Language experience shapes early electrophysiological responses to visual stimuli: The effects of writing system, stimulus length, and presentation duration

How language experience affects visual word recognition has been a topic of intense interest. Using event-related potentials (ERPs), the present study compared the early electrophysiological responses (i.e., N1) to familiar and unfamiliar writings under different conditions. Thirteen native Chinese speakers (with English as their second language) were recruited to passively view four types of scripts: Chinese (familiar logographic writings), English (familiar alphabetic writings), Korean Hangul (unfamiliar logographic writings), and Tibetan (unfamiliar alphabetic writings). Stimuli also differed in lexicality (words vs. non- words, for familiar writings only), length (characters/letters vs. words), and presentation duration (100 ms vs. 750 ms). We found no significant differences between words and non-words, and the effect of language experience (familiar vs. unfamiliar) was significantly modulated by stimulus length and writing system, and to a less degree, by presentation duration. That is, the language experience effect (i.e., a stronger N1 response to familiar writings than to unfamiliar writings) was significant only for alphabetic letters, but not for alphabetic and logographic words. The difference between Chinese characters and unfamiliar logographic characters was significant under the condition of short presentation duration, but not under the condition of long presentation duration. results suggest that N1 response might not reliably differentiate familiar and unfamiliar writings. More importantly, our results suggest that N1 is modulated by visual, linguistic, and task factors, which has important implications for the visual expertise hypothesis.


Introduction
Learning to read is accompanied by significant behavioral changes such as increased processing speed and attenuated or diminished effect of stimulus length (Perfetti and Hogaboam, 1975;Reicher, 1969;Weekes, 1997). Neural changes associated with reading acquisition have also been well documented under both natural (e.g., Brem et al., 2006;Burgund et al., 2006;Maurer et al., 2005bMaurer et al., , 2006 and experimental conditions (e.g., Brem et al., 2005;Chen et al., 2007;McCandliss et al., 1997;Perfetti et al., 2005;Xue et al., 2006). Nevertheless, there is little consensus with regard to the exact mechanisms underlying these neural changes. Researchers have been debating about the degree to which these changes reflect increased visual expertise with a particular writing system, and about the specific underlying neural indices of such expertise (e.g., McCandliss et al., 2003;Xue et al., 2006).
Two types of common techniques, namely neurophysiology (e.g., ERPs and MEG) and functional imaging (e.g., fMRI and PET), have been used to explore the neural mechanisms of reading acquisition. Using functional imaging techniques, researchers have mainly studied whether the left midfusiform area, the so-called visual word form area (VWFA), is specific to the processing of visual word forms and whether years of reading experience increase its sensitivity to visual word forms (Cohen and Dehaene, 2004;Price and Devlin, 2003;Xue et al., 2006). With ERPs and MEG, researchers have primarily explored whether the first negative component with a latency around 150-200 ms localized in the occipito-temporal cortex (i.e., N1) is more sensitive to visual words than to other visual objects (Barber and Kutas, 2007;Maurer et al., 2005b). For both techniques, the important theoretical question is whether these differences, if they exist, could be attributed to visual expertise. In the present study, we focused on neurophysiological studies. N1 and visual word form processing N1, also known as N170, is characterized by the first negativity at bilateral occipito-temporal electrodes and the central positivity at the fronto-parietal electrodes. It has been consistently revealed within a latency between 150 and 200 ms, in all types of object processing, including words, faces, tools, and so on (e.g., Schendan et al., 1998). Source localization indicated relatively hemispheric specificity between words and faces, with left-hemisphere dominance for words and right-hemisphere dominance for faces (Rossion et al., 2003).
N1 has been proposed to be an index of expertise in object recognition. For example, compared to other objects, a stronger N1 response has been revealed when car experts are processing pictures of cars (Tanaka and Curran, 2001), and bird experts are processing pictures of birds (Gauthier et al., 2003). People who were trained to be experts of birds (Scott et al., 2006) or of novel objects (e.g., 'Greeble') (Rossion et al., 2002) also showed a stronger N1 to those training materials. A recent monkey study also showed an increased N1 response to familiar objects as opposed to novel objects (Peissig et al., 2007). Similarly, in the area of visual word recognition, it has been posited that reading experience increases visual expertise of a given writing system, and the enhanced N1 response to that writing compared to simple symbols might reflect a "coarse" specialization of N1 in visual word processing (Brem et al., 2005;Maurer et al., 2005aMaurer et al., ,b, 2006Wong et al., 2005).
Many ERP studies have been conducted to examine the sensitivity or specialization of N1 in visual word processing. So far, researchers have compared native words/letters with false fonts (Eulitz et al., 2000;Schendan et al., 1998;Wong et al., 2005), symbols Brem et al., 2005;Gros et al., 2002;Maurer et al., 2005b;McCandliss et al., 1997; Moscoso del Prado Martin et al., 2006;Tarkiainen et al., 2002Tarkiainen et al., , 1999, foreign writings (Huang et al., 2004;Pernet et al., 2003;Wong et al., 2005), letter strings (Bentin et al., 1999;Cohen et al., 2000;Martin et al., 2006;Proverbio et al., 2004), pseudowords (Proverbio et al., 2004;Shirahama et al., 2004;Wydell et al., 2003), and second-language words/letters (Liu and Perfetti, 2003;Wong et al., 2005). Furthermore, there was a study that compared native and foreign speakers on the same stimuli (Proverbio et al., 2006). The results were mixed: Some studies revealed a stronger N1 response for native words/letters than for other stimuli, but it was also common to see a lack of differences or even reversed patterns of differences (see Table 1). Such inconsistent results point to the possibility that the effect of language experience on the N1 response may be modulated by factors that varied across previous studies. Such factors include, but are not limited to, the experimental tasks (e.g., passive vs. active, perceptual vs. lexical), materials (e.g., letters vs. words, false fonts vs. foreign writings, alphabetical vs. logographic orthography), and stimulus presentation parameters (e.g., short vs. long duration, masked vs. unmasked). In order to have a comprehensive understanding of the relation between the N1 response and language experience, it is imperative to systematically investigate the potential moderators of that relation. Such an investigation could also help to describe the characteristics of N1 and elucidate its underlying cognitive mechanisms.
The present study examined language experience (familiar vs. unfamiliar writings) and lexicality effect (words vs. non-words) across different types of stimuli and experimental conditions, including alphabetic vs. logographic writing systems, short vs. long stimulus length, and short vs. long stimulus presentation duration. First, we review the literature on these potential moderators.

