There has been a long-standing and intense interest in alphabetic scripts regarding how positional information of letters in a word is coded (see review of different models in Grainger and Van Heuven, 2003). This is understandable as about 34% of all four-letter words in English and French can form other words by rearranging their letters (Shillcock et al., 2001). Therefore for alphabetic scripts, spatial specification of letter position is vital to correct word recognition and production (Grainger et al., 2006). While the letters are arranged linearly, radicals (also referred to as components or constituents) in a Chinese character are arranged in a two-dimensional square shape. The same question can be raised whether spatial information of components is similarly necessary. That is, are radicals specified for position of occurrence in the orthographic lexicon? The present study investigated the relevance of positional information of orthographic units during the early stages of visual character recognition using event-related potential (ERP) with a primed-lexical decision task.

The Chinese writing system is a non-alphabetic script with words composed of characters that represent morphemes. It has often been characterized as morphosyllabic. Each character occupies a constant square-shaped space that is constructed by combination of stroke patterns. Particular grouping of strokes form radicals that may exist as characters. These radicals may combine to form complex characters (Hoosain, 1991). Over 80% of complex characters contain radicals carrying probabilistic phonetic cues (e.g. , /je4/¹, is the phonetic radical of ,/je4/) or semantic cues (e.g., “wood/plant” is the semantic radical of “tree”), and their reliability can vary across characters (Chen et al., 1996). It has been reported that up to 10 different spatial arrangement of radicals, or configurations, can be found in complex characters (Fu, 1993), such as horizontal (AB , ABC ), vertical (), and semi-enclosed configuration ().

Studies investigating the properties of radicals have shown that they constitute an important level of representation in the orthographic processing system (e.g., Li and Chen, 1997; Taft et al., 2000), and influence response latencies in lexical decision (e.g., Feldman and Siok, 1997, 1999; Taft and Zhu, 1997), naming (e.g., Taft et al., 1999; Zhou and Marslen-Wilson, 1999; Lee et al., 2005), semantic judgment (e.g., Chen and Weekes, 2004; Chen et al., 2006), and priming tasks (e.g., Ding et al., 2004). They show that character activation is achieved via radical activation in complex characters. For example, Taft and Zhu (1997) found an effect of radical type frequency (number of characters containing that radical irrespective of function) and radical status (real or invented radical) in a lexical decision task.

Growing evidence from electrophysiological studies using ERP has also shown that radicals modulate the N1/N170, frontal P200, and N400 components. Focusing on the phonetic radical, Hsu et al. (2009) found interactive effects of phonetic combinability (neighborhood size of the phonetic radical) and phonological consistency (degree of agreement in pronunciation among orthographic neighbors having the same phonetic radical) at N170 in the occipital region, and main effects of phonetic combinability and phonological consistency at the frontal P200 component. Similar findings were also reported by Lee et al. (2006a,b, 2007) showing that phonetic radical consistency and regularity of the (phonetic) radical was reflected in the frontal P200 component, which Liu et al. (2003) suggested was related to phonological processing. Consistency effects were also found at the late N400 indicating that radicals are involved in later lexical processing via what the authors proposed to be competition of other phonetically similar neighbors (Lee et al., 2006a,b, 2007).

While a number of studies show that the radical acts as a sub-lexical representational unit in the Chinese script, the theoretical question we put forward is to test whether it is necessary to have separate position-specific representations for radicals as argued by Taft and his colleagues in their Multilevel Interactive-Activation framework (Taft and Zhu, 1997; Taft et al., 1999; Taft, 2006). For example, a radical () can be located at different positions in a character, such as on the left (e.g., ), right (e.g., ), top (e.g., ), or bottom (e.g., ; Hoosain, 1991). According to Taft and colleagues, the radical in each of the characters above have their own position-specific representation activated (position-specific view), so that the character would activate a left- radical, and the character would activate a top- radical. Although all models of Chinese character recognition allow for the representation of radicals, they differ in opinion about how spatial information of these radicals is characterized (e.g., Perfetti et al., 2005; Perfetti and Liu, 2006; Yang et al., 2009).

