Phonological Preparation in Korean: Phoneme, or Syllable or Another Unit?

Abstract

With three experiments, the present study investigated the primary phonological preparation unit in spoken word production in Korean. Adopting the form preparation paradigm, 23 native Korean speakers named pictures in homogeneous or heterogeneous lists. In homogeneous lists, the names of the pictures shared the same initial phoneme (Experiment 1), initial consonant + vowel (i.e., CV) body (Experiment 2), or initial CVC syllable (Experiment 3); in heterogeneous lists, the names did not share any phonological components systematically. Compared to naming pictures in heterogeneous lists, participants’ naming speed was significantly faster when the initial body or the initial syllable of target names was shared. However, this form preparation effect was not shown in Experiment 1, when only the initial phoneme was shared. These results suggested that the body serves as the primary phonological preparation unit in Korean, that is, native Korean speakers plan spoken words in a body-coda fashion, probably due to a joint contribution from the strong prevalence of the CV structure and early literacy instructional approach.

Introduction

Spoken word production requires a series of processing steps, including concept encoding, lexical retrieval, mapping the target lexical item to its phonological representation, selecting the correct sounds, and articulation (Ferreira, 2010). At the stage of phonological retrieval, the phoneme plays a fundamental role in Indo-European languages such as Dutch and English (Meyer, 1990, 1991; O’Seaghdha et al., 2010; Schiller, 1998, 2000). Phonemes of target words are retrieved in a parallel fashion, then linearized in a syllabified organization that guides articulation (Levelt et al., 1999). In other languages, however, speakers may not retrieve phonological information in a phonemic fashion. For example, in Chinese speakers retrieve phonological information in a syllabic fashion (Chen et al., 2002; Chen & Chen, 2013; O’Seaghdha et al. 2010; You et al., 2012), and in Japanese speakers plan phonological information based on moras (Kureta et al., 2006; Verdonschot et al., 2011, 2015, 2019). Therefore, the phoneme, syllable, and mora have been considered as the primary phonological preparation unit (or proximate unit/functional phonological unit in some literature) in spoken word production in English, Chinese, and Japanese, respectively. This cross-language difference could be explained by multiple reasons, such as phonological features, orthographic features, and speakers’ language learning approach. The present study aimed to further investigate the primary factors among various possibilities by examining the primary phonological preparation unit in Korean, a language that has been less studied in this line of research.

The primary phonological preparation unit of a language has typically been examined with the form preparation paradigm, or implicit priming paradigm (Chen et al., 2002; Chen & Chen, 2013; Kureta et al., 2006, 2014; Li et al., 2015, 2017, 2018; Meyer, 1990, 1991; O’Seaghdha et al., 2010). The task may involve an associative-learning session followed by a naming session, or only involve a picture naming session. For the former, participants memorize some prompt-response word pairs (e.g., night-day, wet-dew, and bread-dough; examples adopted from O’Seaghdha et al., 2010). In the following associate-naming session, they see prompts and produce the corresponding response words as soon as possible. For the latter, participants name pictures that are presented one at a time. For both associate-naming and picture naming tasks, the critical manipulation is that the target response words always share the same initial phonological component (e.g., the same initial phoneme or onset, like in day, dew, dough) in homogeneous lists, while in heterogeneous lists the response words do not share any phonological components systematically (e.g., say, dew, and bow). Speakers’ response time of homogeneous lists may be faster than heterogeneous lists as a result of the fore knowledge of the shared initial phonological components, a difference referred as the form preparation effect or implicit priming effect (i.e., speakers are able to prepare the initial phonological component before the stimulus is presented). The smallest shared phonological components that facilitate speakers’ naming speed would be the primary phonological preparation unit of the language by default. For example, native English and Dutch speakers benefited from the shared initial phoneme (e.g., Meyer, 1990, 1991; O’Seaghdha et al., 2010), while native Chinese speakers instead only benefited from the shared initial atonal syllable (O’Seaghdha et al., 2010). Note that these results do not suggest that other phonological units do not play an important role in language production — they may play a secondary role. For example, studies that used event-related potentials (ERPs) suggested Chinese speakers are also sensitive to shared initial phoneme when producing words or phrases, though this sensitivity was not revealed in behavioral data (Qu et al., 2012, 2020). The primary phonological preparation unit and other units may even be processed in a similar time course, as in the case when shared phonemes and syllables elicited ERP components in similar time windows (230–300ms for phonemes and 200–280ms for syllables).

