Content area
Elucidating the interaction between lexical processing and word learning is essential for a complete understanding of the underlying mechanisms of each of them. Long-term priming for words reflects an interplay between lexical processing and word learning. Although robust long-term priming effects have been found between two occurrences of the same word and between semantically similar words, it remains unclear whether long-term priming between orthographically similar words (i.e., long-term form priming) is a reliable effect. Following the theoretical analysis based on the connectionist framework, we articulated the possibility that long-term form priming might be modulated by the phonological congruency between the prime and target words, and that if this modulator was under control, reliable effects of long-term form priming would emerge. However, this hypothesis has not been adequately tested empirically. The present study tested this hypothesis by using Chinese phonograms and the phonetic radicals embedded in them as the prime and target items. In three experiments that varied in the types of stimuli and testing tasks, we consistently found that when the prime and target had the same phonology, naming the prime facilitated later processing of the target, while when they had different phonologies, the priming effect was inhibitory. These observations were consistent with the connectionist account of long-term priming for words. Our findings help confirm the reliability, generalizability, and robustness of long-term form priming and elucidate its underlying mechanisms, and suggesting promising future directions on the interactions between lexical processing and word learning.
Abstract
Elucidating the interaction between lexical processing and word learning is essential for a complete understanding of the underlying mechanisms of each of them. Long-term priming for words reflects an interplay between lexical processing and word learning. Although robust long-term priming effects have been found between two occurrences of the same word and between semantically similar words, it remains unclear whether long-term priming between orthographically similar words (i.e., long-term form priming) is a reliable effect. Following the theoretical analysis based on the connectionist framework, we articulated the possibility that long-term form priming might be modulated by the phonological congruency between the prime and target words, and that if this modulator was under control, reliable effects of long-term form priming would emerge. However, this hypothesis has not been adequately tested empirically. The present study tested this hypothesis by using Chinese phonograms and the phonetic radicals embedded in them as the prime and target items. In three experiments that varied in the types of stimuli and testing tasks, we consistently found that when the prime and target had the same phonology, naming the prime facilitated later processing of the target, while when they had different phonologies, the priming effect was inhibitory. These observations were consistent with the connectionist account of long-term priming for words. Our findings help confirm the reliability, generalizability, and robustness of long-term form priming and elucidate its underlying mechanisms, and suggesting promising future directions on the interactions between lexical processing and word learning.
Keywords Long-term form priming * Phonological congruency * Connectionist * Lexical processing * Word learning
(ProQuest: ... denotes non-USASCII text omitted.)
Introduction
Long-term priming for words refers to the phenomenon that processing a word could influence later processing of the same or a related word (Bowers et al., 2002). As the interval between two lexical processing episodes could be minutes, hours, or sometimes even longer, long-term priming for words reflects learning rather than transient changes of lexical activation (Bowers et al., 2002; Oppenheim et al., 2010). Long-term priming for words is important because it bridges the gap between lexical processing and word learning, two processes that have typically been examined independently. In principle, word learning is usually seen as a process of acquiring new information about a word, such as visual word form, sound pattern, and meaning, extensively examined with novel words whose representations need to be built up from scratch (Kapnoula & McMurray, 2016). On the other hand, lexical processing is usually seen as a process of mere usage of known words. However, by using known words, findings on long-term priming for words suggest that experiences of using these words might constantly tune their representations (Oppenheim et al., 2010), thus enriching our understanding of the mechanisms of both lexical processing and word learning.
The present study focuses on long-term form priming, i.e., long-term priming between orthographically similar words. This type of long-term priming for words is much less understood compared to other types, such as long-term repetition priming between two occurrences of the same word (e.g., Gupta & Cohen, 2002; Stark & McClelland, 2000), and long-term priming between semantically similar words (e.g., Becker et al., 1997). Fundamentally, it is unclear whether long-term form priming is a reliable effect, because inconsistent results have been found across studies (e.g., Pexman et al., 2005; Rueckl & Mathew, 1999; Seidenberg et al., 1984; Wagenmakers & Raaijmakers, 2006). After a theoretical analysis based on the connectionist framework (McClelland & Rumelhart, 1985; McLeod et al., 1998), we suggested that long-term form priming might be modulated by the phonological congruency between the prime and target words (i.e., whether their shared orthography is associated with the same pronunciation). We hypothesize that stable priming effects would emerge when this factor was under control. However, this hypothesis has not been adequately tested empirically, in either alphabetic or non-alphabetic languages. The present study tested it using Chinese characters. Our findings could not only inform the fundamental question of whether long-term form priming indeed exists, but also could shed light on its underlying mechanisms and mechanisms of Chinese language processing.
The connectionist account of long-term priming for words
In fact, a coherent account of long-term priming for words could be derived from the connectionist framework. According to this framework, long-term priming for words would arise naturally from the changes in lexical representations after processing a word (Hinton et al., 1986; McClelland & Rumelhart, 1985; Stark & McClelland, 2000). In a connectionist model, processing a word involves activation spreading from the input nodes (e.g., orthography) to the output nodes (e.g., phonology) along the weighted connections. As a result, learning occurs as exhibited by gradual changes in the connection weights, which in turn influences future processing of the same word (i.e., repetition priming). Such a process might also have an impact on similar words that overlap with the processed word in orthography, phonology, and semantics, thus leading to semantics-related and formrelated long-term priming (Hinton et al., 1986; McClelland & Rumelhart, 1985; McLeod et al., 1998; Oppenheim et al., 2010; Stark & McClelland, 2000).
Importantly, the connectionist account would also predict that the direction of long-term form priming would be modulated by whether the shared orthography between the prime and target is mapped onto the same output, such as phonological segment or semantic features (Hinton et al., 1986; Stark & McClelland, 2000). For example, in models that implemented a word naming task (e.g., Seidenberg & McClelland, 1989), when the prime and target words whose shared orthography is mapped onto the same phonology (e.g., in "Cat" and "Coat", the letter C is pronounced as /k/ in both instances), changes in the shared connections ("C""/k/") as resulting from processing the prime would also benefit the processing of the target. On the other hand, if the shared orthography between the prime and target is mapped onto different phonologies (e.g., in "Cat" and "Cell", where the letter C is pronounced as /k/ and /s/, respectively), different connections are required to represent these conflicting mappings. Consequently, changes in the connections as a result of processing the prime would lead to forgetting the target. These opposing effects imply that long-term form priming should be scrutinized separately for different types of prime-target pairs.
Empirical studies on long-term form priming
Current studies have revealed that long-term form priming could be facilitative (e.g., Feustel et al., 1983; Pexman et al., 2005; Rueckl, 1990; Wagenmakers & Raaijmakers, 2006), inhibitory (e.g., Burt & Humphreys, 1993; Seidenberg et al., 1984; a trend in Napps & Fowler, 1987) or nonexistent (e.g., Murrell & Morton, 1974; Pexman et al., 2005; Rueckl & Mathew, 1999). The mixed findings summarized above give one the impression that long-term form priming is not a robust or reliable effect. However, it is also likely that the discrepancies were driven by some modulating factors that varied from study to study, for example, the phonological congruency between the prime and target words, as suggested by the connectionist models. Depicting the effects of long-term form priming requires that such factors are well controlled (e.g., are manipulated or kept constant across experimental conditions).
Only one study (Bowers et al., 2002) has directly examined the role of phonological congruency in modulating long-term form priming. It is worth mentioning that this work aimed to test whether long-term form priming could be simulated by Seidenberg and McClelland's (1989) model of word identification that instantiated back-propagation and distributed representations. While the simulation did not yield significant priming effects across all the data, post hoc analysis revealed facilitative priming when the target word (e.g., MINT) rhymed with the prime word (e.g., HINT), and inhibitory priming when the target and prime did not rhyme (e.g., PINT and HINT). Further, in the following behavioral experiment, Bowers et al. (2002) manipulated whether the prime and target rhymed (i.e., whether the shared orthographic segment had the same pronunciation), and used a naming task in the priming phase (in which the primes were processed) and a lexical decision task in the subsequent testing phase (in which the targets were processed). The results of this experiment replicated those of the simulation.
However, a later study (Pexman et al., 2005) cast doubt on whether the phonological congruency between the prime and target words would matter for long-term form priming. Specifically, this study used the exception words (e.g., PINT) as primes and their non-rhyming word body neighbors (e.g., MINT) as targets, thus providing a partial test of the modulating role of the phonological congruency in long-term form priming. They found no long-term priming when a naming task or a lexical decision task was used, but longterm facilitation (lasting up to 51 s) when a more difficult phonological lexical decision task was used (PLD, does it sound like a word?). Based on these findings, the authors argued that long-term form priming might not be a common effect, but instead would only occur in difficult lexical processing tasks (see also Hughes & Whittlesea, 2003). However, this contradicted some earlier findings of inhibitory long-term form priming between the exception words and their non-rhyming word body neighbors in naming and lexical decision tasks (Burt & Humphreys, 1993; Seidenberg et al., 1984). Pexman et al. (2005) speculated that these discrepancies might be due to some methodological differences between these studies (e.g., proportions of regular fillers). Additionally, it should be noted that as the primes used in Pexman et al. (2005) were only exception words, a special subset of English words, and thus the effectiveness of their findings might be limited in scope. In sum, the modulating role of the phonological congruency in long-term form priming remains to be further examined, by using different types of prime and target words and by making a full manipulation of the phonological congruency between them.
