each character
image, principal component analysis (PCA) was then conducted on this set of images.
PCA is a generic, biologically plausible linear compression technique for dimensionality
reduction (Sanger, 1989), and it has been commonly used to simulate possible informa-
tion extraction processes in the brain in computational models of human cognition (e.g.,
Hsiao & Lam, 2013; Hsiao et al., 2008; Tong, Joyce, & Cottrell, 2008). Fig. 5 shows the
representations of the traditional and simplified characters based on the first two principal
components. (The shared characters were not plotted in order to better illustrate the
difference between the traditional and the simplified characters.)
The PCA plot showed that the representations of the traditional and simplified charac-
ters overlapped with each other, suggesting a high visual similarity between the two. In
addition, characters in the traditional character set (including both traditional and shared
characters) were less spread out than those in the simplified character set (including both
simplified and shared characters): The traditional character set had a smaller variance
(0.0080) as compared with the simplified character set (0.0084); similar results were
obtained when we changed the font from Ming font (the most frequently used font in
print) to Kai (less frequently used in print) (traditional character set: 0.0077; simplified
character set: 0.0081) or Feng fonts (not commonly used in print) (traditional character
set: 0.0082; simplified character st: 0.0083). To quantitatively test the difference between
T. Liu et al. / Cognitive Science (2016) 19
the two sets in the spread of distribution, for each character in a set, we calculated the
average distance between this character and all the other characters from the same set.
Thus, in total we had 4,623 observations (one for each character) from the traditional set
and 4,623 observations from the simplified set. We then compared these distances in the
two character sets using paired t-test. The comparisons were done on Ming, Kai, and
Feng font characters separately. Table 6 summarizes the results. The results showed that
across all three font types, the average distance from each character to all the other char-
acters in the character set was significantly larger in the simplified Chinese character set
than in the traditional Chinese character set. This result was consistent with the PCA plot
(Fig. 5), indicating that the representation of the traditional characters was more con-
densed (i.e., had a smaller variance), whereas that of the simplified characters was more
spread out (i.e., had a larger variance).
To quantify the proportion of the distributions that were overlapped with each other,
for the distribution of each character set, we estimated a Gaussian distribution from the
Table 6
The average distance between each character and all the other characters in the same character set, separately
for the simplified and traditional character set. In all character sets with different font types, the simplified
character set had a significantly larger average distance among characters as compared with the traditional
character set (Paired t-test, MING: t(4622) = 56.85, d = 0.52, CI [.0029, .0031], p < .001; FENG: t(4622) = 27.93, d = 0.69, CI [.0011, .0012], p < .001; KAI: t(4622) = 65.86, d = 0.25, CI [.0032, .0034],
p < .001)
Simplified Chinese Traditional Chinese pMing Font .13 .12 <.001Kai Font .13 .12 <.001Feng Font .13 .12 <.001
Fig. 5. Results of the PCA of 1,175 simplified and 1,175 traditional Chinese characters in Ming font, the
most common font used in prints. The two dimensions represent the first two principal components (i.e., with
the highest eigenvalues, or in other words, the dimensions with the largest variance). The red dots represent
simplified Chinese characters, and the blue dots represent traditional Chinese characters.
20 T. Liu et al. / Cognitive Science (2016)
data, and measured the percentage of overlap between the two Gaussian distributions in
the area within two standard deviations from the mean. More specifically, for the area
within two standard deviations from the mean in the two Gaussian distributions, we cal-
culated the percentage of overlap as the intersection of the two distributions divided by
the union of the two distributions. The union represented the area occupied by either tra-
ditional or simplified character distributions, whereas the intersection represented the area
occupied by both traditional and simplified character distributions. Using this method, we
found that the proportion of overlap between the two distributions was 0.79. Thus, the
two distributions were highly overlapped, with the distribution of traditional characters
having a smaller variance than that of simplified characters.
Fig. 6 shows a similar PCA plot with only the characters used in the current study. It
can be seen that the distributions of the used characters resembled those of the whole
character sets: The two distributions were largely overlapped, with the distribution of the
simplified characters being more spread out than that of the traditional characters.