The effect of writing system
One factor that has received little attention is writing system. Writing systems differ greatly in their visual features. Logographic (e.g., Chinese) and alphabetic writing systems (e.g., English) provide a good illustration of this contrast. For example, relatively consistent differences have been reported in the primary visual system, with Chinese showing more bilateral activation and English showing more leftward asymmetry (Xue et al., 2005). But how different orthographies affect word sensitivity remains obscure. Though several studies on alphabetic languages reported word sensitivity in the left fusiform cortex (e.g., Cohen et al., 2002;Polk and Farah, 2002), two recent fMRI studies revealed no word sensitivity in this region for either native Chinese or English speakers when their native language was compared to logographlike foreign characters, i.e., Korean Hangul (Xue et al., 2006;Xue and Polrack, 2007). Presumably, the compact and complex visual pattern of logographic characters requires extensive visual analysis and thus reduces its differences with native writings. In the present study, this hypothesis was examined by comparing the N1 response to logographic and alphabetic unfamiliar writings, and by examining word sensitivity in logographic and alphabetic writing systems.

The effect of stimulus length
Another important factor that affects N1 is word length. Behaviorally, it has been repeatedly shown that high-frequency words of different lengths are processed at a comparable speed (i.e., no word length effect), whereas low-frequency words and non-words showed the word length effect (Weekes, 1997). Such results are consistent with the notion that high-frequency words, but not low-frequency words and non-words, are processed holistically (but see Pelli et al., 2003). Several studies have examined the effect of stimulus length on the N1 response, but no clear picture has been obtained. For example, some researchers have found a stronger N1 response to longer stimuli than to shorter stimuli (Wong et al., 2005;Wydell et al., 2003) whereas others found no differences on P150 at the parietal electrodes (Schendan et al., 1998), the equivalent of N1 observed at the posterior electrodes. Several studies have further revealed a stronger ERP response to longer words at around 100 ms Hauk et al., 2006;Hauk and Pulvermuller, 2004), but a reversed pattern after 150 ms (Hauk and Pulvermuller, 2004). The word length effect has also been reported in a later time window, again with mixed results (see Hauk and Pulvermuller, 2004 for a review).
These mixed effects may have been exacerbated by the use of both lexical (e.g., Hauk and Pulvermuller, 2004) and non-lexical materials (e.g., Wong et al., 2005). Wydell et al. (2003) found similar length effects for words and non-words at around 100 ms, but stronger length effects for non-words after around 200 ms. Moreover, significant differences in other experimental parameters, including stimulus presentation and task requirements might contribute to these different findings. The present study addressed this issue by examining the length-by-language (familiar vs. unfamiliar writings) interaction under the same experimental conditions.