It is also unclear whether the phonological congruency would modulate long-term form priming in non-alphabetic languages, such as Chinese. The Chinese writing system differs from the alphabetic systems in many aspects. Specifically, the mapping from spelling to sound in Chinese is syllable-based with no constituent parts of a character corresponding to phonemes (Li, 1993; Zhou & MarslenWilson, 1999). This means that Chinese has a deep orthography where the pronunciation of a character cannot be assembled sound-by-sound from its constituent parts (Yang et al., 2013). While more than 80% of Chinese characters are phonograms that are semantic-phonetic compounds, it is important to note that the phonetic radical does not reliably indicate the pronunciation of the phonogram. Only about less than 30% of the phonograms are regular ones (e.g., "...", /qiu2/, "ball"), which have exactly the same pronunciation with their phonetic radical (e.g., "...", /qiu2/, "pursue"). About one-third are irregular ones (e.g., "...", / sa3/, "spray") that have a completely different pronunciation from their radical (e.g., "...", /xil/,"west") (Li, 1993; Zhou & Marslen-Wilson, 1999). The remaining phonograms are semi-regular ones, and could be further classified into those whose pronunciation had the same initial phoneme (e.g., "...", /bo2/, "white") with their phonetic radical (e.g., "...", /bai2/, "white"), those that had the same rhyming segment (e.g., "...", /choul/, "take out") with their radical (e.g., "...", /you2/, "reason"), and those that shared both of these segments but not a lexical tone (e.g., "...", /mal/, "mother") with it (e.g., "..." /ma3/, "horse") (Li, 1993; Zhou & Marslen-Wilson, 1999). Thus, different from the alphabetic languages, the irregular and semi-regular phonograms are not the minority in Chinese. They take up about 55% of Chinese characters in total, which is much more than the regular ones (about 25%). Additionally, the phonetic radicals themselves are generally characters (e.g., "..." /qiu2/, "pursue") that cannot be further decomposed and are classified as simple characters, taking up about 5% of Chinese characters. The rest characters are compound characters that are not phonograms (e.g., "...", /lin2/, "woods").
Presumably, due to the less transparent orthography-tophonology mappings, it has been shown that phonological processing plays a less important role in Chinese reading than in alphabetic languages (e.g., Liu et al., 2007; Pan et al., 2015; Shu et al., 2006; Yang et al., 2013; Zhao et al., 2014; Zhou & Marslen-Wilson, 2009). Studies using word recognition tasks, such as lexical decision and phonological decision, have found that making judgments purely based on phonological information was very difficult in Chinese (Zhou & Marslen-Wilson, 2009). Even in character-naming tasks, which are biased towards activating phonological representations, phonological processing does not seem to play a greater role than semantic processing (e.g., Liu et al., 2007; Yang et al., 2013; Zhao et al., 2014). For example, using a short-term primed naming paradigm, one study found equivalent facilitative effects for semantic and homophonic primes (Zhou & Marslen-Wilson, 2000). While no studies to date have specifically examined long-term form priming using Chinese materials, the prediction of the modulating role of the phonological congruency does not necessarily hold given the relatively lower weight of phonological processing in Chinese reading, as summarized above.
Therefore, it remains to be examined whether long-term form priming would be modulated by the phonological congruency between the prime and target, not only for alphabetic languages, but also for the non-alphabetic Chinese language. The present study aimed to test this hypothesis by using different types of Chinese phonograms and their embedded phonetic radicals as the prime and target words, as they are naturally orthographically similar, and their level of phonological congruency varies depending on the specific type of phonograms used.
The present study
The present study aimed to examine whether the phonological congruency between the prime and target items would modulate long-term form priming by using Chinese characters. To note, this was the first study of long-term form priming using Chinese materials.
We manipulated whether the shared orthography between the prime and target characters were pronounced the same or not, and created two conditions: "same-phonology" versus "different-phonologies." For the condition of same-phonology, pairs of regular phonograms (e.g., "..." /qiu2/, "ball") and their phonetic radicals (e.g., "..." /qiu2/, "pursue") were used. For the different-phonologies condition, we included various types of phonograms and their phonetic radicals to ensure the generalizability of our findings. Specifically, we included totally irregular ones (e.g., "...", /sa3/, "spray") that have a completely different pronunciation from their phonological radical (e.g., "...", /xil/, "west"), the rhyming semiregular ones (e.g., "...", /choul/, "take out") that have the same rhyming segment with their phonetic radical (e.g., " ...", /youll, "reason"), and the alliteration semi-regular ones (e.g., "...", /bo2/, "uncle") that have the same initial consonant with their phonetic radical (e.g., "...", /bai2/, "white").
A two-phase paradigm was implemented in this study. In the priming phase, the prime characters of both the samephonology and different-phonologies conditions were presented. Subsequently in the testing phase, the target characters were presented, which were orthographically similar to those prime characters. This two-phase design allowed for a lengthy and variable prime-target interval, preventing participants from strategically anticipating related forms, which was less unavoidable with a single list design where each prime-target pair was interspersed with filler characters (e.g., Forster & Davis, 1984). To measure the priming effect, control characters were included in the testing phase. Those control characters were matched with the targets on the psycholinguistic properties, but were not orthographically similar to any of the primes. Thus, performance differences between the targets and the controls in each condition indexed the long-term form priming effects specific to that condition.
Three experiments were conducted in total. In Experiment 1, phonograms were used as the primes and phonetic radicals served as the targets (see Table 1). A characternaming task was used in both the priming and testing phases. This task was chosen as a starting point for the present investigation mainly because it is biased for the activation of phonological representation and thus instantiates a situation where the modulating role of the phonological congruency would be relatively more readily observable. In addition, it is also a simple lexical processing task that characterizes one of the most common ways in which Chinese characters are used daily, and thus would be beneficial to establish the generalizability of our findings.
Experiment 2 aimed to replicate the long-term form priming effects observed in Experiment 1 by using different types of characters as primes and targets.To be specific, Experiment 2 used phonetic radicals as the primes and phonograms as the targets, opposite of those used in Experiment 1 (see Table 2). As the phonograms and radicals differed in several psycholinguistic properties, such as the visual complexity, frequency, and number of homophones (see Table 3), obtaining consistent results across these two experiments would help establish the generalizability of long-term form priming and the role of the phonological congruency in modulating long-term form priming across different types of characters.
Experiment 3 aimed to further test whether the longterm form priming effects found in the first two experiments could be generalized when different tasks were used between the priming and testing phases. In both Experiment 1 and Experiment 2, a naming task was used in both the priming and testing phases. So, one might argue that the observed priming effects might be only driven by changes in the articulatory processing, but not necessarily by changes in the lexical processing, particularly considering the existing models of speech production that propose separate stages for lexical processing and articulatory implementation (e.g., Dell, 1986;Levehet al., 1999).
To address this possibility, Experiment 3 maintained the naming task in the priming phase but implemented a Chinese Pinyin to written character matching task in the testing phase. Pinyin is an alphabetic encoding system used in China to aid in learning the pronunciations of Chinese characters, employing English letters, letter groups, and lexical tones as sound symbols (Yin et al., 2011). Pinyin reading is highly proficient among adult readers in China and can automatically activate the pronunciations of characters (Chen et al., 2019; Xia et al., 2022). In each trial of our Pinyin to character matching task, participants were sequentially presented with a Pinyin (e.g., Zpei4Z) and a written character (e.g., "...") on the computer display, and were asked to judge whether the former denoted the pronunciation of the latter. As this task involved no articulatory processing, any observed priming effects could be attributed solely to changes in lexical processing rather than changes in articulation. In addition, as the tasks and specific characters used in this experiment differed from those in the previous experiments, it could further extend the generalizability of long-term form priming and the modulatory role of the phonological congruency in such priming.
To preview, in three experiments we consistently found that the phonological congruency between the prime and target characters modulated the long-term form priming effects, with facilitative priming observed when they had the same phonology, and inhibitory priming observed when they had different phonologies.
Experiment 1
Experiment 1 asked whether a prior naming of a Chinese character would have an impact on the later naming of the character that was orthographically similar to it and whether such a long-term priming effect would depend on whether such pairs of characters had the same pronunciation or not.