The claim for separate position-specific representations was based on a study by Taft and Zhu (1997) showing that character decision latency was influenced by radical position frequency (number of characters containing that radical in that position). Specifically, for two characters of equal frequency, the one with high radical position frequency (e.g., the character containing which occurs on the right-hand side of many characters) was easier to be recognized than the one with low radical position frequency (e.g., containing which occurs on the right-hand side of few characters). Taft et al. (1999) also argued that activation of the appropriate position-specific radical unit within a character would lead to lateral inhibition of other inappropriate position-specific radical units as no interference effects were found when recognizing transposable characters using a lexical decision or naming tasks. Thus, characters containing transposable radicals (e.g., and ) had response latencies and error rates that were comparable to those containing non-transposable radicals (e.g., ). This suggested to them that the same radical occurring at different positions were represented independently. To further support their claim, Ding et al. (2004) show significant priming effects when prime-target characters shared a radical in the same position (e.g., ) but not when they shared a radical in different positions (e.g., ) using a primed-lexical decision task. It was argued that pre-activation of the radical from a similar radical position prime led to faster target lexical decision latencies. However, the priming effect that Ding et al. found is problematic as facilitation in trials with radicals in similar position also shared more visual overlap than when in different positions.

Furthermore, contradictory findings have also been reported, leading some to argue that radicals are not coded for position (position-general view). For instance, the Lexical Constituency Model (Perfetti et al., 2005; Perfetti and Liu, 2006) claims that radicals do not require position information, and postulates instead configuration “slots” to allocate the position-free radicals. This implies that configural information, represented as input units in the model (e.g., left-right or top-bottom), is activated independently from the radicals themselves. Note that Taft’s model also includes position-general radicals that consequently activate position-specific radicals. However, the activation of a complex character is achieved via the activation of its position-specific radicals. This line of reasoning has mainly risen from studies using illusory conjunction and visual search tasks. Tsang and Chen (2009) found using a illusory conjunction task that participants would mistakenly perceive the target character as being one of the two preceding “source” characters when target and source characters have shared radical(s). Importantly, no difference was observed when the shared radical occurred in the same (e.g., preceded by and ) or different ( preceded by and ) positions across source-target. This finding was taken to argue that radicals were position-general and would activate all characters containing it irrespective of their position. Yet, it is unclear in that study why source characters with radicals in the same position as the target were less error prone than ones in different positions, and a latency difference of approximately 40 ms did not reach statistical significance. Using a visual search paradigm, Yeh and Li (2002) found that target characters took longer to identify when embedded in an array of characters sharing a radical and the same configuration but with the common radical in same or different positions than when they appeared in an array of control (unrelated) characters. Nonetheless, the generalizability of its findings may be challenged with only two target characters manipulated to form eight distractor-target pairs, which were not balanced for character frequency (i.e., 5.18–422.53 per million).

Given the limitations and inconsistencies of previous findings, this study separated the effects of visual overlap resulting from position similarity across prime and target in a primed-lexical decision task, and more significantly took a different conceptual approach to specification of radical position. Spatial position of radicals was explicitly conceived in terms of relative frequency across positions of a radical. As mentioned before, a radical can occupy different positions in characters, with some position encompassing more characters than other positions. For example, the radical can be found in at least four different positions in characters but it occurs more frequently on the right (76%, e.g., ) than on the left (9%, e.g., ), top (2%, e.g., ), or bottom (4%, e.g., ), taking into account type-token frequency. Therefore a radical can have a preferred or dominant position (high type-token frequency), while other positions that the radical can also occur in may be considered less preferred or subordinate positions (low type-token frequency). We argue that this conceptualization provides a more controlled test for position-specific radical representations as it relies on the relative frequency of distribution within the radical neighborhood. The contrast between high and low radical position frequency while controlling for character frequency and overall radical frequency in Experiment 2 of Taft and Zhu (1997) can be considered similar to our concept of dominant vs. subordinate positions, although our current design is superior in some important aspects. First, response latencies to characters containing the same radicals were compared as a function of dominant vs. subordinate positions. Unlike in Taft and Zhu where different radicals for high and low radical position frequency were used, without explicitly controlling for factors that may well influence lexical decision latency, including neighborhood size, phonological consistency, and orthographic complexity. Second, our manipulation took into consideration not only type frequency (i.e., number of characters in which a radical appears in a particular position) as in Taft and Zhu, but also token frequency (i.e., the frequency of each of the characters containing the radical in that position), which has been shown to affect the speed of lexical decision (Lee et al., 2005). Finally, the contrast of dominant vs. subordinate positions equally involved the left and right positions or the top and bottom positions of horizontally or vertically structured characters in this study (see details in Materials and Methods), dissimilar to the exclusive focus on the right or bottom position in Taft and Zhu, which may have inadvertently confounded radical position with function. For example, radicals that occur in the right and bottom position are more likely to be phonetic radicals and thus loosely linked to the phonological function. If spatial position information is inherently specified at the radical level, such information should be sensitive to its relative position distribution. Thus, characters with a radical in its dominant position would be recognized faster than those containing a radical in subordinate position, as characters with radicals in preferred (high type-token frequency) locations may require less effort to be activated relative to less preferred (low type-token frequency) locations.

