Neural division of labor in reading is constrained by culture: A training study of reading Chinese...
Transcript of Neural division of labor in reading is constrained by culture: A training study of reading Chinese...
This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies areencouraged to visit:
http://www.elsevier.com/authorsrights
Author's personal copy
Research report
Neural division of labor in reading is constrainedby culture: A training study of reading Chinesecharacters
Jingjing Zhao a,b,*, Xiaoyi Wang c, Stephen J. Frost b, Wan Sun c,Shin-Yi Fang a,b, W. Einar Mencl b, Kenneth R. Pugh a,b,d,e, Hua Shu c,1 andJay G. Rueckl a,b,1
aDepartment of Psychology, University of Connecticut, Storrs, CT, USAbHaskins Laboratories, New Haven, CT, USAcState Key Laboratory for Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, ChinadDepartment of Linguistics, Yale University, New Haven, CT, USAeDepartment of Diagnostic Radiology, Yale University School of Medicine, New Haven, CT, USA
a r t i c l e i n f o
Article history:
Received 27 February 2013
Reviewed 13 August 2013
Revised 11 November 2013
Accepted 8 January 2014
Action editor Roberto Cubelli
Published online 15 January 2014
Keywords:
Reading
Chinese
Division of labor
fMRI
Learning
a b s t r a c t
Word reading in alphabetic language involves a cortical system with multiple components
whose division of labor depends on the transparency of the writing system. To gain insight
about the neural division of labor between phonology and semantics subserving word
reading in Chinese, a deep non-alphabetic writing system, functional magnetic resonance
imaging (fMRI) was used to investigate the effects of phonological and semantic training on
the cortical circuitry for oral naming of Chinese characters. In a training study, we
examined whether a training task that differentially focused readers’ attention on the
phonological or semantic properties of a Chinese character changes the patterns of cortical
activation that was evoked by that character in a subsequent naming task. Our imaging
results corroborate that the cortical regions underlying reading in Chinese largely
overlap the left-hemisphere reading system responsible for reading in alphabetic lan-
guages, with some cortical regions in the left-hemisphere uniquely recruited for reading in
Chinese. However, in contrast to findings from studies of English word naming, we
observed considerable overlap in the neural activation patterns associated with phono-
logical and semantic training on naming Chinese characters, which we suggest may reflect
a balanced neural division of labor between phonology and semantics in Chinese character
reading. The equitable division of labor for Chinese reading might be driven by the special
statistical structure of the writing system, which includes equally systematic mappings in
the correspondences between written forms and their pronunciations and meanings.
ª 2014 Elsevier Ltd. All rights reserved.
* Corresponding author. Laboratoire de Sciences Cognitives et Psycholinguistique, Departement d’Etudes Cognitives, Ecole NormaleSuperieure, 29 rue d’Ulm, 75005 Paris Cedex 05, France.
E-mail addresses: [email protected] (J. Zhao), [email protected] (H. Shu), [email protected] (J.G. Rueckl).1 Address correspondence also to Jay G. Rueckl, Department of Psychology, University of Connecticut, Storrs, CT 06269, USA, and Hua
Shu, State Key Laboratory for Cognitive Neuroscience and Learning, Beijing Normal University, Beijing 100875, China.
Available online at www.sciencedirect.com
ScienceDirect
Journal homepage: www.elsevier.com/locate/cortex
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6
0010-9452/$ e see front matter ª 2014 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.cortex.2014.01.003
Author's personal copy
1. Introduction
Extant evidence indicates that in skilled readers of alphabetic
writing systems, word reading involves a highly organized
and specialized multi-component cortical system distributed
primarily in left-hemisphere (LH) language areas [inferior
frontal gyrus, superior, middle and inferior temporal gyri,
superior and inferior parietal lobule, and fusiform gyrus
(hereafter, IFG, STG, MTG, ITG, SPL, IPL, and FG, respectively)].
These cortical regions are differentiated by their contribution
to orthographic, phonological, and lexical-semantic process-
ing (Cattinelli, Borghese, Gallucci, & Paulesu, 2013; Pugh et al.,
2000; Taylor, Rastle, & Davis, 2013). The relative contribution
of these cortical regions to printed word recognition also ap-
pears to vary systematically as a function of the properties of
the writing system. For example, Paulesu et al. (2000, 2001)
found that readers of Italian, an orthography with a trans-
parent or largely univalent mapping between letters and
phonemes, show greater activation in left STG (phonological
processing areas) than English readers. In contrast, readers of
English, an opaque orthography with multivalent mappings
between letters and phonemes (e.g., ‘I’ is pronounced differ-
ently in ‘PINT’ and ‘MINT’), showed greater activations in the
left posterior ITG and anterior IFG (lexical-semantic process-
ing areas) than Italian readers. In other words, within a
common reading network, the division of labor among its
component processes is differently weighted depending on
specific characteristics of the orthography (Seidenberg, 1992,
2011).
Although alphabetic systems vary in the structure of the
mapping from spelling to sound, differences in this mapping
are particularly pronounced in the contrast between alpha-
betic writing systems and Chinese, in which the mapping
from spelling to sound is syllable-based with no constituent
parts of a character corresponding to phonemes. In addition,
the statistical structure of the mapping from spelling to
meaning also differs substantially between Chinese and
alphabetic languages. Chinese, as one of the oldest writing
systems in the world, is commonly described as an ‘ideo-
graphic’ or morphosyllabic writing system with an extremely
deep orthography. It is true that logographic characters, as the
basic units of Chinese, are typically corresponded to mor-
phemes. In fact, however, only a small percentage of Chinese
characters (those that are most ancient, dating back more
than 3000 years) are aptly termed ideographs (DeFrancis,
1989). A large percentage (80e90%) of modern Chinese char-
acters are “phonograms”, semanticephonetic compounds
with one element (phonetic radical) suggesting its
pronunciation and the other element (semantic radical) indi-
cating the general category of its meaning, e.g.,湖 (/hu2/, lake)
which contains a phonetic radical胡 pronounced as /hu2/ and
a semantic radical 氵meaning water. For example, Chinese
regulareconsistent phonograms1 have exactly (39%) or
approximately the same (26%) pronunciations as their pho-
netic radicals as evaluated by Shu, Chen, Anderson, Wu, and
Xuan (2003) from a total of 2570 Chinese characters taught
in Chinese elementary school. Similarly, the meanings of a
large percentage (88%) of Chinese phonograms are trans-
parently (58%) or semi-transparently (30%) related to the
meanings of their semantic radicals.2 Thus, the structure of
Chinese phonograms is not “absolute-ideographic”, but in-
cludes substantial regularities (although not perfectly pre-
dictable) in the correspondences between written forms and
both their pronunciations and their meanings.
In other words, although the differences between Chinese
and alphabetic writing systems are illustrated remarkably in
the structure of written forms (logographs vs alphabets),
substantial differences between the two writing systems can
also be understood in terms of the statistical properties of
orthography-to-phonology (OeP) and orthography-to-
semantics (OeS) mappings. The statistical structure of the
Chinese writing system and that of alphabetic systems might
differ in two important ways. On the one hand, the OeP
mapping is less systematic in Chinese than in alphabetic
systems. In alphabetic systems, an alphabet of letters can
correspond to individual speech sounds, although English is
somewhat an “outlier” in alphabetic systems but letters or
combinations of letters in English still roughly correspond to
phonemes or combinations of phonemes. In contrast, in
Chinese, although phonetic radicals can provide cues for the
pronunciations of the characters, phonetic radicals are also
logographs, per se, and the pronunciations of phonetic radi-
cals correspond to syllables, the global phonological units for
the pronunciations of the characters, not the constituent parts
of the syllables. Thus, computation of pronunciation of a
Chinese character is not a process of sound-by-sound
assembling as in alphabetic systems in essence, but is a pro-
cess of addressed direct access from logographic forms to
phonology in syllables. In addition, although about two-thirds
of the phonograms have the same or approximately the same
pronunciations as their phonetic radicals, this is far from
consistent as in alphabetic systems. All together, the relations
between orthography and phonology in Chinese are more
arbitrary than in alphabetic scripts. One the other hand, the
OeS mapping is more systematic in Chinese than in alpha-
betic systems. Other than the morphosyllabic characteristics
of simple Chinese characters, semantic radicals in Chinese
phonograms indicate general semantic categories of the
characters and aid in the computation from orthography to
semantics. In contrast, alphabetic systems rarely contain se-
mantic information in the way that Chinese do by grouping
characters into different semantic categories. Although there
1 A regulareconsistent (ReC) phonogram has the same pro-nunciation to its phonetic radical and all other phonogramscontaining the same phonetic radical. An irregulareinconsistent(IReIC) phonogram has different pronunciation to its phoneticradical and other phonograms containing the same phoneticradical. Naming latency and accuracy were found longer and lessaccurate for naming IReIC phonograms than for ReC phono-grams, which termed as regularityeconsistency effects (e.g., Lee,Tsai, Su, Tzeng, & Hung, 2005), similar to the regular-ityeconsistency effects found in reading alphabetic languagessuch as English (e.g., Jared, 2002).
2 Phonograms also vary in semantic transparencydthe degreeto which the meaning of a phonogram is related to the ‘core’meaning of the radical. The more semantic information a se-mantic radical contributes to the meaning of the phonogram thatcontains it, the more transparent is the phonogram.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 91
Author's personal copy
are morphological words in English (e.g., final -s and -ed in
English), they are somewhat encoded phonologically too and
hence not orthogonal to orthography-to-semantics mapping.
To the end, the correspondence between orthography and
semantics in Chinese is certainly less arbitrary than in
alphabetic systems. In such away, the Chinesewriting system
conveys a relatively symmetric statistical structure between
OeP mapping and OeS mapping comparing to that of the
alphabetic systems, which is substantially asymmetrical be-
tween the twomappings with highly consistent OeP mapping
and arbitrary OeS mapping.
