Visual Similarity of Words Alone Can Modulate Hemispheric Lateralization in Visual Word Recognition: Evidence From Modeling Chinese Character Recognition.

Cognitive Science 40 (2016) 351–372 Copyright © 2015 Cognitive Science Society, Inc. All rights reserved. ISSN: 0364-0213 print / 1551-6709 online DOI: 10.1111/cogs.12233

Visual Similarity of Words Alone Can Modulate Hemispheric Lateralization in Visual Word Recognition: Evidence From Modeling Chinese Character Recognition Janet H. Hsiao,a Kit Cheungb a

Department of Psychology, University of Hong Kong Department of Computing, Imperial College London

b

Received 20 January 2013; received in revised form 5 November 2014; accepted 19 December 2014

Abstract In Chinese orthography, the most common character structure consists of a semantic radical on the left and a phonetic radical on the right (SP characters); the minority, opposite arrangement also exists (PS characters). Recent studies showed that SP character processing is more left hemisphere (LH) lateralized than PS character processing. Nevertheless, it remains unclear whether this is due to phonetic radical position or character type frequency. Through computational modeling with artificial lexicons, in which we implement a theory of hemispheric asymmetry in perception but do not assume phonological processing being LH lateralized, we show that the difference in character type frequency alone is sufficient to exhibit the effect that the dominant type has a stronger LH lateralization than the minority type. This effect is due to higher visual similarity among characters in the dominant type than the minority type, demonstrating the modulation of visual similarity of words on hemispheric lateralization. Keywords: Hemispheric asymmetry; Word type frequency; Chinese character recognition; Computational modeling

1. Introduction Phonetic compounds are a major type of characters in Chinese orthography. They comprise about 81% of the 7,000 frequent characters in a Chinese dictionary (Li & Kang, 1993). A phonetic compound consists of a semantic radical and a phonetic radical. The semantic radical usually has information about the character meaning, whereas the phonetic radical usually has information about the character pronunciation. Most phonetic Correspondence should be sent to Janet Hui-wen Hsiao, Department of Psychology, University of Hong Kong, Pokfulam Road, Hong Kong. E-mail: [email protected]

352

J. H. Hsiao, K. Cheung / Cognitive Science 40 (2016)

compounds have a left-right structure; the most common (dominant) structure consists of a semantic radical on the left and a phonetic radical on the right (SP characters). This character type comprises about 90% of all left-right structured phonetic compounds. The opposite, minority (10%) arrangement also exists (PS characters; Hsiao & Shillcock, 2006; Fig. 1). Recent studies have suggested that SP and PS characters are processed differently in the brain. For example, Hsiao, Shillcock, and Lee (2007) showed that in a homophone judgment task, SP characters elicited larger N170 ERP amplitude in the left hemisphere (LH) than the right hemisphere (RH), whereas PS characters elicited similar N170 amplitude in both hemispheres. Consistent with this finding, Hsiao and Liu (2010) showed a right visual field (RVF)/LH advantage in naming SP characters and no hemispheric lateralization in naming PS characters. This result suggests that SP character processing is more LH lateralized than PS character processing. The authors argued that this effect may be due to the dominance of SP characters in the lexicon that makes the readers opt to obtain phonological information from the right side of a character, and thus the automaticity of phonological processing in the recognition of SP characters is superior to PS characters; consequently SP character processing involves more LH phonological modulation (Maurer & McCandliss, 2007; see also Weekes & Zhang, 1999). Alternatively, this effect may also be simply a perceptual effect due to hemispheric asymmetry in perception, rather than LH-lateralized phonological processing.1 In recent years, hemispheric asymmetry in processing local and global features has been consistently reported. In general, there is a LH advantage for processing local features, and a RH advantage in processing global features (e.g., Sergent, 1982; Ivry & Robertson, 1998; see Van Kleeck, 1989, for a review). To account for this difference, it has been proposed that the LH has an advantage in processing high spatial frequency (HSF) information, whereas the RH has an advantage in processing low spatial frequency (LSF) information (e.g., Han et al., 2002; Hsiao, Cipollini, & Cottrell, 2013; Ivry & Robertson, 1998; Keenan, Whitman, & Pepe, 1989; Kitterle, Christman, & Hellige, 1990; Sergent, 1982).2 According to this observation, it is possible that characters in the most common, dominant type (i.e., SP characters) look visually more similar to one another than those in the minority type (i.e., PS characters), and thus more HSF information is required to distinguish SP characters as compared with PS characters, which leads to a stronger LH lateralization in SP character processing (Hsiao & Lam, 2013).

Fig. 1. Examples of Chinese SP and PS characters. The two characters have the same phonetic radical and the same pronunciation [cai3] in Pinyin; their phonetic radicals also have the same pronunciation [cai3]. SP characters are a dominant character type in Chinese orthography, which comprise about 90% of all left-right structured phonetic compounds.


353

Thus, to investigate whether this lateralization difference between SP and PS characters is due to LH-lateralized phonological processing or simply a perceptual effect due to hemispheric asymmetry in perception, here we examine whether this effect can emerge in a computational model that implements hemispheric asymmetry in perception, without assuming LH-lateralized phonological processing. We use a model (Hsiao, Shieh, & Cottrell, 2008) that implements a theory of hemispheric asymmetry in perception, the Double Filtering by Frequency (DFF) theory (Ivry & Robertson, 1998). The theory posits that visual information coming into the brain goes through two frequency-filtering stages: the first stage involves selection of a task-relevant frequency range, and at the second stage, the LH amplifies HSF information, whereas the RH amplifies LSF information. We introduce our model below. In the human visual pathway, the visual field is divided vertically into two hemifields, which are initially contralaterally projected to different hemispheres. Hsiao, Shieh, et al. (2008) conducted a computational modeling study aiming to examine at which processing stage this split information converges. They proposed three models with different timings of convergence: early, intermediate, and late convergence models, and examine whether the model could account for the left side bias effect in face perception. This effect refers to the phenomenon that a chimeric face made from two left half-faces (from the viewer’s perspective) is usually judged more similar to the original face than the one made from two right half-faces, and it has been argued to be an indication of RH involvement in face processing (e.g., Brady, Campbell, & Flaherty, 2005; Gilbert & Bakan, 1973). The authors showed that both the intermediate and late convergence models were able to account for this effect, whereas the early convergence model was not (Fig. 2A shows the architecture of the intermediate convergence model). Hsiao, Shieh, et al. (2008) model adopted several known observations about visual anatomy and neural computation. First, Gabor filter responses over the input images were used to simulate neural responses of cells in the early visual system (Lades et al., 1993). Gabor filters are linear filters used in image processing. They have been commonly used in computational models of visual processing because their spatial frequency and orientation representations are similar to those observed in the early visual system (e.g., Dailey, Cottrell, Padgett, & Adolphs, 2002; Keator et al., 2014). Principal Component Analysis (PCA), a biologically plausible linear compression technique to reduce dimensionality of data (e.g., Sanger, 1989), was used to simulate possible information extraction processes in the lateral occipital region beyond the early visual area. This PCA representation was then used as the input to a two-layer neural network. They also implemented the DFF theory in the model by using two sigmoid weighting functions to assign different weights to the Gabor filter responses in the two hemispheres (Fig. 2B). As shown in Fig. 2B, the Gabor filter responses in the RH was biased to low spatial frequency information by assigning more weights to the responses in the low spatial frequency range, and vice versa for those in the LH. Here, we apply Hsiao, Shieh, et al. (2008)’s intermediate convergence model to Chinese character recognition and examine whether the hemispheric lateralization difference between SP and PS character processing can emerge purely due to their difference in

