N170 and Letter Expertise 1 Running Head: N170 AND LETTER EXPERTISE An early electrophysiological response associated with expertise in letter perception

Expertise with print is likely to optimize visual processes for recognizing characters of a familiar writing system. While brain activations have been identified for words and letter strings compared to other stimuli, relatively little work has focused on the neural basis of single letter perception. English readers and Chinese-English bilinguals participated in an ERP study and performed a 1-back identity judgment on Roman letters, Chinese characters, pseudofonts, and their string versions. The ChineseEnglish bilinguals showed an enhanced N170 for both Roman letters and Chinese characters compared to pseudofonts. For the non-Chinese readers, the N170 amplitude was larger for Roman letters relative to Chinese characters and pseudofonts. Our results suggest that changes in relatively early visual processes underlie expert letter perception. N170 and Letter Expertise 3 Perceptual expertise with letters is one of the results of our prolonged experience with print and reading. The extensive reading experience taking place over the years after we become literate likely modifies the way we process and perceive individual letters. For instance, expert readers are used to seeing print in a coherent style, and are thus able to extract font information to aid letter recognition. They perform a letter identification task better with letter strings of the same font than of mixed fonts (Sanocki, 1987, 1988). Novice readers (e.g., English readers viewing Chinese characters), however, are not as efficient in using this font information (Gauthier, Wong, Hayward, & Cheung, submitted). Likewise, expert readers are accustomed to seeing letters in the context of words. When they fixate on a part of a word, they obtain not only highresolution information of the letters in the fovea but also low-resolution information of the parafoveal letters. With experience they develop a strong tendency to use low-resolution information of the parafoveal letters, such that even when high-resolution information is artificially made available (by magnifying the parafoveal letters), the readers are unable to utilize this extra information (Nazir, Jocobs, & O'Regan, 1998). Such behavioral phenomena suggest that our perception of letters is influenced by our reading experience. N170 and Letter Expertise 4 Neural selectivity can develop as a result of perceptual expertise with certain categories of objects (Gauthier, 2000). There are two neural hallmarks of the kind of expertise we acquire with identifying objects within homogeneous classes (e.g., faces, cars, dogs, birds, and computergenerated novel objects). Compared with common objects these objects of expertise elicit a larger event-related potential (ERP) component, N170, in posterior brain regions (Rossion, Gauthier, Goffaux, Tarr, & Crommelinck, 2002; Tanaka & Curran, 2001), and greater recruitment of a small region in the fusiform gyrus, mainly on the right (Gauthier, Skudlarski, Gore, & Anderson, 2000; Gauthier, Tarr, Anderson, Skudlarski, & Gore, 1999). Because the kind of expertise we have with letters differs in several ways from the kind of expertise we have with faces, cars or dogs, letters would be expected to recruit a different part of the extrastriate cortex. Indeed, as described below, words and letterstrings (Cohen et al., 2002; Polk & Farah, 1998) and more recently single letters (James, James, Jobard, Wong, & Gauthier, submitted) elicit greater activity in parts of the left fusiform gyrus compared to control stimuli, including digits and unfamiliar characters. ERPs would also be expected to reveal this neural selectivity for letters, and there is in fact some evidence for an early selective response for letters, although the emphasis N170 and Letter Expertise 5 has been on higher-level stimuli such as words (Bentin, MouchetantRostaing, Giard, Echallier, & Pernier, 1999). The majority of neural studies about print have focused on selectivity to words and pronounceable strings (Assadollahi & Pulvermuller, 2003; Bookheimer, 2002; Cohen et al., 2000; Cohen et al., 2002; Dehaene, Le Clec'H, Poline, Le Bihan, & Cohen, 2002; Hauk & Pulvermuller, 2004; McCandliss, Posner, & Givon, 1997; Petersen, Fox, Snyder, & Raichle, 1990; Proverbio, Vecchi, & Zani, 2004). These studies therefore address the linguistic more than the perceptual aspect of reading. More relevant to the question of neural selectivity for letters per se are studies showing more activity for unpronounceable letter strings than control stimuli. For example, the amplitude of the N170 is greater for words, pseudowords, and unpronounceable consonant strings than for strings formed by alphanumeric symbols and forms (Bentin et al., 1999). A greater P150 component has also been found not only for words and letter strings, but also for strings of letter-like stimuli, compared with object icon strings (Schendan, Ganis, & Kutas, 1998). The P150, maximal at the central top electrode (Cz) when recorded with respect to a mastoid reference, may be the positive counterpart of the N170 maximal at occipito-temporal electrodes. A larger intracranial N200 has also been found bilaterally in the posterior fusiform gyrus for words and nonwords N170 and Letter Expertise 6 (pronounceable or not) compared with objects like cars and butterflies (Allison, McCarthy, Nobre, Puce, & Belger, 1994; Nobre, Allison, & McCarthy, 1994). An fMRI study showed more activity for letter strings than textures