The cross-language differences in the primary phonological preparation unit may be attributed to multiple factors, and the primary one might be the features of the phonological system in each language. Roelofs (2015) proposed that one important feature is the word-shape frame of each language. In phonological encoding, phonological units and the word-shape frame are activated in a parallel fashion, followed by a serial unit-to-frame association (see the WEAVER++ model, Levelt et al., 1999). In Chinese each syllable carries its own lexical tone¹ and in Japanese each mora carries a pitch (high vs. low). As a result, phonological encoding may be more efficient for Chinese and Japanese speakers if they retrieve the stored phonological units that are directly associated to the corresponding tonal frame (i.e., atonal syllable ~ tone; mora ~ pitch). In contrast, in atonal languages like English and Dutch, stress functions as the metrical frame (i.e., an absence of tonal frame). Although stress patterns are associated with syllables instead of phonemes, how to assign stress would only be a concern in multisyllabic words. That is, unlike Chinese and Japanese in which a single syllable/mora may be associated with any one out of the four lexical tones/out of the two pitches, a single syllable (e.g., a monosyllabic word) is always a stressed syllable. In addition, syllabification is required before associating syllables to stress, and given that re-syllabification (another important feature in phonological encoding) is common in English (see Vroomen & de Gelder, 1999), it may be less efficient to plan sounds in syllables than in phonemes then syllabify them. Examples of re-syllabification are as follows: the past tense of the word pat is pronounced as pa/tted instead of patt/ed; the phrase escort us is produced as es/cor/tus instead of es/cort/us in connected speech (Levelt et al., 1999). In contrast, re-syllabification never occurs in Chinese, making it easy to retrieve sounds in syllabic chunks.

Orthographic features can affect the primary phonological preparation unit temporarily, though orthographic knowledge is not required in language production. Although the syllable is the primary phonological preparation unit in Chinese by default, when native Chinese speakers were instructed to memorize words written in Pinyin, a writing system that uses Roman alphabets to transcribe the pronunciation of Chinese characters (e.g., ma1 for 妈), they benefited from the shared initial phoneme (Li et al., 2015). However, this benefit disappeared if the same stimuli were presented in Chinese characters. This might be due to the strict one-to-one letter-sound correspondence in Pinyin, which encourages speakers to encode words in a phoneme-by-phoneme fashion. In contrast, each Chinese character represents a syllable and does not represent each phonemic unit transparently. Similarly, Japanese speakers also showed significant phoneme facilitation when memorizing words in Romaji, a phonetic system to write Japanese using the Latin alphabet (Kureta et al., 2014). Explicit orthographic information may temporarily change speakers’ phonological preparation unit in a task, while orthographic experience may exert a longer influence. For example, still in the form preparation paradigm, 7-year-old native Chinese-speaking children benefited from the fore knowledge of the shared onset of utterances in picture naming, an effect that was absent in older children and adults (Li & Wang, 2017). This seems to be relevant to their extensive exposure to Pinyin throughout the first-grade academic year. It is unclear whether the influence is due to the alphabetic nature of Pinyin symbols, or because the children are instructed to read Pinyin in an onset-rime fashion during the classroom instruction, or both. Either way, the results suggested that orthographic experience may have a relatively longer-term influence on speakers’ primary phonological preparation unit. Likewise, Inagaki et al. (2000) showed that orthographic experience affected the way Japanese children segmented auditory words – they found that word segmentation shifted from being a mixture of syllable- and mora-based to being predominantly mora-based as children acquired kana orthography, which is a mora-based writing system. For Pinyin in Chinese and kana letters in Japanese, the instructional approach clearly matched the orthographic features of these scripts (i.e., onset-rime-based that emphasizes the initial phoneme and mora-based respectively). However, for Korean children, the instruction of Hangul emphasizes body-coda units more than phonemes or syllables, allowing us to differentiate the effect of instructional/learning approach versus orthographic features (see more discussion later).

Few studies examined the primary phonological preparation unit in Korean. Korean has a relatively simple phonological system compared to English. Unlike English, Korean does not have consonant clusters (Wang et al., 2017).² In addition, each syllable is equally stressed in multisyllabic words, and there are clear syllabic boundaries in language production in Korean (Lee & Hahn, 1996). As an atonal language, unlike Chinese or Japanese, no lexical tone or pitch is available in Korean.³ However, re-syllabification exists in Korean, just like in English. For example, the sound of the final consonant of a first syllable moves to the first sound of a second syllable if the second syllable starts with a vowel (e.g., 군인/gun.in/, meaning soldier, is be pronounced as 구닌 /gu.nin/).

In terms of the Korean writing system, Hangul is alphabetic syllabary, and syllable is its distinctive unit (Pae, 2011). There are clear syllable boundaries among syllables in a Hangul word, similar to the boundary between Chinese characters (see the above example 군인). Each Hangul syllable block is built from two to four letters that represent phonemes (e.g., 발 /pal/, meaning foot, consists of three phonemic symbols on the upper left ㅂ/p/, the upper right ㅏ/a/, and the bottom ㄹ/l/), and there is consistent letter-to-phoneme correspondence in this shallow orthography.