354


(A)

xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx

(B)

Fig. 2. (A) Details of the model architecture, based on Hsiao, Shieh, et al.’s (2008) hemispheric processing model. Note that the visual word form area (VWFA) is typically left-lateralized, and recent evidence has shown that the right homologue of the VWFA is also involved in visual word recognition (Rauschecker, Bowen, Parvizi, & Wandell, 2012). In the current model the hidden layer (by analogy with the VWFA) was not split into two halves. How a split hidden layer here influences Chinese character processing requires further examinations (i.e., the split fovea model; see Hsiao & Shillcock, 2005). (B) Sigmoidal weighting functions to give different weights to the spatial frequency scales in the RH and the LH respectively (unbiased condition, a = 0; biased conditions, a = 1.5).

character type frequency in the lexicon, instead of difference in their involvement in LH lateralized phonological processing. Hsiao and Lam (2013; see also Cheung & Hsiao, 2010) recently used the same model and demonstrated that visual and task characteristics of a writing system alone could account for differences in hemispheric asymmetry in visual word recognition between different languages, without assuming LH-lateralized phonological processing. More specifically, by using nonexisting English words, they showed that (a) the more similar the words look in the lexicon, the more HSFs are required to distinguish them, leading to a stronger LH lateralization and (b) alphabetic mapping from orthography to phonology requires more HSFs than logographic mapping, resulting in a stronger LH lateralization. In alphabetic mapping, visual word recognition


355

involves a systematic mapping from word components to corresponding components in the pronunciation (i.e., grapheme-phoneme correspondence). In this condition, word recognition requires decomposition of visual word input into components in order to map them to corresponding phonemes, and thus requires more HSFs. In contrast, in logographic mapping, each word input is mapped to a pronunciation at the syllable level and there is no systematic mapping between visual word components and pronunciation components. Consequently, word recognition in this condition does not require decomposition of word input into components, and thus is more RH (LSF) lateralized. This result is consistent with the literature: Visual word form processing/recognition in alphabetic languages usually shows a RVF advantage and LH lateralization (e.g., English: Bryden & Rainey, 1963; Measso & Zaidel, 1990; Rossion, Joyce, Cottrell, & Tarr, 2003; McCandliss, Cohen, & Dehaene, 2003; Spanish: Hernandez, Nieto, & Barroso, 1992; Hebrew: Bentin, 1981; Japanese phonetic script kana: Hanavan & Coney, 2005; Korean Hangul words: Endo, Shimizu, & Nakamura, 1981), in contrast with the processing of Chinese characters, a logographic language, which usually shows a left visual field (LVF) advantage in orthographic processing (e.g., Cheng & Yang, 1989; Leong, Wong, Wong, & Hiscock, 1985; Tzeng, Hung, Cotton, & Wang, 1979; Yang & Cheng, 1999) and bilateral/ RH lateralization in the visual system in brain imaging studies (according to a metaanalysis by Tan, Laird, Li, & Fox, 2005; see also Chen, Fu, Iversen, Smith, & Matthews, 2002; Fu, Chen, Smith, Iversen, & Matthews, 2002; Kuo et al., 2001; Tan et al., 2000, 2001; Siok, Spinks, Jin, & Tan, 2009). Thus, because of the dominance of the SP structure in the Chinese lexicon, it is possible that SP characters look visually more similar to one another since they in general share more common radicals as compared with PS characters, leading to a higher demand on HSF information and thus a stronger LH lateralization. Here, we test this hypothesis through computational modeling, since models give us good control over variables that are difficult to tease apart in human studies. In the Chinese orthography, in addition to character type frequency (i.e., the SP structure is a much more common structure than the PS structure), SP and PS characters differ in at least three other aspects: (1) Position of the phonetic radical: The phonetic radical of an SP character appears on the right, whereas that of a PS character is on the left. (2) Character configuration in terms of visual complexity: Since semantic radicals usually have fewer strokes than phonetic radicals, SP characters tend to have a right-heavy (i.e., more strokes/features on the right) configuration, whereas PS characters usually have a left-heavy configuration. (3) Character information structure (in terms of entropy in information theory, a measure of uncertainty/unpredictability of a variable): Since there are more phonetic radical types than semantic radical types (the ratio is about 8 to 1; Hsiao & Shillcock, 2006), the phonetic radical of a character typically has more information about the identity of the character. Consequently, in SP characters there is more information on the right, whereas in PS characters there is more information on the left (see note 2). Thus, to examine the influence of character type frequency on hemispheric lateralization in visual word recognition, in Experiment 1 we create artificial lexicons with nonexisting SP and PS characters consisting of real Chinese character components and manipulate their character type fre-

356


quency (equal frequency vs. SP dominant vs. PS dominant) with all the other three factors controlled. Since most Chinese characters are left-right structured phonetic compound characters, to examine how position of the phonetic radical influences lateralization (factor (1)), in Experiment 2 in addition to character type frequency, we also manipulate position of the phonetic radical, with both radical visual complexity (factor (2)) and character information structure (factor (3)) controlled. The model performs a pronunciation task, that is, to map a character visual input to its pronunciation in the output. In Experiment 1, in order to understand whether the observed difference between the processing of the dominant and minority character types only applies to phonetic compound characters, in the simulations, two mapping conditions are created: In the regular mapping condition, characters with the same phonetic radical have the same pronunciations; that is, similar to (regular) phonetic compound pronunciations, there is a systematic mapping between the pronunciation of a character and the pronunciation of its phonetic radical. In contrast, in the logographic mapping condition, we map each character to a randomly assigned pronunciation. By comparing these two conditions, we are able to examine whether the lateralization difference between the two character types is related to the presence of a regular phonetic radical in the characters, which may encourage the model/reader to decompose a character into radicals for sublexical phonological processing. According to Hsiao and Lam (2013), this decomposition may lead to a higher demand on HSF information/LH perceptual processing. If similar lateralization differences between SP and PS character processing are observed in the regular and logographic mapping condition, this result will suggest that the lateralization difference is mainly driven by visual similarity among characters and thus is not restricted to phonetic compounds. In Experiment 2, since we aim to manipulate position of the phonetic radical in the character pronunciation task, only the regular mapping condition is used.