The relatively symmetric statistical structure of thewriting
system of Chinese compared to the alphabetic systems im-
plies that the division of labor between phonology and se-
mantics may be different for readers of Chinese than for
readers of alphabetic systems, with readers of Chinese relying
more on semantics than alphabetic readers (Harm &
Seidenberg, 2004; Seidenberg, 2011; also see Frost, 2012). For
example, the arbitrariness in spelling-to-sound mapping of
Chinese characters may make the computation from orthog-
raphy to phonology much less efficient in Chinese and the
consistency in spelling-to-meaning mapping of Chinese
characters may make the computation from orthography to
semantics more efficient in Chinese compared to alphabetic
systems. Thus, semantics should contribute more in naming
(a typical task for examining computation from orthography
to phonology) Chinese characters than in naming alphabetic
words. In fact, extant behavioral work has supported this
hypothesis and shown that shallow alphabetic orthographies
(e.g., Italian) demonstrated negligible effects of semantic
variables (e.g., imageability and concreteness) in skilled
reading (Barca, Burani, & Arduino, 2002; Bates, Burani,
D’Amico, & Barca, 2001) and only in deep orthographies such
as English, semantic variables play some role in naming aloud
words with atypical pronunciations especially if those words
are relatively infrequent (Shibahara, Zorzi, Hill, Wydell, &
Butterworth, 2003; Strain, Patterson, & Seidenberg, 1995,
2002; Woollams, 2005), a finding that is also borne out by
computational modeling (Harm & Seidenberg, 2004; Plaut,
McClelland, Seidenberg, & Patterson, 1996) and neuro-
imaging (Frost et al., 2005; Graves, Desai, Humphries,
Seidenberg, & Binder, 2010) measures.
In contrast, a large corpus analysis of the naming responses
of 2423 Chinese characters indicates that semantic variables
seem to play substantial role in reading aloud all Chinese
characters regardless of types (Liu, Shu, & Li, 2007). Yang, Shu,
McCandliss, and Zevin (2012) further provided quantitative
evidence for semantic contribution in naming different types of
Chinese characters in their connectionist modeling work. They
specifically compared semantic and phonological effects in
naming different types of characters via impairments of se-
mantic and phonological units of the naming model. The au-
thors found that semantics not only played substantial role in
naming Chinese irregulareinconsistent (IReIC) phonograms
but also made as much contribution as phonology in naming
regulareconsistent (ReC) phonograms. They additionallymade
direct comparison of semantic and phonological effects in
naming Chinese phonograms with those of naming English
words using the same model architecture and training pro-
cedures. The naming results between the two writing systems
contrast dramatically. For ReC words, semantics showed
relatively little effects in English reading, but made substantial
effects as phonology in naming Chinese ReC phonograms. For
IReICwords, semantics showed some effects in English reading
but the effects were weaker than phonology and much more
modest than the effects in naming Chinese IReIC phonograms.
In summary, previous studies coherently suggest that se-
mantics plays limited effects in alphabetic systems and only
contributes in naming IReIC words in relatively deeper
alphabetic orthography such as English, whereas in the
reading of a non-alphabetic script such as Chinese, semantics
plays important role in naming all types of Chinese charac-
ters, including both IReIC phonograms and ReC phonograms.
Thus, investigating reading of Chinese, especially for ReC
phonograms,might offer an important approach to examine if
the division of labor between phonology and semantics in
reading Chinese characters is different than in reading English
words. Although extant behavioral and computational
modeling findings have examined this issue and suggest dif-
ferences in division of labor in reading between the two
writing systems, few studies have explored this issue in the
neural level.
The purpose of the present study was to provide neuro-
imaging evidence for the division of labor between phonology
and semantics for Chinese readers. We used an adaptive
learning paradigm developed by Sandak et al. (2004) that has
been shown to provide a sensitive measure of the division of
labor between the neural circuits underlying OeP and OeS
processes in readers of English. By using a paradigm and
scanning parameters previously used in this study of reading
in English, the results of the present studywould also allow us
to compare the division of labor of Chinese reading with the
findings in English reading. In the present Chinese adaptive
learning experiment, native Mandarin speakers were first
trained on Chinese low-frequency phonograms with regu-
lareconsistent pronunciations and transparent semantic
radicals outside the functional magnetic resonance imaging
(fMRI) scanner, half in a training condition that emphasized
attention to phonological form and half in a training condition
that emphasized semantics. Immediately following training,
participants named aloud the trained phonograms and a
comparable set of untrained phonograms in an event-related
fMRI paradigm. This design allowed us to observe how the
neural activation associated with naming a phonogram is
modulated by recent prior exposures to that phonogram by
contrasting the response to trained and untrained phono-
gram. Critically, the training-task manipulation allowed us to
further determine whether the effects of prior exposure differ
depending on the focus (phonological or semantic) of the
training task.
2. Methods
2.1. Participants
A total of 18 right-handed native Mandarin speakers (seven
males and 11 females) participated in the fMRI study after
giving informed consent. The subjects ranged in age from 19
to 27 years with a mean of 22.8 years. All participants were
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 692
Author's personal copy
undergraduate or graduate students from Beijing Normal
University. None of the subjects reported a history of neuro-
logical disorders or reading impairments. All experimental
protocols were approved by State Key Laboratory of Cognitive
Neuroscience and Learning at Beijing Normal University.
Becausewewere not able to obtain behavioral naming data
during fMRI scanning, we further recruited a separate cohort
(hereafter, the behavioral cohort) of 37 right-handed partici-
pants (13 males and 24 females, 19e28 years of age with a
mean of 22.1 years) from undergraduate and graduate stu-
dents of Beijing Normal University in order to collect naming
latencies and accuracy outside scanner. The participants were
all native Mandarin speakers. None of the participants re-
ported a history of neurological disorders or reading impair-
ments. Training was identical for this cohort; however,
instead of receiving an fMRI scan we collected their naming
latencies and accuracy for the Chinese phonograms outside
scanner.
2.2. Materials
The critical items in our adaptive learning paradigm included
a total of 126 Chinese low-frequency phonograms that ranged
in frequency from 1 to 30 occurrences per million (mean
(M) ¼ 10.99) according to the Modern Chinese frequency dic-
tionary (Wang, 1986). All critical phonograms were semantic-
transparent regulareconsistent (ReC) phonograms, whose con-
sistency ratios3 ranged from .5 to 1 and averaged .91 (Liu, Shu,
et al., 2007). For regularity of pronunciation of a phonogram,
tone difference between the pronunciation of a phonogram
and the pronunciation of its phonetic radical was not
controlled due to the limited number of possible stimuli, but
each phonogram and its phonetic radical shared the same
onset and rime. All phonograms were semantically trans-
parent with explicit meaning cues from their semantic
radicals.2
For counterbalancing purposes, the critical phonograms
were partitioned into three lists of 42 items each. The three
lists were matched on frequency (M ¼ 10.78, 11.74, 10.46,
standard deviation (SD) ¼ 7.24, 8.99, 7.30), consistency ratio
(M ¼ .92, .90, .90, SD ¼ .15, .14, .13), number of strokes (M ¼ 11,
12, 11, SD¼ 3, 3, 3), frequency of phonetic radical (M¼ 629, 596,
606, SD ¼ 856, 923, 825), and number of characters that had
different toneswith their phonetic radicals (13, 12, 12). Each list
was assigned to the phonological training, semantic training,
and untrained conditions equally across subjects.
To construct the materials for the training tasks, each
target phonogram (e.g., 筏 /fa2/, raft) was paired with three
different higher frequency comparison characters: a phono-
logically related character (e.g., 乏 /fa2/, tired), a semantically
related character (e.g., 舟 /zhou1/, boat), and a phonologically
and semantically unrelated character (e.g., 斗 /dou4/, fight).
The phonologically related characters were homophones of
their corresponding targets. The semantic relationship be-
tween the targets and their semantically related and unre-
lated characters was rated on a 1e7 Likert scale (1 e least
relevant; 7 e most relevant) by a group of 20 native Chinese
students from Beijing Normal University, none of whom
participated in the main experiments. Themeans of semantic
relatedness scores for semantically related and unrelated
characters were respectively 5.82 (SD¼ .72) and 1.54 (SD¼ .33).
No obvious orthographic clues from the paired characters
were able to be used to perform the training tasks.
In addition to the 126 ReC phonograms, the stimuli pre-
sented during the naming task also included 42 low-
frequency, semantic-transparent, irregulareinconsistent
(IReIC) phonograms. The mean consistency ratio for these
items was .23 (SD ¼ .39). They were matched with the critical
phonograms in frequency (M ¼ 11.97, SD ¼ 5.83) and number
of strokes (M¼ 11, SD¼ 4). These IReIC phonograms served for
two purposes: (a) equating the number of trained versus un-
trained items in the naming task, and (b) allowing us to
examine neural effects of orthographicephonological consis-
tency,1 which can also be used to validate the credibility of the
present fMRI results by comparing with extant neuroimaging
results of regularity and consistency effects (Lee et al., 2004;
Tan, Feng, Fox, & Gao, 2001).
Finally, in order to directly compare the neural circuits
underlying phonological and semantic processes in reading
Chinese phonograms, participants were also required tomake
in-scanner homophone judgments and semantic association
judgments on a new set of phonograms after they completed
the naming task. The stimuli included 72 low-frequency ReC
phonograms. The means of frequency and number of strokes
were 21.72 (SD ¼ 14.64) and 11 (SD ¼ 3), respectively. Each
phonogram was assigned to the semantic task for half the
participants and to the phonological task for the remaining
participants and counterbalanced across participants be-
tween the two tasks. As in the training task, the phonograms
were paired with one-character comparison words such that
half of the trials in each task required a ‘yes’ response.
2.3. Procedure
The experiment was performed in two phases: a behavioral
training session followed by an fMRI session. During behav-
ioral training, participants were reinforced their attention to
different features for the low-frequency Chinese phonograms
through repeated presentationsdhalf in a training condition
that emphasized attention to the phonological form and the
other half in a training condition that emphasized attending
to semantics. Immediately following behavioral training, we
examined transfer of training effects in an event-related fMRI
session in which participants named aloud the trained pho-
nograms as well as an additional set of untrained phono-
grams. Following the naming scans, participants were
scanned in a block design paradigm while making phonolog-
ical and semantic judgments on a new set of Chinese pho-
nograms. A computer with Windows XP and Eprime 1.1
software (Psychology Software Tools Inc., Pittsburgh, PA, USA)
was used for all stimulus presentation and response
collection.
2.3.1. Behavioral trainingEach participant was trained on two sets of phonograms with
feedback. In the phonological training condition, participants
3 A consistency ratio is the number of phonograms with thesame pronunciation divided by the total number of phonogramswith the same phonetic radical.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 93
Author's personal copy
made homophone judgments on targets and corresponding
comparison items [e.g., “Do筏 and乏 sound the same?”, where
筏 (/fa2/, raft) is the target item and 乏 (/fa2/, tired) is the
comparison item]. In the semantic training condition, partic-
ipantswere asked to indicatewhether targets and comparison
items were semantically related [e.g., “Are鞍 and马 related in
meaning?”, where 鞍 (/an1/, saddle) is the target item and 马
(/ma3/, horse) is the comparison item].