The effect of presentation duration
Although a wide variety of presentation durations have been used in the literature to examine the visual expertise hypothesis (see Table 1), surprisingly few studies have attempted to directly examine its effect on the N1 response. In one study , words and non-words were presented for 50 ms or 66 ms, and a stronger and faster (i.e., shorter peak latency) N1 response was found under the short duration condition than under the long duration condition for both types of stimuli. There was a significant duration-by-stimulus interaction: That is, there was a significant difference between words and non-words at the 66 ms duration but not at the 50 ms duration. This clearly suggests that presentation duration is an important factor for word sensitivity. So far, most studies used either a short duration of about 100 ms or less, or a long duration of 750 ms or longer, but there were no direct comparisons between these two conditions.
The importance of presentation duration is further supported by PET and fMRI studies. Studies with a 150 ms or shorter duration generally reported more activation for words and word-like stimuli, whereas studies with a duration as long as 750 ms did not show such a pattern (Indefrey et al., 1997) and sometimes showed a reversed pattern (Xue et al., 2006). In two PET studies (Price et al., 1996(Price et al., , 1994, Price and her colleagues systematically explored the effect of presentation duration on word processing. In one study (Price et al., 1994), they used a word-reading task with presentation durations of 150 ms and 1000 ms and used false fonts for the baseline condition. They found that the shorter duration was associated with stronger activations in the left medial fusiform area and the posterior temporal lobe. The other PET study used resting condition as the baseline, and found that activation in the fusiform cortex increased with the longer duration (Price et al., 1996). These results seem to imply a stimulus-by-duration interaction in modulating the activation in the visual system: That is, with the increase of presentation duration, false fonts show more response increase than do words. This possible interaction was examined with ERP in the present study.
Taken together, existing results suggest that the N1 response is modulated by a number of physical, lexical, and learning factors. However, a clear and consistent picture has not been obtained. The apparent differences in experimental details, including the task requirements, data acquisition, and analysis, may simply have exaggerated the discrepancies in results. Moreover, little effort has been made to include all these factors in one study and examine their possible interactions. It is thus necessary to take an integrative approach to examining all these factors within the same methodological framework.
The present study adopted such an approach. Chinese college students who had been learning English as their second language for more than 6 years were asked to passively view four types of scripts: Chinese (familiar logographic writings), English (familiar alphabetic writings), Korean Hangul (unfamiliar logographic writings), and Tibetan (unfamiliar alphabetic writings). To examine the effects of stimulus length, in addition to single characters and letters, non-words were constructed, with two characters in the logographic writings and three letters in the alphabetic writings. For Chinese and English, non-words were compared to visually matched words to examine the effect of lexicality. Each type of stimuli was presented for 100 ms and 750 ms to examine the effect of presentation duration. We were interested in both the main effects of these factors (i.e., stimulus length, presentation duration, and writing system) on N1 as well as the interactions among these factors.

Participants
Sixteen undergraduate subjects were recruited from Beijing Normal University for this study. All participants were righthanded and had normal eyesight in both eyes. They had been learning English as second language since middle school (N 6 years) and had self-reportedly attained a medium level of English fluency. Three subjects were removed from analysis due to substantial eye blink artifacts or signal drift. The remaining 13 subjects (7 males) had a mean age of 22.6 years (ranged from 20 to 27 years). Subjects gave written informed consent before the experiment.