2. Modeling methods The model was trained to perform a print-to-sound classification task. Following Hsiao, Shieh, et al. (2008), in our model we first applied a 10 9 10 rigid grid of 2D Gabor filters (Daugman, 1985) to the images of SP and PS characters we created; the image size was 60 9 60 pixels. At each grid point we applied Gabor filters with eight orientations and five frequency scales (see also Hsiao & Lam, 2013). As introduced earlier, Gabor filters are linear filters for image processing, and have been commonly used in computational modeling of visual processing to simulate responses of neurons in the early visual areas (e.g., Dailey et al., 2002; Lades et al., 1993). To simulate the low and high spatial frequency biases in the RH and the LH, respectively (i.e., the DFF theory), Gabor filter responses in different frequency scales were subsequently given different weights. Two conditions were created: (a) the unbiased condition, in which Gabor responses in all scales were given equal weights in both visual hemifields, and (b) the biased condition, in which larger weights were given to HSF responses in the RVF/LH, and larger weights


357

were given to LSF responses in the LVF/RH (the second stage of the DFF; the weights were determined by a sigmoid function; see Fig. 2B). The difference in the model’s behavior between the two conditions thus reflected the influence of the spatial frequency bias (based on the DFF theory) on the lateralization effects. The LVF and RVF Gabor representations were then compressed by PCA (a biologically plausible linear compression technique) separately into two 50-element representations (100 elements in total, following Hsiao, Shieh, et al. (2008); Hsiao & Lam, 2013), to reduce the dimensionality of the Gabor representations. The PCA representations were then normalized to equalize the contribution of each element. The normalized PCA representation was then used as the input to a two-layer neural network. The network was trained to recognize input images by mapping them to corresponding pronunciations at the syllable level (the pronunciation of each Chinese character only has one syllable). There were 15 nodes in the output layer, corresponding to pronunciations of 15 phonetic radicals. Each character had the same pronunciation as its phonetic radical.3 The training continued until the performance on the training set reached mean-square-error 0.001 or until it could not be further improved. Gradient descent with an adaptive learning rate was used as the training algorithm (implemented by Matlab Neural Network Toolbox): The learning rate was increased by 5% if the performance improvement (measured by summed squared errors) between the two most recent runs was smaller than the previous one, and decreased by 30% if it was larger than the previous one. To measure lateralization effects, following previous studies with the same model (Hsiao & Lam, 2013; Lam & Hsiao, 2014; cf. Hsiao, Shieh, et al., 2008), we presented either only the LH PCA or only the RH PCA representation of the test images by setting the other representation to zeros (note that this is different from the divided visual field paradigm used in behavioral studies such as in Hsiao & Liu, 2010). A LH lateralization index was defined as the accuracy difference between pronouncing correctly a LH/RVF lateralized character and pronouncing correctly a RH/LVF lateralized character. A higher index hence corresponded to a stronger LH lateralization and indicated the model’s reliance on HSF information.

3. Experiment 1 Here, we examined the influence of character type frequency on hemispheric lateralization of character processing, with all the other three factors controlled: (1) position of the phonetic radical, (2) character configuration (i.e., visual complexity of the radicals), and (3) character information structure (defined by predictability of the radicals). We created artificial lexicons with nonexisting characters that consisted of existing radicals and manipulated the proportions of SP and PS characters. Each character was generated by concatenating two mirror-symmetric radicals. To control for factor (1), characters formed a top-bottom configuration (Fig. 3) so that the phonetic radical was not biased to either the LVF or RVF. One of the two radicals was randomly chosen from a “P list” (phonetic radicals), while the other was randomly chosen from an “S1 list” (semantic radicals) if it

358


Fig. 3. Examples of the stimuli used in Experiment 1 (top-bottom structured characters) and Experiment 2 (left-right structured characters).

was placed on the top (SP characters, for the current purpose), or from an “S2 list” if it was placed on the bottom (PS characters). Each list contained 15 radicals, so there were 15 9 15 (225) possible combinations in total for either SP or PS characters. This was to simulate the Chinese orthography in which SP and PS characters may have the same phonetic radicals, but the semantic radicals are usually different. Each artificial lexicon contained 100 characters. The SP characters were randomly selected from different combinations of S1-list radicals and P-list radicals; the semantic radicals of the PS characters (S2-list) had the same combinations with P-list radicals as those in the SP characters. Thus, the radical combinability in the SP and the PS characters was well matched. When a majority type of characters existed, the combinations used for the minority type was randomly selected from those for the majority type. Note that as the result of this random sampling, the radical combinability (i.e., how many different radicals can be combined with a given radical to form a character in the lexicon) in the minority type in general was reduced as compared with that in the majority type. The radical choices in S1 and S2 lists were counterbalanced across simulation runs. The radicals used were all existing stroke patterns in Chinese, and the numbers of strokes of the radicals in the three lists were matched (about 6.4 strokes on average; factor (2)). The same number of radicals was used for the P list and S lists, so that the amount of information for recognition defined by entropy on either side of the characters was balanced (i.e., factor (3)). Mirror images were used in half of the simulation runs to counterbalance possible low-level featural differences between the two sides of the characters. There were three variables in the design: Spatial frequency (SF) bias (unbiased vs. biased), character type (SP vs. PS), and type frequency (SP Dominant [90% SP, 10% PS] vs. PS Dominant [10% SP, 90% PS] vs. equal frequency [50% SP, 50% PS]).4 The dependent variable was LH lateralization index. For each condition, we ran the model with 20 different lexicons; for each character there were eight images in different fonts, with four of them used for training and the other four for testing (counterbalanced across simulation runs). For each lexicon, 10 different sets of network weight initialization were used, and the average accuracy over the 10 runs was used for data analysis. This was to minimize the fluctuation in accuracy caused by the random weight initialization. These 10 sets of weight initialization were the same across different conditions to closely match the models.


359

3.1. Results The results showed a significant SF bias effect (F(1, 39) = 3,364.903, p < .001; Fig. 4, top row): In the biased condition, the model exhibited less LH lateralization (i.e., stronger RH/LSF preference) than in the unbiased condition, consistent with the literature showing RH (LSF) lateralization in Chinese character visual word form processing (e.g., Hsiao & Cottrell, 2009; Tan et al., 2005; Tzeng et al., 1979). In addition, there was an interaction between SF bias, character type, and type frequency (F(1.558, 60.753) = 168.235, p < .001; Fig. 4, top row); in the unbiased condition, there was no effect of character type or type frequency (F < 1). In contrast, in the biased condition, there was an interaction between character type and type frequency (F(1.411, 55.016) = 222.974, p < .001): When SP and PS characters had an equal type frequency, there was no difference in lateralization between SP and PS character processing; when SP type was dominant, SP character processing was more LH lateralized than PS character processing (t(39) = 16.081, p < .001), whereas when PS type was dominant, PS character processing was more LH lateralized than SP character processing (t(39) = 12.031, p < .001). Thus, consistent with our prediction, the dominant type had a stronger LH (HSF) lateralization. To test whether the effects observed here only applied to phonetic compound characters, which had a phonetic radical that biased phonological information to one component and thus may encourage character decomposition, in a separate simulation, instead of mapping each character systematically to the pronunciation of its phonetic radical, we mapped each character to a randomly assigned pronunciation in the output layer (i.e., logographic mapping). The results (Fig. 4, bottom row) showed a similar interaction between SF bias, character type, and type frequency (F(2, 72) = 79.881, p < .001). No effect of character type or type frequency was observed in the unbiased condition; in contrast, in the biased condition, there was an interaction between character type and type frequency (F(2, 78) = 107.903, p < .001): The dominant character type had a stronger LH lateralization than the minority type. This result suggested that the type frequency effect does not require the existence of a phonetic radical; it emerged when the phonological information was not biased to one radical and thus no decomposition was required. This result suggests that the lateralization difference between the dominant and minority character types is not restricted to phonetic compound characters. When we compared the two mapping methods (regular vs. logographic mappings) in the biased condition, there was a significant main effect of mapping method (F(1, 78) = 64.989, p < .001): The regular mapping condition had a stronger LH lateralization than the logographic mapping condition. This effect suggests that when the phonological information was biased to one radical, and thus decomposition of a character into radicals might be required, the model in general had a stronger LH (HSF) lateralization.5 To examine whether the observed lateralization difference between SP and PS character processing was related to visual similarity among characters in the lexicon, we conducted a further analysis to compare the visual similarity among characters in the dominant type with that in the minority type. The single linkage (shortest distance) method (e.g., Mardia, Kent, & Bibbi, 1980) was used to calculate the shortest distance