With all the aforementioned linguistic features, either the syllable or the phoneme might be the primary phonological preparation unit in Korean. The absence of tonal frame and presence of re-syllabification would make the phoneme a more robust preparation unit, while the clear syllabic boundary in spoken words and simple syllable structure may make the syllable a more salient, cohesive unit for preparation in Korean. If orthographic features play a role, the transparent mapping between letters and phonemes in each syllable block may encourage phonemic encoding, whereas the clear visual syllabic boundary may encourage syllabic encoding. Therefore, the investigation of primary phonological preparation unit in Korean would allow us to get a better understanding about what factors weigh more in determining the grain size of phonological processing in language production.

Kim and Davis (2002) may be the first study to report relevant data. In a masked priming naming task, native Korean speakers read Korean monosyllables aloud, with a prime that shared some phonological components or nothing systematically with the target. While participants named the targets significantly faster when the prime shared the CV body or the whole CVC syllable with the target, the priming effect was only marginally significant (p = .06) when a prime shared the initial phoneme (onset) with the target. In contrast, using the same paradigm, two other studies showed significant priming effects when a nonword prime shared the same initial phoneme with the bi-syllabic Korean target nonword (e.g., 댄소/tæn.so/ - 독가/tok.ka/) (Han & Verdonschot, 2019; Witzel et al., 2013). Han and Verdonschot (2019) also showed a significant phoneme onset effect in a phonological Stroop task, in which participants named the ink color of Hangul nonwords. Korean speakers’ color naming speed was significantly improved if the nonword shared the initial phoneme with the name of the target color (e.g., 낙문/nak.mun/ - 노랑 /no.laŋ/, meaning yellow). While these studies seemed to suggest that the phoneme plays an important role in spoken word preparation (although the phoneme effect in Kim and Davis (2002) was not as robust as that in other studies), Hangul was presented at least as a part of the stimuli, thus involving some bottom-up processing. In masked priming naming tasks, all responses were initiated by orthographic decoding. However, traditionally, we consider language production a top-down process that is driven by concepts. In the phonological Stroop task, although color naming was driven by the color concept, the presence of Hangul symbols (as distractors) still involve some bottom-up processing, thus may still affect phonological processing in the task (see Li & Wang, 2015).

Choi, Oh, and Han (2017) showed some evidence that the phonological preparation pattern differs among spoken responses that are initiated by top-down vs. bottom-up processing. In an associate-naming task with the form preparation paradigm, Korean speakers did not benefit from the shared initial phoneme; however, in a word reading task with the form preparation paradigm (participants read aloud a word in each trial), Korean speakers’ reading speed was significantly improved when words in a list shared the initial phoneme. When response words shared the initial syllable, both tasks showed a robust form preparation effect. The researchers suggested that the cross-task contrast in the phoneme condition might be a result of different attention allocation in low vs. high demanding tasks (O’Seaghdha & Frazer, 2014). Word reading is more automatic than associate-naming, thus allowing speakers to allocate attention to phonemic units. In contrast, it is difficult to do so in a task that requires speakers to recall a different word when seeing a prompt.

In a less demanding picture naming (rather than associate-naming) task which does not involve a memorization session, again using the form preparation paradigm, Han and Choi (2016) found that Korean speakers showed a significant form preparation effect of syllable and also a marginally significant effect of phoneme. More shared phonological components may lead to larger form preparation effects, probably a result of sequential assembly of phonological components (Meyer, 1991; also see the Segmental Overlap Hypothesis for similar argument in Schiller 1998, 2000). Therefore, it is not surprising that the form preparation effect is larger in the syllable condition than the phoneme condition in Korean. However, the marginally significant effect in the phoneme condition might be a repetition rather than preparation effect given the absence of such effect for the first several trials in that study. Therefore, the aforementioned two studies that adopted the form preparation paradigm seemed to jointly suggest that the syllable could be the primary phonological preparation unit in Korean language production.

Other than the phoneme and the syllable, a third possibility we considered was that Korean speakers may prepare spoken words in a body-coda fashion, given that the body-coda unit plays an important role in Korean (e.g., Cho et al., 2008; Kim, 2007; Kim & Davis, 2002; Yoon et al., 2002). Within a syllable, the body-coda boundary (e.g., ca/t) is more salient than onset-rime boundary (e.g., c/at) for Korean children, and Korean children’s body-coda awareness is an important predictor of word reading in Korean (Kim, 2007). Similarly, Korean kindergarteners read nonwords more accurately when clue words and the target nonwords shared the same consonant + vowel (i.e., CV) body than when they shared the same vowel + consonant (i.e., VC) rime (Yoon et al., 2002), and also completed coda deletion tasks better than onset deletion (Cho et al., 2008). The body-coda preference was also shown in written word recognition among Korean adults in masked primed lexical decision tasks, that overlapping body components elicited a significant priming effect (Kim & Bolger, 2016). The preference of the CV body unit might be due to the prevalence of CV syllables in Korean — a corpus study showed that out of all monosyllabic Korean words, more than 80% have a CV structure (Kim, 2007). Another possibility is that Korean children acquire CV knowledge early, that is, the language approach. Cho (2009) showed that Korean children develop knowledge about the CV body first, then individual letter names, and finally letter sounds. This may encourage Korean speakers to process phonological information in a body-coda fashion in future life. Consistently, Cho (2009) also found that the CV body knowledge explained unique variance of Korean children’s reading performance longitudinally, while letter name and letter sound knowledge (i.e., phonemic level knowledge) did not provide unique contribution.