Materials
Ten types of materials were used in this study (see Fig. 1). The five logographic stimuli included Chinese words (CW), Chinese non-words (CNW), Chinese characters (CC), unfamiliar logographic (i.e., Korean Hangul) words (ULW), and unfamiliar logographic characters (ULC). The five alphabetic stimuli included English words (EW), English non-words (ENW), English letters (EL), unfamiliar alphabetic (i.e., Tibetan) words (UAW), and unfamiliar alphabetic letters (UAL). Each category, except for EL and UAL, had 120 stimuli, which were divided into two matched groups for the short-(i.e., 100 ms) and long-duration presentation (i.e., 750 ms) conditions, and counterbalanced across subjects. Only twenty ELs and UALs were used and repeated three times in each condition. The selection criteria for each type of stimuli are detailed below.
All Chinese characters (CC) were high-frequency characters (higher than 90 per million according to the Chinese word frequency dictionary) (Wang and Chang, 1985), with 3-10 strokes, and 2-3 units according to the definition by Chen et al. (1996). The ULCs were strictly matched with Chinese characters in visual complexity (i.e., number of strokes and units). The Chinese words (CW) were all high-frequency words, with each character complying with the above criteria. To construct the Chinese nonwords (CNW), 240 different characters according to the above criteria were selected and randomly paired. Pairs were replaced if they (e.g., AB) or the reversed pattern (i.e., BA) were real words, or homophones of real words. These stimuli were further evaluated by five research assistants in the laboratory to make sure they did not elicit clear meanings (i.e., no items scored more than 2 on a 5-point scale with 1 representing "meaningless" and 5 representing "meaningful"). Because the subjects in the present study could not read Korean, the ULWs were random combinations of two different Korean Hangul characters selected based on the above criteria regarding visual complexity. Visual complexity of each character was strictly matched across the five visual categories and the two experimental conditions. The word frequency of Chinese characters was also matched across conditions. Twenty English letters (excluding all vowels and the letter "Y") and twenty Tibetan letters were selected. Tibetan letters were visually dissimilar to English letters and other scripts familiar to the subjects (e.g., Roman numerals). All English words (EW) were high-frequency 3-letter words chosen from MRC Psycholinguistic (2) using a pure visual task, including passive viewing (with a feature detection task to assure attention) or visual comparison; (3) comparing words or PW with other visual symbols, like non-words, foreign writings, false fonts, geometry shapes, and simple symbols; or comparing native and foreign readers. Results from developmental studies and training studies were not included here but are discussed in the text. b They found a significant difference between words and symbols at the late N1 response (183-256 ms post stimulus) for adolescents, but not for adults. c No quantification or statistical comparison was available.
database. Because the frequency statistics might not apply to Chinese speakers for whom English is only a second language, five undergraduate students with a similar level of English fluency as the subjects were asked to evaluate the familiarity of these English words on a 5-point scale with 1 representing "never seen it before" and 5 representing "very familiar." Words with a mean score less than 4 were excluded. Because the subjects could not read Tibetan, UAWs were created by randomly combining three different Tibetan letters. Three-letter ENWs were also created randomly.
It should be noted that the Korean and Tibetan words used in the present study should be functionally equivalent to Korean and Tibetan non-words because the subjects did not know these languages. Nevertheless, they were so named to facilitate the comparisons with real words in Chinese and English.

Procedure
Subjects were seated 105 cm away from the computer screen in a dimly lit, sound-attenuated room. A passive viewing task was adopted to reduce the effect of higher-level task requirements and possible strategic differences between familiar and unfamiliar writings in the comparison task. All stimuli were presented visually in white against black background at the center of the screen. The logographic stimuli (300 trials in total) and alphabetical stimuli (also 300 trials) were presented in separate sessions to reduce the length of each session. Both alphabetic and logographic stimuli were presented with two presentation durations: a long-and a shortduration condition. In other words, there were four separate sessions: alphabetic stimuli/long duration, alphabetic stimuli/short duration, logographic stimuli/long duration, and logographic stimuli/short duration. Within each session, all stimuli (familiar letters/characters, familiar words, familiar non-words, unfamiliar letters/characters, and unfamiliar words) were randomly presented. This design allowed us to directly examine the critical contrast (familiar vs. unfamiliar writings, words vs. non-words, and letters/ characters vs. words) within one session. For each trial, the stimulus (151 × 151 pixels for letters and characters, and 302 × 151 pixels for words and strings) was presented for 750 ms (i.e., the long-duration condition) or 100 ms (i.e., the short-duration condition), followed by a blank screen for 1250 ms or 1900 ms, respectively. Random time delay (i.e., "jitter"), created according to the exponential distribution (mean: 500 ms, range from 6 ms to 3000 ms), was added to the end of each trial to reduce the effect of expectation. Subjects were asked to fixate on the screen and passively view these stimuli during the whole session. To guarantee subjects' attention, for 5% of the trials (i.e., 15 trials in each session), a picture was flashed (100 ms duration) shortly after the writings (100 ms delay). Subjects were required to judge whether the picture depicted was an animal or an object by pressing one of two buttons. These trials were excluded from analysis. A 3-s blink symbol ("∼ ∼") appeared every four trials signaling that subjects could blink their eyes. The next trial began after an interval of 1 s. Subjects took a short break of 1-2 min in the middle of each session and resumed the task by pressing a button. They took a longer rest between two sessions. The order of the four sessions was counterbalanced across subjects.
Before the formal test, subjects were presented 20 practice trials. During the practice period, subjects were instructed to avoid eyeblinks within a trial. Subjects were given feedback if they blinked, made a wrong response, or had obvious head movements.