360

J. H. Hsiao, K. Cheung / Cognitive Science 40 (2016) (A)

(B)

Fig. 4. Results of Experiment 1: (A) The regular mapping condition, in which the pronunciation of a character was the same as its phonetic radical. (B) The logographic mapping condition, in which there was no systematic mapping between the pronunciation of a character and its phonetic radical.


361

between the Gabor responses of a character and all the other characters in a character type, as a measure of how dissimilar this character was to the other characters (Hsiao & Lam, 2013). The average of this shortest distance over all characters in a character type was used as the dissimilarity measure for the character type in the lexicon. The results showed that in our artificial lexicons, characters in the dominant type had a lower dissimilarity (i.e., higher similarity) among one another compared with those in the minority type (SP dominant: t(60.348) = 7.829, p < .01; PS dominant: t(60.391) = 9.686, p < .001). In addition, in both the regular and logographic mapping conditions, the LH lateralization effect in the model (in the SF biased condition) was significantly correlated with the dissimilarity measure: The more similar the characters were in a character type, the stronger the LH lateralization was in processing the character type (Regular Mapping: r = .691, p < .001; Logographic Mapping: r = .737, p < .001). This effect suggests that the level of LH lateralization in different character types was mainly due to visual similarity among characters. Consistent with this speculation, we found that SP characters in the real Chinese lexicon (Hsiao & Shillcock, 2006) have higher visual similarity than PS characters using the same similarity measure (t(239.119) = 6.181, p < .001; in total there are 1,942 SP and 217 PS characters in the database). These results suggest that the difference in lateralization between SP and PS character processing observed in Chinese readers may be simply due to the fact that SP characters have higher visual similarity to one another as compared with PS characters, instead of the automaticity of LH phonological processing in SP character recognition as previously speculated.

4. Experiment 2 Here, we aimed to examine the influence from position of the phonetic radical (left vs. right; i.e. factor (1)) on the lateralization effect in addition to character type frequency. We used the same experiment settings and radicals as Experiment 1, except that the radicals were arranged horizontally to form left-right structures (Fig. 3). Since the position of the phonetic radical was manipulated, no mirror images were used in the current simulation, and only the regular mapping condition was used (i.e., characters consistently had the same pronunciation as their phonetic radicals). 4.1. Results Similar to Experiment 1, the results showed a significant SF bias effect (F(1, 39) = 2,434.803, p < .001): In the biased condition, the model exhibited less LH lateralization than in the unbiased condition. In contrast with Experiment 1, there was a significant character type effect (F(1, 39) = 6,804.565, p < .001). As shown in Fig. 5A, in all SF bias and character type frequency conditions, SP character processing was strongly lateralized to the LH, whereas PS character processing was strongly lateralized to the RH. This effect was observed in both SF biased and unbiased conditions, suggesting that the effect did not rely on hemispheric difference in SF processing. Since this effect was

362

J. H. Hsiao, K. Cheung / Cognitive Science 40 (2016) (A)

Unbiased

Biased

Mean LH Lateralization

0.6 0.4 0.2 0.0

-0.2 -0.4 -0.6 -0.8

(B)

Fig. 5. Results of Experiment 2: (A) Overall results. (B) LH lateralization difference between SP and PS characters in the SP dominant and PS dominant conditions: While no significant difference between the SP and PS dominant conditions was observed in the SF unbiased condition, in the biased condition the LH lateralization difference between SP and PS characters was reduced in the PS dominant condition, demonstrating the influence from character type frequency, that is, reduced LH lateralization in the minority character type due to reduced visual similarity among characters.

not observed in Experiment 1, it was likely due to the modulation of position of the phonetic radical: When an SP character was centrally fixated, the phonetic radical, which contained information about the character pronunciation, was initially projected to and


363

processed in the LH, resulting in a strong LH lateralization effect (vice versa for PS characters). This effect showed that position of the phonetic radical can also modulate hemispheric lateralization in character processing.6 In addition, similar to Experiment 1, there was a significant interaction between SF bias, character type, and type frequency (F(2, 78) = 16.560, p < .001; Fig. 5A). While a significant interaction between character type and type frequency was observed in both the SF biased and unbiased conditions (the unbiased condition: F(2, 78) = 310. 931, p < .001; the biased condition: F(2, 78) = 311.812, p < .001), this three-way interaction between SF bias, character type, and type frequency was driven mainly by the effect that in the PS dominant condition, the LH lateralization difference between SP and PS characters in the biased condition was reduced (as compared with the unbiased condition) the most among the three type frequency conditions (Fig. 5A). To better illustrate the effect, Fig. 5B contrasts the LH lateralization difference between SP and PS characters in the SP dominant and PS dominant conditions. As shown in the figure, in the unbiased condition, the LH advantage in SP characters over PS characters was similar between the two conditions. In contrast, in the biased condition, the LH advantage in SP characters over PS characters was reduced in the PS dominant condition as compared with the SP dominant condition. According to the results in Experiment 1, this effect may be because in the PS dominant condition, the LH lateralization effect due to phonetic radical position was counteracted by the type frequency effect observed in Experiment 1: The effect of phonetic radical position resulted in a stronger LH lateralization in SP than in PS characters (as demonstrated in the unbiased condition in Fig. 5A), whereas the effect of type frequency resulted in a weaker LH lateralization in SP than in PS characters because SP characters were the minority type (as demonstrated in the biased condition in Fig. 4A). This effect suggests that both phonetic radical position and character type frequency can modulate the lateralization effect.