In the same-phonology condition, the primes were regular phonograms (e.g., ... /qiu2/, ball) and the targets were phonetic radicals that had the same pronunciations as the primes (e.g., /qiu2/, ask); and in the-different phonologies condition, the primes were irregular and semi-regular phonograms (e.g., ... Zsa3Z, spray) and the targets were phonetic radicals that had different pronunciations with them (e.g., ... ZxilZ, west). The testing phase also included the control characters for each condition. The same as the targets, the controls were also phonetic radicals (e.g., ... /Ian2/, orchid), but they were not orthographically similar to any of the phonograms presented in the priming phase. Thus, the performance differences between the targets and the controls in each condition indexed the long-term form priming effects in that particular condition.
Methods
Participants
Twenty-eight undergraduate students were recruited and received monetary compensation for their participation. Participants were all native speakers of Chinese, had normal hearing and normal or corrected-to-normal vision, and provided written informed consent. The sample size was determined by performing a power analysis using G·Power Version 3.1 (Faul et al., 2007). The effect size (Cohen's f = 0.23) used for this estimation was obtained in our pilot study (n=16). The alpha level and power were set to 0.05 and 80% respectively.
Apparatus and stimuli
Stimulus presentation and data collection were controlled using a notebook PC using E-Prime 2.0 (www.pstnet.com/ eprime) and responses were collected using an SR-Box button box.
The critical stimuli (those for the primes, targets, and controls) included 80 phonograms, and 80 simple characters that appeared as the phonetic radical of these phonograms (Liu et al., 2007) (see Appendix 1, Tables 5 and 6). The phonograms included two sets of 20 regular (Set A-Regular & Set B-Regular) and two sets of 20 irregular ones (Set A-Irregular & Set B-Irregular). Accordingly, there were four sets of phonetic radicals, i.e., the phonetic radicals of Set A regular phonograms (Set A-Regular's Radical), the phonetic radicals of Set B regular phonograms (Set B-Regular's Radical), the phonetic radicals of Set A irregular phonograms (Set A-Irregular's Radical), and the phonetic radicals of Set B irregular phonograms (Set B-Irregular's Radical).
Two versions of stimuli (i.e., Version 1 and Version 2) were designed (see Table 1), which were randomly assigned to the participants. In Version 1, the Set A- Regular phonograms and Set A-Irregular phonograms were presented in the priming phase, serving as the primes in the same-phonology and different-phonologies conditions respectively. In the testing phase, all four sets of phonetic radicals were presented, serving as the targets and controls in these two conditions. To be specific, for the same-phonology condition, the phonetic radicals of Set A-Regular phonograms (Set A-Regular's Radical) were the targets, as their orthographically similar character had been presented in the priming phase, and the phonetic radicals of Set B regular phonograms (Set B-Regular's Radical) were the control stimuli, as they were not orthographically similar to any of the characters in the priming phase. Similarly, for the different-phonologies condition, the targets were the phonetic radicals of Set А-Irregular phonograms (Set A-Irregular's Radical) and the controls were those phonetic radicals of Set B-Irregular phonograms (Set B-Irregular's Radical).
Version 2 was designed in the same way as Version 1 except that the Set B instead of Set A regular and irregular phonograms were presented in the priming phase, and thus in the testing phase the target or control roles of the phonetic radicals were opposite of those in Version 1.
By designing two versions of the stimulus list this way and randomly assigning the two versions to the participants, the targets and controls in each condition were essentially the same stimuli, and thus were perfectly matched on the psycholinguistic properties.
For both Versions, 40 filler complex characters were interleaved with the primes in the priming phase, so as to minimize the chances for the participants to become aware of the relationship between the critical characters (primes, targets, and controls) in the two phases. These fillers characters shared no phonetic radical with any of the critical characters, and were not similar to any of the critical characters in orthography, phonology, or semantics. In both phases, the stimuli were presented in a randomized order.
The phonograms and the phonetic rad 'cals' are of both low and high frequencies (range: 1-1,309 per million) (Cai & Brysbaert, 2010), the phonograms had significantly lower frequency than the radicals (see Table 3 for more details), F(l,158) = 36.65, p < .001. However, the subjective ratings of 25 additional participants verified that all characters were of high familiarity. The rating was made on a Likert scale of 1-7 (1 = "very unfamiliar"; 7 = "very familiar") and the average rating scores of all sets of characters were above 6, and there was no significant difference between the phonograms and the phonetic radicals, F(1,158) = 1.24, p = .35. In addition, compared to the radicals, the phonograms were of significantly greater visual complexities (indexed by the number of strokes), F( 1,158) = 79.42, p < .001, and slightly fewer number of homophones, F( 1,158) = 2.46, p = .12.
The four sets of phonograms were matched on character frequency and familiarity (F(3,76) = 0.13, p = 0.94; F(3,76) = 0.53, p = 0.67), and so were the four sets of phonetic radicals (F(3,76) = 0.3l,p = 0.82; F(3,76) = 0.49, p = 0.69). Based on the psycholinguistic norms of Chinese characters (Liu et al., 2007), the four sets of phonograms were also matched on the visual complexity (indexed by the number of strokes) and the number of homophones (see Table 3) (F(3,76) = 1.03, p = 0.38, F(3,76) = 1.17, p = 0.33), and so were the four sets of phonetic radicals (F(3,76) = 1.21, p = 0.31, F(3,76) = 0.09, p = 0.93).
Procedures
The experiment consisted of two phases. The priming phase consisted of 80 trials, including one set of regular and irregular phonograms that served as the primes for the same-phonology and different-phonologies conditions respectively, and 40 filler characters. In this phase, the 80 characters were named in a randomized order. The testing phase also consisted of 80 trials, in which all four sets of phonetic radicals were named in a randomized order. These two phases lasted for about ten minutes. Thus, the mean interval between the prime character and the target character that was orthographically similar to it was about five minutes. For both phases, each trial began with a fixation presented at the center of the screen for 500 ms, and then after a blank screen of 500 ms, one character was presented for 3,000 ms. Participants were required to say it aloud into a microphone as quickly and as accurately as they could and their vocal responses were automatically recorded by the computer. The recordings were used for the coding of the accuracy of the naming responses.
At the beginning of the experiment, participants received eight practice trials to get familiarized with the procedures. The characters used in practice had no overlap with those in the priming and testing phases. At the end of the experiment, participants were interviewed and asked whether they noticed any relationship among the characters they named, and only two out of 28 reported that they found that some of the characters they named later in the experiment were embedded in the characters they named earlier.
Reliability
A trained coder who was blind to the study's purpose coded the naming accuracy. Another independently trained coder coded 50% of the data, and the point-by-point agreement between them was 98%.
Data analysis
Generalized linear mixed-effects models (GLMMs) and linear mixed-effects models (LMMs) were used to analyze the accuracies and reaction times (RTs) of the naming responses in Phase 2 respectively, using the lme4 package (Bates et al., 2015) in R (R Core Team, 2018). Since accuracy was binomial (correct or incorrect, scored as 1 or 0), a binomial linking function was used. As for the RTs, we excluded the incorrect trials (0.67% of all trials across 28 participants) and the trials (2.6% of all trials across 28 participants) with RTs lower than 100 ms or higher than three standard deviations of the mean of each subject.
Analyses for both the accuracies and RTs included phonological congruency and stimulus identity as the fixed factors, and subjects and items as the random intercepts. Using the log-likelihood ratio test (Baayen et al., 2008), we compared this simple model with more complex models that included not only subjects and items as random intercepts, but also by-subject and by-item random slopes for phonological congruency and stimulus identity. None of the more complex models produced more significantly better fit (ps > .50), so our analyses used this simple model. As both fixed factors were binary, levels of each were coded as -1 versus 1.
To note, the results remained essentially the same when the two participants who had some awareness of the prime - target relationship were removed. Therefore, in the following, we reported the results of all 28 participants.
Results and discussion
As shown in Fig. 1, although the naming accuracies of all conditions were close to the ceiling, the difference between the target and control stimuli was more sizable in the different-phonologies condition than that in the same-phonology condition. The generalized linear mixed-effects model analysis showed a significant main effect of phonological congruency (ß = 0.005, SE = 0.002, z = 3.11, p = 0.002), a significant main effect of stimulus identity (ß = -0.004, SE = 0.002, z = -2.54, p = 0.01), and a significant interaction of these two factors (ß = 0.005, SE = 0.002, z = 3.13, p = 0.002). To understand the interaction, we analyzed the effect of stimulus identity within each level of phonological congruency using a similar model structure. It turned out that for the different-phonologies condition, the naming accuracies of targets were significantly lower than those of the controls, demonstrating an inhibitory long-term priming effect (ß = 0.009, SE = 0.003, z = -2.99, p = 0.002). But for the same-phonology condition, the accuracies of targets were not significantly different from those of the controls (ß = 0.0009, SE = 0.0009, z = 1.00, p = 0.32).