Electroencephalogram (EEG) recording and analysis
Scalp voltages were recorded by a NeuroSCAN system, using a 64-channel Quick-cap with silver chloride electrodes (Neurosoft, Inc. Sterling, USA). The impedance of all electrodes was kept below 5 kΩ. Linked ears served as reference, and the middle of the forehead served as ground. Two channels were placed at the outer canthi of both eyes to record the horizontal electrooculogram (HEOG), another two channels above and below the left eye for vertical electrooculogram (VEOG). Electroencephalogram (EEG) was amplified on-line with a high-pass frequency filter of 0.05 Hz and a low-pass frequency filter of 100 Hz. The sampling rate was 1000 Hz.
The continuous EEG was transformed to the average reference (Lehmann and Skrandies, 1980). Trials were rejected for movement artifacts, eye movements, blinks, or amplifier saturation. Trials with artifacts exceeding ±100 μV were also excluded. The remaining trials were averaged for each type of stimuli separately for each subject. Due to our special effort to minimize the eye-blink artifacts, the valid trials used for averaging were 48 for each type of stimuli. The continuous EEG data were segmented into epochs from 200 ms pre-stimulus until 800 ms post-stimulus. The 200 ms pre-stimulus served as the baseline, which had been subtracted out before grand average. Averaged waveform was filtered with a low-pass filter of 30 Hz (zero-phase, 12 dB/octave). The grand average was obtained by averaging across subjects' averages separately for each type of stimuli. Scalp topographies were visualized with EEGLAB (http:// sccn.ucsd.edu/eeglab/).

N1 quantification and statistical analysis
Based on visual inspection of each subject's data, N1 was identified as the first visible negative peak 120-180 ms poststimulus in occipito-temporal lobe. Because the N1 response was most obvious in PO5 and PO6, further data quantification and statistical analysis focused on these two electrodes. The mean amplitude of N1 was calculated for an epoch comprising 20 time points (20 ms), 10 before and 10 after the peak for each type of stimuli, based on the grand mean response (the center of time windows identified for each condition was summarized in Table 2). A series of within-subject repeated-measures ANOVA was conducted to examine the effects of experimental factors. Unless otherwise noted, within-subject standard error is reported (Loftus and Masson, 1994). Due to our focus on the sensitivity hypothesis, no further statistical analysis was conducted on the peak latency.

Results
Overall, the early N1 and P1 responses to all stimulus types occurred at a latency of around 150-170 ms. The topographic map at 120-180 ms post-stimulus for each type of stimuli is shown in Fig. 2. ERPs from selected 10-20 locations with the strongest N1 response (i.e., PO5 and PO6) are shown in Fig. 3 (for logographic writings) and Fig. 4 (for alphabetic writings).

The effect of lexicality
In the first analysis, we examined the lexicality effect by testing whether words and non-words showed different N1 responses. Three-way ANOVA with lexicality (words vs. non-words), duration (short vs. long), and hemisphere (left vs. right) as within-subject factors were conducted for Chinese and English materials separately (Fig. 5). (Lexicality effect was not examined for unfamiliar writings because the subjects did not know those languages.) Results showed no significant differences between Chinese words and non-words [F(1,12) = .004, p = .948], nor between English words and non-words [F(1,12) = .85, p = .375]. Also, there was no effect of presentation duration or hemisphere, nor were there any significant interactions (all p values N.133).

The effect of language experience
The second analysis was conducted to examine the N1 effects in both hemispheres (left vs. right) of language experience (familiar vs. unfamiliar writings), stimulus length (letters/characters vs. words), presentation duration (long vs. short), and writing system (alphabetic vs. logographic). A five-way ANOVA with the abovementioned variables as within-subject factors was conducted. Results are shown in Fig. 6.
This analysis revealed significant main effects of language experience [F(1,12) = 12.207, p = .004] and stimulus length [F(1,12) = 14.187, p = .003]. In addition, there was a significant interaction between language experience and stimulus length [F(1,12) = 10.424, p = .007], a significant three-way interaction among language experience, stimulus length, and writing systems [F(1,12) = 12.221, p = .004], and a marginally significant four-way interaction among stimulus length, hemisphere, language experience, and presentation duration [F(1,12) = 4.070, p = .066]. No other main effects or interactions were significant (all p values N .16). To investigate the significant interactions, we did further simple effect analyses to examine the language experience effects separately by their significant moderators (stimulus length and writing system). This analysis revealed a significantly stronger N1 response to English letters than to unfamiliar alphabetic letters [F(1,12) = 20.190, p = .001], and to Chinese characters than to unfamiliar logographic characters [F(1,12) = 11.315, p = .006]. However, there were no language experience effects for either logographic or alphabetic words (both F values b .36). Direct comparisons revealed that the language experience effect was stronger for alphabetic letters [t(12) = 3.42, p = .005] and logographic characters [t(12) = 2.78, p = .017] than for their word versions. Furthermore, the language experience effect was stronger for alphabetic letters than for logographic characters [t(12) = 3.12, p = .009], but there were no significant differences between alphabetic words and logographic words [t(12) = −.46, p = .653], reflecting the language experience by stimulus length by writing system three-way interaction. Although the overall interaction between language experience and presenta-