Few previous studies examined the possibility that the CV body served as the primary phonological preparation unit in Korean. Choi et al. (2017) only included the shared initial phoneme and initial CVC syllable conditions, but not a shared initial CV body condition. In Han and Choi (2016), open syllables were used, making it difficult to differentiate the syllable vs. body effects. In a recent study, Verdonschot and colleagues (2020) showed some evidence about body-coda phonological encoding in spoken word production in Korean. Adopting the picture-word interference (PWI) paradigm, Korean speakers’ picture naming latency was shown to be significantly faster when the written pseudoword (i.e., distractor) shared the same initial syllable with the name of the target picture, compared to the condition that the distractor and the picture name did not share any phonological unit. However, no facilitation was found when the distractor only shared the same initial phoneme with the picture name. Importantly, if the target picture name started with a CVC syllable (e.g., 국수/kuk.su/, meaning noodle), a significant facilitation effect was found when the distractor started with the same CV (e.g., 구툴 /ku.t^hul/). This seemed to suggest that participants encode phonological information in a body-coda fashion. However, if the target picture name started with a CV syllable (e.g., 자석/ca.sʌk/, meaning magnet), no significant facilitation effect was shown when the distractor started with a CVC syllable that shared the same CV (e.g., 잠진 /cam.cin/). Therefore, the researchers concluded that the syllable was the proximate unit (i.e., the primary phonological preparation unit) in Korean.

The present study investigated the primary phonological preparation unit in Korean among three possibilities: phoneme, body, or syllable. To our best knowledge, the present study is one of the first that took CV body into consideration in the form preparation paradigm. The investigation would shed some light on the relative weight of the different factors in phonological preparation pattern, including phonological features, orthographic/visual features, language learning approach, and the strong prevalence of certain phonological units. While the first two prefer phonemic or syllabic phonological preparation, depending on the specific features (e.g., re-syllabification and syllabic boundary), the latter two prefer a body-coda fashion phonological preparation pattern. To avoid the potential influence from bottom-up processing (i.e., orthographic information) that is not typically involved in language production and confounds driven by the memorization process in associative-naming (O’Seaghdha & Frazer, 2014), we used the picture naming task with the form preparation paradigm. Native Korean speakers would name pictures, and the critical manipulation was that in homogeneous lists the target response words shared the same initial phoneme, initial body, or initial syllable. The smallest unit that elicits a form preparation effect should be the primary phonological preparation unit in Korean.

Experiment 1: Phoneme as the Preparation Unit

In Experiment 1 we examined whether the phoneme serves as the phonological preparation unit among native Korean speakers.

Method

Participants

A total of 23 native Korean speakers (10 males, M age = 26.8, SD = 5.2) were recruited from University of Maryland, College Park for cash compensation. All participants completed all the three experiments in one session with the order counterbalanced across participants. It took about one hour for each participant to complete all experiments.

Materials & Design

In the form preparation task, participants were presented with three sets of pictures. The names of all pictures were disyllabic, with the initial phoneme /b/, /g/, and /n/ for the first syllable, which was always in the CVC structure. See Table 1 for the full stimuli. As shown in Table 1, three homogeneous lists (words sharing the initial phoneme, e.g., 백조 (baek jo, swan), 반지 (ban ji, ring), 봉투 (bong tu, envelope)) and three heterogeneous lists (words sharing nothing systematically, e.g., 반지 (ban ji, ring), 낙타 (nak ta, camel), 김치 (gim chi, kimchi)) were created. Each list consisted of three unique pictures that repeated 12 times, which led to 36 trials per list. In each list, trials were pseudorandomized so that the same picture did not appear twice in a row (i.e., to avoid repetition priming). The order to finish the six lists rotated across participants. Half of the participants completed the homogeneous lists first, while the other half completed the heterogeneous lists first.

Table 1.

