A central question in psycholinguistic research concerns the types of information stored in the mental lexicon. Whatever theoretical perspective one takes, it is evident that the formal properties of words, such as how words sound or look, must be lexically represented. An issue still in debate is whether and how sublexical information might be represented in the mental lexicon. Among the studies examining this issue, the role of morphological structure has been most extensively studied in the visual domain. The accumulated evidence is far from conclusive.#

In general, there are two theories of morphological processing. One is the semantic dependency hypothesis which proposes that complex words share lexical representations with their morphemic constituents only when there is a semantically transparent relationship between the complex words and their constituents (Marslen-Wilson et al., 1994). Semantic transparency is defined as whether a word has a transparent meaning relationship with its constituent words. For example, the meanings of semantically transparent words can be derived from the meanings of their constituents (e.g., milkman, cleaner). Words are semantically opaque if their meanings are unrelated to the meanings of their constituents (e.g., butterfly, corner). This hypothesis is supported by evidence from priming studies where the primes were semantically associated with the constituents of semantically transparent compounds (e.g., women–MILKMAN), semantically opaque compounds (e.g., bread–BUTTERFLY), or pseudo-compounds (e.g., girl–BOYCOTT), which serve as targets. A facilitative semantic priming effect was found for the constituents of semantically transparent compounds, but no reliable semantic priming was found for the components of semantically opaque compounds and pseudo-compounds (Sandra 1990; Zwitserlood 1994).#

An alternative hypothesis proposes that any input string comprised of a morphological surface structure may be subject to decomposition irrespective of semantic transparency (Taft and Forster, 1975). This hypothesis is supported by Longtin et al. (2003), who found that the recognition of French stems was accelerated significantly and equivalently by semantically transparent and opaque words. Rastle et al. (2004) investigated what information is used to segment a complex word into its morphemic constituents during visual word recognition and found that letter strings comprising a morphological surface (e.g., cleaner, corner), no matter whether they are semantically transparent or opaque words, activate sublexical morphemic units (e.g., clean, corn, -er). Their results suggest that there is a purely orthographic structural decomposition, which occurs in early visual word recognition independent of semantic transparency.#

A fundamental difference in sublexical units is inherent between alphabetic (such as English) and non-alphabetic (such as Chinese) writing systems. In an alphabetic system, letters and words represent two distinct levels, where letters are constituent parts that are wholly contained within the word. In Chinese, approximately 80% of the characters are phonetic compounds that are made up of a semantic radical (usually on the left) and a phonetic radical (usually on the right). Very often, the semantic and phonetic radicals are stand-alone characters with their own pronunciations and meanings. In other words, the radicals are orthographic units that may have value as morphemes (unlike letters). To illustrate, the phonetic compound is pronounced as feng1 (the number represents one of the four tones in Mandarin Chinese) and has the meaning “maple”. Both its semantic radical (mu4, “wood”) and the phonetic radical (feng1, “wind”) are simple characters with their own meanings and pronunciations. There is increasing evidence to indicate that reading a complex character involves the processing of its radicals. For example, Yeh and Li reported a repetition blindness effect for radicals (with no other intervening character) when the duration of exposure is less than 50 ms and suggested that sublexical processing occurred relatively early in Chinese character recognition (Yeh and Li, 2004). The repetition of the semantic radical facilitated target identification, especially when prime and target were formed by the same semantic radical and when the meaning of the semantic radical was consistent with the meaning of the whole prime character (Feldman and Siok, 1999). In addition, researchers have tried to describe how well the phonetic radical provides cues to the pronunciation of the whole character in terms of its regularity, which is defined as whether the complex character is pronounced in the same way as its phonetic radical. For example, feng1 is pronounced the same as its phonetic radical feng1 and is defined as a regular character, whereas tsai1 is pronounced differently from its phonetic radical qing1 and is thus defined as an irregular character. Previous studies have also demonstrated a frequency by regularity interaction in naming Chinese phonograms: the speed of naming a regular character is much faster than that of naming an irregular character, especially when those characters are low in frequency (Hue 1992; Lee et al. 2005; Seidenberg 1985). These findings suggest that semantic and phonetic radicals play different functions in compounding phonograms. Semantic radicals usually indicate the semantic category of morphemes corresponding to the phonograms, whereas phonetic radicals provide the phonological clues for the whole characters.#