As shown in Fig. 2, the RTs of the target stimuli were shorter than those of the control stimuli in the same-phonology condition, but the difference between them in the different-phonologies condition seemed to be negligible. The linear mixed-effects model analysis showed that the main effect of phonological congruency (ß= -1.83, SE = 2.76, t = -0.66, p = 0.51) and that of stimulus identity (ß= -2.78, SE = 2.71, t = -1.03, p = 0.32) were both non-significant, but the interaction of these two factors (ß= -6.19, SE = 2.71, t = -2.28, p = 0.02) was significant. To understand this interaction, we analyzed the effect of stimulus identity within each level of phonological congruency using a similar model structure. It turned out that for the same-phonology condition, the targets were named significantly faster than the controls, demonstrating a facilitative long-term priming effect (ß= 9.04, SE = 3.26, t = 2.78, p = 0.005). But for the different-phonologies condition, the naming latencies of the targets and those of the controls did not differ significantly (ß= 3.07, SE = 4.31, t = 0.71, p = 0.48).
Taken together, this experiment revealed a facilitative effect of long-term form priming when the prime and target characters had the same phonology as indicated by RTs, and an inhibitory effect when they had different phonologies as indicated by accuracies. These results were consistent with one previous study using English words (e.g., Bowers et al., 2002), and supported the connectionist's prediction that the phonological congruency between the prime and target words would modulate the direction of long-term form priming.
However, our findings of inhibitory long-term form priming in the condition of different-phonologies were not consistent with the null results found in Pexman et al. (2005). In their study, no long-term priming effects were observed between exception-word prime (e.g., PINT) and their nonrhyming word body neighbors (e.g., MINT) in the naming and lexical decision tasks. Thus, contrary to the argument of Pexman et al. (2005) that long-term form priming could only occur in difficult testing tasks (e.g., Pexman et al., 2005; see also Hughes & Whittlesea, 2003), our findings of significant long-term form priming in a naming task suggested that long-term form priming could be a general effect that constantly occurs in the common word-use situations.
Our results were the first evidence of long-term form priming in Chinese materials, and also the first evidence that the phonological congruency between the primes and targets modulated long-term form priming in Chinese materials. These findings challenge the notion that phonological processing had a lower weight in Chinese reading compared to alphabetic languages, as suggested by previous studies (e.g., Liu et al., 2007; Pan et al., 2015; Yang et al., 2013; Zhao et al., 2014). Our results demonstrated that a phonological factor, specifically the phonological congruency between the prime and target characters, could modulate how processing one character might influence the future processing of an orthographically similar one. This implies that the phonological representations of Chinese phonograms and their phonetic radicals could be sufficiently activated in written character naming, and that these activated representations could be sufficiently altered to influence the future processing of orthographically similar characters.
Experiment 2
This experiment aimed to test whether the results of Experiment 1 could be replicated when the roles of primes vs. targets were switched between the phonograms and phonetic radicals (see Table 2), i.e., using the phonetic radicals as the primes and the phonograms as the targets. Thus, the targets used in this experiment (i.e., phonograms) were of higher visual complexity, lower character frequency, and slightly lower homophone density than those used in Experiment 1 (i.e., phonetic radicals). Consistent results across these two experiments would help further establish the generalizability of long-term form priming and the role of the phonological congruency in modulating long-term form priming across different character types.
Methods
Participants
Thirty undergraduate students were recruited and received monetary compensation for their participation. Participants were all native speakers of Chinese, had normal hearing and normal or corrected-to-normal vision, and provided written informed consent. They did not participate in Experiment 1.
Stimuli and procedures
The stimuli and procedures were the same as those used in Experiment 1 except that the roles of prime vs. target were switched from those in Experiment 1 (see Table 2). Thus, in Version 1, in the priming phase, the phonetic radicals of Set А-Regular and Set A-Irregular phonograms (Set A-Regular's Radical and Set A-Irregular's Radical) were presented, serving as the primes for the same-phonology and different-phonologies conditions respectively. In the testing phase, all four sets of phonograms were presented. Therefore, for the samephonology condition, the targets were the Set A-Regular phonograms, and the controls were the Set B-Regular phonograms. For the different-phonologies condition, the targets were the Set А-Irregular phonograms, and the controls were the Set B-Irregular phonograms. Version 2 was designed in the same way as Version 1 except that the phonetic radicals of B sets were presented in the priming phase, and thus in the testing phase the target or control identities of the phonograms were opposite of those in Version 1. These two versions of stimuli were randomly assigned to the participants. None of the 30 participants reported that they found that any of the characters they named earlier in the experiment were embedded in the characters they named later.
Reliability
A trained coder who was blind to the study's purpose coded the naming accuracy. Another independently trained coder coded 50% of the data, and the point-by-point agreement between them was 97%.
Data analysis
The same analysis methods were used as those in Experiment 1. To analyze the RTs of the naming responses, we excluded the incorrect trials (2.5% of all trials across 30 participants) and the trials (2.4% of all trials across 30 participants) with RTs lower than 100 ms or higher than three standard deviations of the mean of each subject.
Results and discussion
As shown in Fig. 3, the difference between the target and control stimuli was more sizable in the different-phonologies condition than that in the same-phonology condition. The generalized linear mixed-effects model analysis showed a significant main effect of phonological congruency (ß = 0.014, SE = 0.003, z = 4.33, p<0.001), a significant main effect of stimulus identity (ß = -0.007, SE = 0.003, z = -2.28, p = 0.02), and a significant interaction of these two factors (ß = 0.007, SE = 0.003, z = 2.03, p = 0.04). To understand the interaction, we analyzed the effect of stimulus identity within each phonological congruency using a similar model structure. It turned out that for the differentphonologies condition, the naming accuracies of targets were significantly lower than those of the controls, demonstrating an inhibitory long-term priming effect (ß = -0.014, SE = 0.006, z = -2.48, p = 0.01). But for the same-phonology condition, the accuracies of targets were not significantly different from those of the controls (ß = 0.008, SE = 0.026, z = 0.26, p = 0.80).
As shown in Fig. 4, the RTs of the target stimuli were shorter than those of the control stimuli in the same-phonology condition, but this pattern was reversed in the differentphonologies condition. The mixed-effects model analysis showed that neither the main effect of phonological congruency (ß = -1.83, SE = 2.76, t = -0.66, p = 0.51) nor that of stimulus identity (ß = -2.78, SE = 2.71, t = -1.03, p = 0.32) was significant, but the interaction of these two factors (ß = -6.19, SE = 2.71, t = -2.28, p = 0.02) was. To understand this interaction, we analyzed the effect of stimulus identity within each level of phonological congruency using a similar model structure. It turned out that for the same-phonology condition, the targets were named significantly faster than the controls, demonstrating a facilitative long-term priming effect (ß = 9.04, SE = 3.26, t = 2.78, p = 0.005); but for the different-phonologies condition, the targets were named significantly more slowly (ß = 20.90, SE = 4.19, t = 4.99,p<0.001), demonstrating an inhibitory long-term priming effect.
Taken together, the above results also demonstrated that the prior naming of a character facilitated the naming of an orthographically similar character with the same pronunciation, and inhibited the naming of an orthographically similar character with a different pronunciation. These results replicated those of Experiment 1 with different prime-target pairings of Chinese characters. As aforementioned, the targets used in this experiment (i.e., phonograms) were of higher visual complexity, lower character frequency, and slightly lower homophone density than those used in Experiment 1 (i.e., phonetic radicals), while the comparisons between the primes in this experiment (i.e., phonetic radicals) and those in Experiment 1 (i.e., phonograms) were of the opposite pattern. Previous studies have shown that higher visual complexity, lower character frequency, and lower homophone density were all associated with slower character naming (e.g., Chen et al., 2009b; Liu et al., 2007; Ziegler et al., 2000). Consistent with these results, the average naming latencies of the targets in Experiment 2 (i.e., the phonograms) (M = 683 ms, SD = 106 ms) were much longer than those in Experiment 1 (i.e., the radicals) (M = 569 ms, SD = 102 ms). However, despite these differences in the psycholinguistic properties and the naming performance, the long-term priming effects were consistently found across these two experiments. That means that naming characters with low-processing efficiency could influence the future naming of orthographically similar characters with highprocessing efficiency, and vice versa, which further suggests that long-term form priming commonly occurs in daily language usage of Chinese.