Thus, due to the overlap between the representations of the simplified and traditional
characters (which suggests high similarity in feature), and a larger variance in the repre-
sentation of the simplified characters, simplified Chinese readers may be more able to
interpolate and generalize their analytic skills developed in reading and writing simplified
characters to reading traditional characters, as compared with traditional Chinese readers
in reading simplified characters (cf. Tong et al., 2008). The results of our visual similarity
analysis supported the speculation that simplification of Chinese characters made them
physically more distinctive than their traditional forms (Chen, 1999). The advantage of
having a visually more distinctive (or in other words, with a larger variance) set of char-
acters in one’s mental lexicon seems to be in the ease of transferring analytic processing
skills to unfamiliar characters with similar structures, as reflected in the generalization
abilities observed in simplified Chinese readers.
Fig. 6. Results of the PCA of 160 simplified and 160 traditional Chinese characters used in the current study.
The red dots represent simplified Chinese characters, and the blue dots represent traditional Chinese characters.
T. Liu et al. / Cognitive Science (2016) 21
Tanaka, Curran, and Sheinberg (2005) trained participants to recognize different types of
birds and showed that training at the subordinate (species) level better facilitates perceptual
expertise transfer to novel exemplars and novel categories than training at the categorical
(family) level. Through computational modeling, Tong et al. (2008) further showed that the
advantage of training at the subordinate level in perceptual expertise transfer lies in the
magnification of differences between exemplars within a category (i.e., a larger variance) in
one’s perceptual representation as the result of the subordinate-level training, making it
easier to discriminate novel exemplars or exemplars in novel categories. Although in Tong
et al.’s (2008) simulation, the difference in variance of perceptual representation was
observed at the hidden layer level of a neural network (i.e., an intermediate perceptual rep-
resentation), whereas in the current study, our analysis was performed at an earlier percep-
tual level based on Gabor filter responses; both cases demonstrate the advantage of a more
spread-out perceptual representation in the transfer of perceptual expertise. This effect sug-
gests that in perceptual expertise acquisition, different learning experiences, such as learn-
ing to recognize objects at the subordinate level versus the categorical level, or learning to
read simplified versus traditional Chinese characters, can lead to different transfer effects
due to differences in the perceptual representation developed as the result of one’s learning
experience; the type of learning experience that enhances exemplar variance in the presenta-
tion seems to facilitate transfer of perceptual expertise effects.
4.3. Other possible explanations of the transfer effect
Recent research has suggested an advantage of training with simpler forms, as opposed
to complex forms, in the transfer effect to new situations in relational learning. This find-
ing might also be related to the asymmetric generalization effect observed in the process-
ing of simplified and traditional Chinese characters. For example, research done in
mathematics and science has revealed that training with simple instances yields better
transfer effects to new situations than training with more complex exemplars (e.g., Gold-
stone & Sakamoto, 2003; Goldstone & Son, 2005; Kaminski, Sloutsky, & Heckler, 2008).
This might be because simple exemplars enable selective attention to the right informa-
tion essential for generalization of knowledge rather than to the irrelevant details in the
complex exemplars. As discussed in the Introduction, the simplification process of Chi-
nese characters involved trimming down some strokes in the components or removing
some components within characters, aiming to ease the recognition and production of the
characters. This simplification process may have implicitly preserved parts of the tradi-
tional characters that are essential for recognition/discrimination while removing less
informative parts. In addition, since simplified characters typically preserve part of their
traditional counterparts, when a simplified Chinese reader attempts to read a traditional
character, a common strategy may be to look for familiar components or parts of the
components within the traditional characters. This strategy may encourage the engage-
ment of local attention, leading to a similar level of analytic processing as traditional Chi-
nese readers. In contrast, when a traditional Chinese reader reads a simplified character,
he or she may take the holistic pattern of the simplified character and try to see whether
22 T. Liu et al. / Cognitive Science (2016)
it matches part of a traditional character, leading to an increased holistic processing effect
as compared with simplified Chinese readers. This asymmetric generalization effect of
reduced holistic processing between traditional and simplified Chinese readers is consis-
tent with a recent finding by Thai and Son (2013). They found that training novices with
simplified Chinese characters produced better generalization to the recognition of the cor-
responding characters in the other script than training with traditional characters. The
ways that simplified characters were created might be related to this asymmetrical gener-
alization effect. Future work will examine this possibility.
In a recent study, McBride-Chang et al. (2011) found that literacy experiences may
influence visual spatial skills, and that literacy experiences with different orthographies
may shape different aspects of visual spatial skills. This finding is consistent with the cur-
rent results, which showed that simplified and traditional Chinese readers had different
transfer effects in holistic/analytic processing of characters in the two Chinese written
systems due to the difference in their experience with the orthographies. McBride-Chang
et al. (2011) also suggested a bidirectional association between visual-spatial skills and
the process of learning to read (see also Zhou, McBride-Chang, & Wong, 2014). Accord-
ingly, it is possible that a better ability to transfer analytic processing of characters in a
script to a new script with similar character structures can facilitate learning to read in
the new script. This possibility requires further investigation.