5. Discussion Here, we investigated the modulation of character type frequency on hemispheric lateralization in Chinese character recognition through computational modeling. We contrasted the processing of SP and PS characters in the Chinese orthography. SP characters are the most common (dominant) character structure type, whereas PS characters are a minority type, and the ratio is about 9 to 1. Previous studies have shown that SP character processing involves a stronger LH lateralization than PS character processing (Hsiao & Liu, 2010; Hsiao et al., 2007). Nevertheless, it remains unclear whether this effect is because the processing of the dominant SP structure has stronger automaticity of LH phonological processing than that of the minority PS structure, or simply due to differences in the spatial frequency content that is important for recognizing a dominant or minority character type in the lexicon. More specifically, since a RH advantage in processing LSF information and an LH advantage in processing HSF information have been consistently reported in visual perception (e.g., Ivry & Robertson, 1998), it is possible that more HSF

364


information is required for processing the dominant SP structure than the minority PS structure due to higher visual similarity among SP characters than PS characters, leading to a stronger LH lateralization in processing SP characters. Through computational modeling, in which we implemented a theory of hemispheric asymmetry in perception (the DFF theory; Ivry & Robertson, 1998) that posits a LSF bias in the RH and a HSF bias in the LH, but did not assume a LH lateralized phonological processing center, in Experiment 1 we showed that when all the other possible factors that may influence lateralization were controlled, including (a) position of the phonetic radical, (b) visual complexity of the radicals, and (c) character information structure defined by entropy, hemispheric lateralization differences could emerge purely due to a difference in character type frequency: The dominant character type had a stronger LH (HSF) lateralization than the minority type (in the biased condition, Fig. 4). In addition, this effect did not require the existence of a phonetic radical that biased phonological information to one component, as this effect also emerged in the logographic mapping condition, in which there was no systematic relationship between phonetic radicals and character pronunciations. This result suggests that the observed lateralization difference between the dominant and minority character types is not restricted to phonetic compound characters. Further analysis suggests that this effect was mainly due to higher visual similarity among characters in the dominant type than those in the minority type. With a fixed number of radicals, when we gradually increased the number of characters in a character type, it became more likely that some characters shared common radicals and thus were visually similar to each other. This result showed that visual similarity of characters alone is sufficient to account for the lateralization difference between SP and PS character processing. More specifically, the lateralization difference between SP and PS character processing may be simply due to higher visual similarity among SP characters than that among PS characters, instead of differential involvement in LH-lateralization phonological processing as previously thought (e.g., Hsiao & Liu, 2010; Maurer & McCandliss, 2007). In Experiment 2, we examined the effect of phonetic radical position (factor (1)) on lateralization, in addition to character type frequency. The data showed that position of the phonetic radical modulated the lateralization effect; nevertheless, this modulation effect did not rely on the hemispheric difference in SF processing (i.e., the DFF theory), as it was observed in both the SF biased and unbiased conditions. This is in contrast with the modulation effect of character type frequency observed in Experiment 1, which was only observed in the SF biased condition but not in the unbiased condition. The effect of character type frequency was due to difference in visual similarity among characters between the minority and the dominant character types, which leads to different requirements on SF processing (i.e., the more similar the characters are, the more HSF information is required, leading to stronger LH lateralization). In contrast, the effect of phonetic radical position was due to phonological information being on only one side of a character, making the processing lateralized to one visual hemifield. More specifically, when an SP character is centrally fixated, the phonetic radical is initially projected to and


365

processed in the LH, leading to a stronger LH lateralization effect. This phonetic radical position effect emerges as a result of the split architecture of the model, instead of the SF content that is required for character recognition. Note that the modulation effect of character type frequency was also observed in Experiment 2: in the SF biased condition, the LH advantage in SP characters over PS characters due to the position of the phonetic radical was reduced when SP characters were the minority type, as compared with the other conditions, demonstrating influences from both phonetic radical position and character type frequency effects (Fig. 5B). The effect that the dominant character type involved a stronger LH lateralization effect than the minority character type was consistent with the previous finding that visual similarity among words in a lexicon can influence hemispheric lateralization in visual word recognition. More specifically, Hsiao and Lam (2013) applied Hsiao, Shieh, et al. (2008) intermediate convergence model for hemispheric processing (the same model as used here) to visual word recognition using artificial lexicons of nonexisting English words. They manipulated visual similarity of words in a lexicon by creating artificial lexicons with the same lexicon size (i.e., number of words in the lexicon) but different alphabet sizes (i.e., number of letters in the alphabet). They showed that when the alphabet size was large, words in the lexicon had low visual similarity because they did not share many common letters. When the alphabet size gradually decreased (while keeping the lexicon size fixed), words in the lexicon started to share more and more common letters and became visually more similar to each other; as the result, the model relied more on HSF information (i.e., LH processing) in recognizing the words. This result shows that visual similarity among words alone may modulate hemispheric lateralization in visual word recognition. Consistent with this finding, Lam and Hsiao (2014) recently found that compared with English monolinguals, bilinguals who learned to read both English and an European language that has a similar alphabet to English (such as Italian and Spanish) had a stronger RVF/LH advantage in English word identification. Using computational modeling, Lam and Hsiao (2014) showed that this lateralization difference between the two reader groups could be accounted for by the factor that these bilinguals had a larger vocabulary size (combining words in both languages) than English monolinguals with a similar alphabet size (since their European languages have a similar alphabet to English). As the result, words in the bilinguals’ vocabulary shared more common letters and thus had higher visual similarities than words in the monolinguals’ vocabulary, which lead to a higher demand on HSF information/LH processing in word recognition. The results in the current study are consistent with this finding, showing that a similar phenomenon can also be observed in character processing in Chinese, a logographic language. More specifically, here we show that with a similar set of radicals (components), since there are a lot more SP characters than PS characters in the Chinese orthography, the visual similarity among SP characters is higher than PS characters, leading to a higher demand on HSF information/LH processing in SP character recognition. Our current modeling results also showed that the regular mapping condition, in which characters with the same phonetic radical had the same pronunciation, resulted in a higher

366


demand on HSF information/LH processing than the logographic mapping condition, in which the pronunciations of the characters were randomly assigned without a systematic mapping between phonetic radicals and character pronunciations (in the SF biased condition, Fig. 4). This result is consistent with the behavioral finding that the processing of Chinese characters with a phonetic radical (i.e., phonetic compounds) involves a stronger LH lateralization than those without a phonetic radical (i.e., integrated characters; see Weekes & Zhang, 1999). Weekes and Zhang (1999) argued that this phenomenon might be due to differential phonological activation in the LH in the identification of the two types of characters. In contrast with this speculation, since our model did not implement LH-lateralized phonological processing, our modeling data suggest that this lateralization difference may be due to the requirement of decomposing a character into radicals for phonetic radical identification in the recognition of phonetic compounds, leading to higher demands on HSP information/LH perceptual processing than the recognition of integrated characters, instead of difference in phonological activation in the LH. Using a similar model, Hsiao and Lam (2013) have previously shown that when there is a systematic mapping between visual word form components and pronunciation components in a visual word recognition task (such as the letter-sound correspondences in alphabetic languages), it encourages the model to decompose words into components to map them to phonemes/pronunciation components. As the result, more HSF information is required as compared with the condition in which no systematic mapping exists between word form and pronunciation components (i.e., logographic mapping). Consistent with this finding, recent research has shown that readers of alphabetic languages that have a consistent grapheme-phoneme correspondence (i.e., having a shallow orthography), such as Italian and Welsh, typically have a stronger LH lateralization in word processing as compared with readers of alphabetic languages that have a less consistent grapheme-phoneme correspondence (i.e., deep orthography) such as English (e.g., Beaton, Suller, & Workman, 2007; Workman, Brookman, Mayer, Rees, & Bellin, 2000; see also Wimmer & Goswami, 1994). Lam and Hsiao (2014) showed that bilinguals who learned a European language and English (both are alphabetic languages) had a stronger RVF/LH advantage in English word identification as compared with Chinese–English bilinguals (Chinese is a logographic whereas English is an alphabetic language). Through computational modeling, they showed that this effect could be accounted for by the difference in their first language experience: In Chinese–English bilinguals’ vocabulary, about half of the words (i.e., Chinese characters) did not have a systematic mapping between visual word form components and pronunciation components, and thus there was less demand on HSF information/LH perceptual processing in general. The current results further showed that even in a logographic language such as Chinese, although there is no systematic mapping between visual word form components and pronunciation components, the existence of a phonetic radical, which provides information about the character pronunciation at the syllable level, is sufficient to encourage decomposition of a character into radicals for phonetic radical identification. This character decomposition leads to a higher demand on HSF information/LH perceptual processing than the recognition of characters without a phonetic radical.