During each trial, a target item was presented in parallel
with a comparison item in themiddle of the computer screen.
Subjects were asked to judge whether or not the target item
matched the comparison item on the targeted feature by
pressing one of two buttons on the keyboard. The items stayed
on the screen until a response was made or elapsed after
2000 msec, followed by an inter-trial interval of 2000 msec.
During the training, response feedback (“Correct” or “Incor-
rect”) was provided on each trial in order to encourage
attention to the targeted feature. Each subject was trained on
sixteen blocks with eight blocks in each training condition.
Training alternated between phonological and semantic
conditions every four blocks and the initial training condition
was counterbalanced across participants. Each phonogram
was presented only once in each block andwas repeated eight
times through the whole training session with either phono-
logical training or semantic training. Within each block,
phonograms were presented randomly. Across all repetitions
within each condition, the number of trials with correct yes
and no responses was equated and balanced across blocks.
Total time for the training session was approximately 2 h.
Both reaction time (RT) and accuracy were recorded during
behavioral training.
2.3.2. fMRI sessionImmediately after the training session, participants per-
formed an overt naming task in an event-related fMRI design
in which they named the phonologically and semantically
trained ReC phonograms, along with untrained ReC phono-
grams, and untrained IReIC phonograms. We employed an
event-related design with interleaved acquisition and “jit-
tered” trial durations (4, 5, 6, 7 sec) with occasional longer trial
durations (i.e., null trials) for the naming task. The target
stimulus in each trial was presented for 1000 msec. Data were
collected from seven functional runs (3 min plus 10 sec for
image stabilization) consisting of 24 items each. Three
different pseudorandom orders of stimuli were created and
rotated equally across subjects and training sets. Due to
scanning environment constrains, we were not able to obtain
behavioral naming responses from this group of subjects
during fMRI scanning.
After performing the overt naming tasks in scanner, par-
ticipants were scanned in a block design with blocks of
phonological judgments and blocks of semantic judgments
alternating with a baseline condition consisting of a line-
orientation judgment, in which participants judged whether
two sets of lines with left or right orientations had the same
pattern of left/right alternation (e.g., Are / / \ / and / / \ / the
same?). The phonological and semantic judgment tasks were
identical to those used during training, but without response
feedback and with a new set of stimuli.
Neuroimaging data for the judgment tasks were collected
from three fMRI runs. Each functional run consisted of four
22 sec activation blocks (two phonological, two semantic) and
three 22 sec baseline (line judgment) blocks. The order of
activation block types was counterbalanced across runs. Each
block consisted of six trials and was preceded by task in-
structions and a letter cue for 4000 msec at the bottom of the
screen to indicate the upcoming task [The English letter “S”
(abbreviation for Sound) was used as a cue for the homophone
judgment; “M” (abbreviation for Meaning) for the semantic
association judgment; “L” for the line-orientation judgment].
The cues remained on the lower part of the screen throughout
the entire block as a reminder of the task. The comparison pair
in each trial was presented for 1500 msec. Behavioral re-
sponses for the judgment tasks (RT and accuracy) were
recorded during fMRI scanning.
2.4. Image acquisition
Functional images were collected on a 3 T Siemens MR im-
aging system (Siemens MAGNETOMTrio, a TIM System) in the
Imaging Center for Brain Research of Beijing Normal Univer-
sity. Prior to functional imaging, 32 axial-oblique anatomic
images (echo time, 2.47 msec; repetition time, 300 msec; field
of view, 220 mm2; 4-mm-thick contiguous slices; matrix size,
256 � 256 � 2 number of excitations) were prescribed parallel
to the intercommissural line based on sagittal localizer im-
ages (echo time, 6.83 msec; repetition time, 20 msec; field of
view, 240/256 mm2; 4-mm-thick contiguous slices; matrix
size, 240 � 256 � 1 number of excitations). The axial-oblique
images were obtained at the same relative slice locations in
each subject, extending from the inferior aspect of the tem-
poral lobes to the parietal convexity. A single shot, gradient
echo, echo planar imaging (EPI) sequence (flip angle, 80�; echotime, 30 msec; repetition time, 2000 msec; field of view,
220 mm2; 4-mm-thick contiguous slices; matrix size,
64 � 64 � 1 number of excitations) was used to acquire acti-
vation images in the same slice locations as those used for
axial-oblique anatomic images, resulting in 90 images ob-
tained per functional run. After functional images were ac-
quired, an additional high-resolution whole brain anatomic
scan was obtained with the following parameters (flip angle,
7�; echo time, 3.66 msec; repetition time, 2530 msec; field of
view, 256 mm2; 1-mm-thick contiguous slices; matrix size,
256� 256� 1 number of excitations). The entire scanning took
about 40 min.
2.5. Image analysis
The imaging data were preprocessed and statistically
analyzed using in-house routines written in MATLAB (Math-
Works, Natick, MA). The first five images of each functional
run were excluded from data processing due to image stabi-
lization. Functional images were first sinc-interpolated to
correct for slice acquisition time, corrected formotion (Friston
et al., 1995), and spatially smoothed with a Gaussian filter of
size 3.125 mm full width at half maximum. For each subject,
an affine transformation to the standardized space defined by
the Montreal Neurological Institute (MNI) was obtained using
Bio-ImageSuite (Papademetris, Jackowski, Schultz, Staib, &
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 694
Author's personal copy
Duncan, 2003; www.bioimagesuite.org), mapping between the
subject-space T1 anatomic and the MNI space “Colin” brain
(available at www.bic.mni.mcgill.ca). Prior to across-subjects
analysis, this transformation was applied to the single-
subject activation maps, with trilinear interpolation, into 2-
mm isotropic MNI space.
For single-subject event-related analysis, a regression-
based method was used that allows direct estimation of the
hemodynamic response for each trial type at each voxel
separately (Miezin, Maccotta, Ollinger, Petersen, & Buckner,
2000). Time course estimates were made for 1-sec intervals
from �5 to þ15 sec relative to trial onset. To create subject
activation maps of the evoked response for each task condi-
tion, we obtained regression estimates of the mean difference
in these time course estimates for an activation period
(4e8 sec after trial onset) relative to a baseline period (0e3 sec
prior to trial onset). For single-subject block design analysis,
linear regression was used to generate contrast images of the
mean signal differences among the experimental and baseline
conditions. Stimulus blocks were entered into a separate
regressionmodel as a simple boxcar function, with a 4-sec lag
to account for the hemodynamic delay. Contrasts for effects of
interest were applied to these regression estimates to obtain
contrast images for each subject. In order to test effects across
subjects for both event-related and block design sections,
each voxel in these contrast images was tested versus zero
with a t-test, implementing a repeated measures analysis of
variance (ANOVA; Holmes & Friston, 1998; Woods, 1996).
3. Results
3.1. Adaptive learning effects
3.1.1. Behavioral training tasksMean RT and accuracy for the behavioral training outside fMRI
scanner (the fMRI group) are presented in Fig. 1. Two
2 � 8 � 2 � 3 (training condition � repetition � block
order � stimulus list) repeated measures ANOVAs were
computed with RT for correct responses and accuracy as
dependent measures separately. Training condition and
repetitionwere entered as within-subjects factors; block order
and stimulus list were entered as between-subjects factors.
Because block order and stimulus list were counterbalanced
across subjects and not theoretically interesting; interactions
involving these two factors are not reported. The main effects
of training condition [F(1,12) ¼ 10.985, p < .01] and repetition
[F(7,84) ¼ 20.009, p < .001] were both significant for RT, such
that RT decreased reliably across repetitions in both phono-
logical and semantic training conditions, and RT was signifi-
cantly faster in the semantic training condition than
phonological training condition. Planned paired-sample t-
tests revealed no RT differences in the first three pre-
sentations and shorter RTs for semantic training condition
than phonological training condition beginning at the fourth
presentation [t(17) ¼ 3.019, p < .01] and continuing to the end
of training [t(17) ¼ 3.941, p < .01]. For accuracy, only the main
effect of repetition [F(7,84) ¼ 10.487, p < .001] was significant,
indicating the accuracy increased comparably in both training
conditions. Neither RT nor accuracy showed significant in-
teractions between training condition and repetition. In sum,
the behavioral performance from training tasks suggests
successful training from both training tasks.
3.1.2. Training effects on behavioral naming responseAlthough we were not able to collect behavioral naming re-
sponses due to scanning constrains from the participants in
the fMRI group, we collected the naming responses from a
comparable behavioral cohort group. Since the two groups
showed similar response pattern in the training tasks, we
expected the fMRI groupwould show similar response pattern
in the naming task as the behavioral cohort group. Naming
latencies less than 200 msec and greater than 1750 msec were
discarded as outliers (2.8% of all the responses). Data from one
male subject were excluded due to excessive naming errors,
microphone failures and naming time outliers. The mean
naming latencies and accuracies for correct responses to
phonologically trained ReC phonograms, semantically
trained ReC phonograms, and untrained ReC phonograms for
the remaining 36 subjects in the behavioral cohort group were
shown in Fig. 2. A repeated measures ANOVA revealed a sig-
nificant training transfer effect on naming latencies
[F(2,70) ¼ 54.250, p < .001]. Post-hoc t-tests showed that both
phonologically and semantically trained ReC phonograms
yielded faster naming latencies relative to untrained ReC
phonograms [t(35) ¼ �9.170, p < .001 and t(35) ¼ �7.731,
p < .001, respectively], indicating that phonological and se-
mantic training facilitated subsequent naming of the trained
Fig. 1 e Mean reaction time (RT) and accuracy by training conditions and presentation numbers in the behavioral training
tasks from the fMRI group before fMRI scanning.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 95
Author's personal copy
phonograms. No difference was observed between naming
latencies for phonologically and semantically trained ReC
phonograms, indicating that semantic and phonological
training produced equivalent levels of behavioral training ef-
fects to naming. The observed differences between trained
and untrained ReC phonograms in naming time did not
reflect a speedeaccuracy tradeoff. Naming accuracies of ReC
phonogramswere high in all conditions, with a slightly higher
accuracy for phonologically trained ReC phonograms than the
other two types of ReC phonograms.