and faces at the left occipito-temporal junction (Puce, Allison, Asgari, Gore, & McCarthy, 1996). Greater fMRI activations have also been found for letter strings than digit strings in a widespread area around the left fusiform gyrus (Polk et al., 2002). These results suggest neural selectivity for strings of letters and letter-like stimuli that do not readily contain linguistic information at a word level. One may intuitively equate the selectivity for unpronounceable strings to selectivity for letters, although this is not necessarily correct for two reasons. First, since letter strings are more word-like perceptually, they are likely to evoke more word-level processes involving orthography, phonology, etc., than single letters (Price, 2000). Second, an interesting dissociation has been obtained between two areas in the occipitotemporal region, one being selective for individual letters but not for letter strings, with the other being selective for strings but not individual letters (James et al., submitted). A few other studies suggested selectivity for individual letters. For example, fMRI activity in bilateral occipito-temporal areas habituates to the same letter in the same font (vs. different fonts) but not to the same face N170 and Letter Expertise 7 (vs. different faces) (Gauthier, Tarr et al., 2000). Also, there is more fusiform gyrus activity for single letters than oblique lines (Longcamp, Anton, Roth, & Velay, 2003). More left middle occipital activations have also been shown for single letters compared with symbols and colors (Flowers et al., in press; Garrett et al., 2000). A concern is that these fMRI activations may be caused by feedback from higher-level processing, e.g., letter naming. However, a number of MEG studies by Tarkiainen and colleagues argue against this alternative account (Tarkiainen, Cornelissen, & Salmelin, 2002; Tarkiainen, Helenius, Hansen, Cornelissen, & Salmelin, 1999). They located a left inferior occipitotemporal region showing more activity at about 150 ms for pronounceable letter strings than for strings of rotated letters. Despite the primary interests of the authors on strings, this region also showed more activity for single upright letters than rotated ones. The early latency of these MEG responses make the feedback from higher-level processes a less likely explanation for the selectivity found in above-mentioned fMRI studies. The current study examines the early neural selectivity associated with letter expertise. Two groups of participants (English readers who cannot read Chinese, and Chinese-English bilinguals) took part in an ERP experiment and saw three types of characters (Roman, Chinese, N170 and Letter Expertise 8 pseudofont). The Group × Stimulus Type design creates expert (nonChinese readers viewing Roman characters, bilinguals viewing Roman and Chinese characters) and novice situations (non-Chinese readers viewing Chinese and pseudofont characters, bilinguals viewing pseudofont characters), allowing a more direct test of the association between expertise and neural selectivity for letters. For example, the same stimuli (Chinese characters) were expected to elicit different levels of activity depending on the amount of expertise, i.e., bilinguals were expected to show comparable activity with Roman letters and Chinese characters, while non-Chinese readers were expected to show more activity with Roman letters than Chinese characters. Such results would not be explained by the feature differences between the stimuli, which are difficult to control perfectly. The use of both alphabetic and logographic characters also improves the generalizability of results. We adopted ERPs to tap into early, visual letter processing relatively isolated from most linguistic processes. Past research shows that the earliest potential reflecting high-level, visual differences among object categories appear as a posterior negative component peaking at about 170 ms after stimulus presentation (Bentin, Allison, Puce, Perez, & et al., 1996; Curran, Tanaka, & Weiskopf, 2002; Rossion, Gauthier et al., 2002; Tanaka & Curran, 2001). This N1/N170 potential has been shown to N170 and Letter Expertise 9 be highly associated with expertise level, among other factors (Gauthier, Curran, Curby, & Collins, 2003; Rossion, Gauthier et al., 2002; Tanaka & Curran, 2001). Therefore, we expected a larger N170 (compared with pseudofont control) at posterior channels for letters of expertise, i.e., Roman letters for non-Chinese readers, and both Roman letters and Chinese characters for bilinguals. It is important to note that any N170 effect for letter expertise would not necessarily reflect the same processes as the N170 effect found for subordinate-level object and face expertise. Since various spatio-temporally overlapping visual processes are likely to contribute to the scalp-recorded N170 (Rossion, Curran, & Gauthier, 2002), it is a reasonable postulate that the N170 can be modulated by different types of perceptual expertise with objects. Our primary aim here is not to equate or dissociate letter expertise from face-like expertise, but to describe properties of the selectivity associated for letters and letterstrings with expertise. Method Participants Thirty-seven undergraduates from University of Colorado at Boulder participated for course credit. Twenty-two Chinese-English bilinguals participated for payment of $15/hour. Because we were unable N170 and Letter Expertise 10 to recruit as many bilingual subjects, the present results included only 18 non-Chinese readers and 18 Chinese-English