In fact, studies of long-term form priming using English words also found that different word types could prime each other (Burt & Humphreys, 1993; but not in Seidenberg et al., 1984). Burt and Humphreys (1993) adopted a word-naming task, and found inhibitory long-term form priming not only between regularly-pronounced primes and exception word targets (e.g., Mint-Pint), but also in the reverse direction, i.e., between exception word primes and regularly-pronounced targets (e.g., Says-Rays). However, in their study, not only the exception word targets were of low frequency (e.g., Pint; Mean Frequency: 10 per million), but also the regular targets (e.g., Rays; Mean Frequency: 17 per million) were. In fact, they chose low-frequency words deliberately for the latter, so as to maximize the long-term priming effects.
In contrast, we did show in Experiment 1 that the highfrequency radical targets (e.g., "Щ", /xil/, west; mean frequency: 234 per million) could be primed by the lowfrequency phonogram primes (e.g., " Д", /sa3/, spray; mean frequency: 23 per million) after a long delay. Although no direct comparison has been made between languages, we could speculate that Chinese high-frequency phonetic radicals might be more likely to undergo long-term form priming than English high-frequency regularly pronounced words. The reason lies in the different distributions of regular and irregular words in the two languages. As aforementioned, the phonetic radicals that are simple characters themselves take up less than 5% of all Chinese characters, the regular phonograms, which have the same pronunciation as their phonetic radicals, take up about 25%, while the semi-regular and completely irregular phonograms take up about 55% in total (Li, 1993). Compared to Chinese, English is a much more transparent language with a greater ratio of regular to irregular words (Chen et al., 2009a). Thus, in daily language uses, the underlying representations of Chinese phonetic radicals as simple characters would be constantly twisted by irregular and semi-regular phonograms that are pronounced differently from them, and could hardly stay in a stable state. And the same would be true for regular phonograms. In contrast, for English, the representations of regular words could be repeatedly consolidated with only occasional interference from exception words, which is relatively negligible.
In sum, the first two experiments showed consistent longterm form priming effects and the modulation of the primetarget phonological congruency, across different types of characters with varying psycholinguistic properties. These convergent findings suggest that long-term form priming commonly occurs in daily language uses of Chinese and that the lexical representations of Chinese characters undergo constant tuning by daily language uses.
Experiment 3
Experiment 3 aimed to replicate the results of the first two experiments when a task that involved no articulatory processing, the Chinese Pinyin to written character matching task, was used in the testing phase.
As aforementioned, the task change in this experiment served two purposes. First, it helped to rule out the possibility that the priming effects found in the first two experiments were due only to changes in the articulatory processing rather than to changes in the lexical processing. Second, by using different tasks, this experiment helped further establish the generalizability of long-term form priming and the modulatory role of the phonological congruency in such priming across different task contexts. In terms of the experimental stimuli, this experiment used more critical characters (those for the primes, targets, and controls) and more filler characters, and only a small subset of them overlapped with those used in the first two experiments. This expansion of stimuli would also help enhance the generalizability of the above effects across materials.
In the Pinyin to written character matching task, participants were presented with one Pinyin (e.g., /pei4/) and one written character (e.g., "0b") sequentially, and were asked to judge whether the former denoted the pronunciation of the latter (see Fig. 5). Thus, in each trial, the participants first accessed the pronunciation denoted by the Pinyin, then they kept it in the working memory briefly, and once the pronunciation of the written character was accessed, they needed to make a comparison between the two and arrive at a decision. Clearly, this task involved no articulatory processing.
In this task, all the target and control trials were made to be the no-response trials in which the Pinyin did dot denote the pronunciation of the character. The yes-response was not opted for the targets, as the Pinyin that denoted the pronunciation of a target might induce some short-term priming effect, which might interact with the long-term priming effect that was of our real interest and contaminate it. In addition, to avoid any repetition priming effect, each target and control character was presented only once. An equal number of filler trials were included and were designed as the yes-response trials, in which the Pinyin matched the pronunciation of the character.
Methods
Participations
Twenty-six undergraduate students were recruited and received monetary compensation for their participation. Participants were all native speakers of Chinese, had normal hearing and normal or corrected-to-normal vision, and provided written informed consent. They did not participate in Experiment 1 or Experiment 2. The sample size was determined by performing a power analysis using G·Power Version 3.1 (Faul et al., 2007). The effect size (Cohen's f = 0.24) used for this estimation was obtained in our pilot study (n = 12). The alpha level and power were set to 0.05 and 80% respectively.
Apparatus and stimuli
Compared to the first two experiments, there was both expansion and modification of the experimental stimuli. The critical stimuli included 120 phonograms and 120 phonetic radicals of these phonograms (see Appendix 2, Tables 7 and 8). Similar to the first two experiments, the phonograms included two sets of 30 regular (Set A-Regular & Set B-Regular) and two sets of 30 irregular ones (Set A-Irregular & Set B-Irregular). Accordingly, there were also four sets of phonetic radicals, i.e., the phonetic radicals of Set A regular phonograms (Set A-Regular's Radical), the phonetic radicals of Set B regular phonograms (Set B-Regular's Radical), the phonetic radicals of Set A irregular phonograms (Set A-Irregular's Radical), and the phonetic radicals of Set B irregular phonograms (Set B-Irregular's Radical).
Two versions of stimuli (i.e., Version 1 and Version 2) were designed in essentially the same way as those in Experiment 2 (see Table 2). That is, in Version 1, the phonetic radicals of Set A regular and Set A irregular phonograms (i.e., Set А-Regular's Radical and Set A-Irregular's Radical) were presented in the priming phase, serving as the primes for the same-phonology and different-phonologies conditions respectively. In the testing phase, all four sets of phonograms were presented. Therefore, for the same-phonology condition, the targets were the Set A regular phonograms, and the controls were the Set B regular phonograms. For the different-phonologies condition, the targets were the Set A irregular phonograms, and the controls were the Set B irregular phonograms. Version 2 was designed in the same way as Version 1 except that the phonetic radicals of B sets were presented in the priming phase, and thus in the testing phase the target or control identities of the phonograms were opposite of those in Version 1. These two versions of stimuli were randomly assigned to the participants.
The descriptive statistics of the psycholinguistic properties of the stimuli used in Experiment 3 were shown in Table 4. The four sets of phonograms were matched on character frequency (F(3,l 16) = 0.45, p = 0.72), familiarity (F(3,l 16) = 0.48, p = 0.69), number of strokes (F(3,l 16) = 0.21, p = 0.89), and number of homophones (F(3,116) = 0.68, p = 0.56), and so were the four sets of simple characters (F(3,116) = 0.24, p = 0.87; F(3,116) = 0.13, p = 0.94; F(3,116) = 0.07, p = 0.98; F(3,116) = 0.19, p = 0.91).
As aforementioned, we made all the target and control trials (120 in total) the no-response trials. For each of these trials, Pinyin that preceded the character did not denote its pronunciation. To minimize any potential influence from Pinyins, all these 120 Pinyins did not denote the pronunciation of any of the target or control characters. In addition, the Pinyin that was presented prior to each target or control character was carefully chosen to make sure that the former did not rhyme with the latter. We also included 120 filler trials be the yes-response trials, for which the Pinyin did denote the pronunciation of the character. These fillers shared no phonetic radicals with any of the critical characters and were not similar to any of the critical characters in orthography, phonology, or semantics. For both Versions of our stimuli, the filler characters were interleaved with the targets and controls in a randomized order in the testing phase.
Procedures
The experiment also consisted of the priming and testing phases. These two phases lasted for about 20 min. Thus, the mean interval between the prime-target character pairs was about 10 min, which was about two times longer than that of the first two experiments.
The priming phase used the naming task, in which one set of phonetic radicals of the regular and irregular phonograms (60 in total) were named in a randomized order. The testing phase used the Pinyin to written character matching task, in which 60 targets, 60 controls, and 120 fillers were presented in a randomized order. Half of the targets were the regular and irregular phonograms whose phonetic radicals had been presented in the priming phase, and thus belonged to the same-phonology condition. The other half belonged to the different-phonologies condition, which were the irregular phonograms whose phonetic radicals presented in the priming phase. The controls also consisted of half in the same-phonology condition and the other half in the different-phonologies condition, which were the other set of regular and irregular phonograms respectively. The controls were not orthographically similar to any of the characters shown in the priming phase.
In the naming task, each trial began with a fixation presented at the center of the screen for 500 ms, and then after a blank screen of 500 ms, one prime character was presented for 1,000 ms. Participants were required to say it aloud into a microphone as quickly and as accurately as they could, and their vocal responses were automatically recorded by the computer. After the offset of the character, there was a blank screen for 1,000 ms, and then the next trial began.
In the Pinyin to written character-matching task used in the testing phase (see Fig. 5), each trial began with a fixation presented at the center of the screen for 500 ms, and then after a blank screen of 500 ms, one Pinyin was presented for 1,000 ms, and then following a blank screen of 1,000 ms, one character was presented. Participants were required to judge whether the Pinyin denoted the pronunciation of the character, and then press the corresponding response key as quickly and as accurately as they could. The character disappeared when the key was pressed or when it had been presented for 2,000 ms. Then after a blank screen of 1,000 ms, the next trial began.