3.1.3. Training effects on neural circuits of namingTo examine whether there were differential effects of the
phonological and semantic training on brain activation, we
directly contrasted the naming maps for phonologically and
semantically trained phonograms (Fig. 3A and Table 1). Only
the inferior aspect of LH IFG Brodmann area 47, or (BA47) in the
well-documented reading networks was less activated for
semantically trained phonograms than for phonologically
trainedphonograms (the regionofyellow/red inFig. 3A). Fig. 4A
presents the mean activation (B-weights in the regression
function) across all the voxels in this regionwhen participants
named aloud the phonologically trained, semantically trained,
and untrained phonograms, suggesting more neural facilita-
tion effect in this region for semantically trained phonograms
than for phonologically trained phonograms. Beside the infe-
rior aspect of LH IFG (BA47), no other regions in the major
identified LH reading systems showed any differences be-
tween phonologically trained and semantically trained pho-
nograms. Several regions outside the standard reading
networks were less activated for phonologically trained than
for semantically trained ReC phonograms, including LH pos-
terior cingulate gyrus, RH cingulate gyrus, andRH superior and
medial frontal gyrus (regions of blue/purple in Fig. 3A).
Interestingly, when phonologically and semantically
trained ReC phonograms were combined as trained ReC
phonograms and compared with the activations of untrained
ReC phonograms, many differences were found (Fig. 3B and
Table 1). Most importantly, trained phonograms showed
reduced activation at bilateral speech-articulatory areas
compared with untrained phonograms (regions of blue/purple
in Fig. 3B), including LH insula/IFG (BA13/45) and right-
hemisphere (RH) IFG (BA45). The reduction in activation in
these regions indicates general neural training-facilitation
effects. In addition, trained phonograms were more active at
bilateral IPL than untrained phonograms (regions of yellow/
red in Fig. 3B). Fig. 4BeE presents the mean activations (B-
weights in the regression function) across all the voxels in
each of the following regions respectively, LH insula/IFG, RH
IFG, LH IPL, and RH IPL, when participants named aloud
phonologically trained, semantically trained, and untrained
phonograms.
3.2. Regularityeconsistency effects
Both naming latency and accuracy in the behavioral
cohort group confirmed the empirical behavioral regular-
ityeconsistency effects.1 Naming responses to untrained
IReIC phonograms were significantly slower [latency:
770 msec (SD ¼ 109), t(35) ¼ �6.058, p < .001] and less accurate
[accuracy: 87% (SD ¼ 7), t(35) ¼ 4.310, p < .001] than for un-
trained ReC phonograms.
Neurobiologically, the contrast between the two types of
phonograms from the fMRI group revealed many differences
in the well identified distributed LH Chinese reading circuitry.
IReIC phonograms showed higher activation than ReC pho-
nograms at LH IFG/middle frontal gyrus (MFG) (including
BA45, 46, 47, and 9), FG, MTG, and SPL extending to precuneus,
IPL, and supramarginal gyrus (SMG) (see Fig. 5 and Table 2).
However, no regions were found with higher activation for
ReC phonograms than IReIC phonograms. The activation
pattern at LH IFG/MFG, FG, SPL, IPL, and SMG in the present
data was similar to a previous study for consistency effect in
reading Chinese phonograms (Lee et al., 2004); however, we
also found higher activation for naming IReIC phonograms
than ReC phonograms at LHMTGwhereas Lee et al. (2004) did
not. Our finding of the differential activation at LH MTG as a
function of consistency squares findings in English with
similar manipulations, which has typically been taken as an
indication that visual word recognition involves the coopera-
tive interaction of phonological and semantic processes, with
stronger recruitment of semantic processes when the
computation of phonology is difficult (Frost et al., 2005; Graves
et al., 2010).
3.3. Judgment tasks
Behaviorally, mean RT and accuracy of the (in-scanner)
phonological and semantic judgment tasks were similar.
Mean RTs in the phonological judgment task, semantic judg-
ment task, and line-orientation judgment task were 951
(132) msec, 950 (134) msec, and 1155 (120) msec, respectively.
Fig. 2 eNaming latency and accuracy for phonologically trained, semantically trained, and untrained ReC phonograms from
the behavioral cohort group.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 696
Author's personal copy
Mean accuracies in the phonological judgment task, semantic
judgment task, and line-orientation judgment task were .95
(.1), .93 (.1), and .88 (.1), respectively. Repeated measures
ANOVA revealed significant main effects of task for both RT
[F(2,34) ¼ 38.223, p < .001] and accuracy [F(2,34) ¼ 10.405,
p < .001]. Post-hoc t-test revealed differences between
phonological task and line-orientation task, as well as se-
mantic task versus line-orientation task [RT: t(17) ¼ �6.221,
p < .001 and t(17) ¼ �8.994, p < .001, respectively; accuracy:
t(17) ¼ 4.106, p < .001 and t(17) ¼ 2.652, p < .05, respectively],
but no difference between phonological task and semantic
task.
Surprisingly, direct comparison of neural activation levels
during the semantic task versus the phonological task also
revealed few differences (Fig. 6A and Table 3). Only one region
of LH IFG extending to MFG was activated higher for the se-
mantic task than for the phonological task (regions of yellow/
red in Fig. 6A). The only region that showed higher activation
for the phonological task was precuneus extending into the
posterior cingulate gyrus (regions of blue/purple in Fig. 6A). In
contrast, combining phonological and semantic tasks
revealed a number of regions that were significantly more
activated for the language (phonological and semantic) tasks
than for the non-language control (line-orientation) task,
including LH IFG (BA47, 44, and 45) extending to MFG, MTG,
middle occipital gyrus (MOG) extending to lingual and FG, and
a focal activation in and along left collateral sulcus (regions of
yellow/red in Fig. 6B and Table 3). It is worth noting that an
area at RH FG, which was suggested to play an important role
in Chinese reading by previous studies (Bolger, Perfetti, &
Schneider, 2005; Liu, Dunlap, Fiez, & Perfetti, 2007; Nelson,
Liu, Fiez, & Perfetti, 2009; Tan, Laird, Li, & Fox, 2005), was
found activated to a higher degree for the non-language con-
trol task compared with the language tasks. In general, the
neuroimaging responses from judgment tasks indicate a
multi-components neural network for reading of Chinese in
widespread regions of the LH, which subserves phonology and
semantics by-and-large equally.
4. Discussion
In this study, we examined the neural system of Chinese
reading, with a particular focus on the division of labor be-
tween phonological and semantic processes of reading Chi-
nese ReC phonograms. We employed an adaptive learning
Fig. 3 e Brain activations where semantic-trained
phonogramswere less activated than phonological-trained
phonograms (yellow/red) and where phonological-trained
phonograms were less activated than semantic-trained
phonograms (blue/purple) (A). Brain activations where
trained phonograms were less activated than untrained
phonograms (blue/purple) and where trained phonograms
were more active than untrained phonograms (yellow/red)
(B). Images were presented at a univariate threshold of
p < .01, corrected for mapwise false discovery rate (FDR;
Genovese, Lazar, & Nichols, 2002). Images from top to
bottom correspond to the following position along the z-
axis in MNI space: D52, D42, D22, D8, D2, and L6,
respectively, with the LH on the right side of the images.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 97
Author's personal copy
paradigm, in which participants were required to focus on
different properties (phonology vs semantics) of Chinese ReC
phonograms during training. This paradigm allowed us to
examine the transfer of phonological and semantic training to
a subsequent naming task. Our training effects showed that
prior exposure had similar effects on the naming of Chinese
phonograms regardless of whether the prior exposure took
place while participants were engaged in a phonological or a
semantic task. Behaviorally, naming latencies were facilitated
to the same extent by phonological and semantic training.
Neurobiologically, the transfer effects of phonological and
semantic training on subsequent naming circuits were by-
and-large rather similar in the major identified LH reading
system for Chinese reading. Our data suggest that the char-
acteristics of the Chinese writing system produce a relatively
equitable division of labor between phonological and seman-
tic processes, specifically when reading Chinese ReC
phonograms.
Beyond our main finding of division of labor between
phonology and semantics in reading of Chinese ReC phono-
grams, our neuroimaging results from judgment tasks also
shed light on the neural circuitry of the Chinese reading. In
general, neural activation patterns in judgment tasks corrob-
orated a multi-componential reading system for Chinese
reading primarily in left-sided reading areas (see Tan et al.,
2005, for reviews), including LH areas that also subserve
alphabetic reading (IFG, MTG, IPL/SPL, and FG), but also MFG
that is thought to play a specific role in the reading of Chinese.
These regions were all engaged in component (both phono-
logical and semantic) judgment tasks and were also activated
higher in naming IReIC phonograms compared to naming
ReC phonograms, again suggesting that semantics and
phonology are somewhat sharing the same neural network.
Our results provided the first-order neurobiological
empirical evidence for cooperation of phonology and seman-
tics to reading of Chinese ReC phonograms. More impor-
tantly, we directly contrasted phonological and semantic
effects in reading of Chinese ReC phonograms in fMRI, which
was rarely done in previous studies. Surprisingly, phonology
and semantics not only illustrated relatively comparable
behavioral effects in naming aloud, but also induced quite
similar effects neurobiologically. All together, the results
imply that Chinese reading might require a combination of
semantic and phonological strategies that are hardly differ-
entiated from each other. The equitable division of labor be-
tween phonological and semantic contribution to reading of
Chinese ReC phonograms that we found in the current fMRI
study corroborated recent computational modeling work
within connectionist framework, which showed that seman-
tics played as much important role as phonology in naming
Chinese ReC phonograms (Yang et al., 2012).
From a broader theoretical background, the current results
are also consistent with findings of reading development and
impairments in Chinese. Although a numerous studies have
reported a strong link between phonological processing and
character recognition in Chinese children (Chow, McBride-
Chang, & Burgess, 2005; Ho, Law, & Ng, 2000; Siok & Fletcher,
2001), the contribution of semantics to the development of
Chinese reading is also critically important (Chen, Hao, Geva,
Zhu, & Shu, 2009; McBride-Chang, Shu, Zhou, Wat, & Wagner,
2003; Shu, McBride-Chang, Wu, & Liu, 2006). In particular, Li,
Shu, McBride-Chang, Liu, and Peng (2012) administrated 184
kindergarteners at age 5e6 and 273 primary school students at
age 7e9 from Beijing with a comprehensive battery of tasks
and found both phonological and semantic skills significantly
explained the performance of character recognition of these
children, which underscores the importance of dimensions of
both phonological and semantic processing for very early and
intermediate acquisition of Chinese reading. Developmental
dyslexic work by Shu, Meng, Chen, Luan, and Cao (2005) also
implies that semantic and phonological impairments have
widespread effects for Chinese character reading. They re-
ported three cases of developmental dyslexia in Chinese, and
found that one case with impairment in semantic processing
(surface dyslexia) and two cases with deficit of phonological
processing (deep dyslexia). However, both children with se-
mantic and phonological deficits were impaired on naming of
all characters relative to their age-matched controls. These
Table 1 e Brain activation regions for training effects.Specific training effects: direct comparison betweenphonologically trained ReC phonograms and semanticallytrained ReC phonograms. General training effects: directcomparison between trained phonograms (collapsed ofphonologically trained and semantically trained ReCphonograms) and untrained ReC phonograms.
