The Primacy of Abstract Syllables in Chinese Word Production

Abstract

Convergent evidence suggests that syllables play a primary and distinctive role in the phonological phase of Mandarin Chinese word production. Specifically, syllables are selected before other phonological components and thus guide subsyllabic encoding. The proximity of phonological syllables to word representations in Chinese languages ensures that they are also activated automatically by word perception. Therefore, in contrast to Indo-European languages, syllables but not necessarily subsyllabic components such as initial consonants can be perceptually primed in production. We tested this prediction in two masked-priming experiments. To isolate relevant phonological activation originating in primes, we used single character masked primes whose corresponding tones and lexical meanings always differed from those of the targets’ first morphemes. Related primes potentially activated the atonal first syllables or the first consonants of target words. To strongly engage production-specific processes, we used pictures as prompts for disyllabic target words. Facilitation relative to unrelated controls was observed only in the syllable sharing condition. If anything, sharing of initial consonants had a negative valence, perhaps indicative of competition among similar co-activated words or syllables. These findings corroborate the view that abstract syllables are the first selected, proximate phonological units in Chinese word production, and that phonemic segments play a subordinate role.

Word production involves several phases of processing (Dell, 1986; Levelt, Roelofs, & Meyer, 1999; Rapp & Goldrick, 2000) whose implementation may vary across languages. In this article, we focus on the distinctive role of syllables at the pivotal transition from pre-phonological to phonological status in Chinese languages. At this phonological juncture, a priori linguistic analysis (Cheng, 1973; Kuo, 1994), speech error patterns (Chen, 1999, 2000), computational work (T.-M. Chen, Dell, & J.-Y. Chen, 2007), and behavioral experiments (see O’Séaghdha, 2015, for review) all indicate that the type of unit playing the primary functional role differs between Chinese and other languages. Specifically, previous studies suggest that phonological encoding begins with syllables in Chinese (J.-Y. Chen, T.-M. Chen, & Dell, 2002; O’Séaghdha, J.-Y. Chen, & T.-M. Chen, 2010; Chen & Chen, 2013), whereas it begins with segments in Dutch, English, and perhaps all Indo-European languages (Meyer, 2000), and with morae in Japanese (Kureta, Fushimi, & Tatsumi, 2006; Kureta, Fushimi, Sakuma, & Tatsumi, 2015; Verdonschot et al., 2011). Thus, it is important to incorporate this variation in general accounts of language production.

Current language production theories, based mainly on data from European languages, assume a language-general segmental currency. The phonemic segments of words are retrieved and then assembled into syllabified phonological words for phonetic encoding and articulation (Dell, 1986; Levelt et al., 1999; O’Séaghdha & Marin, 2000; Rapp & Goldrick, 2000; Roelofs, 1997). These accounts do not work well for Chinese languages. In Chinese languages, evidence suggests that stored syllables are retrieved directly, and that the processes of segmental retrieval and assembly unfold later under the auspices of syllables (Chen, 1999, 2000; J.-Y. Chen et al., 2002; O’Séaghdha et al., 2010; see Roelofs, 2015). In response to this language-theory mismatch, we have proposed a proximate units principle (O’Séaghdha & Chen, 2009; O’Séaghdha et al., 2010; O’Séaghdha, 2015) whereby the initial transition to phonological form in production is enacted by different types of explicitly selectable phonological units across languages. Specifically, in Chinese languages, syllables are the proximate units and segmental encoding is therefore subordinated to stored syllables, whereas in Indo-European languages phonemic segments are the proximate units and syllables are assembled flexibly from these segments. A third configuration in which morae (phonological timing units of varying segmental composition; Kubozono, 1989; Labrune, 2012) are the proximate units has been identified in Japanese (Kureta et al., 2006; Verdonschot et al., 2011). The proximate units rubric preserves language-general phonological encoding operations of unit activation and selection, but emphasizes that the type of the primary units that implement these operations varies across languages (see O’Séaghdha, 2015, for additional discussion). Several lines of experimental evidence are relevant to these conclusions.

Form preparation

Perhaps the most compelling evidence for cross-language variation in phonological encoding comes from the form preparation task (Meyer, 1990, 1991) in which speakers repeat small sets of words that share (homogeneous condition) or do not share (heterogeneous condition) a designated phonological property. With this task, Dutch speakers exhibit a homogeneity benefit when the first consonant of a set of words is shared, and generally a larger benefit for any additional contiguous word-initial phonemes (Meyer, 1991). In contrast, there is a syllable benefit but no first consonant preparation benefit under equivalent circumstances in Chinese (J.-Y. Chen et al., 2002; O’Séaghdha et al., 2010), suggesting that Chinese speakers engage syllables but do not directly engage individual segments in the form preparation task. Crucially, O’Séaghdha et al. (2010) found no preparation of Mandarin Chinese onsets even for simple monosyllables, showing that the inaccessibility of initial consonants to preparation cannot be due merely to the need to select syllables before phonemic segments in complex words. In addition, Chen and Chen (2013) replicated the data pattern in separate experiments with auditory as well as written cues, and with pictures rather than printed characters as prompts for target names. These extensions show that the form preparation outcomes do not depend on the syllabic grounding of the writing system.

Similar to the Mandarin Chinese findings, Kureta et al. (2006; see also Kureta et al., 2015) showed form preparation of morae (e.g., a word initial CV unit), but not of the word initial consonants of the same units, in Japanese. This provides important corroboration in a very different phonological system of the non-universality of onset accessibility in production.