The effect of stimulus length
Because a comparison between alphabetic letters and alphabetical words may be confounded by the different number of repetitions of the same stimuli (letters, but not words, were repeated three times due to the small number of letters), we limited our analysis of the effect of stimulus length to the comparison between logographic words and characters. Four-way ANOVA revealed a significant stimulus length effect [F(1,12) = 18.481, p = .001], and a length by language experience interaction [F(1,12) = 7.74, p = .017], reflecting a much stronger length effect for the unfamiliar logographic writing than for Chinese. Simple effect analysis revealed a significant length effect for the unfamiliar logographic writing [F(1,12) = 21.045, p = .001], but only a marginally significant effect for Chinese [F(1,12) = 3.80, p = .075].

The effect of writing systems
Also to avoid the repetition effect for alphabetic letters, we examined the effect of writing systems by comparing only logographic words and alphabetic words. Four-way ANOVA revealed no significant main effects or interactions (all p values N .15).

Discussion
To achieve a better understanding of the relation between N1 response and language experience and to elucidate possible mechanisms of N1 response, the present ERP study examined the effect of language experience on the amplitude of the early N1 response and how it is modulated by writing system, stimulus length, and presentation duration. In this section, we will first discuss the language experience effect, as revealed by the comparisons between familiar and unfamiliar writings as well as the lexicality effect. We will then discuss the characteristics of the N1 response. Finally, we will try to integrate these data to postulate possible mechanisms of the N1 response during visual word recognition and suggest questions for future research.
How does language experience change the N1 response?
The effect of language experience on the N1 response has generally been examined by comparing stimuli with different lexical status, like non-words or consonant strings, pseudowords, and words, or by comparing language with nonlinguistic materials, including false fonts, symbols, and foreign writings. Whereas the former approach can examine the effects of higher-level factors (e.g., orthography, phonology, and semantics) on word recognition, the latter can examine the combined effects of both higher-level factors and lower-level factors (e.g., letter identification).
The present study examined the lexicality effect by comparing words with non-words (Chinese language) as well as words with consonant strings (English language). The absence of the lexicality effect is consistent with many previous studies (Bentin et al., 1999;Cohen et al., 2000;Khateb et al., 2002;Maurer et al., 2005b;Nobre et al., 1994;Proverbio et al., 2004;Schendan et al., 1998;Shirahama et al., 2004;Wydell et al., 2003). Nevertheless, a significant N1 difference had been revealed in previous studies between orthographic and non-orthographic stimuli (Compton et al., 1991;Dehaene, 1995;McCandliss et al., 1997), words and pseudowords McCandliss et al., 1997), verbs and nouns (Koenig and Lehmann, 1996), words belonging to dif-  ferent semantic classes (Koenig and Lehmann, 1996;Pulvermuller et al., 2001;Skrandies, 1998Skrandies, , 2004Skrandies and Chiu, 2003), and words with different levels of semantic coherence . These results suggest that linguistic processes can affect the early N1 response, but whether or not this leads to a significant lexicality effect depends on many factors. We will return to this issue in a later section.
We also examined the differences between familiar and unfamiliar writings. We found a complex pattern of N1 sensitivity due to the modulation of orthography and stimulus length. A stronger N1 response to familiar writings than to unfamiliar writings was significant for alphabetic letters, but not for alphabetic and logographic words. Further, the language experience was bigger for alphabetic letters and logographic characters. We also found a marginally significant moderating effect of presentation duration. That is, the difference between Chinese characters and unfamiliar logographic characters was significant only under the short presentation duration, but not under the long presentation duration. These results suggest that N1 might not reliably differentiate familiar and unfamiliar writings, and it is significantly modulated by writing system and visual complexity (i.e., stimulus length).
Our results (especially those from the comparisons between familiar and unfamiliar alphabetic words) should be discussed in the context of the findings from previous research comparing language with symbol strings and false fonts. Most of previous studies that used symbol strings (e.g., familiar simple geometric shapes such as squares, rectangles, and diamond shapes) have found stronger activations to words than to simple symbol strings (Bentin et al., 1999;Brem et al., 2005;Maurer et al., 2005a,b;Tarkiainen et al., 2002Tarkiainen et al., , 1999. On the other hand, consistent with our results, studies using visually matched false fonts showed no robust differences between language and false fonts (Eulitz et al., 2000;Schendan et al., 1998). One interpretation of these results is that the N1 differences reflect "coarse" specialization of visual word processing due to visual expertise with writings, letter strings, false fonts, but not symbol strings. Alternatively, however, the differences in the N1 effect could be attributed to several other factors, including differences in experimental tasks (one-back comparison vs. passive viewing and feature detection), systematicness (symbols vs. pseudo/real writings), complexity (simple vs. visually matched), and familiarity (familiar vs. unfamiliar) of the visual stimuli. As will be discussed below, some of these factors might have a significant effect on the N1 response.