Region BA x y z Volume(mm3)
p Value(at peak)
Semantic-trained < Phonological-trained
L. IFG 47 �26 34 �6 528 .0006
Phonological-trained < Semantic-trained
R. MFG 32 22 6 51 2824 .0001
L. Posterior
cingulate gyrus
23 �4 �32 22 1472 .0004
R. Cingulate gyrus 24 6 �18 44 1296 .0001
R. STG 9 22 48 42 960 .0002
R. Cingulate gyrus 24 8 �2 34 832 .0002
R. Cerebellum 35 20 �36 �44 320 .0006
L. Cerebellum 19 �18 �64 �42 248 .0018
L. Cerebellum 19 �24 �70 �46 240 .0031
L. Thalamus e �16 �26 16 216 .0024
L. STG 6 �16 0 74 216 .0014
Trained < Untrained
R. IFG 45 46 22 8 656 .0008
L. Insula 13 �42 16 2 312 .0034
Trained > Untrained
R. IPL 19 44 �72 42 2288 <.0001
L. IPL 19 �42 �76 42 2240 .0010
L. Putamen e �10 4 �8 640 .0007
L. Cuneus 7 �8 �68 32 632 .0033
R. Lingual gyrus 18 4 �72 �1 448 .0049
L. Cerebellum 37 �32 �48 �44 432 .0007
L. STG 6 �34 12 60 432 .0016
R. MFG 6 14 �16 56 352 .0007
R. MFG 10 20 60 6 296 .0038
For clusters with multiple anatomical locations, only the peaks of
activations are labeled. Volumes bigger than 200 mm3 are reported.
Thresholds for the training effects in the naming task are set at
p < .01 FDR corrected. Coordinates are MNI space at activation
peaks. IFG: inferior frontal gyrus, MFG: middle frontal gyrus, STG:
superior temporal gyrus, IPL: inferior parietal lobule.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 698
Author's personal copy
cases were successfully simulated and confirmed by con-
nectionist models recently (Yang et al., 2012). It is also worth
noting that a recent neuroimaging study observed similar al-
terations in brain activations for phonological and semantic
processing to visual characters in a group of Chinese devel-
opmental dyslexic children (Liu et al., 2012), again suggesting
that access to phonology and semantics from the visual
orthography for Chinese readers might use the same neural
network.
Even though our neurobiological results have shown that
the neural network that underlies reading in Chinese largely
overlapped the network responsible for reading in English, the
neural division of labor between phonology and semantics in
Chinese reading contrasted tremendously from English find-
ings. Althoughwe did not conduct comparable experiments in
an alphabetic writing system in the present study, it is still
worth doing a qualitative comparison between the current
study and those studies from the literature of alphabetic
writing systems to illuminate how structure of the writing
system drives the differences in division of labor between
phonology and semantics in reading.
First, for the adaptive learning experiments, the most
comparable study that has been done in alphabetic writing
systems is the work conducted by Sandak et al. (2004) in En-
glish reading. Although the two studies employed slightly
different training procedures (phonological training: rhyming
judgment in English vs homophone judgment in Chinese; se-
mantic training: category judgment in English vs semantic
association judgment in Chinese) necessitated by the differ-
ences in training materials (English pseudowords vs Chinese
ReC phonograms), in general both studies employed the same
adaptive learning paradigm and both aimed to examine how
different strategies of training change the responses of
naming words. In addition, the behavioral pattern of adaptive
learning in the present study was similar to that observed in
readers of English. In both studies, phonological and semantic
training resulted in faster naming latencies than control
condition and the facilitation effects in naming latencies were
comparable between phonological training and semantic
training, which again suggests the comparability of the two
studies.
However, the neurobiological patterns associated with di-
vision of labor between phonological and semantic processes
in reading Chinese and English seemed quite different. In the
results of Sandak et al. (2004) with readers of English, the
neurobiological effects of phonological and semantic training
differed substantially. Reinforcing different dimensions
(phonological vs semantic) facilitated naming via different
neural components: phonological training resulted in de-
creases in left FG, SMG, and posterior IFG activation, whereas
semantic training produced an increase at left MTG. In
contrast, in the present study phonological and semantic
training had largely similar neurobiological effects on the
neural correlates of reading in Chinese characters, including
increased activations in bilateral IPL and reduced activations
in the well-known speech output areas at bilateral IFG (BA45),
Fig. 4 e Activation values for naming phonologically trained, semantically trained, and untrained ReC phonograms in the
regions showed specific training effects: LH IFG (A) and general training effects: LH insula (B), RH IFG (C), LH IPL (D), and RH
IPL (E).
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 99
Author's personal copy
which may reflect training-related increased processing effi-
ciency to naming at articulatory-motor circuits (Katz et al.,
2005; Poldrack & Gabrieli, 2001; Pugh et al., 2008; Sandak
et al., 2004). Perhaps the only meaningful difference between
the two training conditions was in one region of the anterior
part of left IFG (BA47), which was reducedmore when naming
semantically trained phonograms than naming phonologi-
cally trained phonograms, suggesting that semantic training
slightly induced more efficient retrieval of semantics than
phonology in naming Chinese phonograms. Although our
findings of IFG (BA47) seem new compared to Sandak et al.
(2004)’s study, in which the authors didn’t observe neural
changes of left IFG responding to semantic training, our result
did corroborate findings from our own component judgment
tasks, in which we observed that the anterior part of IFG
(BA47) was responsible more in semantic processing, paral-
leling various findings from English studies (Gough, Nobre, &
Devlin, 2005; Poldrack et al., 1999). In sum, our results
showed that phonological and semantic training have rather
similar effects on the patterns of brain activation associated
with naming aloud Chinese charactersdin contrast to the
differential effects of phonological and semantic training in
English. In other words, neural OeP and OeS pathways for
reading of Chinese are not as easily to differentiate by focused
learning as in English.
Second, the effects of the task manipulation (phonological
vs semantic) observed in the judgment tasks appear to differ
substantially from those previously observed in English too.
We found that phonological and semantic tasks induced quite
similar neural activation patterns in the reading system for
Chinese, indicating that the weights for the multi-
components of Chinese reading system do not change sub-
stantially according to the task demands. Although semantic
judgment task did induce higher activation than phonological
judgment in left IFG (BA47), most of the components in the LH
(BA45 of IFG, MFG, MTG, FG, and IPL/SPL) were activated
equivalently between phonological judgment and semantic
judgment. In contrast, previous studies in readers of English
revealed that phonological and semantic tasks result in sub-
stantially different patterns of neural activation. Relative to
phonological tasks, semantic tasks are associatedwith greater
activation in LH anterior IFG, angular gyrus, MTG, and anterior
FG; phonological tasks evoke greater activation in LH poste-
rior/dorsal IFG, insula, SMG, and posterior FG (Booth et al.,
2002a, 2002b; Demonet et al., 1992; Devlin, Matthews, &
Rushworth, 2003; McDermott, Petersen, Watson, & Ojemann,
2003; Poldrack et al., 1999; Price, Moore, Humphreys, & Wise,
1997; Pugh et al., 1996). The comparison between results
from our component judgment tasks and previous English
findings again supports our hypothesis that Chinese character
reading has more equitable division of labor than English
reading. The similar network for OeP and OeS processes in
Fig. 5 e Brain activations where untrained IReIC
phonograms were greater than untrained ReC
phonograms. Images were presented at a univariate
threshold of p < .01, corrected FDR. Images from top to
bottom correspond to the following position along the z-
axis in MNI space: D52, D42, D32, D22, D8, D2, L6, and
L12, respectively, with the LH on the right side of the
images.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6100
Author's personal copy
reading of Chinese characters suggests that even though
subjects were asked to focus on phonological information,
semantic processes are recruited to facilitate the computation
of phonology and vice versa. In sum, the cortical regions
subserving phonological and semantic processing seem to be
more interdependent when reading Chinese characters than
when reading English words.
The different neurobiological effects of training from En-
glish and Chinese adaptive learning experiments suggest that
reading system of Chinese might illustrate a more equitable
division of labor between phonology and semantics than that
of English. The differences in the neural division of labor be-
tween readers from the two writing systems might reflect the
differences in the structure of the writing systems and might
be explained in the framework of connectionist models, with
the assumption that different statistical regularities from
print to sound andmeaning drive the differences of division of
labor between phonology and semantics across different
writing systems (Harm & Seidenberg, 2004; Seidenberg, 1992,
2011; Yang et al., 2012). As we described in Introduction, for
Chinese, the OeP and OeS mappings are relatively equally
systematic, although both are not as much consistent as OeP
mapping in English. In contrast, although the OeP mapping in
English is largely systematic (albeit not to the extent of
‘shallower’ orthographies such as Italian or Spanish), the OeS
mapping is largely arbitrarydwith the exception of morpho-
logical regularities (Rueckl, 2010; Seidenberg & Gonnerman,
2000) that are for the most part redundant with the structure
of the phonologicalesemantic mapping [and hence, available
to the reader via the phonologically mediated (orthogra-
phicephonologicalesemantic) pathway]. Thus, it is reason-
able to conjecture that reading in Chinese should involve a
more equitable division of labor between phonological and
semantic processes than reading in English (Harm &
Seidenberg, 2004; Yang et al., 2012). The comparison of the
division of labor between OeP and OeS processes in the
reading process of computing phonology from orthography is
deciphered in Fig. 7. As shown in the figure, the two processes
(direct mapping from orthography to phonology and indirect
mapping from orthography to semantics then to phonology)
both contribute to the computation of phonology in the two
writing systems. However, the weighting for the two map-
pings differ substantially between the two writing systems. In
English, there is greater reliance on OeP mapping than OeS
mapping, whereas in Chinese, the twomappings by-and-large
equally contribute to the computation of phonology.