367

Thus, hemispheric lateralization in visual word recognition, including both alphabetic and logographic languages, can be influenced by both visual similarity among words in the lexicon, and the requirement of decomposing a word into components for sublexical processing. These two factors correspond well to the two routes in the dual route approach to orthographic processing in visual word recognition (Grainger & Ziegler, 2011; Grainger & Jacobs, 1994; see also Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001; Ellis & Young, 1988; Perry, Ziegler, & Zorzi, 2010; Zorzi, 2010): a direct, lexical route which adopts a whole word approach and uses minimal visual information to identify a word, and an indirect, nonlexical route which uses fine-grained visual information to chunk a word into sublexical orthographic components for mapping them to corresponding phonological representations. Visual similarity among words is likely to pose constraints to word identification in the direct, whole-word route. In contrast, the requirement of decomposing a word into components for sublexical processing, which leads to reliance on HSF information, corresponds to the indirect, sublexical route of word processing. Thus, our results here are consistent with this dual route approach to orthographic processing, demonstrating that both factors/routes can influence hemispheric lateralization of visual word processing. These two factors are able to explain hemispheric lateralization differences between the recognition of faces and words (e.g., Maurer & McCandliss, 2007; Rossion et al., 2003), between word recognition in different languages (e.g., Beaton et al., 2007; Maurer & McCandliss, 2007; Siok et al., 2009; Tan et al., 2005; Workman et al., 2000), between word recognition in monolinguals and bilinguals of different languages (e.g., Hsiao & Lam, 2013), and between the recognition of different character types in the Chinese orthography (e.g., Hsiao & Liu, 2010; Hsiao et al., 2007; Weekes & Zhang, 1999). In addition to these two factors, there are also other factors that may contribute to lateralization effects in visual word recognition. For example, as demonstrated in Experiment 2, position of the phonetic radical may also account for the lateralization difference between SP and PS character processing in Chinese character recognition. In a recent fMRI study, Van der Haegen, Cai, and Brysbaert (2012) observed a positive correlation between the lateralization indices in a language production task and a reading (lexical decision) task among left-handers. This result suggests that hemispheric lateralization in visual word recognition may also be influenced by lateralization of language production within individuals. In summary, through computational modeling, here we show that the stronger LH lateralization in SP character processing than PS character processing in Chinese character recognition can emerge without assuming phonological processing being LH lateralized (cf., Hsiao & Liu, 2010). More specifically, this effect may be due to higher visual similarity (i.e., a perceptual effect) among SP characters than PS characters because of a much greater number of SP characters than PS characters in the lexicon, instead of the difference between SP and PS characters in position of the phonetic radical, automaticity of using the phonetic radical, or LH-lateralized phonological processing in the brain. In addition, we show that when characters have a phonetic radical that consistently conveys information about the character pronunciation, the model relies more on HSF/LH

368


perceptual processing as compared with characters without a consistent phonetic radical. This result is consistent with the behavioral findings in the literature (Weekes & Zhang, 1999), suggesting that the difference in lateralization between processing phonetic compound and integrated characters may be due to the requirement of decomposing a phonetic compound character into radicals for phonetic radical identification, instead of differential phonological activation in the LH as speculated before. These results demonstrate that lateralization effects in visual word recognition may be driven mainly by perceptual processes instead of phonological processes. It also shows that visual similarity among words and the requirement of word decomposition for sublexical processing are two important factors that can account for hemispheric lateralization in visual word recognition in readers of both alphabetic and logographic languages. Further investigation will examine whether these two factors can explain other lateralization phenomena in visual word recognition, and whether there are other factors that may influence lateralization effects in visual word recognition.

Acknowledgments We are grateful to the Research Grant Council of Hong Kong (project code: HKU 744509H and HKU 745210H to J.H. Hsiao). We thank Tin Yim (Tim) Chuk for his help on the character similarity analysis, and Dr. Antoni B. Chan for helpful comments. We thank the editor and two anonymous reviewers for helpful comments.

Notes 1. Recent research from eye movement studies of reading continuous Chinese texts (e.g., Sun, Morita, & Stark, 1985) or isolated Chinese characters/words (e.g., Hsiao & Cottrell, 2009; Liu & Li, 2013) has suggested that expert Chinese readers can recognize Chinese characters/words with a single eye fixation at around the center of the characters/words. 2. Note that this hemispheric asymmetry in spatial frequency processing has been proposed to emerge at a later stage of perceptual processing beyond the sensory level (e.g., Sergent, 1982; Hsiao, Shahbazi, & Cottrell, 2008; Hsiao, Cipollini, & Cottrell, 2013; Heinze, Hinrichs, Scholz, Burchert, & Mangun, 1998), since the asymmetry effect was not usually observed in grating detection or contrast sensitivity tasks (e.g., Di Lollo, 1981; Fendrich & Gazzaniga, 1990; Kitterle et al., 1990; Peterzell, Harvey, & Hardyck, 1989; Rijsdijk, Kroon, & Van der Wildt, 1980). 3. Another difference between SP and PS characters is that the percentage of regular characters is higher in SP characters than in PS characters (a regular character has the same pronunciation as its phonetic radical). Here, we focused on structural differences and did not aim to examine the factor of pronunciation regularity; in the artificial lexicons we assumed all characters were regular.


369

4. Greehouse–Geisser correction was applied whenever the assumption of sphericity was not met. 5. Although our model assumed split architecture (i.e., consistent with the split fovea claim; see Ellis & Brysbaert, 2010), some have argued that the fovea representation for reading is bilaterally projected (e.g., Jordan & Paterson, 2009). Thus, in separate simulations we projected the same word input to both hemispheres (i.e., assuming word input is bilaterally projected) during training, and during testing we projected the word input to only one hemisphere (similar to the divided visual field paradigm used in behavioral studies; e.g., Hsiao & Liu, 2010). We obtained similar results, since the characters had a symmetric configuration. 6. In separate simulations, we projected the same word input to both hemispheres during training to simulate a bilateral projection of the fovea representation (Jordan & Paterson, 2009). In contrast with the results obtained in Experiment 2, there was no character type main effect, and the results were similar to that in Experiment 1 in which phonetic radical position was not manipulated. This finding suggests that the effect of phonetic radical position on hemispheric lateralization observed in Experiment 2 relies on the assumption of split architecture (i.e., the split fovea claim; Ellis & Brysbaert, 2010).