In conclusion, here we show that reduced holistic processing of Chinese characters can
be influenced by transfer effect, in addition to sensorimotor experience as reported previ-
ously (Tso et al., 2014). Expertise in reading and writing simplified Chinese characters
seems to facilitate readers to transfer their analytic processing skills to the processing of
traditional Chinese characters, as compared with traditional Chinese readers in processing
simplified characters. This asymmetric transfer effect between the two groups of readers
may be related to a larger variance in the visual forms of simplified characters than those
of traditional characters. This finding suggests that transfer of perceptual expertise effects
in visual recognition may be constrained by both the similarity in feature and the differ-
ence in exemplar variance between visual categories. It also has important implications
for teaching and learning in perceptual expertise acquisition. For example, a learning pro-
gram that implicitly enhances exemplar variance in one’s perceptual representation, such
as the subordinate-level training used in Tanaka et al. (2005), or with explicit instructions
to direct learners’ attention to largest variations among exemplars, may facilitate transfer
of perceptual expertise effects to novel exemplars or stimuli with similar features in a
novel category.
Acknowledgments
We are grateful to the Research Grant Council of Hong Kong (project #HKU 745210H
and #HKU 758412H to J.H. Hsiao). We thank Nick Wang for his suggestions about com-
paring simplified and traditional Chinese reading. We thank the Editor and three anony-
mous reviewers for helpful comments.
T. Liu et al. / Cognitive Science (2016) 23
Notes
1. Wong et al. (2012) showed that Chinese readers (experts) had a stronger holistic
processing effect in processing real Chinese characters than noncharacters (i.e.,
nonexisting characters consisting of real Chinese character components in illegal
positions), whereas non-Chinese readers (novices) did not show this effect; this
suggests that holistic processing in Chinese character recognition may also be influ-
enced by familiarity with the component positions.
2. The two groups of readers did not differ significantly in accuracy in naming
(F(1, 46) = 3.61, p = .07), copying (F(1, 46) = .11, p = .74), or dictating the
shared Chinese characters (F(1, 46) = 3.29, p = .08). However, Mainland China
simplified Chinese readers were faster than Hong Kong traditional Chinese readers
in naming the shared characters (F(1, 46) = 4.83, p < .05). This difference may be
due to the pronunciation differences between Cantonese (spoken by Hong Kong
participants) and Mandarin (spoken by Mainland China participants). It may also
be related to greater confusion in pronunciation in Hong Kong participants due to
their knowledge of both Cantonese and Mandarin pronunciations of the characters
as Mandarin has become more popular in Hong Kong than before. Note that two
shared characters were removed from analysis based on the item analysis of partici-
pants’ reading accuracy of each character, since the average accuracy of the two
characters were below 2 standard deviations from the mean accuracy of Mainland
China and Hong Kong participants’ performance (0.99).
3. When we compared the two groups, Mainland China participants were more accu-
rate (F(1, 46) = 18.76, p < .001) and faster (F(1, 46) = 13.77, p < .01) in naming
simplified Chinese (one outlier was removed from the analysis according to the item
analysis of participants’ reading accuracy of each character), whereas Hong Kong
participants were more accurate (F(1, 46) = 11,31, p < .01), but not significantly
faster (F(1, 46) =.73, p = .40) in naming traditional Chinese characters. Note, how-
ever, this effect was due to a main effect of group in naming RT of all three types of
characters (F(1, 46) = 6.22, p < .05): Mainland China participants in general had
shorter naming RT than Hong Kong participants (see note 2, for reference). When
we examined their naming performance within each participant group, Mainland
China simplified Chinese readers were more accurate (t(23) = 4.04, p < .01) and
faster (t(23) = 3.57, p < .01) in naming simplified than traditional Chinese charac-
ters, whereas Hong Kong participants were faster (t(23) = 4.25, p < .001) and more
accurate (t(23) = 4.93, p < .001) in naming traditional than simplified Chinese char-
acters. Thus, both groups of participants were more proficient in reading the script
they are more familiar with. In the writing tasks, Mainland China participants were
more accurate in copying (F(1, 46) = 12.83, p < .01) and dictating (F(1,46) = 37.09, p < .001) simplified Chinese characters than Hong Kong participants,
while Hong Kong participants were more accurate in copying (F(1, 46) = 91.75,
p < .001) and dictating (F(1, 46) = 442.29, p < .001) traditional Chinese characters
24 T. Liu et al. / Cognitive Science (2016)
than Mainland China participants. Thus, both groups of participants were more profi-
cient in writing the script they are more familiar with.