Since many semantic and phonetic radicals can be characters, having both meaning and pronunciation, they may participate at the lexical level and the sublexical level in different ways, that is, as a word at the higher level and as a potential cue to meaning or pronunciation at the sublexical level (Perfetti et al., 2005). However, according to the semantic dependency hypothesis, even when phonetic radicals are legal characters with their own semantic representations at the lexical level, they do not activate the semantic representations when they serve as sublexical units of the phonograms since they have nothing to do with the meaning of the whole character (i.e., they are semantically opaque). Zhou and Marslen-Wilson (1999) found a facilitative priming effect when a target (e.g., yu3, “rain”) was semantically related to the phonetic radical (e.g., , feng1, “wind”) embedded in the prime (e.g., , feng1, “maple”), even though the prime itself was not semantically related to the target. This evidence of semantic activation of a phonetic radical in complex characters appeared to last at least 200 ms regardless of whether the prime was a regular or irregular character. This suggested that the sublexical processing of phonetic radicals is, like lexical-level processing, not only a phonological event but also a semantic event and, thus, that there are no fundamental differences between the sublexical processing of phonetic radicals and the lexical processing of simple and complex characters (Zhou and Marslen-Wilson, 1999).#

A remaining question is how the dual processes, lexical and sublexical, interact with each other, especially for the phonetic radials. Because the main function of a phonetic radical is to serve as a clue to the pronunciation of the whole character rather than its meaning, Perfetti et al. (2005) proposed an interactive constituency model for Chinese character reading. In this model, the representation of a word consists of three interlocking constituents: orthographic, phonological, and semantic. The input units are the radicals and the spatial relationship between the radicals. Successful lexical retrieval (word identification) requires full specification of all three constituents. According to this model, the validity of a phonetic radical (whether a character in which the phonetic radical appears has the same pronunciation as the radical does as a stand-alone character) affects the sublexical semantic activation. Unfortunately, Zhou and Marslen-Wilson's study examined the sublexical semantic activation of regular and irregular characters in separate experiments, leaving this question unanswered.#

The present study aims to further explore how and when phonetic radicals can be manifested via lexical/sublexical processing, along with providing phonological information. In order to probe this question, different prime types and stimulus onset asynchronies (SOAs) were manipulated to replicate the sublexical semantic priming effect found by Zhou and Marslen-Wilson (1999). However, instead of using traditional reaction time measurements, we used event-related potentials (ERPs). The measurement of reaction time reflects the summation of all the steps that elapse prior to the behavioral response to a language task. In contrast, ERPs provide insight into the time course of language comprehension because they can provide data that reflect processing at each millisecond from the onset of the language stimuli. Most importantly, ERPs can reflect how cognitive operations are processed in the brain. For example, the N400 is a negative polarity ERP component that is maximal over centro-parietal electrode sites; it usually emerges at about 250 ms after the onset of word stimulation and reaches its maximal amplitude at around 400 ms. It is widely accepted that the N400 component elicited by words is particularly sensitive to the processing of semantics, both in prime–target word pairs (Kutas and Federmeier, 2000) and in sentential contexts (Kutas and Hillyard 1980a; Kutas and Hillyard 1980b). The amplitude of the N400 component is usually inversely related to the degree of fit between the word and its preceding semantic context. That is, the N400 amplitude is attenuated for words that are preceded by a semantically related word or semantically congruous context, compared to words that are preceded by an unprimed word or semantically incongruous context. Recent studies have also shown that indirectly related targets can elicit intermediate N400 amplitudes (Chwilla et al. 2000; Kiefer et al. 1998; Weisbrod et al. 1999).#