References Beaton, A. A., Suller, S., & Workman, L. (2007). Visual laterality effects in readers of a deep and a shallow orthography. Laterality, 12, 199–215. Bentin, S. (1981). On the representation of a second language in the cerebral hemispheres of right handed people. Neuropsychologia, 19, 599–603. Brady, N., Campbell, M., & Flaherty, M. (2005). Perceptual asymmetries are preserved in memory for highly familiar faces of self and friend. Brain and Cognition, 58, 334–342. Bryden, M. P., & Rainey, C. A. (1963). Left-right differences in tachistoscopic recognition. Journal of Experimental Psychology, 66, 568–571. Chen, Y., Fu, S., Iversen, S. D., Smith, S. M., & Matthews, P. M. (2002). Testing for dual brain processing routes in reading: A direct contrast of Chinese character and pinyin reading using fMRI. Journal of Cognitive Neuroscience, 14, 1088–1098. Cheng, C. M., & Yang, M. J. (1989). Lateralization in the visual perception of Chinese characters and words. Brain and Language, 36, 669–689. Cheung, K. C. F., & Hsiao, J. H. (2010). Visual and Task characteristics may explain hemispheric asymmetry in visual word recognition. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society (pp. 1441–1446). Austin, TX: Cognitive Science Society. Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). Review DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108(1), 204–256. Dailey, M. N., Cottrell, G. W., Padgett, C., & Adolphs, R. (2002). EMPATH: A neural network that categorizes facial expressions. Journal of Cognitive Neuroscience, 14(8), 1158–1173. Daugman, J. (1985). Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters. Journal of the Optical Society of America-A, 2(7), 1160–1169. Di Lollo, V. (1981). Hemispheric symmetry in visible persistence. Perception & Psychophysics, 11, 139–142.

370


Ellis, A. W., & Brysbaert, M. (2010). Split fovea theory and the role of the two cerebral hemispheres in reading: A review of the evidence. Neuropsychologia, 48(2), 353–365. Ellis, A. W., & Young, A. (1988). Human cognitive neuropsychology. London: Erlbaum. Endo, M., Shimizu, A., & Nakamura, I. (1981). Laterality differences in recognition of Japanese and Hangul words by monolinguals and bilinguals. Cortex, 17(3), 391–399. Fendrich, R., & Gazzaniga, M. (1990). Hemispheric processing of spatial frequencies in two commissurotomy patients. Neuropsychologia, 28, 657–663. Fu, S., Chen, Y., Smith, S., Iversen, S. D., & Matthews, P. M. (2002). Effects of word form on brain processing of written Chinese. NeuroImage, 17, 1538–1548. Gilbert, C., & Bakan, P. (1973). Visual asymmetry in perception of faces. Neuropsychologia, 11, 355–362. Grainger, J., & Jacobs, A. M. (1994). A dual read-out model of word context effects in letter perception: Further investigations of the word superiority effect. Journal of Experimental Psychology: Human Perception and Performance, 20, 1158–1176. Grainger, J., & Ziegler, J. C. (2011). A dual-route approach to orthographic processing. Frontiers in Psychology, 2, 54. Han, S., Weaver, J. A., Murray, S. O., Kang, X., Yund, E. W., & Woods, D. L. (2002). Hemispheric asymmetry in global/local processing: Effects of stimulus position and spatial frequency. NeuroImage, 17, 1290–1299. Hanavan, K., & Coney, J. (2005). Hemispheric asymmetry in the processing of Japanese script. Laterality, 10 (5), 413–428. Heinze, H. J., Hinrichs, H., Scholz, M., Burchert, W., & Mangun, G. R. (1998). Neural mechanisms of global and local processing: A combined PET and ERP study. Journal of Cognitive Neuroscience, 10, 485–498. Hernandez, S., Nieto, A., & Barroso, J. (1992). Hemispheric specialization for word classes with visual presentation and lexical decision task. Brain Cognition, 20, 399–408. Hsiao, J. H., Cipollini, B., & Cottrell, G. (2013). Hemispheric asymmetry in perception: A differential encoding account. Journal of Cognitive Neuroscience, 25(7), 998–1007. Hsiao, J. H., & Cottrell, G. W. (2009). Not all expertise is holistic, but it may be leftist: The case of Chinese character recognition. Psychological Science, 20(4), 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., & Liu, T. (2010). Position of phonetic components may influence how written words are processed in the brain: Evidence from Chinese phonetic compound pronunciation. Cognitive, Affective, & Behavioral Neuroscience, 10, 552–559. Hsiao, J. H., Shahbazi, R., & Cottrell, G. W. (2008). Hemispheric asymmetry in visual perception arises from differential encoding beyond the sensory level. In B. C. Love, K. McRae, & V. M. Sloutsky (Eds.), Proceedings of the 30th Annual Conference of the Cognitive Science Society (pp. 857–862). Austin, TX: Cognitive Science Society. 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. (2005). Foveal splitting causes differential processing of Chinese orthography in the male and female brain. Cognitive Brain Research, 25, 531–536. Hsiao, J. H., & Shillcock, R. (2006). Analysis of a Chinese phonetic compound database: Implications for orthographic processing. Journal of Psycholinguistic Research, 35, 405–426. Hsiao, J. H., Shillcock, R., & Lee, C. (2007). Neural correlates of foveal splitting in reading: Evidence from an ERP study of Chinese character recognition. Neuropsychologia, 45, 1280–1292. Ivry, R., & Robertson, L. C. (1998). The two sides of perception. Cambridge, MA: MIT Press. Jordan, T. R., & Paterson, K. (2009). Re-evaluating split-fovea processing in visual word recognition: A critical assessment of recent research. Neuropsychologia, 47, 2341–2353.