Event-related potential in addition to response latency was collected, as ERP provides excellent temporal resolution and can reveal the unfolding of graphic, phonological, and semantic processes in visual word recognition online (e.g., Perfetti and Tan, 1998; Liu et al., 2003; Hauk et al., 2006; Holcomb and Grainger, 2006). Although previous studies in Chinese character recognition had not investigated ERP components that functionally reflected the processing of position-specific radical representation, ERP components known to reflect semantic and phonetic radical analysis were selected as components of interest, specifically the N1/N170, P200, and N400 components. Using silent naming, Lee and colleagues (Lee et al., 2006a,b, 2007; Hsu et al., 2009) showed that radical processing is associated with the N1/N170, and frontal P200 component, which they suggested reflects activation of radical processing at the visual word form area during the mapping of orthography-to-phonology. Greater N1/N170 negativity was found at electrodes P5/P6, P7/P8, PO5/PO6, and PO7/PO8 for characters that encompassed radicals with high combinability/neighborhoods size (Hsu et al., 2009). They suggest that the N170 component is an index of orthographic detection during the early perceptual categorization stage (see also Bentin et al., 1999), and greater visual experience from highly combinable characters lead to more efficient and specialized processing, thus, showed greater activation at the N170 than low combinability characters.

Lee and colleagues suggested that this early stage of visual word recognition shapes later orthographic-to-phonological conversion of the character’s radical that was reflected in the N170 and more robust at the frontal P200 component (see also Sereno et al., 1998). Sensitivity to the consistency of the (phonetic) radical’s pronunciation within a character showed smaller N170 effects and smaller P200 effects were found at electrodes F3/F4, FC3/FC4, C3/C4, CP3/CP4 Fz, FCZ, FC3, Cz, and CPz for characters that were highly consistent, with the greatest significant difference at the left frontal site F3 (Lee et al., 2007; see also Lee et al., 2006b; Hsu et al., 2009). In a separate task, Liu et al. (2003) showed with a primed pronunciation task that characters sharing similar radicals (graphical similarity) between the prime and target elicited smaller P200 at frontal and central electrodes sites. The N170 and (frontal) P200 components were further found to be sensitive to differences between two types of characters having opposite arrangement of semantic and phonetic radicals (one with semantic radical on the left and phonetic radical on the right vs. one with the opposite alignment; Hsiao et al., 2007). Note though that, this finding could also suggest that character recognition may be sensitive to the positions in which radicals are more likely to occupy as 89.9% of phonetic radicals occur on the right side of characters with semantic radicals on the left of horizontally structured complex characters (Hsiao and Shillcock, 2006).

Radical processing has also been shown to affect the N400 component. Lee et al. (2007) suggests that the N400 reflects a later stage of lexical processing after the P200. Greater N400 component was found for high phonologically consistent radicals at the central region electrodes Fz, FCz, Cz, CPz, and Pz. They argued that due to more homophone characters found in the high consistency condition, greater lexical competition would occur leading to an enhanced N400. In addition, Hsu et al. (2009) found using similar electrodes of interest that highly combinable radicals also elicit a greater N400 component, suggesting that highly combinable radicals increase semantic competition at the N400 (see also Holcomb et al., 2002). Liu et al. (2003) also identified that characters preceded by visually similar primes showed smaller N400 amplitudes at the central region during a semantic relatedness judgment task.

In light of previous ERP findings, it is assumed that position information is encoded early (Taft, 2006), and will modulate the early stages of character recognition in which visual-perceptual analysis proceeds to the orthographic stage as reflected in the N1/N170 and P1 visual components, but may also influence the frontal P200 and N400 (Liu et al., 2003; Lee et al., 2007; Hsu et al., 2009). It is predicted that characters with similar radical position to their primes will show a more negative P1, N1/N170, and reduced N400 due to prior exposure from the prime, while the position effect of dominance may be reflected in the later components, including the N170 and N400 components.