Although we have discussed in general the differences
between Chinese and English reading in terms of division of
labor between OeP and OeS processes under the framework
of connectionist models, these differences may also partly
depend on the structure of written and phonological forms
themselves. The greater orthographic complexity and simpler
monosyllabic phonological structure of Chinese may also
contribute to the differences in reading processes between the
two writing systems.
In terms of orthographic structure, although Chinese in-
cludes both phonological radicals and semantic radicals, the
radicals may be further divided into about 600 sub-
components, which have fixed internal structures. The com-
plex visuospatial organization of these subcomponents may
make relatively greater demands on basic visual or ortho-
graphic analysis in Chinese reading than in alphabetic
reading. Also, many of the radicals or components have their
legal positions, thus awareness of these orthographic legal-
ities is important in character recognition (e.g., Li et al., 2012).
Studies have also found that orthographic processing deficit
may also directly relate to reading disability (Ho, Chan, Chung,
Lee, & Tsang, 2007; Ho, Chan, Lee, Tsang, & Luan, 2004).
With regard to phonological structure, Chinese has a
relatively simpler phonological structure than English does.
Chinesemorphemes always correspond tomonosyllables and
the syllable only allows structures such as CV, CVN, V, and VN
(C ¼ Consonant, V ¼ Vowel, and N ¼ Nasal). Indeed, syllable
awareness was found more important than phonemic
awareness in spoken word processing in Chinese, a finding
contrasted substantially with English studies (Zhao, Guo,
Zhou, & Shu, 2011). Siok, Jin, Fletcher, and Tan (2003) also
observed that distinct brain regions associate with syllable
and phoneme processing for Chinese readers. Thus, it is
plausible to infer that the specialty of the phonological
structure of Chinese might also produce some effects in Chi-
nese reading compared to English reading too (Siok, Perfetti,
Jin, & Tan, 2004; Tan et al., 2005).
Further, some brain regions have also been argued to be
specifically involved in identifying Chinese characters for the
increased demand on visuospatial processing (LHMFG and RH
FG in Bolger et al., 2005; Liu, Dunlap, et al., 2007; Nelson et al.,
2009; Tan et al., 2005) and “addressed” phonological process-
ing at the syllable level (LH MFG in Siok et al., 2003, 2004; Tan
et al., 2005). In our study, we observed that LH MFG was more
engaged for the linguistic judgment task as comparedwith the
non-linguistic judgment task, indicating its role for general
visuospatial analysis of Chinese characters (Tan, Liu, et al.,
2001; Tan et al., 2003). However, LH MFG was also recruited
more by the semantic judgment task than the phonological
judgment task as illustrated by our direct comparison map
between the two tasks and more for naming IReIC
Table 2 e Brain activation regions forregularityeconsistency effects: direct comparison betweennaming untrained IReIC phonograms and untrained ReCphonograms.
Region BA x y z Volume(mm3)
p Value(at peak)
Irregulareinconsistent > Regulareconsistent
L. IFG 9 �56 14 30 13,296 <.0001
R. Cerebellum 37 38 �70 �48 5216 .0002
L. SPL 7 �30 �56 50 4928 .0001
L. FG 19 �52 �72 �12 4248 .0003
L. Cingulate gyrus 8 �8 16 56 1448 .0001
R. Cerebellum 19 34 �82 �21 1120 .0004
L. MFG 6 �32 �4 62 720 .0005
L. MFG 6 �20 �4 52 472 .0007
R. IFG 47 34 26 �2 312 .0014
For clusters with multiple anatomical locations, only the peaks of
activations are labeled. Volumes bigger than 200mm3 are reported.
Thresholds for the regularity-consistency effects in the naming
task are set at p < .01 FDR corrected. Coordinates are MNI space at
activation peaks. IFG: inferior frontal gyrus, SPL: superior parietal
lobule, FG: fusiform gyrus, MFG: middle frontal gyrus.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 101
Author's personal copy
phonograms than ReC phonograms. Hence, we tend to sug-
gest that LH MFG may broadly serve roles for orthographic,
phonological, and semantic processing. In terms of RH FG, it
should be noted that our results do not support a reading-
specific role of it for Chinese reading as proposed by some
previous studies (Bolger et al., 2005; Liu, Dunlap, et al., 2007;
Nelson et al., 2009; Tan et al., 2005), because in our study,
the activation of this region was higher in a non-language
control task (line-orientation) than in the language (seman-
tic association and homophone judgment) tasks. Thus, we
suggest that RH FG may not be overly specialized for Chinese
character processing. Our claim is consistent with a recent
Chinese and English reading comparison study by Hu et al.
(2010) with the most tightly matched participants and ana-
lyses, which did not show any RH FG activation more for
Chinese readers than English readers. To sum, our data sug-
gest that the fundamental differences in orthographic and
phonological representation between Chinese and alphabetic
systems seem drive differences in neural activation patterns
in LH MFG but not in RH FG for the reading network. However,
additional workwill be required for a complete understanding
of whether these fundamental representational differences
have a comparable, bigger, or smaller impact than the differ-
ential statistical structure of weighting between OeP and OeS
mappings in driving differences in division of labor between
the two writing systems.
5. Limitations
In the current study, we have made our primary claim (equi-
table division of labor between phonology and semantics for
Chinese reading) based on relatively similar responses for
phonological and semantic strategies. However, we believe
that the results that we found in this paper were not null re-
sults but rather reflect important and specific characteristics
of Chinese reading. It might be argued that the lack of large
differences produced by phonological training and semantic
training could likely be due to the basics of the adaptive
learning paradigm and the naming task that we used in the
study, which may not be sensitive to detect the differences
across the two training conditions. This argument is not
tenable, because of the following reasons. Firstly, we found
many differences in cortical activation in the naming task
between trained and untrained conditions regardless of
training types. The most important neural facilitation effect
Fig. 6 e Brain activations greater in the semantic judgment
task than in the phonological judgment task (yellow/red)
and greater in the phonological judgment task than in the
semantic judgment task (blue/purple) (A). Brain activations
greater in the linguistic judgment task (collapsed across
the phonological judgment and the semantic judgment)
than in the line-orientation judgment task (yellow/red) and
greater in the line-orientation judgment task than in the
linguistic judgment task (blue/purple) (B). Images were
presented at a univariate threshold of p < .0001, corrected
FDR. Images from top to bottom correspond to the
following position along the z-axis in MNI space: D42,
D28, D20, D12, D4, L4, L12, and L18, respectively, with
the LH on the right side of the images.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6102
Author's personal copy
for trained ReC phonograms compared to untrained ReC
phonograms was reduced activation at bilateral IFG/insula,
which was consistent with many previous training-related
studies in English (Katz et al., 2005; Poldrack & Gabrieli, 2001;
Pugh et al., 2008; Sandak et al., 2004). This suggests that
neural response of the task of naming aloud is actually sen-
sitive to training and the reason that it is not sensitive to
different training conditions might be mainly driven by the
inherent characteristics of the training focus rather than the
naming task.
Secondly, we also observed robust results with regard to
the regularityeconsistency effects in the naming task: The
less consistent (IReIC) phonograms induced higher activation
at distributed LH reading circuits than the phonograms that
are more consistent (ReC phonograms). Themain locus of the
regularityeconsistency effect at left IFG in naming Chinese
phonograms was similar to the region associated with
phonological consistency effects in English (Fiez, Balota,
Raichle, & Petersen, 1999; Frost et al., 2005; Graves et al.,
2010) and overlapped with the loci reported in two previous
studies respectively for regularity effects (Tan, Feng, et al.,
2001) and consistency effects (Lee et al., 2004) in Chinese
reading. We also observed the differential activation of pre-
dominately semantic regions (LH MTG) as a function of the
orthographicephonological consistency in Chinese reading, a
similar neuroimaging finding as parallel manipulations in
English (Frost et al., 2005; Graves et al., 2010). The
Table 3 e Brain activation regions for judgment tasks. Specific linguistic effects: direct comparison between phonologicaljudgment and semantic judgment. General linguistic effects: direct comparison between linguistic judgment (collapsed ofphonological and semantic judgments) and line-orientation judgment.
Region BA x y z Volume (mm3) p Value (at peak)
Semantic judgment > Phonological judgment
L. IFG 47 �40 41 �6 4232 <.0001
Phonological judgment > Semantic judgment
R. Cingulate gyrus 31 20 �52 24 7128 <.0001
Linguistic judgment > Line judgment
L. IFG 47 �38 38 �4 12,984 <.0001
R. Cerebellum 18 14 �88 �30 6816 <.0001
L. MTG 21 �58 �42 0 3496 <.0001
L. MOG 18 �26 �84 �6 3120 <.0001
L. Cuneus 18 �10 �98 18 2840 <.0001
L. Parahippocampus 20 �38 �13 �22 1880 <.0001
L. Collateral sulcus 37 �30 �36 �18 872 <.0001
R. Cuneus 18 38 �96 0 720 <.0001
L. MFG 6 �38 10 52 712 <.0001
L. STG 8 �8 34 47 600 <.0001
R. MFG 13 36 �18 22 520 <.0001
R. Cuneus 18 16 �100 18 384 <.0001
L. MFG 13 �36 �26 24 304 <.0001
Line judgment > Linguistic judgment
R. IPL 7 26 �64 34 24,200 <.0001
R. Precentral gyrus 6 32 �2 50 7504 <.0001
L. SPL 7 �22 �60 56 5528 <.0001
L. Cerebellum 37 �36 �46 �28 3136 <.0001
R. FG 37 54 �62 �14 3072 <.0001
L. Cerebellum 37 �30 �60 �42 1536 <.0001
L. Precentral gyrus 6 �22 �6 50 1336 <.0001
R. MTG 19 38 �78 18 1320 <.0001
R. MTG 9 51 6 28 912 <.0001
R. Cerebellum 37 40 �46 �32 568 <.0001
L. SPL 19 �36 �80 14 328 <.0001
For clusters with multiple anatomical locations, only the peaks of activations are labeled. Volumes bigger than 200 mm3 are reported.
Thresholds for the specific and general linguistic effects in the judgment task are set at p < .0001 FDR corrected. Coordinates are MNI space at
activation peaks. IFG: inferior frontal gyrus, MTG:middle temporal gyrus, MOG:middle occipital gyrus, MFG:middle frontal gyrus, STG: superior
temporal gyrus, IPL: inferior parietal lobule, SPL: superior parietal lobule, FG: fusiform gyrus.