371

Keator, D. B., Fallon, J. H., Lakatos, A., Fowlkes, C. C., Potkin, S. G., & Ihler, A. (2014). Feed-forward hierarchical model of the ventral visual stream applied to functional brain image classification. Human Brain Mapping, 35(1), 38–52. Keenan, P. A., Whitman, P. D., & Pepe, J. (1989). Hemispheric asymmetry in the processing of high and low spatial frequencies: A facial recognition task. Brain & Cognition, 11, 229–237. Kitterle, F. L., Christman, S., & Hellige, J. B. (1990). Hemispheric differences are found in the identification, but not the detection, of low versus high spatial frequencies. Perception & Psychophysics, 48, 297–306. Kuo, W.-J., Yeh, T.-C., Duann, J.-R., Wu, Y.-T., Ho, L.-T., Hung, D., Tzeng, O., & Hsieh, J.-C. (2001). A left-lateralized network for reading Chinese words: A 3T fMRI study. NeuroReport, 12, 3997–4001. Lades, M., Vorbruggen, J. C., Buhmann, J., Lange, J., von der Malsburg, C., Wurz, R., & Konen, W. (1993). Distortion invariant object recognition in the dynamic link architecture. IEEE Computer, 42, 300–311. Lam, S. M., & Hsiao, J. H. (2014). Bilingual experience modulates hemispheric lateralization in visual word processing. Bilingualism: Language and Cognition, 17 (30), 589–609. Leong, C. K., Wong, S., Wong, A., & Hiscock, M. (1985). Differential cerebral involvement in perceiving Chinese characters: Levels of processing approach. Brain and Language, 26, 131–145. Li, Y., & Kang, J. S. (1993). Analysis of phonetics of the ideophonetic characters in Modern Chinese. In Y. Chen (Ed.), Information analysis of usage of characters in modern Chinese (pp. 84–98). Shanghai: Shanghai Education Publisher (in Chinese). Liu, P., & Li, X. (2013). Optimal viewing position effects in the processing of isolated Chinese words. Vision Research, 81, 45–57. Mardia, K. V., Kent, J. T., & Bibbi, J. M. (1980). Multivariate analysis (probability and mathematical statistics). Waltham, MA: Academic Press. Maurer, U., & McCandliss, D. (2007). The Development of visual experticse for words: The contribution of electrophysiology. In E. L. Grigorenke, & A. Naples (Eds.), Single-word reading: Cognitive, behavioral and biological perspectives. Mahwah, NJ: Lawrence Erlbaum Associates. McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The visual word form area: Expertise for reading in the fusiform gyrus. Trends in Cognitive Sciences, 7, 293–299. Measso, G., & Zaidel, E. (1990). Effect of response programming on hemispheric differences in lexical decision. Neuropsychologia, 28, 635–646. Perry, C., Ziegler, J. C., & Zorzi, M. (2010). Beyond single syllables: Large-scale modeling of reading aloud with the Connectionist Dual Process (CDP++) model. Cognitive Psychology, 61(2), 106–151. Peterzell, D. H., Harvey Jr., L. O., & Hardyck, C. D. (1989). Spatial Frequencies and the cerebral hemispheres: Contrast sensitivity, visible persistence, and letter classification. Perception & Psychophysics, 46, 433–455. Rauschecker, A. M., Bowen, R. F., Parvizi, J., & Wandell, B. A. (2012). Position sensitivity in the visual word form area. Proceedings of the National Academy of Sciences of the United States of America, 109 (24), E1568–E1577. Rijsdijk, J. P., Kroon, J. N., & Van der Wildt, G. J. (1980). Contrast sensitivity as a function of position on retina. Vision Research, 20, 235–241. Rossion, B., Joyce, C. A., Cottrell, G. W., & Tarr, M. J. (2003). Early lateralization and orientation tuning for face, word, and object processing in the visual cortex. NeuroImage, 20, 1609–1624. 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. Sergent, J. (1982). The cerebral balance of power: Confrontation or cooperation? Journal of Experimental Psychology: Human Perception and Performance, 8, 253–272. Siok, W. T., Spinks, J. A., Jin, Z., & Tan, L. H. (2009). Developmental dyslexia is characterized by the coexistence of visuospatial and phonological disorders in Chinese children. Current Biology, 19(19), R890– R892. Sun, F., Morita, M., & Stark, L. W. (1985). Comparative patterns of reading eye movement in Chinese and English. Perception & Psychophysics, 37, 502–506.

372


Tan, L. H., Laird, A. R., Li, K., & Fox, P. T. (2005). Neuroanatomical correlates of phonological processing of Chinese characters and alphabetic words: A meta-analysis. Human Brain Mapping, 25, 83–91. Tan, L. H., Liu, H. L., Perfetti, C. A., Spinks, J. A., Fox, P. T., & Gao, J. H. (2001). The neural system underlying Chinese logograph reading. NeuroImage, 13, 826–846. Tan, L. H., Spinks, J. A., Gao, J. H., Liu, H. L., Perfetti, C. A., Xiong, J., Stofer, K. A., Pu, Y., Liu, Y., & Fox, P. T. (2000). Brain activation in the processing of Chinese characters and words: A functional MRI study. Human Brain Mapping, 10, 16–27. Tzeng, O. J. L., Hung, D. L., Cotton, B., & Wang, S. Y. (1979). Visual lateralization effect in reading Chinese characters. Nature (London), 282, 499–501. Van der Haegen, L., Cai, Q., & Brysbaert, M. (2012). Colateralization of Broca’s area and the visual word form area in left-handers: fMRI evidence. Brain and Language, 122(3), 171–178. Van Kleeck, M. H. (1989). Hemispheric differences in global versus local processing of hierarchical visual stimuli by normal subjects: New data and a meta-analysis of previous studies. Neuropsychologia, 27, 1165–1178. Weekes, B. S., & Zhang, B. Y. (1999). Chinese character recognition in the left and right visual fields. Brain & Cognition, 40, 269–272. Wimmer, H., & Goswami, U. (1994). The influence of orthographic consistency on reading development: Word recognition in English and German children. Cognition, 51, 91–103. Workman, L., Brookman, F., Mayer, P., Rees, V., & Bellin, W. (2000). Language laterality in English/Welsh bilinguals: Language-acquisitional and language-specific factors in the development of lateralisation. Laterality, 5(4), 289–313. Yang, M. J., & Cheng, C. M. (1999). Hemisphere differences in accessing lexical knowledge of Chinese characters. Laterality, 4, 149–166. Zorzi, M. (2010). The connectionist dual process (CDP) approach to modelling reading aloud. European Journal of Cognitive Psychology, 22, 836–860.

Computational Modeling of Morphological Effects in Bangla Visual Word Recognition.

Preserved mid-fusiform activation for visual words in a patient with a visual word recognition impairment.

Functional Anatomy of Recognition of Chinese Multi-Character Words: Convergent Evidence from Effects of Transposable Nonwords, Lexicality, and Word Frequency.

Phonology and orthography in visual word recognition: evidence from masked non-word priming.

Hemispheric asymmetry in visual recognition of words and motor response in schizophrenic and depressive patients.

Individual differences in involvement of the visual object recognition system during visual word recognition.

Visual word recognition across the adult lifespan.

What can we learn from learning models about sensitivity to letter-order in visual word recognition?

Resolving the locus of cAsE aLtErNaTiOn effects in visual word recognition: Evidence from masked priming.

Aging and the optimal viewing position effect in visual word recognition: Evidence from English.

An ERP investigation of the co-development of hemispheric lateralization of face and word recognition.

The resolution of visual noise in word recognition.

What can we learn about visual attention to multiple words from the word-word interference task?

Rapid modulation of spoken word recognition by visual primes.

Does bold emphasis facilitate the process of visual-word recognition?

Real-time functional architecture of visual word recognition.

Effective connectivity of brain regions related to visual word recognition: An fMRI study of Chinese reading.

Lexical ambiguity, semantic context, and visual word recognition.

Suprasegmental lexical stress cues in visual speech can guide spoken-word recognition.

Tracking the emergence of the consonant bias in visual-word recognition: evidence with developing readers.

Does N200 reflect semantic processing?--An ERP study on Chinese visual word recognition.

Hemispheric lateralization of linguistic prosody recognition in comparison to speech and speaker recognition.

Decomposition, lookup, and recombination: MEG evidence for the full decomposition model of complex visual word recognition.

Evidence for Separate Contributions of High and Low Spatial Frequencies during Visual Word Recognition.