4. Mainland China simplified Chinese readers and Taiwan traditional Chinese readers
did not differ in accuracy in naming (F(1, 46) = 3.85, p = .06) or dictating shared
characters (F(1, 46) = 3.43, p = .07). However, in the copying task, Taiwan partic-
ipants performed slightly but significantly better than Mainland China participants
(F(1, 46) = 6.05, p < .05; mean accuracy: Mainland China = 96%, Tai-
wan = 98%). Similar to the results in Experiment 1, Mainland China simplified
Chinese reader were faster than Taiwan traditional Chinese readers in naming
shared Chinese characters (F(1, 46) = 13.13, p < .01). This effect might be attribu-
ted to the widely used Minnan dialect in Taiwan in addition to Mandarin, which
may lead to greater confusion in pronunciation in Taiwan participants because of
their knowledge of both the Minnan and Mandarin pronunciations of the characters.
Note that three shared characters were removed from the analysis based on the item
analysis of participants’ reading accuracy of each character, since the average accu-
racy of the three characters was below two standard deviations from the mean
accuracy of Mainland China and Taiwan participants’ naming performance (0.99).
5. In between group comparisons, Mainland China participants were more accurate (F(1,46) = 34.24, p < .001) and faster (F(1, 46) = 42.84, p < .001) than Taiwan partici-
pants in naming simplified Chinese characters. In contrast Taiwan participants were
more accurate (F(1, 46) = 4.92, p < .05), but slower (F(1, 46) = 42.84, p < .001) in
naming traditional Chinese characters (two characters were identified as outliers and
removed from the analysis according to the item analysis of participants’ reading
accuracy of each character). Note, however, that there was a main effect of group in
naming RT of all three types of characters (F(1, 46) = 13.65, p < .01): Mainland
China participants in general named characters faster than Taiwan participants (see
note 10 for reference). When we examined their naming RT within each group sepa-
rately, Mainland China participants named simplified characters more accurately (t(23) = 4.16, p < .001) and faster (t(23) = 3.61, p < .01) than traditional charac-
ters, and Taiwan participants were more accurate (t(23) = 4.79, p < .001) and faster
(t(23) = 3.60, p < .01) in naming traditional than simplified characters. In the writing
tasks, Mainland China participants were more accurate in copying (F(1, 46) = 21.44,
p < .001) and dictating (F(1, 46) = 421.34, p < .001) simplified Chinese characters
than Taiwan participants, while Taiwan participants were more accurate in copying
(F(1, 46) = 133.62, p < .001) and dictating (F(1, 46) = 443.29, p < .001) traditional
Chinese characters than Mainland China participants. Thus, both groups of partici-
pants were more proficient in writing the script they are more familiar with.
References
Busey, T. A., & Vanderkolk, J. R. (2005). Behavioral and electrophysiological evidence for configural
processing in fingerprint experts. Vision Research, 45, 431–448.
T. Liu et al. / Cognitive Science (2016) 25
Chen, P. (1999). Modern Chinese: History and sociolinguistics. Cambridge, UK: Cambridge University Press.
Chen, Y. P., Allport, D. A., & Marshall, J. C. (1996). What are the functional orthographic units in Chinese
word recognition: The stroke or the stroke pattern? Quarterly Journal of Experimental PsychologySection A: Human Experimental Psychology, 49, 1024–1043.
Cheung, H., & Ng, L. (2003). Chapter 1: Chinese reading development in some major Chinese societies: An
introduction. In C. McBride-Chang & H. C. Chen (Eds.), Reading development in Chinese children(pp. 3–17). Westport, CT: Praeger.
Gao, D.-G., & Kao, H. S. R. (2002). Psycho-geometric analysis of commonly used Chinese characters. In H.
S. R. Kao, C.-K. Leong, & D.-G. Gao (Eds.), Cognitive neuroscience studies of the Chinese language (pp.
195–206). Hong Kong: Hong Kong University Press.
Gardner, M. F. (1996). Test of visual-perceptual skills (Non-motor): Revised manual. Hydesville, CA:
Psychological and Educational Publications.