N1 response is modulated by writing system and visual complexity
First, we found that writing system can modulate the language experience effect. Although there was a difference between English letters and unfamiliar alphabetic letters, the difference between Chinese characters and unfamiliar logographic characters was more subtle and less reliable. The absence of difference between familiar and unfamiliar logographic characters under the 750 ms presentation duration is consistent with previous fMRI results (Xue et al., 2006). This suggests that unfamiliar logographic writings might elicit an N1 response comparable to or even stronger than that elicited by familiar logographic writings. Consistently, a previous ERP study revealed a stronger N1 for Asiatic characters than for letters even for subjects who had no experience with Asiatic characters (Pernet et al., 2003). Stronger fusiform activation has also been found for Chinese than for English stimuli even when the subjects were English speakers and had only limited exposure to the Chinese language (Liu et al., 2007). These results seem to be inconsistent with those of Wong et al. (2005), who found a stronger N1 response to Roman letters than to Chinese characters among English speakers. However, in their study, only eight simple Chinese characters (matched with Roman letters in visual complexity) were used and repeated many times. Their results might be confounded by different priming effects for familiar and unfamiliar writings. Furthermore, the writing-system effect we found cannot be attributed to a transfer between Chinese and Korean (due to their visual similarity) because it has been found that English speakers who had no experience with any logographic writings also showed similar middle fusiform activation between English and Korean Hangul (Xue and Polrack, 2007). Nevertheless, further studies need to examine this issue by studying participants with no experience of any logographic language (Xue and Polrack, 2007), or by comparing the N1 response to an unfamiliar logographic writing (e.g., Korean Hangul) by subjects with different language backgrounds (e.g., Chinese vs. English readers).
The present study also revealed a significant effect of stimulus length on the N1 response. A stronger N1 response was found with long stimuli than with short stimuli, which replicates previous findings (Wong et al., 2005;Wydell et al., 2003). A further examination of our data suggested that stimulus length (or visual complexity in general) had a stronger effect on unfamiliar writings than on familiar writings. These results are consistent with behavioral data indicating a decreased word length/complexity effect for familiar writings. This result has important implications for our understanding of the mechanisms of the N1 response, which will be discussed below.
So far, relatively few studies have examined the effect of presentation duration on the N1 response. Although the present study did not reveal a consistent presentation duration effect, the significant interaction between presentation duration and language experience is very important. This pattern is consistent with our previous fMRI data that showed no differences between familiar and unfamiliar writings under a long presentation duration condition (Xue et al., 2006). One reason for this interaction might be that under the short presentation duration condition, familiar writings attracted more attention than did unfamiliar writings, which in turn increased the synchrony of the LFPs  and the ERP response (see Engel et al., 2001 for a review). In contrast, under the long-duration presentation condition, the differences between familiar and unfamiliar writings might have been greatly reduced. Consistent with this speculation, a larger N1 response under the short duration than under the long duration condition was found only for familiar writings [F(1,12) = 3.39, p = .09], but not for unfamiliar writings (p = .94).