At the beginning of both phases, participants received eight practice trials to get familiarized with the procedures. The characters used in practice had no overlap with those in the priming and testing phases. At the end of the experiment, participants were interviewed and asked whether they noticed any relationship among the characters they named, and none of the 26 reported that they found that any of the characters they saw later in the experiment were embedded in the characters they named earlier.
Data analysis
The same analysis methods were used as those in Experiment 1 and Experiment 2. To analyze the RTs of the decision responses, we excluded the incorrect trials and the trials with RTs lower than 100 ms or higher than three standard deviations of the mean of each subject (3.2% of all trials across 26 participants).
Results and discussion
As shown in Fig. 6, the decision accuracies of all conditions were close to the ceiling. The generalized linear mixedeffects model analysis showed that the main effect of phonological congruency (ß = -0.161, SE = 0.171, z = -0.95, p = 0.34) and the main effect of stimulus identity (ß = 0.018, SE = 0.144, z = 0.13, p = 0.90) were both insignificant. Besides, the interaction of these two factors was not significant (ß = 0.016, SE = 0.144, z = 0.11, p = 0.91).
As shown in Fig. 7, the RTs of the target stimuli were shorter than those of the control stimuli in the same-phonology condition, but this pattern was reversed in the differentphonologies condition. The mixed-effects model analysis showed that neither the main effect of phonological congruency (ß = 3.72, SE = 3.81, t = 0.97, p = 0.34) nor that of stimulus identity (ß = 0.178, SE = 2.64, t = 0.068, p = 0.95) was significant, but the interaction of these two factors (ß = -7.38, SE = 2.41, t = -3.06, p = 0.002) was. To understand this interaction, we analyzed the effect of stimulus identity within each level of phonological congruency using a similar model structure. It turned out that for the same-phonology condition, the targets were named significantly faster than the controls, demonstrating a facilitative long-term priming effect (ß = -7.2, SE = 3.44, t = -2.09, p = 0.04); but for the different-phonologies condition, the targets were named significantly more slowly (ß = 7.57, SE = 3.4, t = 2.23, p = 0.03), demonstrating an inhibitory long-term priming effect.
Taken together, this experiment revealed that when the priming task and the testing task were different, the longterm form priming effect was still modulated by the phonological congruency of the prime-target character pairs. The same as in the first two experiments, here we found a facilitative effect of long-term form priming when the prime and target had the same phonology, and an inhibitory effect when they had different phonologies. However, different from these two experiments, here participants were required to produce the phonology of the characters only in the priming phase, but not in the testing phase. Thus, if the long-term form priming effects lay only in changes in the articulatory processing but not in the lexical processing at all, no effects would have been detected. Put in another way, these results suggested that the long-term priming effects induced by a single trial of naming a Chinese character could have a locus in lexical processing. This is consistent with the connectionist idea that long-term priming for words lies in the small changes in the lexical representations resulting from lexical processing, which in a connectionist model would be instantiated as the changes in the connection weights between the relevant orthographic to the phonological nodes (e.g., Hinton et al., 1986; McClelland & Rumelhart, 1985; Stark & McClelland, 2000).
However, based on the existing models and some empirical findings of word production, we would like to argue that such long-term form priming effects might lie in both lexical processing and articulatory processing, rather than be restricted to one single stage. First, although the existing models of speech production assume discrete levels of lexical processing (e.g., semantic selection and phonological encoding) and articulatory implementation (Dell, 1986; Level! et al., 1999), they also admit that the earlier lexical processing sends output to articulation. Therefore, if prior naming led to changes in the lexical processing, it would propagate to the following articulatory processing. Second, some empirical evidence suggests that there are in fact cascaded real-time interactions between lexical and articulatory processing (e.g., Baese-Berk & Goldrick, 2009; Goldrick & Blumstein, 2006; Benham & Goffman, 2020; McMillan et al., 2009; Heisler et al., 2010). For example, fine-grained speech kinematic analysis showed that the wrong productions in fact contained acoustic traces of the intended target (e.g., Goldrick & Blumstein, 2006), and that linking a phonological word form with a visual referent improved the articulatory stability of word production (e.g., Heisler et al., 2010). Such evidence supports a new perspective of lexical representation that the semantic, phonological, orthographic, and motor-articulatory components are all associated, which altogether represent a word (e.g., Heisler et al., 2010; Pierre-humbert, 2001). This new perspective would also suggest that an instance of word production would not only lead to changes in the orthography-to-phonology connections, but also changes in the phonology-to-articulation connections. It is beyond the scope of the present study, but future studies could employ both behavioral experiments and computational modeling to test this hypothesis directly.
Finally, we could like to note that despite changes in the tasks and stimuli, the results of the present experiment converged with those of the first two experiments, which helped extend the generalizability of long-term form priming and the modulatory role of the phonological congruency across tasks and stimuli.
General discussion
In three experiments, we consistently found that long-term form priming was modulated by the phonological congruency between the prime and target characters, with facilitative priming observed when they had the same phonology, and inhibitory priming observed when they had different phonologies. Such modulation was found when different types of prime-target pairs were used, when the prime was processed only one time in a simple and common lexical processing task (i.e., naming), and when the priming and testing tasks had different processing requirements. Taken together, these observations helped confirm the reliability, robustness, and generalizability of long-term form priming.
It is important to recall that the previous studies (e.g., Napps & Fowler, 1987; Pexman et al., 2005; Rueckl & Mathew, 1999; Seidenberg et al., 1984; Wagenmakers & Raaijmakers, 2006) revealed mixed effects of long-term form priming, i.e., long-term form priming could be facilitative, inhibitory or nonexistent. The present study suggested that this was likely due to the mixture of the different kinds of materials (e.g., the prime-target pairs whose overlapped orthographic segments had the same phonology or different phonologies) in different proportions across these studies. That is, failure to consider the modulation of the phonological congruency between the prime and target words could have obscured the genuine effect of long-term form priming in the previous studies. This further suggested that the appropriate way to probe the long-term form priming is to delineate how it occurs specifically in different contexts defined by its modulators. In fact, one previous study (Pecher et al., 2005) manipulated whether the orthographically similar prime and target words were from the semantic category (e.g., semantic congruency), and found similar modulating patterns of long-term form priming. Specifically, Pecher et al. (2005) found that when the prime and target were semantically congruent (e.g., CAT and RAT), making a semantic-category decision on the former facilitated a later decision on the latter, but when they were semantically incongruent (e.g., CAT and HAT), the long-term priming effect was inhibitory. Future studies could explore further how these factors might interact in modulating the long-term form priming effects.
Our findings of the long-term form priming effects and the modulation by the phonological congruency between the prime and target were consistent with the predictions derived from the connectionist account of long-term priming (e.g., Hinton et al., 1986; McClelland & Rumelhart, 1985; Stark & McClelland, 2000). In particular, we revealed these effects by using a simple Character-naming task in both the priming and testing phases, which lent support to the connectionist idea that long-term priming for words would arise naturally from common lexical processing experiences, rather than would only occur in difficult lexical processing tasks (Hughes & Whittlesea, 2003; Pexman et al., 2005). Further, we replicated these effects when the testing task was replaced with one that required no articulatory processing, thus ruling out the possibility that these priming effects were not due to changes in lexical processing, but due only to changes in articulation. Thus, these findings lent support to the connectionist idea that long-term priming for words arises from the incremental tuning of the lexical representations after processing a word (Hinton et al., 1986; McClelland & Rumelhart, 1985; Stark & McClelland, 2000). Bowers et al. (2002) adopted Seidenberg and McClelland's (1989) connectionist model of word identification that instantiated back-propagation and distributed representations, and demonstrated that long-term form priming in the model was modulated by whether the prime and target words rhymed (i.e., phonological congruency). Future studies could adopt the connectionist models of Chinese character reading (e.g., Perfetti et al., 2005; Taft, 2006; Yang et al., 2009) to simulate our findings of long-term form priming here, so as to directly test the connectionist account of longterm priming for words. Such a test would also introduce new constraints and help select and develop more appropriate models of Chinese character reading.