Some researchers, however, have questioned the generality of the proximate units account. For example, Wong, Huang and Chen (2012) reported a partial form preparation benefit in Cantonese Chinese when sets of form-related words shared the first consonant and vowel (CV) but not the whole first syllable.ⁱ This result seems to challenge the restriction of preparation to whole syllable units. Other evidence suggests that, under some conditions, Chinese speakers may show preparation of initial consonants. Verdonschot et al. (2013) reported such a form preparation benefit in proficient Chinese-English bilinguals provided that the target words all shared the same syllable structure. With this qualification, it appears that bilinguals who had mastered the segmental phonological system of English showed a form preparation benefit in their first language as well. This raises interesting questions about the flexibility of form preparation. However, form preparation of initial consonants under some circumstances in Chinese languages is not incompatible with the proximate units principle. It merely suggests that preparation may be deployed to various units once those units are metalinguistically accessible and thus available to directed attention (O’Séaghdha, 2015; see also O’Séaghdha & Frazer, 2014). In Wong et al.’s experiments, the CV components constituted syllables or potential syllables, for which preparation is natural (O’Séaghdha, 2015). A benefit of such preparation could accrue during encoding of the shared segments of fully or partly consistent syllables. In Verdonschot et al., the shared component was a nonproximate initial consonant that likewise may yield a benefit during segmental spellout of syllables for bilingual speakers. Nonetheless, these findings suggest that form preparation effects depend on how speakers attend to phonological components. Therefore, convergent evidence from other tasks is needed to bolster conclusions about default phonological encoding in Chinese.

Picture-word facilitation

Wong, Chen and colleagues also conducted more direct tests of initial consonant priming in Cantonese Chinese in a picture-word distraction task. In the picture-word task, participants name pictures in the presence of distractor words. Wong and Chen (2008) used both written and spoken distractors with monosyllabic picture names, showing no initial consonant sharing benefit in either format. These findings contrast with well attested facilitation effects with equivalent conditions in European languages (e.g., Damian & Martin, 1999; Meyer & Schriefers, 1991). The absence of facilitation in the auditory condition is important because a null effect in this situation cannot be due to failure to activate input phonology. Rather, it indicates failure of subsyllabic input phonology to influence the output process. As expected, Wong and Chen observed facilitation with syllable distractors. However, somewhat problematically for the proximate units account, they also found facilitation with rhyme (VC), and especially rhyme+tone distractors. Wong and Chen (2009) additionally found facilitation with CV distractors for CVC targets, and Wong and Chen (2015) reported benefits of any two segments of a CVC syllable (including noncontiguous C_C consonants) if tone was also congruent. Thus there is a challenge in reconciling the absence of initial consonant priming (a key finding for the proximate units account) with the presence of priming in a variety of other subsyllabic conditions. In contrast to Wong and Chen who suggested that the latter findings challenge the generality of the proximate units account, we are more persuaded by the consistent absence of initial consonant priming. Because of this absence of initial consonant priming, we suggest that the part-syllable effects in the picture-word task must have a nonsegmental explanation. Specifically, they may arise in syllable selection through a process where the distractors reduce uncertainty concerning the pronunciation of spoken target syllables. Thus, the part-syllable facilitation effects do not necessarily speak to the question of how phonology is initially retrieved (see O’Séaghdha, 2015, for additional discussion).

Masked priming

In contrast to the form preparation and picture-word tasks, masked priming provides evidence concerning automatic activation of phonology in a procedure that limits or precludes deliberate strategies. Masked priming has been extensively studied in alphabetic languages, often with a focus on onset priming as a distinct perceptual phenomenon, the “masked onset priming effect” or MOPE (Forster & Davis, 1991). According to Forster and Davis, masked priming of onsets could occur through grapheme-phoneme conversion which provides a jump-start to target production, perhaps pre-selecting the first phoneme of the eventual target. Subsequent work has suggested that onset priming occurs later during word production (Kinoshita & Woollams, 2002; Malouf & Kinoshita, 2007; see also Schiller, 2008) through activation of the onset but without pre-insertion.

A good deal of masked phonological priming research has been devoted to nonlexical priming, for example from single, word-initial letters (see Schiller, 2008). However, because standard Chinese writing and printing are nonalphabetic, visual letter priming cannot be implemented. This statement must be qualified by the existence of pinyin. Pinyin is a special alphabet-like code that is used in support of character generation but is not used in fluent reading. It thus cannot be equated with alphabetic writing-and-reading systems. The smallest natural visual prime units in Chinese writing are characters or character components (called radicals). In the current work, primes and targets almost never share radicals, and so the discussion is centered on phonological activation triggered by whole characters.

To test for onset effects with visual primes, one must use characters that share a phonological onset with the target word. At the level of syllables, masked priming studies of both European and Chinese languages have examined whether syllables make a difference when segmental overlap is controlled (Chen, Lin, & Ferrand, 2003; Schiller, 1998). In such studies, primes and targets share several segments that undershoot, coincide with, or exceed a syllable boundary (see Chinese example below) so that the effect of segmental overlap can be segregated from the effect of the presence of a syllable boundary. With this design, masked priming is insensitive to syllable boundaries in Dutch (Schiller, 1998) and English (Schiller, 2000; but see Ferrand, Segui, & Humphreys, 1997). The status has been controversial in French. The evidence initially suggested a distinct role of syllables in phonological encoding (Ferrand, Segui, & Grainger, 1996), but later studies point to the conclusion that differences between French (and other Romance languages) and Germanic languages may not extend to the initial choice of fundamental units of encoding (Brand, Rey, & Peereman, 2003; Perret, Bonin, & Méot, 2006; Schiller, Costa, & Colomé, 2002). Rather, these language groups may differ in other respects such as the strength of segmental affiliation with syllable structures (e.g., Romani, Galluzzi, Bureca, & Olson, 2011), or the salience of post-lexical phonetic syllables (e.g., Carreiras & Perea, 2004; Cholin, Schiller, & Levelt, 2004; Cholin, Dell, & Levelt, 2011).