Stimuli used in Experiments 1, 2, and 3

	homogeneous lists	heterogeneous lists
Experiment 1 (shared initial onset)	백조 (baek jo, swan) 반지 (ban ji, ring) 봉투 (bong tu, envelope )	반지 (ban ji, ring) 낙타 (nak ta, camel) 김치 (gim chi, kimchi)
	늑대 (neuk dae, wolf) 낙타 (nak ta, camel) 농구 (nong gu, basketball)	백조 (baek jo, swan), 농구 (nong gu, basketball 경찰 (gyeong chal, police)
	김치 (gim chi, kimchi) 경찰 (gyeong chal, police) 감자 (gam ja, potato)	봉투 (bong tu, envelope) 늑대 (neuk dae, wolf) 감자 (gam ja, potato)
Experiment 2 (shared initial body)	달력 (dal yeok, calendar) 담배 (dam bae, cigarette) 당근 (dang geun, carrot)	달력 (dal yeok, calendar) 잡채 (jap chae, vegetable stir-fried noodles) 심장 (shim jang, heart)
	잔디 (jan di, grass) 잡채 (jap chae, vegetable stir-fried noodles) 작두 (jak du, straw cutter)	담배 (dam bae, cigarette) 잔디 (jan di, grass) 신문 (shin mun, newspaper)
	신문 (shin mun, newspaper) 심장 (shim jang, heart) 식탁 (shik tak, dining table)	당근 (dang geun, carrot) 작두 (jak du, straw cutter) 식탁 (shik tak, dining table)
Experiment 3 (shared initial syllable)	공주 (gong ju, princess) 공룡 (gong ryong, dinosaur) 공사 (gong sa, construction)	공주 (gong ju, princess) 선물 (seon mul, present) 장갑 (jang gab, gloves)
	선반 (seon ban, shelf) 선물 (seon mul, present) 선녀 (seon nyeo, fairy)	공사 (gong sa, construction) 선녀 (seon nyeo, fairy) 장독 (jang dok, Korean jar)
	장미 (jang mi, rose) 장독 (jang dok, Korean jar) 장갑 (jang gab, gloves)	공사 (gong sa, construction) 선반 (seon ban, shelf) 장미 (jang mi, rose)

Procedure

First, each participant was presented with nine flash cards, each of which had one target picture on it, and was instructed to use a disyllabic word to name each picture. They were corrected if the names were different from our target name. A practice session followed, in which each picture was presented twice in a randomized order (i.e., 18 trials in total). Each trial began with a 200 ms, 1,000 HZ beep and an 800 ms cross fixation (“+”) presented at the center of the computer screen, which was then replaced by the picture that stayed for 2,000 ms or until a response was provided. Participants were instructed to name each picture as quickly and accurately as possible. In the testing session, all participants completed all the six lists. Before each list, participants were informed which three pictures would be presented. For each list, each trial was presented in the same way in the practice session.

Results and Discussion

Analyses were carried out in R, an open source programming environment for statistical computing (R Core Team (2013) with the lme4 package (Bates et al., 2015)) for linear mixed effects modeling (LMM) and general linear mixed effects modeling (GLMM). For all the pictures, reaction times (RTs) data for incorrect responses were excluded. Correct RTs were trimmed if any one or more of the following conditions were met: hesitation, disfluency, the correct answer failed to trigger the voice key, or noise or the utterance of the previous word triggered the voice key. Reaction times that were 2.5 standard deviations larger or smaller than the means for each participant were also removed from analyses. The remaining RT data were log-transformed and then submitted for analyses. In both RT and error rate analyses, contrast-coded fixed effects included context (homogeneous vs. heterogeneous), sequence (homogeneous lists first vs. heterogeneous lists first), centered trial number of each list, and all of the two-way and three-way interactions. Participant and picture were entered as two random intercepts with related random slopes. Random slopes and picture random intercept were removed for error rate analyses due to the failure to converge. The significance of each fixed effect was assessed via likelihood ratio tests (Barr et al., 2013). Figure 1 plots participants’ mean RTs (Panel A) and mean error rates (Panel B) in each context.

Figure 1.

Mean picture naming reaction time (Panel A) and error rates (Panel B) of homogeneous vs. heterogeneous lists in Experiment 1 (shared initial onset in homogeneous lists), Experiment 2 (shared initial body in homogeneous lists), and Experiment 3 (shared initial syllable in homogeneous lists).

In the analysis of RTs, we removed 0.82% incorrect responses, and another 4.84% of trials due to trimming procedures. Figure 2 plots the mean RTs of each context and sequence across all trials. The most critical result was that participants named pictures in homogeneous and heterogeneous lists equally fast, an absence of the form preparation effect (M = 543 ms vs. 551 ms; β = −.012; SE β = .019; χ² (1) < 1). Although they named pictures faster in homogeneous than heterogeneous lists when homogenous lists were completed first (M = 533 ms vs. 557 ms; β = −.043; SE β = .017; χ² (1) = 6.049, p = .014), no difference was shown when heterogeneous lists were completed first (M = 557 ms vs. 544 ms; β = .017; SE β = .032; χ² (1) < 1), a significant interaction between context and sequence (β = −.060; SE β = .028; χ² (1) = 4.576, p = .032). In other words, whichever type of lists that were completed first were named more quickly, which suggested that participants did not show a reliable form preparation effect in this experiment. Results also showed a significant interaction between sequence and trial number (β = −.001; SE β = .0007; χ² (1) = 4.559, p = .033). While participants’ RTs became slower toward the end of a list when heterogeneous lists were completed first (β = −.001; SE β = .0006; χ² (1) = 5.060, p = .024), such effect was not shown for participants who completed homogeneous lists first (β = .0001; SE β = .0001; χ² (1) < 1). None of the other effects were significant (ps > .10).