In this study, whether the semantic properties associated with a phonetic radical embedded in the two types of phonograms (regular and irregular characters) are automatically activated will be indexed by the N400 component. Four types of prime–target relationships, including lexical and sublexical semantic associations, are manipulated (see Table 1, the prime–target relationship will be explained in more detail in the Experimental design and materials section). We expect that both the lexical and sublexical semantic activation can be captured by the manifestation of the N400 component. In addition, in order to trace the time course of sublexical semantic activation, different SOAs between the prime and the target (50 ms, 100 ms, and 300 ms) were manipulated as a between-subject variable. We expect that the sublexical semantic activation of a phonetic radical will fade away over time since the meaning of the phonetic radical is usually irrelevant to, or may eventually interfere with, the meaning of the character. In addition, Tan and Perfetti (1997) demonstrated that the naming time for a Chinese target character (e.g., , kuan, “wide”) was facilitated by a prime (e.g.,, kuo, “include”) that was homophonically related to a synonym of the target, as well as by the synonym itself (e.g., , kuo, “wide”). This is the so-called phonologically mediated priming effect. Since the main function of a phonetic radical is to provide a phonological clue to the whole character, its phonological information should persist. Thus, when the phonetic radical is embedded in a regular character prime, which has the same sound as the phonetic radical, it has a greater chance of retaining its meaning through phonological mediation (Tan and Perfetti, 1997), relative to when it is embedded in an irregular character prime. In other words, regular and irregular characters may have different time courses or strength for sublexical semantic activations.#

The main purpose of this study was to examine whether there is sublexical semantic activation for phonetic radicals and how this process interacts with the function of the phonetic radical in providing phonological information for the whole phonogram. This was done by using the N400 ERP component as an index of semantic processing in order to examine the sublexical semantic priming effect for both regular and irregular phonograms. The results revealed a temporal sequence of sublexical semantic activation in relation to the phonetic radical representing the phonological clue. First of all, we observed a typical semantic priming effect of N400 for the primes (e.g., , fu1, “husband”) that were semantically related to the targets (e.g., , che1, “wife”) across the three SOAs of 50 ms, 100 ms, and 300 ms. Second, when the prime was a regular phonogram (e.g., , feng1, “maple”) whose phonetic radical (e.g., , feng1, “wind”) was semantically related to the target (e.g., , yu3, “rain”), significant N400 effects were found when the SOAs between the prime and the target were 50 ms and 100 ms, but the N400 effect disappeared at an SOA of 300 ms. Finally, when the prime was an irregular phonogram (e.g., , du2, “read”) whose phonetic radical (e.g., , mai4, “sell”) was semantically related to the target (e.g., , mai3, “buy”), a significant N400 effect could be found at the SOAs of 50 ms and 100 ms. However, this effect was significantly smaller at the SOA of 100 ms compared to that at 50 ms and totally disappeared at 300 ms. Furthermore, the regional analyses of difference wave over successive time periods revealed that, in both SOAs of the 50 ms and 100 ms conditions (see Table 2), the time periods showing significant sublexical semantic priming effects elicited by irregular phonograms (40 ms and 60 ms) seem to be shorter than that elicited by regular phonograms (60 ms and 100 ms). These results suggest that phonetic radicals automatically activate their corresponding semantic representations in the early stage of visual word recognition, which is congruent with Zhou and Marslen-Wilson's findings. However, in contrast to their findings, our results suggest that the sublexical semantic activation of the phonetic radical does not last long and that the decay function of the sublexical semantic activation is modulated by how well the phonetic radical represents the pronunciation of the whole phonogram, i.e., its level of regularity, as defined in the present study.#

In alphabetic languages, there is no clear visual cue for the separation between intralexical orthographic units. Researchers have aimed to understand what information is used to segment a complex word into its morphemic constituents during visual word recognition and, in particular, whether semantic information plays a role in the segmentation. The absence of sublexical semantic activation in reading monomorphemic words (e.g., boycott) or semantically opaque words (e.g., butterfly) in English (Sandra, 1990) or Dutch (Zwitserlood, 1994) has been used to support the semantic dependency hypothesis. However, recent studies have demonstrated a statistically equivalent semantic priming effect from semantically transparent and semantically opaque primes, especially at short SOAs (Rastle and Davis 2003; Rastle et al. 2000). These findings indicate that the decomposition operating in early visual word recognition is not meaning-dependent but instead is structurally based (Rastle et al., 2004).#