bilinguals. Subject selection was based upon absence of EEG artifact (6 monolingual and 1 bilingual subjects were excluded for excessive artifact), maintaining high accuracy levels and minimizing group differences in accuracy (subjects with less than 90% accuracy were excluded: non-Chinese = 6, Chinese-English = 3), maintaining counterbalancing, and equating the sex distribution of the two groups (9 males and 9 females per group). The Chinese-English bilinguals, who were mostly graduate students, were older (MN = 24, range = 19 29) than the undergraduate non-Chinese readers (MN = 19, range = 18 22). All of the Chinese-English bilinguals were born in China, learned English in China (mean age = 11, range = 4 to 15), had known English for a long time (mean years = 13, range = 5 to 21), and recently moved to the US (mean years in US = 3, range = 1 to 10). Stimuli, design, and procedure There were six types of stimuli (Roman, Chinese, and pseudofont characters, and their string versions). Figure 1 shows the eight Roman consonants, eight Chinese characters, and eight pseudofont characters used, and one example of each type of trials. Each character was about 1 × 1 cm large (0.57 degree at a viewing distance of 100 cm). Each string N170 and Letter Expertise 11 consisted of 5 characters and was about 7 cm wide (4 degree at a viewing distance of 100 cm). The Roman strings were formed by first randomly picking and assembling Roman letters to form 100 different 5-character strings, and then replacing characters for certain strings according to the following rules: (a). There is no repetition of letters within each string. (b). All letters occur at approximately the same frequency in the 100 strings (mean = 62; range = 58-65). (c). All letters occur at approximately the same frequency in the central, underlined position (12 or 13). (d). There are no familiar or potentially meaningful 2-letter character combinations (e.g., HP, HB, BP, HK). (e). There are no valid graphemes (e.g., BL, PH). (f). All 2-letter combinations (e.g., “DF”), except the removed ones, occur at similar frequencies (mean = 7.96; range = 6-12). The Chinese and pseudofont strings were formed by taking the 100 Roman strings and replacing them with the corresponding Chinese or pseudofont characters. We also checked to ensure that there was no meaningful character combination in Chinese strings (e.g., , which means “dry soil”). -----------------------------Figure 1 inserted here -----------------------------There were 100 trials for each of the six types of stimuli, separated into 5 blocks, each containing 20 trials. Participants performed a 1-back N170 and Letter Expertise 12 identity-matching task. Each trial started with a fixation cross at the center for a random period between 250 and 750 ms. A stimulus (a character or string) then appeared for 750 ms, followed by a 500-ms blank screen, and the fixation for the next trial. Participants were instructed to press the key “1” on the number key pad when the character shown was identical to the previous one, or when the central, underlined character of the current string repeated that of the previous string (flanking characters were always different in both same and different trials). These “same” trials amounted to 10% of all trials for each stimulus (i.e., 10 out of 100). The numbers of same trials were 1, 2, 2, 2, and 3 for the five blocks. The six types of stimuli formed a total of 600 trials presented in 30 blocks. Each block only contained one type of stimuli. The different stimulus blocks alternated with each other, such that the six types of stimulus blocks were each presented once before any one of them was presented the second time, and so on. The order of block presentation was counterbalanced across participants. Forty trials (20 for Roman letters, 20 for Roman strings) were introduced at the beginning as practice. EEG/ERP methods Scalp voltages were collected with a 128-channel Geodesic Sensor NetTM (Tucker, 1993) connected to an AC-coupled, 128-channel, highN170 and Letter Expertise 13 input impedance amplifier (200 MΩ, Net AmpsTM, Electrical Geodesics Inc., Eugene, OR). Amplified analog voltages (0.1-100 Hz bandpass, -3 dB) were digitized at 250 Hz. Individual sensors were adjusted until impedances were less than 50 kΩ. The EEG was digitally low-pass filtered at 40 Hz. Trials were discarded from analyses if they contained incorrect responses, eye movements (EOG over 70 μV), or more than 20% of channels were bad (average amplitude over 100 μV or transit amplitude over 50 μV). The mean number of trials per subject per condition was 90 (range = 63 to 100). Individual bad channels were replaced on a trial-bytrial basis with a spherical spline algorithm (Srinivasan, Nunez, Silberstein, Tucker, & Cadusch, 1996). EEG was measured with respect to a vertex reference (Cz), but an average-reference transformation was used to minimize the effects of reference-site activity and accurately estimate the scalp topography of the measured electrical fields (Bertrand, Perin, & Pernier, 1985; Curran, Tucker, Kutas, & Posner, 1993; Dien, 1998; Lehman & Skrandies, 1985; Picton, Lins, & Scherg, 1995; Tucker, Liotti, Potts, Russell, & Posner, 1994). Average-reference ERPs were computed for each channel as the voltage difference between that channel and the average of all channels. The average reference was corrected for the polar average reference effect (Junghöfer, Elbert, Tucker, & Braun, 1999). N170 and Letter Expertise 14 ERPs were baseline-corrected with respect to a 100-ms prestimulus recording interval.