Distinctly syllabic priming, however, is clearly present in Chinese (Chen, Lin, & Ferrand, 2003; see also Chen & Dell, 2006; Lin & Chen, 2003; You, Zhang & Verdonschot, 2012). In the most pertinent condition of their study, Chen et al., (2003, Experiment 3) used masked priming with a word naming task in which the masked prime characters and the target characters were always different. A CV-GV word (C stands for consonant, V for vowel, G for glide; e.g., 霸業 /ba/4-/ye/4) was named faster when preceded by a visually masked CV character (e.g., 把 /ba/3) than by a masked CVG character (e.g., 百 /bay/3), even though the glide comprised the onset of the second syllable of the target. Likewise, a CVG-CVX word (X stands for any legal segment; e.g., 敗類 /;bay/4-/ley/4) was named faster when preceded by a masked CVG character (e.g., 百 /bay/3) than by a masked CV character (e.g., 把 /ba/3). In combination, these results show a syllable sharing benefit that does not depend on number of segments. These findings were subsequently replicated with CV/CVC first syllables (Lin & Chen, 2003), providing further generality. However, whereas these studies convincingly show a benefit of syllable priming independent of the number of shared segments in Chinese, they do not provide a test of segmental onset priming. Moreover, because they did not include an unrelated character baseline, they do not provide a definite estimate of the syllable priming effect size (See You, Zhang & Verdonschot, 2012). Finally, because word naming involves reading, the observed syllabic effect could be promoted by perceptual priming of characters in the writing system rather than by production-specific processes.

The Current Study

In this article, we test a strong prediction of the proximate units perspective, that perceptually driven syllable activation necessarily intersects with selection of syllables in the production of Chinese words. That is, we predict that syllable priming originating in perception will occur in the production pathway even if segmental priming does not. We test this prediction using masked priming, because this task is not subject to the explicit strategies that inevitably complicate the interpretation of form preparation and picture-word data. Our reasoning is that because syllables are the primary, first activated phonological units for production, they are automatically activated by any lexical input.

In Chinese languages, syllables are not only proximate units, adjacent to non-phonological lexical lemma representations (see Figure 1) but also discrete controlling units for subsequent phonological processes. Before a Chinese syllable can be spoken, its segments must be impleted and the resulting phonological syllable must be tonally specified (O’Séaghdha et al., 2010). We use the term impleted here to express the idea that phonemic segments are straightforwardly filled in to positions in syllables. In contrast, phonemic encoding in Indo-European languages is more complex. Phonemes are activated, selected and linearized in emergent, syllabified phonological words (Levelt et al., 1999). Thus, phonemes are proximate units in these languages, but they do not by themselves control phonological encoding. In Chinese, syllables are selected and linearized, and the segments of each syllable are impleted without a need for resyllabification (see Roelofs, 2015, for a similar view). Proximate syllable units are activated by acoustic or visual word recognition, and because syllable selection for production applies to these units, this perceptually-driven activation causes facilitation in picture naming.

Figure 1.

Simplified schematic of relevant processing in masked priming conditions. Tone processing is not shown. All connections between linguistic units are bidirectional.

In contrast to syllables, subsyllabic phonological elements may not be activated enough in Chinese word perception to influence word production for any of several reasons. First, activation of phonological segments in production is mediated by syllables and so may be weak. Second, because of the priority of syllables in production, any pre-activation of segments through perception may dissipate before they become relevant. Third, there is increasing consensus that input and output phonology are separate at more peripheral stages of perception and production (see e.g., Nozari, Kittredge, Dell, & Schwartz, 2010 for discussion). Accordingly, in Chinese, visual character inputs may activate corresponding auditory input phonology but this need have no effect on subsyllabic production units. Finally, in our account of Chinese production, onset-related primes do automatically activate proximate output syllables. Because these syllables are similar to the target syllables, there is potential for interference from such primes during selection of the first syllables of the targets. All of these considerations point to weak or nonexistent net facilitation of phonological onset consonants in Chinese under masked priming conditions.

There is another important theoretical motivation for masked priming tests in Chinese. Although the bulk of evidence suggests that word-initial consonant priming does not influence naming time of Chinese words in form preparation or picture-word tasks, it is still possible that these components are activated despite their behavioral invisibility. Qu, Damian and Kazanina (2012) argued for this view based on event-related potential (ERP) evidence. In their study, participants named pictures with color-adjective + noun phrases (for example, 紅盒 hong2 he2 ‘red box’ versus 藍盒 lan2 he2 ‘blue box’). In contrast to corresponding English language studies (Damian & Dumay, 2007, 2009), there was no behavioral facilitation from the shared consonants, seemingly supporting the conclusion that phonological segments are not functional at the outset of Chinese word production. However, posterior ERP signatures of the same conditions were clearly different in the shared consonant conditions and in unrelated controls about 200 ms after picture display. Moreover, the conditions were also distinct (though with reversed polarity) in a separate anterior ERP signature in a 300–400 ms time window. Qu et al. proposed that relatively weak phonemic activation (posterior pattern) is balanced out by a monitoring process (anterior pattern) that oversees sequencing of similar words in Chinese. That is, they proposed that phonemic segments are functional despite their behavioral invisibility very early in Chinese word production. Masked priming provides a good way to evaluate this hypothesis. Under masked priming, speakers are unaware of the prime and so there should be little or no monitoring, with the result that facilitation should occur under Qu et al.’s view. In contrast, as explained earlier in this introduction, the proximate units view suggests that any effect of onset segment sharing between primes and targets is mediated in the first instance by syllables. For this reason, there may be interference between similar syllables, but onset facilitation may not occur.