To note, however, the inhibitory long-term form priming in our different-phonologies condition might also arise from increased lexical interference resulting from these representational changes. That is to say, the strengthened orthography-phonology connections would make the prime character (and those that shared these connections with it) stronger competitors and thus would exert greater interference on the target with conflicting connections (e.g., Dahan et al., 2001; Kapnoula & McMurray, 2016; Luce & Pisoni, 1998; Magnuson et al., 2007; Mirman, 2011). This is indirectly supported by one previous study on the long-term phonological priming for spoken words (Monsell & Hirsh, 1998). This study found an inhibitory priming effect when the prime and target words were cohorts (sharing the initial sound, such as CHAP and CHAT), but no priming when they rhymed (e.g., GEM and HEM). As in this study, there is no reason to expect differential learning in these two conditions, such results could only be explained by the wellknown effect that a strengthened cohort neighbor could exert stronger interference on the processing of the target than a rhymed one (e.g., Magnuson et al., 2007; McClelland & Elman, 1986; Vitevitch & Luce, 2004). In addition, in some related domains, such as the studies of semantic interference (Howard et al., 2006) and those of phonological interference (Qu et al., 2021), there have also been suggestions that increased lexical interference should be considered together with incremental learning to account for these inhibitory effects. A direct test on whether the inhibitory long-term form priming might have a locus in increased lexical interference could be one promising direction for future studies.
As the first evidence of long-term form priming and the modulation of the phonological congruency on long-term form priming of Chinese characters, our results also added to our understanding of how Chinese characters are represented and processed. Most critically, they lent support to the hypothesis that Chinese characters could be represented and processed analytically (Yang et al., 2009; Zhou & Marslen-Wilson, 1999). In other words, they supported the existence of sub-lexical representations and processing at the radical level for Chinese characters. For example, in our Experiment 2, if the phonology of the phonetic radical within irregular phonograms was not activated in the testing phase, learning resulting from naming the radical in the priming phase would not influence the naming of the phonogram that contained it. To note, the present study for the first time provided support for the existence of the sublexical representations and processing of Chinese characters by using a long-term priming paradigm. The previous studies that supported this idea all used short-term paradigms such as short-term priming (e.g., Ding et al., 2004; Zhou et al., 2014; Zhou & Marslen-Wilson, 1999), Stroop (e.g., Luo et al., 2014), or attentional blindness (e.g., Chen & Yeh, 2015; Yeh & Li, 2004). For example, one study used a primed naming paradigm and revealed that the naming of an irregular phonogram (e.g.,%, /cail/, "guess") was hindered when it was preceded by its phonetic radical (e.g., W, /qing2/, "blue") (Zhou & Marslen-Wilson, 1999). Therefore, our findings not only suggested that the sub-lexical phonological representations could be activated when a character is being processed, but also suggested that these temporarily activated representations could then be sufficiently altered to influence future lexical processing in a much-extended time scale. This reveals a different facet of phonological processing in Chinese character reading, and also adds to its importance in Chinese processing, which as aforementioned, has been deemed to be relatively low (e.g., Liu et al., 2007; Yang et al., 2013; Zhao et al., 2014).
Our findings have important implications for the interactions between lexical processing and word learning. Specifically, we found that a single instance of using a known character could influence the future lexical processing of its orthographically similar character, even for high-frequency ones. Similarly, previous studies also found that a single instance of lexical processing of one known word could influence future processing of the same word and its phonologically and semantically similar words (e.g., Becker et al., 1997; Gupta & Cohen, 2002; Monsell & Hirsh, 1998; Stark & McClelland, 2000; Qu et al., 2021). Taken together, all these results of long-term priming for words converged on the suggestion that even for highly familiar words, learning is not turned off, and that their lexical representations undergo continuous tuning via daily language uses. This implies that lexical processing is also a word-learning event. To date, studies on lexical processing have been focusing intensively on the effect of lexical interactions on the realtime processing (e.g., Dahan et al., 2001; Magnuson et al., 2003; McClelland & Rumelhart, 1986; Shen et al., 2016), but largely ignoring whether such processes might lead to any learning effects. As far as we know, only a handful of studies have examined the effect of in-the-moment lexical processing on the formation of the long-term lexical representations of novel words (e.g., Apfelbaum & McMurray, 2017; McMurray et al., 2012; for a review, see Kucker et al., 2015). The present study and those on long-term priming for words extended these investigations on the influence of lexical processing on the learning of novel words to the tuning of the representations of known words. Taken together, all these studies point to a potential direction of examining the interaction between real-time language use and learning on a more extended time scale (McMurray et al., 2012).
Conclusions
To conclude, the present study helped build the reliability, generalizability and robustness of long-term form priming, which was largely owing to the manipulation of the phonological congruency between the prime and target characters. Our findings lend further support to the suggestion that long-term priming for words might be one common way by which the moment-by-moment experience of word uses tune the lexical representations, including the well-learned ones (Oppenheim et al., 2010). Ours and the previous findings on long-term priming for words suggest that models of lexical processing should incorporate learning mechanisms, and call for studies to examine the interaction between real-time lexical processing and word learning on a more extended time scale (McMurray et al., 2012).
References
Apfelbaum, K. S., & McMurray, B. (2017). Learning during processing: Word learning doesn't wait for word recognition to finish. Cognitive Science, 41, 706-747.
Baese-Berk, M., & Goldrick, M. (2009). Mechanisms of interaction in speech production. Language and Cognitive Processes, 24(4), 527-554.
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Scftware, 67(1):1-8.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59(4), 390-412.
Becker, S., Moscovitch, M., Behrmann, M., & Joordens, S. (1997). Long-term semantic priming: A computational account and empirical evidence. Journal of Experimental Psychology Learning Memory & Cognition, 23(5), 1059-1082.
Benham, S., & Goffman, L. (2020). Lexical-semantic cues induce sound pattern stability in children with developmental language disorder. Journal of Speech, Language, and Hearing Research, 63(12), 4109-4126.
Bowers, J. S., Damian, M. F., & Havelka, J. (2002). Can distributed orthographic knowledge support word-specific long-term priming. Journal of Memory & Language, 46, 24-38.
Burt, J. S., & Humphreys, M. S. (1993). Delayed priming of the pronunciation of inconsistent words and pseudowords. Journal of Memory and Language, 32, 743-765.
Cai, Q., & Brysbaert, M. (2010). SUBTLEX-CH: Chinese word and character frequencies based on film subtitles. PLoS One, 5(6), el0729.
Chen, Y. C., & Yeh, S. L. (2015). Binding radicals in Chinese character recognition: Evidence from repetition blindness. Journal of Memory and Language, 78, 47-63.
Chen, C., Xue, G., Mei, L., Chen, C., & Dong, Q. (2009a). Cultural neurolinguistics. Progress in Brain Research, 178, 159-171.
Chen, H. C., Vaid, J., & Wu, J. T. (2009b). Homophone density and phonological frequency in Chinese word recognition. Language and Cognitive Processes, 24(1-8), 967-982.
Chen, L., Perfetti, C. A., Fang, X. R, Chang, L. Y., & Fraudorf, S. (2019). Reading Pinyin activates sublexcial character orthography for skilled Chinese readers. Language, Cognition and Neuroscience, 34(6), 736-746.
Dahan, D., Magnuson, J. S., Tanenhaus, M. K., & Hogan, E. M. (2001). Subcategorical mismatches and the time course of lexical access: Evidence for lexical interference. Language and Cognitive Processes, 16(5-6), 507-534.
Dell, G. S. (1986). A spreading activation theory of retrieval in sentence production. Psychological Review, 93, 283-321.
Ding, G. S., Peng, D. L., & Taft, M. (2004). The nature of the mental representation of radicals in Chinese: A priming study. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30(2), 530-539.
Faul, F., Erdfelder, E., Lang, A. G., & Buchner, A. (2007). G·Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175-191.
Feustel, T. C., Shiffrin, R. M., & Salasoo, A. (1983). Episodic and lexical contributions to the repetition effect in word identification. Journal of Experimental Psychology: General, 772(3), 309-346.
Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(4), 680-698.
Goldrick, M., & Blumstein, S. E. (2006). Cascading activation from phonological planning to articulatory processes: Evidence from tongue twisters. Language and Cognitive Processes, 21(6), 649-683.
Gupta, R, & Cohen, N. J. (2002). Theoretical and computational analysis of skill learning, repetition priming, and procedural memory. Psychological Review, 109, 401-448.
Heisler, L., Goffman, L., & Younger, B. (2010). Lexical and articulatory interactions in children's language production: Lexical and articulatory interactions. Developmental Science, 13(5), 722-730.
Hinton, G. E., McClelland, J. L., & Rumelhart, D. E. (1986). Distributed Representations. In D. E. Rumelhart & J. L. McClelland (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition. (Vol. I). MIT Press.
Howard, D., Nickels, L., Coltheart, M., & Cole-Virtue, J. (2006). Cumulative semantic inhibition in picture naming: Experimental and computational studies. Cognition, 99(3), 464-482.
Hughes, A. D., & Whittlesea, B. (2003). Long-term semantic transfer: An overlapping-operations account. Memory & Cognition, 31(3), 401-111.
Kapnoula, E. C., & McMurray, B. (2016). Training alters the resolution of lexical interference: Evidence for plasticity of competition and inhibition. Journal of Experimental Psychology: General, 145(1), 8-30.