What can N1 tell us about visual word recognition?
So what can the above results tell us about the mechanisms of the N1 response to visual word processing? There is a long-standing assumption about the visual expertise effect in visual word form processing. However, this assumption has not been supported by compelling evidence (see Xue et al., 2006 for a discussion on behavioral and fMRI evidence). In the following paragraphs, we will argue that, based on the above characteristics of the N1 response, the connection between the enhanced N1 amplitude and visual expertise is questionable.
One salient finding of the present study was that the effect of language experience was modulated by stimulus length. Further analysis suggested that stimulus length enhanced the N1 response, more so for unfamiliar writings than for familiar writings. This result is inconsistent with the visual-expertise-induced sensitivity hypothesis, which would expect only a main effect of language experience, not its interaction with stimulus length. Moreover, we found that language experience reduced the visual complexity effect, which again is not consistent with this hypothesis. We think our results need to be explained by multiple processes involved in visual word processing. Some of those processes may counteract one another. For example, functional imaging studies have shown that visual familiarity or visual expertise decreases the neural response (Xue et al., 2006;Xue and Polrack, 2007), whereas linguistic components such as phonology and semantics associated with the visual words increase the neural response (Xue et al., 2006). Consistently, preliminary ERP data have shown that visually unfamiliar text (e.g., mirror text) elicited stronger N1 responses than did visually familiar text (e.g., normal text), suggesting that increased visual familiarity reduces the N1 response (Proverbio et al., 2007).
The top-down modulation of the early N1 response has been supported by many observations (see Introduction and above Discussion). Particularly, it is well known from previous literature that linguistic, task-related, and attentional factors can have a substantial effect on the amplitude of N1 (Luck et al., 2000). Cumulating evidence suggests that the speed of processing and information flow through the visual system and the time-window for possible top-down modulation is much faster than it was assumed. For example, the first afferent volley could reach the frontal lobe 80 ms post-stimulus (Foxe and Simpson, 2002;Thorpe et al., 1996). It has been shown that the ERP components might be indicative of recurrent feedback-driven processes rather than the first information sweep through the system (Buchner et al., 1997; also see Sereno and Rayner, 2003 for a review).
The above hypothesis can explain not only the results observed in this study but also those of previous studies. That is, for visually complex stimuli (e.g., words vis-à-vis letters), the N1 component is mainly modulated by visual complexity, and the neural decreases due to increased visual familiarity or visual expertise will outweigh the neural increases caused by the acquisition of linguistic components (e.g., sounds and meanings), resulting in an overall decrease in the N1 response. Whereas for visually simple stimuli (i.e., single letters), the effect of visual expertise would be small and thus could be outweighed by the effect of linguistic components, resulting in an overall increase in the N1 response. This idea fits well with the significant interactions found between language experience and visual complexity-that is, a smaller stimulus length effect for familiar writings than for unfamiliar writings and the absence of language sensitivity for visually complex stimuli, like alphabetical and logographic words.
Our hypothesis might also explain the absence of the lexicality effect in this study. The lexicality effect (comparisons between logographic words and non-words and between alphabetic words and consonant strings) usually reflects an integrated influence of visual familiarity, orthography, phonology, and semantics (Xue et al., 2006). Because these factors have differential effects on the neural response to visual words, as shown by previous fMRI data on fusiform activation, they might counteract one another, resulting in an absence of the lexicality effect under certain circumstances. For example, different task requirements might have differentially emphasized visual forms, phonological, or semantic processing, resulting in discrepant findings (see Table 1).
Although our speculation provides a potentially promising way to resolve the discrepancies in existing research, further studies are necessary to directly examine this hypothesis. Particularly, artificial language training paradigms that can separate these effects will be very helpful (Xue et al., 2006). Furthermore, our hypothesis also points to the importance of examining factors such as stimulus presentation duration  and task requirements. Finally, the combination of ERP and fMRI techniques will also be useful for characterizing the neural responses associated with the early stage (spatially and temporally) of visual word recognition, as well as to provide important information on the role of bottom-up physical factors and top-down linguistic factors in this process.

Conclusion
Reading represents an important type of visual object processing. Although many studies have emphasized the role of visual expertise in shaping the cognitive and neural changes in the early stage of visual word recognition, its exact effects on neural response as indexed by the N1 response has been controversial. Our study suggests that there are a number of factors aside from visual expertise, such as writing system and visual complexity, could affect the N1 response. Moreover, emerging evidence has suggested that linguistic factors could affect the processes that occur in the early stage of the neural circuitry, such as the primary visual system, and within 150-200 ms of the stimulus presentation. Thus, a careful examination and separation of the different effects of visual and linguistic factors on the neural response will help us to obtain a better understanding of how language experience affects reading.