The aims of this study were to assess Taft and colleagues’ claim for independent representation of position-sensitive radicals and to identify ERP components that may reflect position-specific radical processing and the stage(s) at which the associated effects take place. Unlike previous studies, this investigation separated the effect of visual overlap from that of position specification in radicals by manipulating whether prime-target pairs shared similar radical positions or whether the target’s radical differed in relative position frequency factorially. While the behavioral results only showed a position similarity effect, the ERP findings revealed a more complex contribution/relationship of radical processing during the stages of lexical processing. Radical dominance effects were found at the occipital N170-frontal P200 and N400, in addition to the observations of visual similarity effects at the early occipital P1 and later N400 components. Neural sensitivity to radical dominance supports the view that position-specific radicals are activated early, and that the spatial relationship among orthographic units within a character can impact on character recognition. The following discussion considers the position similarity effect reflected in response latency, and examines each of the significant ERP components and attempts to integrate the temporal dynamics of radical processing into Taft’s model of character processing (Taft et al., 1999; Taft, 2006).

Based on the behavioral effects, when a target character shared a radical in similar position to its prime, participants were faster to recognize the target due to pre-activation of the relevant position-specific constituent radical in the prime facilitating the recognition of the target character. Previously, this has been taken as evidence for position-specific information of radical representation (Taft and Zhu, 1997; Ding et al., 2004), but such an interpretation may be problematic. First, the ERP findings confirmed that the facilitation was primarily driven by visual overlap between the prime and target radical rather than from independent representations of position-sensitive radicals per se. This was because radical position effects were found in early components known to reflect visual analysis at the P150 component in the occipital regions (Grainger and Holcomb, 2009). Less activation was needed when primes and targets shared a radical in the same position compared to different positions, particularly in the left hemisphere. Moreover, although the radical position effects found at the N400 could be indicative of position-specific radical representational processing as radicals in different positions required more effort for integration as they elicited greater N400, one may nevertheless argue that such an observation would suggest that the independent radical representations only influence lexical processing.

Contrasting the effects of radical position and radical dominance in relation to the time course of Chinese character recognition, the findings suggest that visual overlap/similarity is initially processed before access to radical representation as radical position effect precedes the radical dominance effect. The effects of radical dominance provide stronger support for independent representation of position-specific radicals, considering that it is unaffected by the degree of visual overlap between prime and target. Greater negativity at occipital N170 and positivity at frontal P200 for characters with radicals in subordinate position may be taken to reflect increased processing effort to activate the less frequently encountered subordinate radical representation. On the other hand, dominant radical representations are frequently activated because they are connected to more characters containing the radicals in the same positions, and thereby have a lower activation threshold. As such, characters with radicals in dominant positions would require less effort to process, at least initially. This pattern of less activation for characters with radicals in dominant position, however, changes at the later N400 component when character recognition proceeds from radical level facilitation to lexical level competition. Note that radical position effects however, continue to show a smaller N400 component for target characters sharing a similar radical position to their primes. At the N400 component, characters with dominant position radicals elicited greater negativity compared with subordinate position. We argue that radical processing at the N400 reflects competition at the lexical level (see also Lee et al., 2007) as dominant radical position characters naturally have more neighbors that are simultaneously activated, and thus may require greater effort and/or lateral inhibition to suppress irrelevant neighboring competitors to select the appropriate lexical entry. Characters with subordinate position radicals, on the other hand, co-activate a smaller set of neighbors and would therefore involve less conflict resolution. An issue, however, arises that the N400 is generally considered to be sensitive to lexico-semantic features and assumed to represent post-lexical semantic integration or access to semantic representations in the long term memory (e.g., Kutas and Hillyard, 1980, 1984; Nobre and McCarthy, 1994; Barber and Kutas, 2007). However, the N400 time window of 300–450 ms in this study is earlier than the typical N400 associated with post-lexical semantic integration; thus, the authors argue that the N400 here may capture the earlier phase of the N400 (see Grainger and Holcomb, 2009, for a similar interpretation of ERP masked repetition priming effects). All models of Chinese character recognition take the view that radicals serve as perceptual input only (Taft et al., 1999; Perfetti et al., 2005; Perfetti and Liu, 2006; Taft, 2006), whether the N400 component continues to reflect lexical competition at the word form level or interference at the semantic level from non-target neighbors requires further investigation. We are, however, inclined to argue that the interference reflected in the earlier N400 in our findings occurs at the word form level as equal numbers of semantic and phonetic radicals served as target radicals, and all prime-target pairs were semantically unrelated. Another pertinent observation is that the main effect at the P200 and N400 show similar topographic difference distribution suggesting that the P200 effect argued to be at the sub-lexical level could be related to the later N400. However, we favor the view that the N400 effect is a lexical event associated with the more conventional interpretation of lexical level competition. Nonetheless, what is clear is that position specified radicals play a crucial role in character recognition in Chinese, and affect this process in a complex manner that behavioral experiments alone cannot reveal. Specifically, changes from facilitation to competition of the position-specific radicals from the radical level to the lexical level may offset each other, and result in a null effect of radical dominance in RT.