Figure 2.

Mean picture naming response time in each condition across all trials in Experiment 1 (shared initial onset in homogeneous lists), Experiment 2 (shared initial body in homogeneous lists), and Experiment 3 (shared initial syllable in homogeneous lists).

Figure 3 shows the mean error rates in each context and sequence across all trials. Overall, participants produced more errors in homogeneous than heterogeneous lists (M = 4.86% vs. 0.82%; β = −1.015; SE β = .038; χ² (1) = 7.656, p = .006), and that they tended to produce fewer errors toward the end of a list, a marginally significant trial number main effect (β = .033; SE β = .018; χ² (1) = 3.354, p = .067). None of the other effects were significant (ps > .73). The higher error rates in homogeneous lists further supported the absence of form preparation effects when to-be-named items shared the same initial onset — at best, the numerical but not significant 8 ms difference was a result of speed-accuracy trade-off. The marginally significant trial number effect might reflect benefits from practice. In summary, there was no clear evidence that showed participants benefited from shared onset in spoken word production.

Figure 3.

Mean picture naming error rates in each condition across all trials in Experiment 1 (shared initial onset in homogeneous lists), Experiment 2 (shared initial body in homogeneous lists), and Experiment 3 (shared initial syllable in homogeneous lists).

Experiment 2: Body as the Preparation Unit

Method

Participants

Participants were same as those who participated in Experiment 1.

Materials, Design & Procedure

Similar to Experiment 1, participants were presented with three sets of pictures. The names of all pictures were disyllabic, with the initial body of /da/, /ja/, and /shi/, and the initial syllable was always in the CVC structure. See Table 1 for the full stimuli. As shown in Table 1, three homogeneous lists (words sharing the initial body, e.g., 신문 (shin mun, newspaper), 심장 (shim jang, heart), 식탁 (shik tak, dining table)) and three heterogeneous lists (words sharing nothing systematically, e.g., 당근 (dang geun, carrot), 작두 (jak du, straw cutter), 식탁 (shik tak, dining table)) were created. The randomization of pictures within each list and the counterbalance of lists across participants followed the same rules in Experiment 1. The procedure of Experiment 2 was also the same as that in Experiment 1.

Results & Discussion

The procedure of data cleaning and analyses was the same as that used in Experiment 1. Figure 1 plots participants’ mean RTs (Panel A) and mean error rates (Panel B) in each context. In the analysis of RTs, we removed 1.05% incorrect responses, and another 5.52% of trials due to trimming procedures. Figure 2 plots the mean RTs of each context and sequence across all trials. Critically, participants named pictures faster in homogeneous than heterogeneous lists, a significant form preparation effect (M = 516 ms vs. 529 ms; β = −.029; SE β = .012; χ² (1) = 6.24, p = .012). Participants also named pictures more slowly toward the end of a list (β = .001; SE β = .0006; χ² (1) = 4.432, p = .035), probably a fatigue effect. There was also a significant three-way interaction (β = .002; SE β = .001; χ² (1) = 4.313, p = .038). For participants who completed homogeneous lists first, form preparation effects tended to be smaller toward the end of a list (β = .002; SE β = .001; χ² (1) = 3.82, p = .057), while such effect was absent for participants who completed heterogeneous lists first (β = −.0005; SE β = .001; χ² (1) < 1). These results suggested that the form preparation effect was not a result of repetition – if anything, the effect was larger at the beginning of a list. Thus, the benefits in homogeneous lists were more likely to be from the fact that all items started with the same body. None of the other effects were significant (ps > .14).

Figure 3 shows the mean error rates in each context and sequence across all trials. The analysis of error rates did not show any significant results, except a marginally significant interaction between context and trial number (β = −.049; SE β = .028; χ² (1) = 2.910, p = .088) – error rates tended to be lower in homogeneous than heterogeneous lists at the beginning, a tendency that was absent toward the end of a list. Other than this, none of the other effects were significant (ps > .16). Thus, the results of Experiment 2 indicated significant form preparation effects when target words shared the same initial body.

Experiment 3: Syllable as the Preparation Unit

Method

Participants

Participants were same as those who participated in Experiments 1 and 2.

Materials, Design, & Procedure

Similar to Experiments 1 and 2, participants were presented with three sets of pictures. The names of all pictures were disyllabic, with the initial syllable of /gong/, /seon/, and /jang/. See Table 1 for the full stimuli. As shown in Table 1, three homogeneous lists (words sharing the initial syllable, e.g., 공주 (gong ju, princess), 공룡 (gong ryong, dinosaur), 공사 (gong sa, construction)) and three heterogeneous lists (words sharing nothing systematically, e.g., 공주 (gong ju, princess), 선물 (seon mul, present), 장갑 (jang gab, gloves)) were created. The randomization of pictures within each list and the counterbalance of lists across participants followed the same rules in Experiments 1 and 2. The procedure of the experiment was also the same as that in Experiments 1 and 2.