In this article, we address the limitations of previous masked priming studies in several ways. We included an onset sharing condition as well as a syllable condition, and we used the same primes in separate counterbalanced control conditions. In addition, to ensure that word production processes were strongly engaged but that individual characters and corresponding word structure were not manifest, we cued the word targets with pictures (see also You et al., 2012). With this procedure, we tested priming of both word-initial consonant onsets and atonal syllables. By atonal syllables we mean abstract syllables defined by their segmental composition but lacking tonal values (see Figure 1 where tone is retrieved at the word level but not transmitted to phonological syllables; see O’Seaghdha et al., 2010, and O’Séaghdha, 2015, Figure 2 for additional consideration of syllable-tone unification). These units are therefore not articulatory. In the experiments, primes always differed in tone (and thus morphemically, semantically, and in corresponding written character) from targets.

Experiment 1: Between Groups Masked Priming of Picture Name Onsets and First Syllables

The first experiment tested for both onset and syllable priming of word production by masked Chinese characters. To maximize engagement of production processes we used picture cues to elicit target production. We used the same picture cues following related primes sharing initial consonant onsets or first syllables (but not tone) with the picture names, and also following unrelated primes. To limit the origin of activation to the intended character, we used single character masked primes. The primes were always words, and so our experiment accommodates nonspecific lexical effects of prime activation in all conditions (see Schiller, 2008). The same primes were used in related and unrelated conditions for each type. If phonological activation in Chinese follows the same channels as in Indo-European languages, then activation of onsets by onset-sharing primes will be strong enough to benefit the initiation of word production, though not as much as in a language with an alphabetic writing system (see O’Séaghdha, Chen & Chen, 2013). Alternatively, because syllables are proximate units in production of Chinese, syllable but not necessarily onset priming will occur.

Method

Participants

Sixty students from National Cheng Kung University, Taiwan, multilingual native speakers of Mandarin Chinese, participated in the experiment. They also spoke Taiwanese and were learners of English since middle school but were not proficient. They were paid 100 Taiwan dollars for participation.

Design

We used a mixed factorial design. Priming condition (onset, syllable) was a between groups and between items factor, and prime relatedness (related, unrelated) and block (1, 2, 3) were within participants factors. Each picture appeared once with a related prime and once with an unrelated prime in each of three blocks. The order of items was randomized in each block. Picture name response times and errors were tabulated.

Materials

The targets were 42 line-drawings with disyllabic names. Each target was paired with two single character primes whose names shared an atonal syllable (same segmental composition but different tone) or a syllable onset consonant (also with a different tone on the prime and the target syllable) with the target’s first syllable (see Appendix). For each priming condition, the same pictures and characters were re-paired to create 42 unrelated character-target pairs, such that the character names shared no systematic phonological overlap with the first syllables of the picture names. We also ensured that there was no accidental overlap with the second syllable. In addition, there were 20 practice trials in which a set of 10 pictures was used twice, once with a related prime (onset or syllable) and once with an unrelated prime.

Procedure

The experiment was programmed in DMDX (Forster & Forster, 2003) and run on an ASUS laptop with a 14-inch LCD display. Participants were familiarized with the names of the 42 target pictures and the 10 practice pictures. Each picture was presented at the center of the screen with the picture name displayed below it. Then the pictures were named one by one. Errors were corrected and the procedure was repeated if there were incorrect or disfluent responses. Finally, the testing blocks were conducted, following the four-field masking procedure of Chen et al. (2003).

For each trial, participants fixated on a plus sign (‘+’) that appeared at the center of the screen to signal the location where the target would be shown. They pressed the left button of the mouse to indicate readiness. The following four events ensued: 1) A forward mask, a double ‘at’ sign (‘@@’) scaled to the size of two Chinese characters, appeared for 504 ms; 2) the character prime (in Standard Kai font, about 1.5 cm wide) appeared for 49 ms.; 3) a backward @@ mask replaced the prime for 14 ms; 4) the target picture sized 8.5 × 8.5 cm, appeared for up to 2000 ms until the participant initiated a naming response. Participants were asked to name the pictures as quickly and accurately as possible. A chinrest was used and the laptop was raised so that eye level was aligned with the center of the laptop display at a distance of about 60 cm. At this distance, each character subtended approximately 1 degree of visual angle horizontally and vertically.

Results

Errors (incorrect names, stuttering, and responses beginning with non-speech sounds, 0.89%) and extreme values (correct responses less than 200 ms, or two standard deviations beyond each participant’s group mean, 4.5%) accounted for 5.4% of the data. These were excluded from further analysis.

Table 1 presents the mean response times for the correct and valid responses as a function of prime type, relatedness and block, together with net priming effects (unrelated RT – related RT). It is clear from the table that there were across-the-board practice effects, especially from the first to the second block, for both prime conditions. The priming effects were slightly but consistently negative in the onset prime condition over the three blocks (−5, −6, −6 ms). The priming effects were facilitatory in the syllable prime condition in the first two blocks but not in the third block (10 , 12, −1 ms).

Table 1.

Experiment 1: Mean response times in milliseconds and net priming effects for correct responses as a function of prime type, relatedness and block.

	Onset Prime			Syllable Prime
	Related	Unrelated	Priming Effect	Related	Unrelated	Priming Effect
Block 1	643 (115)	638 (109)	−5	642 (116)	652 (125)	10
Block 2	598 (88)	592 (84)	−6	594 (98)	607 (107)	12
Block 3	593 (94)	587 (87)	−6	590 (104)	589 (99)	−1
Overall	611 (99)	606 (93)	−6	609 (106)	616 (110)	7

The response times were inverse transformed and submitted to linear mixed-effect modeling (LMM, Baayen, 2008; Bates, 2005) using the REML procedure in R Version 3.2.1.