The presence of radical dominance effects at the occipital N170 could be seen as reflecting neighborhood size effects (Chen and Weekes, 2004; Chen et al., 2006; Hauk et al., 2009; Hsu et al., 2009) or a neighborhood-by-position interaction (Grainger et al., 2006). The observation of greater processing effort for characters with subordinate radicals is consistent with findings of neighborhood size effects reported from behavioral, functional imaging, and ERP studies in the form of facilitative effects for words with larger neighborhoods (Hsiao et al., 2007; Hsu et al., 2009, 2011; Li et al., 2010). Hsiao et al.’s (2007) study showing position effects at occipital N170 and frontal P200 components can also be described as a type of neighborhood-by-position bias effects between two types of characters having opposite distribution of semantic and phonetic radicals of left-right configuration characters. The dominant radical positions in this study can be analogous to their SP condition where the phonetic radical occurs on the right side of character and left for semantic radical (S and P denote the semantic and phonetic radical, respectively). The less preferred or subordinate position has the opposite alignment (PS). In this case, greater negativity at the occipital N170 or positivity at the frontal P200 for the less preferred positions of radicals is similar to our findings. However, our study differs significantly from Hsiao et al. as it takes a more parsimonious approach of not conflating radical position with functional specificity (see also Zhou and Marslen-Wilson, 1999). As radicals assuming particular functions are more likely to be located in particular positions, the relationship between radical position and functional units still requires delineating. If indeed the radical dominance effects can be conceptualized to be associated with the neighborhood size effect, we suggest that the N170 could reflect a lexico-orthographic component in which the representation of the word form neighborhood is processed. Based on Grainger and Holcomb (2010), the time course of the N170 component in the current study, corresponding to their N250, suggests that activation from orthographic units (e.g., radicals) to words (e.g., character) may take place in this time window with its neural generator located at the left fusiform gyrus region associated with the visual word from area (VWFA). In relating the N170 to N250 in Grainger and Holcomb, it is notable that the time course of our N170 (as well as P1) peaks at a similar time to Grainger and Holcomb (i.e., P150 and N250) which use a similar masked priming design. This may account for the delay of our N170 compared to previous ERP studies with Chinese radicals which typically used covert naming (e.g., Lee et al., 2007) or semantic judgment tasks (e.g., Liu et al., 2003) without a mask.

Finally, the early N1 component observed in this study may be a consequence of the prime’s N1 component overlapping with the target character, and reflect a delayed N1 activity of the prime. This may explain the null effects and more crucially its early peak latency of the N1 at 75 ms (prime + backward mask = 112 ms). The typical visual N1 component peaks at around 120–150 ms.

To integrate the present results within the context of Taft’s (2006) interactive-activation model in terms of the temporal dynamics of radical processing, we propose that position-general radicals are initially activated at the occipital P1 component at approximately 150 ms, and subsequently spread activation to their position-coded radicals. Around 280 ms in the occipital N170/frontal P200 component, character representations containing position-specific radicals are activated and lexical selection of the word form (or semantics) occurs at approximately 380 ms at the N400, with greater negativity reflecting greater lexical competition.

In conclusion, our results support the role of position-specific radicals in orthographic processing as proposed by Taft (2006) and Taft et al. (1999). This study importantly separates facilitative effects due to visual overlap from position information of radicals and proposes a temporal framework for radical processing. Distinct position-specific representations are conceptualized in terms of dominance of radical position via relative type-token frequency, and we demonstrated that independent position-specific radicals are activated within the first 250 ms of character recognition. Such early access is revealed via modulation of the occipital N170/frontal P200 component, followed by the retrieval of lexico-orthographic or lexico-semantic information at the N400. Our evidence for position-specific radical representations, therefore, highlights the over simplication of the Lexical Constituency Model (Perfetti et al., 2005) as an account of Chinese character processing. While character configuration may be relevant, it alone may not be adequate to access word form representation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.