Gauthier, I., & Bukach, C. (2007). Should we reject the expertise hypothesis? Cognition, 103, 322–330.Gauthier, I., Curran, T., Curby, K. M., & Collins, D. (2003). Perceptual interference supports a non-modular
account of face processing. Nature Neuroscience, 6, 428–432.Gauthier, I., & Tanaka, J. W. (2002). Configural and holistic face processing: The whole story. Journal of
Vision, 2(7), 616.Gauthier, I., & Tarr, M. J. (1997). Becoming a “Greeble” expert: Exploring the face recognition mechanism.
Vision Research, 37, 1673–1682.Ge, L., Wang, Z., McCleery, J. P., & Lee, K. (2006). Activation of face expertise and the inversion effect.
Psychological Science, 17, 12–16.Goldstone, R. L., & Sakamoto, Y. (2003). The transfer of abstract principles governing complex adaptive
systems. Cognitive Psychology, 46, 414–466.Goldstone, R. L., & Son, J. Y. (2005). The transfer of scientific principles using concrete and idealized
simulations. Journal of the Learning Sciences, 14, 69–114.Ho, H.-H., & Kwan, T. W. (2001). Hong Kong, Mainland China & Taiwan: Chinese character frequency—A
trans-regional, diachronic survey. Available at http://arts.cuhk.edu.hk/Lexis/chifreq/. Accessed October 30,
2011.
Ho, C. S. H., Ng, T. T., & Ng, W. K. (2003). A “radical” approach to reading development in Chinese: The
role of semantic radicals and phonetic radicals. Journal of Literacy Research, 35(3), 849–878.Hodge, B., & Louie, K. (1998). The politics of Chinese language and culture (pp. 63–64). London: Routledge.Hole, G. J. (1994). Configurational factors in the perception of unfamiliar faces. Perception, 23, 65–74.Hsiao, J. H., & Cottrell, G. W. (2009). Not all visual expertise is holistic, but it may be leftist: The case of
Chinese character recognition. Psychological Science, 20, 455–463.Hsiao, J. H., & Lam, S. M. (2013). The modulation of visual and task characteristics of a writing system on
hemispheric lateralization in visual word recognition—A computational exploration. Cognitive Science,37, 861–890.
Hsiao, J. H., Shieh, D., & Cottrell, G. W. (2008). Convergence of the visual field split: Hemispheric
modeling of face and object recognition. Journal of Cognitive Neuroscience, 20, 2298–2307.Hsiao, J. H., & Shillcock, R. (2006). Analysis of a Chinese phonetic compound database: Implications for
orthographic processing. Journal of Psycholinguistic Research, 35, 405–426.James, K. H., & Atwood, T. P. (2009). The role of sensorimotor learning in the perception of letter-like
forms: Tracking the causes of neural specialization for letters. Cognitive Neuropsychology, 26, 91–110.Kaminski, J. A., Sloutsky, V. M., & Heckler, A. F. (2008). The advantage of abstract examples in learning
math. Science, 320, 454–455.Maurer, D., Le Grand, R., & Mondloch, C. J. (2002). The many faces of configural processing. Trends in
Cognitive Sciences, 6, 255–260.McBride-Chang, C., Chow, B. W. Y., Zhong, Y., Burgess, S., & Hayward, W. (2005). Chinese character
acquisition and visual skills in two Chinese scripts. Reading and Writing, 18, 99–128.
26 T. Liu et al. / Cognitive Science (2016)
McBride-Chang, C., Zhou, Y., Cho, J.-R., Aram, D., Levin, I., & Tolchinsky, L. (2011). Visual spatial skill:
A consequence of learning to read. Journal of Experimental Child Psychology, 109, 256–262.McCleery, J. P., Zhang, L., Ge, L., Wang, Z., Christiansen, E. M., Lee, K., & Cottrell, G. W. (2008). The
roles of visual expertise and visual input in the face inversion effect: Behavioral and neurocomputational
evidence. Vision Research, 48, 703–715.Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory.
Neuropsychologia, 9, 97–113.Peng, G., Minett, J. W., & Wang, W. S.-Y. (2010). Cultural background influences the liminal perception of
Chinese characters: An ERP study. Journal of Psycholinguistics, 23, 416–426.Richler, J. J., Tanaka, J. W., Brown, D. D., & Gauthier, I. (2008). Why does selective attention to parts fail
in face processing? Journal of Experimental Psychology: Learning, Memory, & Cognition, 34, 1356–1368.Sanger, T. (1989). An optimality principle for unsupervised learning. In D. Touretzky (Ed.), Advances in
neural information processing systems (Vol. 1, pp. 11–19). San Mateo, CA: Morgan Kaufmann.