Participants and items were random factors. Prime type (onset, syllable), prime relatedness (related, unrelated), block (first, second, third) and all possible interactions were fixed-effects. We used a maximal model with random intercepts and by-subject random slopes for block, relatedness, and their interaction (see Barr, Levy, Scheepers & Tily, 2013). t-values and p-values for fixed factors and interactions were obtained via the lmerTest package, which provides p-values calculated through Satterthwaite approximation.

The analysis showed nonsignificant main effects of relatedness (t = 0.65, p = .52) and prime type (t = 0.09, p = .93). Crucially, the interaction of relatedness and prime type was significant (t = −2.12, p < .04), reflecting the contrasting effects of relatedness in the onset and syllable priming conditions. The simple main effects of relatedness in the onset prime condition (− 6 ms; t = 1.99, p = .046) and in the syllable prime condition (7 ms; t = −3.39, p = .0007) were significant. In addition, there was a significant acceleration between blocks 1 and 2 (t = 8.95, p < .001) and between blocks 2 and 3 (t = 8.37, p < .001). However, block did not interact with prime type or relatedness.

Because syllable priming was not evident in the third block, we conducted an additional analysis on the data of the first two blocks only. This analysis also showed no main effects of relatedness (t = 0.64, p =0.53) or prime type (t = 0.09, p =0.93). Relatedness was modulated by prime type (t = −2.11, p < .04). In contrast to the full analysis, the negative effect of relatedness was not significant in the onset prime condition (t = 1.41, p =0.17) in this analysis. However, the positive effect in the syllable prime condition was clearly significant (t = −4.13, p < .001). As in the first analysis, there was a main effect of block (t = 8.99, p < .001), but block did not interact with the other factors. Analysis of the third block data confirmed that prime type and relatedness were nonsignificant and did not interact. The reasons for the loss of syllable facilitation in the third block are not clear, but they may involve an increase in prime visibility with multiple exposures.

Discussion

The results support the hypothesis that syllables constitute the starting point of phonological activation in Chinese word production. The key finding is facilitation in the syllable prime condition but not in the onset prime condition in the first two blocks. Because the primes were masked and because there was no visual or semantic relation between primes and targets or target first syllables, the syllable priming effect must be phonological.

The slight negative effect of onset priming, significant overall but not in the secondary analysis of the first two blocks, was unexpected. Because onset priming is typically facilitatory (Schiller, 2008), the negative tendency in our data may have a different origin. In the general discussion, we propose that it is a syllable level effect reflecting co-activation of similar prime and target syllables. In addition, the syllable priming effect unexpectedly disappeared in the third block. This contrasts with Chen et al.’s (2003) word naming study where the effects did not vary with repetition. We suspect that the loss of syllable priming in the third block of Experiment 1 may be a result of masking losing its effectiveness due to the multiple repetitions of primes across conditions and blocks.

Although none of these aspects of the data contradict the presence of distinctively syllabic priming, a replication is warranted. In Experiment 2, we made several adjustments to address the uncertainties of interpretation left by Experiment 1. The primary change, adoption of a within-subjects manipulation of prime type, also addresses the slight baseline difference in the unrelated condition latencies (see Table 1).

Experiment 2: Within Subjects Forward Mask Priming of Picture Name Onsets and First Syllables

This experiment was designed to replicate Experiment 1 and to further resolve the status of syllable and first consonant priming. Relative to Experiment 1, we made several changes. First, we used a within-subjects design to maximize sensitivity. Second, to reduce repetition of primes and prime-target pairings, each target was presented only once in each of two blocks, with relatedness counterbalanced over four lists and reassigned in each block so that no particular prime-target combination occurred more than once. We also made two adjustments to the masking procedure. The mask was changed to a double pound (##) and sized to the 1.5 cm single-character prime rather than to the target picture. This made the mask smaller and, more importantly, it was impossible for any central properties of primes to be visible between the two components of the mask, as may possibly have been the case with the larger double at (@@) mask in Experiment 1. In addition, following studies such as Kinoshita and Woollams (2002), we dispensed with the backward mask, thus potentially increasing the effectiveness of the primes. We also shortened the prime duration from 49 ms to 33 ms to reduce the possibility of primes becoming visible.

Method

Participants

Twenty-four students from National Taiwan Normal University, Taiwan, native speakers of Mandarin Chinese and otherwise similar to the Experiment 1 participants, participated in the experiment. They were paid 110 Taiwan dollars for their participation.

Materials

The targets were 39 of the 42 line-drawings with disyllabic names we used in Experiment 1. Two items were removed to enable fully systematic counterbalancing of lists (see Appendix). One other item was removed due to an error in unrelated prime assignment. As in Experiment 1, each target was paired with single character atonal syllable and onset-related primes. The overall frequencies of the two prime types were identical at 479/million (Mandarin Promotion Council, Taiwan, 1998). Except for adjustments to one item resulting from the deletions, the unrelated condition pairings were the same as in Experiment 1. There were three warm-up trials and seven practice trials with items that were not used in the main experiment.

Design

Prime type (onset, syllable), relatedness (related, unrelated), and block were manipulated within subjects. Four lists were created using a Latin square procedure by rotating the targets across the four prime conditions in sectors of ten. The sectors were balanced such that mean target latency from Experiment 1 was almost identical in each sector (639 ms). Unrelated prime-target pairs appeared equally often across lists, but unrelated primes could range across sectors (see Appendix). Six participants were randomly assigned to each of the four lists. Each participant received two blocks of 40 trials with different prime-target assignments in each block. The order of target items was randomized in each block.