Kucker, S., McMurray, B., & Samuelson, L. (2015). Slowing down fast mapping: Redefining the dynamics of word learning. Child Development Perspectives, 9(2), 74-78.
Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22,1-75.
Li, D. (1993). A study of Chinese characters. Peking University Press.
Liu, Y. Y, Shu, H., & Li, P. (2007). Word naming and psycholinguistic norms: Chinese. Behavior Research Methods, 39, 192-198.
Luce, P. A., & Pisoni, D. B. (1998). Recognizing spoken words: The neighborhood activation model. Ear and Hearing, 79(1), 1-36.
Luo, C., Proctor, R. W., Weng, X., & Li, X. (2014). Spatial Stroop interference occurs in the processing of radicals of ideogrammic compounds. Psychonomic Bulletin & Review, 21, 715-720.
Magnuson, J. S., Tanenhaus, M. K., Aslin, R. N., & Dahan, D. (2003). The microstructure of spoken word recognition: Studies with artificial lexicons. Journal of Experimental Psychology: General, 132,202-227.
Magnuson, J. S., Dixon, J. A., Tanenhaus, M. K., & Aslin, R. N. (2007). The dynamics of lexical competition during spoken word recognition. Cognitive Science, 31(1), 133-156.
McClelland, J. L., & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1-86.
McClelland, J. L., & Rumelhart, D. E. (1985). Distributed memory and the representation of general and specific information. Journal of Experimental Psychology: General, 114(2), 159-188.
McLeod, R, Plunkett, K., & Rolls, E. T. (1998). Introduction to connectionist modeling of cognitive processes. Oxford University Press.
McMillan, C., Corley, M., & Lickley, R. J. (2009). Articulatory evidence for feedback and competition in speech production. Language and Cognitive Processes, 24(1), 44-66.
McMurray, B., Horst, J. S., & Samuelson, L. K. (2012). Word learning emerges from the interaction of online referent selection and slow associative learning. Psychological Review, 119(4), 831-877.
Mirman, D. (2011). Effects of near and distant semantic neighbors on word production. Cognitive, Affective, and Behavioral Neuroscience, 77(1), 32-43.
Monsell, S., & Hirsh, K. W. (1998). Competitor priming in spoken word recognition. Journal of Experimental Psychology Learning Memory & Cognition, 24(6), 1495-1520.
Murrell, G. A., & Morton, J. (1974). Word recognition and morphemic structure. Journal of Experimental Psychology, 102(6), 963-968.
Napps, S. E., & Fowler, C. A. (1987). Formal relationships among words and the organization of the mental lexicon. Journal of Psycholinguistic Research, 16(3), 257.
Oppenheim, G. M., Dell, G. S., & Schwartz, M. F. (2010). The dark side of incremental learning: A model of cumulative semantic interference during lexical access in speech production. Cognition, 114, 227-252.
Pan, J., Song, S., Su, M., Mcbride, C, Liu, H., Zhang, Y, et al. (2015). On the relationship between phonological awareness, morphological awareness and Chinese literacy skills: Evidence from an 8-year longitudinal study. Developmental Science, 19(6), 982-991.
Pecher, D., Zeelenberg, R., & Wagenmakers, E. J. (2005). Enemies and friends in the neighborhood: Orthographic similarity effects in semantic categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 121-128.
Perfetti, C. A., Liu, Y., & Tan, L. H. (2005). The lexical constituency model: Some implications of research on Chinese for general theories of reading. Psychological Review, 772(1), 43-59.
Pexman, P. M., Trew, J. L., & Holyk, G. G. (2005). How a PINT can hurt you now but help you later: The time course of priming for word body neighbors. Journal of Memory and Language, 53, 315-341.
Pierrehumbert, J. (2001). Word frequency, lenition, and contrast. In J. Bybee & P. Hopper (Eds.), Frequency effects and the emergence of lexical structure (pp. 137-157). John Benjamins.
Qu, Q. Q., Feng, C., & Damian, M. E (2021). Interference effects of phonological similarity in word production arise from competitive incremental learning. Cognition, 272(5255), 104738.
R Core Team. (2018). R: A language and environment for statistical computing [Computer software], R Foundation for Statistical Computing.
Rueckl, J. G. (1990). Similarity effects in word and pseudoword repetition priming. Journal of Experimental Psychology Learning Memory & Cognition, 16(3), 374-391.
Rueckl, J. G., & Mathew, S. (1999). Implicit memory for phonological processes in visual stem completion. Memory & Cognition, 27(1), 1-11.
Seidenberg, M. S., & McClelland, J. L. (1989). A distributed developmental model of visual word recognition and naming. Psychological Review, 96, 523-568.
Seidenberg, M. S., Waters, G. S., Barnes, M. A., & Tanenhaus, M. K. (1984). When does irregular spelling or pronunciation influence word recognition? Journal of Verbal Learning and Verbal Behavior, 23, 383-404.
Shen, W., Qu, Q., & Li, X. (2016). Semantic information mediates visual attention during spoken word recognition in Chinese: Evidence from the printed-word version of the visual-world paradigm. Attention, Perception, & Psychophysics, 78(5), 1267-1284.
Shu, H., McBride-Chang, C., Wu, S. N., & Liu, H. Y. (2006). Understanding Chinese developmental dyslexia: Morphological awareness as a core cognitive construct. Journal of Educational Psychology, 98(1), 122-133.
Stark, C., & McClelland, J. L. (2000). Repetition priming of words, pseudowords, and nonwords. Journal of Experimental Psychology Learning Memory & Cognition, 26(4), 945-972.
Taft, M. (2006). Processing of characters by native Chinese readers. In P. Li, E. Bates, L. H. Tan, & O. J. L. Tzeng (Eds.), Handbook of East Asian psycholinguistics (Vol. 1, pp. 237-249). Cambridge University Press [Chinese Psycholinguistics],
Vitevitch, M. S., & Luce, P. (2004). A web-based interface to calculate phonotactic probability for words and nonwords in English. Behavioral Research Methods, Instruments, & Computation, 36(3), 481-187.
Wagenmakers, E. J., & Raaijmakers, J. G. W. (2006). Long-term priming of neighbours biases the word recognition process: Evidence from a lexical decision task. Canadian Journal of Experimental Psychology, 60(4), 275-284.
Xia, Z. C., Yang, T., Cui, X., Hoeft, E, Liu, H., Zhang, X. L., Shu, H., &Liu, X. P. (2022). Neurofunctional basis underlying audiovisual integration of print and speech sound in Chinese children. European Journal of Neuroscience, 55(3), 806-826.
Yang, J., McCandliss, B. D., Shu, H., & Zevin, J. D. (2009). Simulating language-specific and language-general effects in a statistical learning model of Chinese reading. Journal of Memory and Language, 61, 238-257.
Yang, J. E, Shu, H., McCandliss, B. D., & Zevin, J. D. (2013). Orthographic influences on division of labor in learning to read Chinese and English: Insights from computational modeling. Bilingulism: Language and Cognition, 16(S2), 354-366.
Yeh, S. L., & Li, J. L. (2004). Sublexical processing in visual recognition of Chinese characters: Evidence from repetition blindness for subcharacter components. Brain and Language, 88(1), 47-53.
Yin, L., Li, W, Chen, X., Anderson, R. C, Zhang, J., Shu, H., & Jiang, W. (2011). The role of tone awareness and pinyin knowledge in Chinese reading. Writing Systems Research, 3(1), 59-68.
Zhao, J., Wang, X., Frost, S., Sun, W, Fang, S.-Y, Mend, E., Shu, H., Pugh, K„ & Rueckl, J. (2014). Neural division of labor in reading is constrained by culture: A training study of reading Chinese characters. Cortex, 53, 90-106.
Zhou, X. L., & Marslen-Wilson, W. (1999). The nature of sublexical processing in reading Chinese characters. Journal of Experimental Psychology: Learning Memory & Cognition, 25(4), 819-837.
Zhou, X. L., & Marslen-Wilson, W. (2000). The relative time course of semantic and phonological activation in reading Chinese. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26,1245-1265.
Zhou, X. L., & Marslen-Wilson, W. (2009). Pseudohomophone effects in processing Chinese compound words. Language and Cognitive Processes, 24(1-8), 1009-1038.
Zhou, L., Fong, C. M., Minett, J. W, Peng, G., & Wang, S. Y. (2014). Pre-lexical phonological processing in reading Chinese characters: An ERP study. Journal of Neurolinguistics, 30, 14-26.
Ziegler, J. C., Tan, L. H., Perry, C., & Montant, M. (2000). Phonology matters: The phonological frequency effect in written Chinese. Psychological Science, 11, 234-238.
(ProQuest: Appendix omitted.)
Copyright Springer Nature B.V. Feb 2024