Fig. 7 e Division of labor between OeP and OeS processes
in the reading process of computing phonology from
orthography in English and Chinese.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 103
Author's personal copy
regularityeconsistency effects in the current results came
from the same imaging runs as the training comparison data,
suggesting a lack of fundamental problems with the data and
the less positive results from the training comparison might
not reflect anything about statistical power of the fMRI design.
In addition, both Frost et al. (2005) and Graves et al. (2010)
have used naming aloud as an experimental task and the
neural OeP andOeS pathwayswerewell separated by amulti-
parametric design of the stimuli (e.g., consistency by image-
ability) in both studies of English readers. Indeed, in the pre-
vious study most closely related to ours, Sandak et al. (2004)
also used the naming task to reveal differential effects of
phonological and semantic training in English. Thus, naming
aloud is certainly a task that has been proved to be sensitive to
examine division of labor between phonology and semantics
and our results should not be due to a problem from the task
thatweused but rather because of theChinesewriting system.
Finally, our judgment tasks also revealed that the phono-
logical and semantic processing for ReC phonograms pro-
duced relatively little differences in ensembles of LH reading
areas, which coincides with an earlier finding by Tan, Liu,
et al. (2001) (although in that study the tasks were not
directly compared and the stimuli included various types of
Chinese characters not specific to ReC phonograms). Using an
explicit task, which specifically tapped into OeP and OeS
processing of Chinese, we still failed to find any differences
between phonological and semantic processes. This result
again suggests that the little differences in our training com-
parison between phonological and semantic reinforcements
may not be due to the paradigm problem, but rather reflect
interdependence of phonological and semantic processes in
Chinese reading, per se. It should be noted that some of the
previous studies (if not all) have reported some differentia-
tions in the neural circuits between Chinese semantic pro-
cessing and phonological processing. However, the subtle
differences might only be able to detect in some special cir-
cumstances, such as increasing task demands and not
restricting the tasks to reading (e.g., Wu, Ho, & Chen, 2012) or
using two-character words as stimuli for reading (e.g., Booth
et al., 2006). Whether the findings from Wu et al. and Booth
et al. can be generalized into single character reading in Chi-
nese seems inconclusive.
Nevertheless, we acknowledged that the comparison in
adaptive learning that we made between Chinese and English
perhaps should be taken conservative, because of the differ-
ences in terms of stimuli and procedure between the two
studies. It might be valuable to conduct a future study on
adaptive learning of English ReC words with similar training
procedure as the current Chinese study, so that the division of
labor between the twowriting systems could be compared in a
more matched way. It might also be interested to look at
cooperative learning effects in reading Chinese pseudo-
characters. Actually, Zhao (2012) have addressed this issue
and found cooperation between phonology and semantics in
learning to read Chinese pseudo-characters, which may
deserve a follow-up neuroimaging study to examine the
neural division of labor effects between phonology and se-
mantics in learning Chinese pseudo-characters.
It is also worth highlighting that the current paper only ad-
dresses the issue related to the division of labor between
phonology and semantics in reading single Chinese characters
and whether division of labor between phonology and seman-
tics in reading two or multi-character words or even sentences
or texts in Chinese is beyond the scope of this paper and is still
an open question which might be valuable to investigate in
future studies. In addition, we have examined the division of
labor between phonology and semantics to a task of reading
aloud in Chinese, but a full account of the division of labor be-
tween phonology and semantics will require consideration of
other tasks (e.g., silent readingwith computation ofmeaning as
the ultimate goal). Indeed, Harm and Seidenberg (2004) have
modeled the division of labor between phonology and seman-
tics in Englishwith silent reading formeaning as the task. None
of previous studies have conducted similar modeling or
empirical research in Chinese reading yet. Future studiesmight
be worth pursuing more in this direction too.
6. Conclusions
In sum, the present study suggests that the division of labor
between phonology and semantics in Chinese reading is equi-
table, which may be driven by similar systematicity between
OeP and OeS mappings in the writing system of Chinese. Even
though the neural network that underlies reading in Chinese
largely overlaps the network responsible for reading in English,
the neural division of labor between phonology and semantics
in Chinese reading contrasts tremendously from English find-
ings. Phonological and semantic processes for English readers
induced great neural differentiations in brain circuits, whereas
the two processes for Chinese readers have largely similar
neurobiological effects. This culture-constrained difference
may be caused by language-specific statistical regularities
embodied in the writing system.
Acknowledgments
This work was supported by the National Institute of Child
Health andHumanDevelopment (grant number P01HD001994)
to Haskins Laboratories, the Natural Science Foundation of
Beijing (grant number 7132119) to Hua Shu, and the Natural
Science Foundation of China (grant number 31271082) to Hua
Shu.
r e f e r e n c e s
Barca, L., Burani, C., & Arduino, L. S. (2002). Word naming timesand psycholinguistic norms for Italian nouns. BehaviorResearch Methods, Instruments, & Computers, 34(3), 424e434.
Bates, E., Burani, C., D’Amico, S., & Barca, L. (2001). Word readingand picture naming in Italian. Memory & Cognition, 29(7),986e999.
Bolger, D. J., Perfetti, C. A., & Schneider, W. (2005). Cross-culturaleffect on the brain revisited: universal structures plus writingsystem variation. Human Brain Mapping, 25(1), 92e104.
Booth, J. R., Burman,D.D.,Meyer, J. R., Gitelman,D. R., Parrish, T. B.,& Mesulam, M. M. (2002a). Functional anatomy of intra- andcross-modal lexical tasks. NeuroImage, 16(1), 7e22.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6104
Author's personal copy
Booth, J. R., Burman, D. D., Meyer, J. R., Gitelman, D. R.,Parrish, T. B., & Mesulam, M. M. (2002b). Modalityindependence of word comprehension. Human Brain Mapping,16(4), 251e261.
Booth, J. R., Lu, D., Burman, D. D., Chou, T. L., Jin, Z., Peng, D. L.,et al. (2006). Specialization of phonological and semanticprocessing in Chinese word reading. Brain Research, 1071(1),197e207.
Cattinelli, I., Borghese, N. A., Gallucci, M., & Paulesu, E. (2013).Reading the reading brain: a new meta-analysis of functionalimaging data on reading. Journal of Neurolinguistics, 26, 214e238.
Chen, X., Hao, M. L., Geva, E., Zhu, J., & Shu, H. (2009). The role ofcompound awareness in Chinese children’s vocabularyacquisition and character reading. Reading and Writing, 22(5),615e631.
Chow, B. W. Y., McBride-Chang, C., & Burgess, S. (2005).Phonological processing skills and early reading abilities inHong Kong Chinese kindergarteners learning to read Englishas a second language. Journal of Educational Psychology, 97(1),81e87.
DeFrancis, J. (1989). Visible speech: The diverse oneness of writingsystems. Honolulu, HI: University of Hawaii Press.
Demonet, J. F., Chollet, F., Ramsay, S., Cardebat, D.,Nespoulous, J. L., Wise, R., et al. (1992). The anatomy ofphonological and semantic processing in normal subjects.Brain, 115(Pt 6), 1753e1768.
Devlin, J. T., Matthews, P. M., & Rushworth, M. F. (2003). Semanticprocessing in the left inferior prefrontal cortex: a combinedfunctional magnetic resonance imaging and transcranialmagnetic stimulation study. Journal of Cognitive Neuroscience,15(1), 71e84.
Fiez, J. A., Balota, D. A., Raichle, M. E., & Petersen, S. E. (1999).Effects of lexicality, frequency, and spelling-to-soundconsistency on the functional anatomy of reading. Neuron,24(1), 205e218.
Friston, K. J., Ashburner, J., Frith, C. D., Poline, J. B., Heather, J. D.,& Frackowiak, R. S. J. (1995). Spatial registration andnormalization of images. Human Brain Mapping, 3(3), 165e189.
Frost, R. (2012). Towards a universal model of reading. Behavioraland Brain Science, 35(5), 263e279.
Frost, S. J., Mencl, W. E., Sandak, R., Moore, D. L., Rueckl, J. G.,Katz, L., et al. (2005). A functional magnetic resonance imagingstudy of the tradeoff between semantics and phonology inreading aloud. NeuroReport, 16(6), 621e624.
Genovese, C. R., Lazar, N. A., & Nichols, T. (2002). Thresholding ofstatistical maps in functional neuroimaging using the falsediscovery rate. NeuroImage, 15(4), 870e878.
Gough, P. M., Nobre, A. C., & Devlin, J. T. (2005). Dissociatinglinguistic processes in the left inferior frontal cortex withtranscranial magnetic stimulation. Journal of Neuroscience,25(35), 8010e8016.
Graves, W. W., Desai, R., Humphries, C., Seidenberg, M. S., &Binder, J. R. (2010). Neural systems for reading aloud: amultiparametric approach. Cerebral Cortex, 20(8), 1799e1815.
Harm, M. W., & Seidenberg, M. S. (2004). Computing the meaningsof words in reading: cooperative division of labor betweenvisual and phonological processes. Psychological Review, 111(3),662e720.
Ho, C. S. H., Chan, D. W., Chung, K. K. H., Lee, S. H., &Tsang, S. M. (2007). In search of subtypes of Chinesedevelopmental dyslexia. Journal of Experimental ChildPsychology, 97(1), 61e83.
Ho, C. S. H., Chan, D. W. O., Lee, S. H., Tsang, S. M., & Luan, V. V. H.(2004). Cognitive profiling and preliminary subtyping inChinese developmental dyslexia. Cognition, 91(1), 43e75.
Ho, C. S. H., Law, T. P. S., & Ng, P. M. (2000). The phonologicaldeficit hypothesis in Chinese developmental dyslexia. Readingand Writing, 13(1e2), 57e79.
Holmes, A. P., & Friston, K. J. (1998). Generalizability, randomeffects, and population inference. NeuroImage, 7, S34.
Hu, W., Lee, H. L., Zhang, Q., Liu, T., Geng, L. B., Seghier, M. L.,et al. (2010). Developmental dyslexia in Chinese and Englishpopulations: dissociating the effect of dyslexia from languagedifferences. Brain, 133(Pt 6), 1694e1706.
Jared, D. (2002). Spelling-sound consistency and regularityeffects in word naming. Journal of Memory and Language, 46(4),723e750.
Katz, L., Lee, C. H., Tabor, W., Frost, S. J., Mencl, W. E., Sandak, R.,et al. (2005). Behavioral and neurobiological effects of printedword repetition in lexical decision and naming.Neuropsychologia, 43(14), 2068e2083.