Procedure

The procedure was almost the same as in Experiment 1, so not all details are repeated here. The experiment was run on an ASUS PC with a larger 24-inch LCD display. Participants were familiarized with the names of all the target pictures, including practice items, in a more systematic way than in Experiment 1. Each picture was presented at the center of the screen with the picture name displayed below it. Then participants were shown cards with the pictures one by one and asked to name them aloud. The procedure was repeated twice. Errors were corrected and the procedure was repeated again if there were incorrect or disfluent responses. The testing blocks used a three-field forward masking procedure. On each trial, participants fixated on a plus sign (‘+’) that appeared at the center of the screen for 1 second (997 ms). The following three events ensued: 1) A double ‘pound’ sign (‘##’) forward mask scaled to the size of one Chinese character (1.5 cm), appeared for 499 ms; 2) the character prime (in New Ximing font) appeared for 33 ms; 3) the target picture (8.5 cm × 8.5 cm in size) appeared for up to 2000 ms until the participant initiated a naming response. Participants were asked to name the pictures as quickly and accurately as possible. A chinrest was not used. Eye level was aligned with the center of the LCD display at a distance of about 65 cm. At this distance, each character subtended approximately 1.1 degrees of visual angle horizontally and vertically.

Results and Discussion

One subject was excluded from the data analysis due to exceptional ability to read masked primes. In addition, trials with the item “ 犀牛 (xi1niu2)” as the target were excluded because this item was accidentally assigned a related prime in the unrelated condition. The remaining data were analyzed using a linear mixed effect model.

Errors (1.62 %) and extreme values (4.2 %) accounted for 5.8 % of the data. Table 1 presents the mean response times for the valid responses as a function of prime type, relatedness and block together with net priming effects (unrelated RT – related RT). Mean RTs were faster than in Experiment 1, and were further reduced over blocks. In the onset prime condition, the priming effect was −10 ms in the first block, and −1 ms in the second block. In the atonal syllable prime condition, a facilitatory effect appeared in both blocks (17, 11 ms). Thus, replicating Experiment 1, facilitatory priming was limited to the syllable condition.

The REML analysis was similar to that of Experiment 1 except that all of the fixed effects, including prime type, were manipulated within participants, and there were two blocks instead of three. Participants and items were random factors. Prime type (onset, syllable), prime relatedness (related, unrelated), block (first, second), and all possible interactions were fixed-effects. The analysis showed nonsignificant main effects of prime type (t = 1.71, p =.09) and relatedness (t = 1.16, p =0.25). There was a significant effect of block (t = 2.40 , p= .02), but block did not modulate any other effects. Crucially, in line with the descriptive summary above and consistent with Experiment 1, the effect of relatedness was modulated by prime type [prime type by relatedness interaction, t = −2.11, p = .03].

Separate analyses of the syllable and onset conditions showed no effect of relatedness in the onset prime condition (t = 0.89, p= 0.38) but a significant effect in the syllable prime condition (t = −2.23, p = .029). Neither of these effects was significantly modulated by block. Thus, the results are very consistent with those of Experiment 1, showing facilitatory priming only in the syllable condition. The marginal negative effect of initial consonant onset in the first block was observed throughout Experiment 1, but occurred only in the first block here.

One slightly puzzling feature of the data is the slower response time to targets in the unrelated syllable (609 ms) than in the unrelated onset (596 ms) conditions despite the use of a within-participants and within-items design. The slight baseline difference in Experiment 1 could be attributed to differences between the aggregated participants serving in the onset and syllable conditions. However, because Experiment 2 used a within-participants design, the baseline difference here is most plausibly attributed to arbitrary properties of the unrelated masked primes in each condition. It follows that the corresponding difference in Experiment 1 may have the same basis. Because the primes are the same in related and unrelated conditions for each prime type in both experiments, this variability is accommodated by the design. Only the effects of the related primes in each condition, relative to their own controls, matter.

General Discussion

We confirmed a key prediction of the proximate units principle (O’Séaghdha et al., 2010), that because word form encoding in Chinese originates in syllables, atonal syllable priming should occur in masked conditions where speakers are unaware of the primes. In both experiments, picture naming times were significantly faster in the syllable priming condition than in the corresponding phonologically unrelated control condition. In contrast, picture naming times were not facilitated in the initial consonant onset priming condition. If anything, they showed a slight reverse tendency. These findings are consistent with but more definitive than previous findings of distinctly syllabic priming in Mandarin Chinese (Chen et al., 2003; You et al., 2012). In addition, they are couched in a more precise, proximate units explanatory framework than previous work.

Because of the design features of Chinese, the syllable priming effect that we report here is not only distinctly syllabic but is definitively phonological (see Chen et al., 2003). Crucially, the syllable priming effect is abstract in that the primes and the first syllables of the targets not only have different meanings, but also look different when written and have different tones. Because prime and corresponding target syllables have different tones, they do not match phonetically, and so it is reasonable to conclude that the masked priming effect occurs at a phonological rather than at an articulatory/phonetic locus (see Figure 1). This attribution of priming to phonological proximate units would be difficult to make in languages such as Japanese (Verdonschot et al., 2011) and Dutch (Levelt et al., 1999; Schiller, 1998). In Japanese, the mora manipulations reported by Verdonschot et al. have the same realizations in primes and targets, and, in conjunction with metrical properties, the morae are advanced directly from phonological to subsequent phonetic realization. Therefore, the priming effect could occur at an earlier or later phase of production. Likewise in alphabetic languages, primes and targets that overlap phonologically usually also coincide in corresponding orthography, and except for extrinsic stress adjustments, target onsets are promoted without systematic modification to articulation. The proximate units framework leaves open the possibility that masked form priming in other languages occurs during later stages of encoding as well as through early proximate phonological unit activation, but our findings indicate that the syllable priming effects we observed in Chinese arise at the early proximate stage.