Seybolt, P. J., & Chiang, G. K.-K. (1979). Introduction. In P. J. Seybolt & G. K.-K. Chiang (Eds.), Languagereform in China: Documents and commentary (pp. 1–10). Oxford: Routledge.
Stanislaw, H., & Todorov, N. (1999). Calculation of signal detection theory measures. Behavior ResearchMethods, Instruments & Computers, 31(1), 137–149.
Taiwan Ministry of Education. (1997). Investigation report on the commonly used words in 1997. Available
at http://www.edu.tw/files/site_content/M0001/86news/ch2.html?open. Accessed October 30, 2011.
Tan, L. H., & Perfetti, C. A. (1998). Phonological codes as early sources of constraint in reading Chinese: A
review of current discoveries and theoretical accounts. Reading and Writing: An Interdisciplinary Journal,10, 165–220.
Tanaka, J. W., Curran, T., & Sheinberg, D. L. (2005). The training and transfer of real-world perceptual
expertise. Psychological Science, 16(2), 145–151.Tanaka, J., & Farah, M. (1993). Parts and wholes in face recognition. The Quarterly Journal of Experimental
Psychology, 46, 225–245.Thai, K. P., & Son, J. (2013). The simple advantage in categorical generalization of Chinese characters. In
M. Knauff, M. Pauen, N. Sebanz, & I. Wachsmuth (Eds.), Proceedings of the 35th Annual Conference ofthe Cognitive Science Society (pp. 3527–3532). Austin, TX: Cognitive Science Society.
Tong, M. H., Joyce, C. A., & Cottrell, G. W. (2008). Why is the fusiform face area recruited for novel
categories of expertise? A neurocomputational investigation. Brain Research, 1202, 14–24.Tso, R. V. Y., Au, T. K., & Hsiao, J. H. (2011). The influence of writing experiences on holistic processing
in Chinese character recognition. In L. Carlson, C. Hoelscher, & T. F. Shipley (Eds.), Proceedings of the33rd Annual Conference of the Cognitive Science Society (pp. 1442–1447). Austin, TX: Cognitive Science
Society.
Tso, R. V. Y., Au, T. K., & Hsiao, J. H. (2012). Writing facilitates learning to read in Chinese through
reduction of holistic processing: A developmental study. In N. Miyake, D. Peebles, & R. P. Cooper
(Eds.), Proceedings of the 34th Annual Conference of the Cognitive Science Society (pp. 2463–2468).Austin, TX: Cognitive Science Society.
Tso, R. V. Y., Au, T. K., & Hsiao, J. H. (2013). Expert marker of Chinese character recognition: Left-side
bias versus holistic processing? In M. Knauff, M. Pauen, N. Sebanz, & I. Wachsmuth (Eds.), Proceedingsof the 35th Annual Conference of the Cognitive Science Society (pp. 1492–1497). Austin, TX: CognitiveScience Society.
Tso, R. V. Y., Au, T. K., & Hsiao, J. H. (2014). Perceptual expertise: Can sensorimotor experience change
holistic processing and left side bias? Psychological Science, 25(9), 1757–1767.Wong, A. C.-N., Bukach, C. M., Hsiao, J., Greenspon, E., Ahern, E., Duan, Y., & Lui, K. F. H. (2012).
Holistic processing as a hallmark of perceptual expertise for nonface categories including Chinese
characters. Journal of Vision, 12(13), 7, 1–15.Yeh, S. L., & Li, J. L. (2002). Role of structure and component in judgments of visual similarity of Chinese
characters. Journal of Experimental Psychology: Human Perception and Performance, 28(4), 933–947.
T. Liu et al. / Cognitive Science (2016) 27
Young, A. W., Hellawell, D., & Hay, D. C. (1987). Configurational information in face perception.
Perception, 16, 747–759.Zhao, S., & Baldauf, R. B. (2011). Chapter 14, Simplifying Chinese characters: Not a simple matter. In J.
Fishman & O. Garcia (Eds.), Handbook of language and ethnic identity: The success-failure continuum inlanguage and ethnic identity efforts (Vol. 2, pp. 154–168). New York: Oxford University Press.
Zhou, Y.-L., McBride-Chang, C., & Wong, N. (2014). What is the role of visual skills in learning to read?
Frontiers in Psychology, 5, 776–776.
28 T. Liu et al. / Cognitive Science (2016)