Lee, C. Y., Tsai, J. L., Kuo, W. J., Yeh, T. C., Wu, Y. T., Ho, L. T., et al.(2004). Neuronal correlates of consistency and frequencyeffects on Chinese character naming: an event-related fMRIstudy. NeuroImage, 23(4), 1235e1245.
Lee, C. Y., Tsai, J. L., Su, E. C. I., Tzeng, O. J. L., & Hung, D. J. (2005).Consistency, regularity, and frequency effects in namingChinese characters. Language and Linguistics, 6(1), 75e107.
Li, H., Shu, H., McBride-Chang, C., Liu, H. Y., & Peng, H. (2012).Chinese children’s character recognition: visuo-orthographic,phonological processing and morphological skills. Journal ofResearch in Reading, 35(3), 287e307.
Liu, Y., Dunlap, S., Fiez, J., & Perfetti, C. (2007). Evidence for neuralaccommodation to a writing system following learning.Human Brain Mapping, 28(11), 1223e1234.
Liu, Y., Shu, H., & Li, P. (2007). Word naming andpsycholinguistic norms: Chinese. Behavior Research Methods,39(2), 192e198.
Liu, L., Wang, W. J., You, W. P., Li, Y., Awati, N., Zhao, X., et al.(2012). Similar alterations in brain function for phonologicaland semantic processing to visual characters in Chinesedyslexia. Neuropsychologia, 50(9), 2224e2232.
McBride-Chang, C., Shu, H., Zhou, A. B., Wat, C. P., &Wagner, R. K.(2003). Morphological awareness uniquely predicts youngchildren’s Chinese character recognition. Journal of EducationalPsychology, 95(4), 743e751.
McDermott, K. B., Petersen, S. E., Watson, J. M., & Ojemann, J. G.(2003). A procedure for identifying regions preferentiallyactivated by attention to semantic and phonological relationsusing functional magnetic resonance imaging.Neuropsychologia, 41(3), 293e303.
Miezin, F. M., Maccotta, L., Ollinger, J. M., Petersen, S. E., &Buckner, R. L. (2000). Characterizing the hemodynamicresponse: effects of presentation rate, sampling procedure,and the possibility of ordering brain activity based on relativetiming. NeuroImage, 11(6 Pt 1), 735e759.
Nelson, J. R., Liu, Y., Fiez, J., & Perfetti, C. A. (2009). Assimilationand accommodation patterns in ventral occipitotemporalcortex in learning a second writing system. Human BrainMapping, 30(3), 810e820.
Papademetris, X., Jackowski, A. P., Schultz, R. T., Staib, L. H., &Duncan, J. S. (2003). Computing 3D non-rigid brain registrationsusing extended robust point matching for compositemultisubject fMRI analysis. In R. E. Ellis, & T. M. Peters (Eds.),Medical image computing and computer assisted intervention (pp.788e795). Berlin, Germany: Springer-Verlag.
Paulesu, E., Demonet, J. F., Fazio, F., McCrory, E., Chanoine, V.,Brunswick, N., et al. (2001). Dyslexia: cultural diversity andbiological unity. Science, 291(5511), 2165e2167.
Paulesu, E., McCrory, E., Fazio, F., Menoncello, L., Brunswick, N.,Cappa, S. F., et al. (2000). A cultural effect on brain function.Nature Neuroscience, 3(1), 91e96.
Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K.(1996). Understanding normal and impaired word reading:computational principles in quasi-regular domains.Psychological Review, 103(1), 56e115.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6 105
Author's personal copy
Poldrack, R. A., & Gabrieli, J. D. (2001). Characterizing the neuralmechanisms of skill learning and repetition priming: evidencefrom mirror reading. Brain, 124(Pt 1), 67e82.
Poldrack, R. A., Wagner, A. D., Prull, M. W., Desmond, J. E.,Glover, G. H., & Gabrieli, J. D. (1999). Functional specializationfor semantic and phonological processing in the left inferiorprefrontal cortex. NeuroImage, 10(1), 15e35.
Price, C. J., Moore, C. J., Humphreys, G. W., & Wise, R. J. S. (1997).Segregating semantic from phonological processes duringreading. Journal of Cognitive Neuroscience, 9(6), 727e733.
Pugh, K. R., Frost, S. J., Sandak, R., Landi, N., Rueckl, J. G.,Constable, R. T., et al. (2008). Effects of stimulus difficulty andrepetition on printed word identification: an fMRI comparisonof nonimpaired and reading-disabled adolescent cohorts.Journal of Cognitive Neuroscience, 20(7), 1146e1160.
Pugh, K. R., Mencl, W. E., Jenner, A. R., Katz, L., Frost, S. J.,Lee, J. R., et al. (2000). Functional neuroimaging studies ofreading and reading disability (developmental dyslexia).Mental Retardation and Developmental Disabilities ResearchReviews, 6(3), 207e213.
Pugh, K. R., Shaywitz, B. A., Shaywitz, S. E., Constable, R. T.,Skudlarski, P., Fulbright, R. K., et al. (1996). Cerebralorganization of component processes in reading. Brain, 119(Pt4), 1221e1238.
Rueckl, J. G. (2010). Connectionism and the role of morphologyin visual word recognition. The Mental Lexicon, 5(3), 371e400.
Sandak, R., Mencl, W. E., Frost, S. J., Rueckl, J. G., Katz, L.,Moore, D. L., et al. (2004). The neurobiology of adaptivelearning in reading: a contrast of different training conditions.Cognitive, Affective, & Behavioral Neuroscience, 4(1), 67e88.
Seidenberg, M. S. (1992). Beyond orthographic depth in reading:equitable division of labor. In R. Frost, & L. Katz (Eds.),Orthography, phonology, morphology, and meaning (pp. 85e118).Oxford, England: North-Holland.
Seidenberg, M. S. (2011). Reading in different writing systems: onearchitecture, multiple solutions. In P. McCardle, B. Miller,J. R. Lee, & O. Tzeng (Eds.), Dyslexia across languages:Orthography and the gene-brain-behavior link (pp. 149e174). PaulBrookes Publishing.
Seidenberg, M. S., & Gonnerman, L. M. (2000). Explainingderivational morphology as the convergence of codes. Trendsin Cognitive Sciences, 4(9), 353e361.
Shibahara, N., Zorzi, M., Hill, M. P., Wydell, T., & Butterworth, B.(2003). Semantic effects in word naming: evidence fromEnglish and Japanese Kanji. Quarterly Journal of ExperimentalPsychology Section A, 56(2), 263e286.
Shu, H., Chen, X., Anderson, R. C., Wu, N., & Xuan, Y. (2003).Properties of school Chinese: implications for learning to read.Child Development, 74(1), 27e47.
Shu, H., McBride-Chang, C., Wu, S., & Liu, H. Y. (2006).Understanding Chinese developmental dyslexia:morphological awareness as a core cognitive construct. Journalof Educational Psychology, 98(1), 122e133.
Shu, H., Meng, X. Z., Chen, X., Luan, H., & Cao, F. (2005). Thesubtypes of developmental dyslexia in Chinese: evidence fromthree cases. Dyslexia, 11(4), 311e329.
Siok, W. T., & Fletcher, P. (2001). The role of phonologicalawareness and visual-orthographic skills in Chinese readingacquisition. Developmental Psychology, 37(6), 886e899.
Siok, W. T., Jin, Z., Fletcher, P., & Tan, L. H. (2003). Distinct brainregions associated with syllable and phoneme. Human BrainMapping, 18(3), 201e207.
Siok, W. T., Perfetti, C. A., Jin, Z., & Tan, L. H. (2004). Biologicalabnormality of impaired reading is constrained by culture.Nature, 431(7004), 71e76.
Strain, E., Patterson, K., & Seidenberg, M. S. (1995). Semantic effectsin single-word naming. Journal of Experimental Psychology:Learning, Memory, and Cognition, 21(5), 1140e1154.
Strain, E., Patterson, K., & Seidenberg, M. S. (2002). Theories ofword naming interact with spelling-sound consistency. Journalof Experimental Psychology: Learning, Memory, and Cognition,28(1), 207e214.
Tan, L. H., Feng, C. M., Fox, P. T., & Gao, J. H. (2001). An fMRI studywith written Chinese. NeuroReport, 12(1), 83e88.
Tan, L. H., Laird, A. R., Li, K., & Fox, P. T. (2005). Neuroanatomicalcorrelates of phonological processing of Chinese charactersand alphabetic words: a meta-analysis. Human Brain Mapping,25(1), 83e91.
Tan, L. H., Liu, H. L., Perfetti, C. A., Spinks, J. A., Fox, P. T., &Gao, J. H. (2001). The neural system underlying Chineselogograph reading. NeuroImage, 13(5), 836e846.
Tan, L. H., Spinks, J. A., Feng, C. M., Siok, W. T., Perfetti, C. A.,Xiong, J., et al. (2003). Neural systems of second languagereading are shaped by native language. Human Brain Mapping,18(3), 158e166.
Taylor, J. S., Rastle, K., & Davis, M. H. (2013). Can cognitive modelsexplain brain activation during word and pseudowordreading? A meta-analysis of 36 neuroimaging studies.Psychological Bulletin, 139(4), 766e791.
Wang, H. (1986). Modern Chinese frequency dictionary. Beijing:Beijing Language Institute Press.
Woods, R. P. (1996). Modeling for intergroup comparisons ofimaging data. NeuroImage, 4(3 Pt 3), S84eS94.
Woollams, A. M. (2005). Imageability and ambiguity effects inspeeded naming: convergence and divergence. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 31(5),878e890.
Wu, C. Y., Ho, M. H. R., & Chen, S. H. A. (2012). A meta-analysis offMRI studies on Chinese orthographic, phonological, andsemantic processing. NeuroImage, 63(1), 381e391.
Yang, J., Shu, H., McCandliss, B. D., & Zevin, J. D. (2012).Orthographic influences on division of labor in learning toread Chinese and English: insights from computationalmodeling. Bilingualism: Language and Cognition, 16(2),354e366.
Zhao, J. (2012). The influence of statistical systematicities on learning toread: Studies with artificial orthographies (Doctoral dissertation).Retrieved from ProQuest Dissertations and Theses database.(UMI No. 3533881).
Zhao, J., Guo, J., Zhou, F., & Shu, H. (2011). Time course of Chinesemonosyllabic spoken word recognition: evidence from ERPanalyses. Neuropsychologia, 49(7), 1761e1770.
c o r t e x 5 3 ( 2 0 1 4 ) 9 0e1 0 6106