Given the positive evidence for the primacy of syllables in Chinese, the absence of the onset priming effect that is consistently found in European language experiments is also striking. In theory, activation could spread from primed syllables to phonemes, producing at least some priming at that level (see Figure 1B). The absence of priming could be due to a failure of activation of the prime onset, to failure of such activation to influence target production, or to over-riding of onset activation by other processes.

If initial consonant onset segments are activated indirectly through syllables in Mandarin production, it follows that single segment activation must be weaker than syllable activation. Therefore, it could be that segments are simply not activated strongly enough by nonalphabetic primes to show any benefit. Testing for alphabetic priming is problematic in Chinese because the alphabetic writing system, Pinyin, is not a fully fledged writing and reading resource, but is possible in the romaji script of Japanese. However, a test with romaji alphabetic onset primes by Verdonschot et al., (2011) proved negative. Thus, unlike in Dutch (Schiller, 2008), letter to phoneme priming appears to be absent in Japanese. This finding, combined with the absence of initial consonant onset effects in both Japanese and Chinese across several tasks reviewed in the introduction, support the possibility that the absence of masked onset priming in both languages could be due to failure of activation or to failure of activation to influence encoding. More precisely, onset consonants may not be activated sufficiently by primes, or their activation may not influence target encoding. For example, target encoding following syllable or mora selection may be highly automatic and thus immune to priming.

The current data in fact indicate not only an absence of onset priming, they suggest a slightly negative effect of onset sharing between primes and targets (a consistent small effect in all blocks of Experiment 1 and a negative though not statistically significant tendency in the first block only of Experiment 2). From the proximate units perspective, a cost of onset sharing could arise due to competition between similar syllables during selection of the first syllables of the targets. Such competition could coincide with and override any concurrent phonemic activation of shared onsets, leaving a small net negative tendency. In this interpretation, both the priming of syllables and the similarity-driven competitive tendency between prime and target syllables, require a syllable-level interpretation. Proximate syllable units mediate the facilitation in the syllable priming condition and may also account for the negative tendency in the onset-sharing condition. Consistent with this interpretation, Sereno and Lee (2014) also observed a cost of onset-sharing in auditory Chinese lexical decisions. This suggests a similar primacy of syllable processes in Chinese production and perception, with syllables preceding segmental encoding in production and superceding segmental processes in word recognition (O’Séaghdha, 2015).

In our theory, abstract syllables are not only proximate phonological encoding units, but also controlling units for subsequent processing. That is, they play a role comparable to that of words or morphemes in languages such as English. If Chinese phonemes are selected through the process we call segmental impletion during elaboration of the phonological forms of selected syllables (Chen et al., 2002; O’Séaghdha et al., 2010; O’Séaghdha, 2015), subsyllabic ingredients may not be available until after syllable selection.

Our theory may also account for the puzzling results of the study by Qu et al. (2012) that we reviewed earlier. Recall that in that study participants produced color adjective + noun phrases. Qu et al. reported effects of onset phoneme sharing in ERP signatures but not in behavioral response times. Because they reflect automatic perception without awareness of primes, the current null or reversed findings for initial consonant onsets argue against Qu et al.’s claim that the ERP data show functional engagement of phonemes at the outset of production. As we suggested in reviewing that study, the presence of shared segments between an adjective and following noun may be registered at the syllable level in Chinese to promote selection of the color adjective before the noun. Thus, although discrete phonemes are indeed selected prior to word production, the repeated phoneme effect in Qu et al.’s early (200 ms) ERPs may reflect syllable selection or monitoring rather than direct engagement with phonemic segments.

It may be useful to recapitulate the implications of our findings in terms of the intersecting perceptual and production pathways involved in character reading and picture naming. Current models of word comprehension and production distinguish between input and output phonology, though this often goes unnoticed because these modalities are usually treated separately in distinct investigations. Our claim that masked character primes activate primary syllabic production units does not contradict evidence of concomitant activation of input phonology during character recognition (e.g., Feng, Miller, Shu, & Zhang, 2001; Perfetti, Liu, & Tan, 2005). Our data do not speak to the functional role of such activation, but they do suggest a need for greater precision in specifying the nature of phonological activation in reading Chinese. We conjecture that phonological input activation during Chinese character recognition is likely to include a similar syllabic grain as output representation (see Ziegler & Goswami, 2005, for relevant context), and may be the point of intersection between distinct input and output operations.

To the extent that either comprehension or production depends functionally on phonological representation of syllabic granularity, direct mapping between perception and production may be constrained to that level. Consider the case of simple repetition of audible words. Nozari et al. (2010) examined the roles of lexically mediated and direct audition-to-speech mapping in English aphasic word repetition, finding that the lexical route was strongly engaged even though a nonlexical option was available. The nonlexical phonological route was also in evidence in a supporting role, perhaps serving to salvage repetition in cases of comprehension failure. If our analysis of Chinese word production is correct, implementation of a nonlexical route in Chinese word repetition would preferentially rely on syllable-to-syllable mapping rather than operating on phonemic segments. More broadly, crosslinguistic data can add further precision and generality to models of aphasic and nonaphasic word production by requiring them to account for implementation through different primary phonological units.

Although onsets by definition correspond to overt word beginnings in auditory perception, our findings suggest that they are not the phonological starting point of word production in Chinese. Rather, phonological encoding begins with syllables and proceeds to a segment impletion phase for each selected syllable. The key contribution of this article is to show that this priority of syllables extends across modalities and to the realm of activation without awareness. We conclude that at the launching point of phonological encoding for word production, syllables rather than segments are the beginnings of Chinese words.

Table 2.