Results and Discussion

The procedure of data cleaning and analyses was the same as that used in Experiments 1 and 2. Figure 1 plots participants’ mean RTs (Panel A) and mean error rates (Panel B) in each context. In the analysis of RTs, we removed 0.64% incorrect responses, and another 6.30% of trials due to trimming procedures. Figure 2 plots the mean RTs of each context and sequence across all trials. Critically, participants named pictures faster in homogeneous than heterogeneous lists (M = 515 ms vs. 547 ms; β = −.069; SE β = .022; χ² (1) = 9.862, p = .002). Participants also named pictures more slowly toward the end of a list (β = .001; SE β = .0004; χ² (1) = 5.972, p = .014), probably a fatigue effect. None of the other effects were significant (ps > .14). Figure 3 shows the mean error rates in each context and sequence across all trials. Participants produced fewer errors in homogeneous than heterogeneous lists (M = 0.32% vs. 0.97%; β = 1.051; SE β = .454; χ² (1) = 5.342, p = .021). None of the other effects were significant (ps > .11). To summarize, the analyses of RTs and error rates consistently showed robust form preparation effects from overlapping initial syllable.

Cross-Experiment Comparison

We further conducted two cross-experiment analyses, between Experiments 1 and 2, and between Experiments 2 and 3, to examine whether context effects differ across experiments, focusing on the interaction between experiment and context. In both comparisons, the analyses included experiment (Experiment 1 vs. 2 or Experiment 2 vs. 3), context (homogeneous vs. heterogeneous), sequence (homogeneous first vs. heterogeneous first), centered trial number, and all the interactions. Participant and picture were entered as two random intercepts with related random slopes. Again, the RTs were log-transformed; random slopes and picture random intercept were removed for error rate analyses due to the failure to converge.

Experiments 1 vs. 2.

In the analysis of RTs, the reaction times were overall similar across Experiments 1 and 2 (M = 547 ms vs. 523 ms; β = .047; SE β = .030; χ² (1) = 2.442, p = .118), and the main effect of form preparation was marginally significant (M = 530 ms vs. 540 ms; β = .021; SE β = .012; χ² (1) = 3.102, p = .078). Recall that in separate analyses the form preparation effect was significant in Experiment 2 (with shared CV body) but was not significant in Experiment 1 (with shared initial phoneme), which explained why the main effect of form preparation was marginal in this new analysis. However, the interaction between experiment and context was not significant (β = .016; SE β = .022; χ² (1) < 1).

In the analysis of error rates though, the only significant result was the interaction between experiment and context (β = 1.355; SE β = .471; χ² (1) = 2.875, p = .004). This was because shared initial phoneme elicited more errors in Experiment 1 while no significant difference was found between homogeneous and heterogeneous lists in Experiment 2, as stated in separate analyses.

Experiments 2 vs. 3.

In the analysis of RTs, participants named pictures equally fast in Experiments 2 and 3 (M= 523 ms vs. 531 ms; β = .009; SE β = .029; χ² (1) < 1). Participants showed a significant form preparation effect (M= 515 ms vs. 538 ms; β = −.049; SE β = .014; χ² (1) = 13.026, p< .001), and tended to be slower toward the end of a list (β = .001; SE β = .004; χ² (1) = 7.752, p = .005). Of greatest interest, the form preparation effect was marginally significantly larger in Experiment 3 than Experiment 2 (M = 32 ms vs. 13ms; β = −.040; SE β = .055; χ² (1) = 3.203, p = .073), suggesting that more shared phonological components could bring more benefits, or that shared syllable could provide an extra boost in phonological preparation.

In the analysis of error rates, participants produced more errors in Experiment 2 than Experiment 3 (M = 1.05% vs. 0.64%; β = .681; SE β = .268; χ² (1) = 6.466, p = .011), and fewer errors in homogeneous than heterogeneous lists (M = 0.60% vs. 1.09%; β = .676; SE β = .267; χ² (1) = 6.412, p = .011). The interaction between experiment and context was not significant (β = 1.49; SE β = 1.07; χ² (1) = 1.962, p = .161).

General Discussion

The present study investigated the primary phonological preparation unit in native Korean speakers. Minimizing the influence of orthographic information (though orthographic knowledge or experience may still play a role), we found that native Korean speakers did not benefit from the shared the initial phoneme when producing concept-driven utterances. However, their picture naming speed was significantly improved with shared initial body and initial syllable. The benefit was not a simple repetition or practice effect, supported by the absence of context × sequence and context × trial number interactions. In addition, the benefit from the shared initial syllable was larger than that from the shared initial body, consistent with Meyer’s (1991) finding that simply sharing longer segments could lead to larger benefits. In other words, the assembly of phonological components is also sequential in Korean. These results suggested that native Korean speakers tend to prepare phonological units in a body-coda fashion by default in spoken word production, which is different from Germanic languages such as English and Dutch (phonological encoding in phonemes), from Chinese (in syllables), and also from Japanese (in mora).