Complex characters in Chinese can be thought of as analogous to compound words, with both semantic and phonetic radicals representing meanings. Visually, there is usually a clear boundary between phonetic and semantic radicals, especially for phonograms with left–right structure. That makes Chinese phonetic radicals much more salient in complex characters than letter clusters are in alphabetic words. However, the phonetic radical usually has nothing to do with the meaning of the whole complex character. The evidence for automatic sublexical semantic activation of Chinese phonetic radicals supports the hypothesis that the sublexical decomposition is purely structural and arises in the early stage of visual word recognition. Furthermore, it also suggests that there are at least two levels of representations in the Chinese orthographic system, a radical level and a character level. This is congruent with the hierarchical model of Chinese visual word recognition proposed by Taft and his colleagues (Ding et al. 2004; Taft and Zhu 1997; Taft et al. 1999) and the lexical constituent model proposed by Perfetti, Liu, and Tan (Perfetti et al., 2005).#

Taft's model adopts the interactive activation framework and assumes that there are three subsystems: orthography, phonology, and semantics. In the orthographic subsystem, radicals are explicitly represented and these representations are position specific. A complex character is recognized via its position-specific radical representations and these, in turn, are activated through a position-free representation of that radical. When these position-free radicals are legal characters on their own, they will in turn connect to their corresponding phonological and semantic representations. This model explains why semantic and phonetic radicals will activate their semantic and phonological representations when these radicals are characters on their own. However, Taft's model also suggests that the character is pronounced only via the mediation of a lemma unit (representing a correlation between form and meaning), which means that the linkage between the phonological and lemma or semantic systems is unidirectional. Therefore, in this model, the phonology is not crucial for the meaning of a character to be generated. This is difficult to reconcile with the current finding, which demonstrates different decay functions of sublexical semantic activation for regular and irregular phonograms. In addition, our findings suggest that it is possible to activate meaning via the mediation of the phonological system. The semantic activations from orthographic and phonological systems are consistent for regular characters, but less consistent for irregular characters, which makes the sublexical semantic activation of the phonetic radical last longer for regular characters than for irregular characters. However, this is not to claim that phonological mediation is obligatory, but to show that a pathway from phonology to meaning is possible.#

On the other hand, the current findings are congruent with the lexical constituent model, which assumes that a word representation consists of the three interlocking constituents of orthography, phonology, and semantics and a successful lexical access needs the activation of all three constituents. According to this model, the input units are radicals and the spatial relationship between the radicals. Those radicals that are characters would have both meaning and syllable-level pronunciation and they each could be connected to the semantic level. That phonetic radical of an IP character did not take longer to activate its semantic value than that of an RP character can be explained by the direct linkage between orthography and semantics, which allows for a rapid retrieval of meaning without phonological involvement. In other words, for both RP and IP characters, their phonetic radical can activate their meaning through the direct linkage between orthography and semantics in the early stage. Meanwhile, the connection between phonology and semantics provides the second route for the mediated access to the meaning. Thus, the phonetic radical of an RP character would receive stronger semantic activation due to the congruent phonological activation of phonetic radical and the whole character. The lexical constituency model thus can capture the regularity effect by representing the pronunciations of radicals that are part of the orthographic character level and connecting their output to phonological units and will explain how phonology interacts with the other constituents of word representation. The present findings of the automatic sublexical semantic activation and how this activation interacts with the phonetic validity are congruent with the prediction of the lexical constituency model for word identification.#

In conclusion, precise meaning is extracted from a very brief viewing of a spelled word or a written character when reading, amid conflicting information embedded in the script. To accomplish this, a reader's brain has to carry out complex neural synthesis in order to resolve the conflict. The present data used the N400 ERP component to demonstrate that the sublexical processing of phonetic radicals in reading Chinese involves activation of semantic properties corresponding to the radicals and the whole character. Readers pick up all the semantic information associated with both lexical and sublexical levels within 50 to 100 ms. Since the meanings of phonetic radicals are usually irrelevant to the meanings of the whole characters, readers will have to solve the conflict between these two levels. Our data showed that the sublexical semantic activation of phonetic radicals were modulated by the validity of the phonological information that the phonetic radical carries. However, none of the sublexical semantic information associated with the phonetic radical was preserved after 300 ms. These data reveal the temporal dynamics of the sublexical semantic and phonological processing and support the framework of the lexical constituent model, which reveals the direct linkages among the orthographic, phonological, and semantic lexicons in reading Chinese.#