Experiment 2: Mean response times in milliseconds (with standard deviations) as a function of prime type, relatedness and block.

	Onset Prime			Syllable Prime
	Related	Unrelated	Priming effect	Related	Unrelated	Priming effect
Block 1	610 (88)	600 (82)	−10	602 (88)	618 (125)	17
Block 2	592 (78)	591 (80)	−1	588 (80)	599 (85)	11
Overall	601 (83)	596 (81)	−6	595 (84)	609 (105)	14

Appendix

Materials showing related syllable and onset primes, corresponding unrelated primes, and picture names with Pinyin transcriptions and English translations. The materials are divided into the 4 sectors of targets used in counterbalancing Experiment 2 assignments. One item (犀牛 xi1niu2) was excluded from Sector 3. This and two other items that were used only in Experiment 1 are at the end of the list.

Prime				Target Picture
Syllable		Onset
Related	Unrelated	Related	Unrelated	Name	Pinyin	English Translation
台 tai2	操 cao1	圖 tu2	粗 cu1	太陽	tai4yang2	‘sun’
組 zu3	惹 re3	早 zao3	染 ran3	足球	zu2qiu2	‘soccer’
港 gang3	計 ji4	狗 gou3	句 ju4	鋼琴	gang1qin2	‘piano’
把 ba3	陣 zhen4	比 bi3	住 zhu4	八卦	ba1gua4	‘eight trigrams’
騷 sao1	母 mu3	森 sen1	米 mi3	掃把	sao4ba3	‘broom’
耐 nai4	入 ru4	念 nian4	讓 rang4	奶瓶	nai3ping2	‘baby bottle’
咖 ka1	七 qi1	哭 ku1	區 qu1	卡車	ka3che1	‘truck’
堂 tang2	起 qi3	頭 tou2	取 qu3	湯匙	tang1chi2	‘spoon’
岔 cha4	舖 pu4	臭 chou4	派 pai4	插座	cha1zuo4	‘electrical outlet’
鳥 niao3	酸 suan1	女 nu3	鬆 song1	尿布	niao4bu4	‘diaper’
醒 xing3	含 han2	許 xu3	湖 hu2	星形	xing1xing2	‘star shape’
收 shou1	茶 cha2	山 shan1	池 chi2	手銬	shou3kao4	‘handcuff’
單 dan1	台 tai2	督 du1	圖 tu2	蛋糕	dan4gao1	‘cake’
傻 sha3	闢 pi4	手 shou3	泡 pao4	鯊魚	sha1yu2	‘shark’
含 han2	醒 xing3	湖 hu2	許 xu3	漢堡	han4bao3	‘hamburger’
北 bei3	乎 hu1	卜 bu3	哈 ha1	貝殼	bei4ke2	‘shell’
夫 fu1	閘 zha2	飛 fei1	折 zhe2	斧頭	fu3tou2	‘ax’
及 ji2	固 gu4	局 ju2	告 gao4	雞腿	ji1tui3	‘drumstick’
入 ru4	耐 nai4	讓 rang4	念 nian4	乳牛	ru3niu2	‘cow’
茶 cha2	妹 mei4	池 chi2	目 mu4	叉子	cha1zi	‘fork’
才 cai2	辣 la4	層 ceng2	路 lu4	彩虹	cai3hong2	‘rainbow’
起 qi3	堂 tang2	取 qu3	頭 tou2	氣球	qi4qiu2	‘balloon’
罪 zui4	收 shou1	奏 zou4	山 shan1	嘴巴	zui3ba1	‘mouth’
闢 pi4	組 zu3	泡 pao4	早 zao3	皮帶	pi2dai4	‘belt’
陣 zhen4	牢 lao2	住 zhu4	梨 li2	針筒	zhen1tong3	‘needle’
七 qi1	咖 ka1	區 qu1	哭 ku1	企鵝	qi4e2	‘penguin’
母 mu3	才 cai2	米 mi3	層 ceng2	木馬	mu4ma3	‘rocking-horse’
奉 feng4	岔 cha4	犯 fan4	臭 chou4	風車	feng1che1	‘windmill’
閘 zha2	北 bei3	折 zhe2	卜 bu3	炸彈	zha4dan4	‘bomb’
乎 hu1	把 ba3	哈 ha1	比 bi3	蝴蝶	hu2die2	‘butterfly’
酸 suan1	鳥 niao3	鬆 song1	女 nu3	算盤	suan4pan2	‘abacus’
計 ji4	港 gang3	句 ju4	狗 gou3	吉他	ji2ta1	‘guitar ’
舖 pu4	底 di3 (exp1) 罪 zui4 (exp2)	派 pai4	等 deng3 (exp1) 奏 zou4 (exp2)	葡萄	pu2tao2	‘grape’
妹 mei4	傻 sha3	目 mu4	手 shou3	玫瑰	mei2gui1	‘rose’
操 cao1	奉 feng4	粗 cu1	犯 fan4	草莓	cao3mei2	‘strawberry’
固 gu4	騷 sao1	告 gao4	森 sen1	骨頭	gu3tou2	‘bone’
辣 la4	及 ji2	路 lu4	局 ju2	喇叭	la3ba1	‘trumpet’
惹 re3	單 dan1	染 ran3	督 du1	熱狗	re4gou3	‘hotdog’
牢 lao2	夫 fu1	梨 li2	飛 fei1	老鷹	lao3ying1	‘eagle’
系 xi4	可 ke3	序 xu4	口 kou3	犀牛	xi1niu2	‘rhinoceros ’
可 ke3	系 xi4	口 kou3	序 xu4	蝌蚪	ke1dou3	‘tadpole’
等 deng3	罪 zui4	底 di3	奏 zou4	燈泡	deng1pao4	‘light bulb’