The biggest concern of this body preference claim was that whether the absence of benefits from shared phoneme just reflected a small effect size. In addition, cross-experiment comparison did not show a significant interaction between Experiment (i.e., shared unit) and context (homogeneous vs. heterogeneous) in RTs either. However, we found shared initial phoneme elicited significant higher error rates in Experiment 1, while the numerical facilitation in RTs was not significant. In addition, comparison between Experiments 1 versus 2 on error rates did show a significant interaction between Experiment and Context, as shared CV body elicited facilitation in Experiment 2. Therefore, it is likely that the numerical phoneme effect in RTs was a result of speed-accuracy trade-off. Additionally, even just for RTs, the shared initial phoneme only elicited numerical benefits when homogeneous lists were completed first; when heterogeneous lists were completed first, it even elicited numerically slower responses. Therefore, the error rate results and the interaction between context and sequence in RTs jointly suggested that the absence of phoneme priming effect in the present study was less likely due to small effect size. More investigation with more sensitive measurement like ERP is needed in future research though, considering that non-primary phonological preparation units may still play a role but was not easy to be detected with behavioral methodology (see Qu et al., 2012).

The absence of form preparation effects in the phoneme condition seemed to be somewhat inconsistent with the marginally significant results in Han and Choi (2016). First, the null effect in the present study should not be due to a small sample size, as the two studies had similar number of participants (23 in the present study vs. 24 in Han and Choi, 2016; 9 unique pictures with 12 repetitions in both studies). Second, although it is unclear what may have driven the inconsistent results across studies, as introduced earlier, in Han and Choi (2016), the onset effect was only marginally significant and was not robust at the beginning of the testing session, thus might be a repetition rather than a form preparation effect. In the present study, even the numerical phoneme benefits were only showed in RTs and when the homogeneous lists were completed first. In error rates, the shared phoneme even showed an interference effect. Therefore, Korean speakers at least do not reliably prepare spoken words in phonemes. A significant phoneme effect had been shown in masked priming (Han & Verdonschot, 2019; Witzel et al., 2013) and phonological Stroop tasks (Han & Verdonschot, 2019), suggesting an important role of phonemes in phonological encoding. Verdonschot et al. (2020), on the other hand, suggested that the syllable plays the primary role in phonological encoding in PWI tasks. However, our results suggest that a third possibility, body-coda might be the primary phonological preparation unit in Korean (other units like phonemes may play a secondary role in concept-driven spoken word production).

The inconsistent findings in studies that adopted different paradigms may reflect a flexible phonological processing mechanism (also see Li et al., 2015; Li & Wang, 2017). This flexibility was also consistent with the findings in Lee and Goldrick (2008), which suggested that Korean speakers’ preference to the body-coda versus onset-rime structure for CVC syllables was modulated by statistical relationships between the onset, vowel, and coda. In addition, different tasks may also tap different linguistic information in language processing (also see Kinoshita & Norris, 2012 for similar argument in word recognition). In the present study we aimed to tap a general/default status of phonological preparation in Korean, which might not apply to all situations.

Then what about the syllable? The presence of re-syllabification in Korean may partially explain why a phonological unit smaller than a syllable would better function as the primary phonological preparation unit in Korean. In addition, there are other phonological processes that occur to coda consonants in Korean, such as assimilation based on the feature of the onset of the following syllable (e.g., 먹는다/muk.neun.da/, meaning eat, is pronounced as 멍는다 /muη.neun.da/), simplification for coda with consonant clusters (e.g., 닭 /dalk/ meaning chicken, is pronounced as /dak/), and palatalization when the following syllable starts with /i/ (e.g., 같이 /gat.i/, meaning together is pronounced as 가치 /ga.chi/; this example entails re-syllabification as well) (Wang et al., 2017). All of these phonological features associated with the initial syllable coda consonants are also likely to make it less reliable to retrieve a syllable (at least for CVC) as a primary unit in production preparation. But why body instead of phoneme is preferred? There is no tonal frame that directly maps to a particular phonological unit in Korean, so the word-shape frame feature does not seem to help explain this preference. Without lexical tone, pitch, or stress function as the word-shape frame, the process of unit-to-frame association might be absent in phonological encoding in Korean. The implicit influence of orthographic features on phonological processing/representation can hardly help explain this preference either. In masked priming read-aloud tasks in which orthographic information was explicitly presented, significant or at least marginally significant priming effects were shown when the prime and the target shared the same initial phoneme (Han & Verdonschot, 2019; Kim & Davis, 2002; Witzel et al., 2013). Therefore, if orthographic features play a role by any chance, it seems more likely that the phoneme would be preferred. Nevertheless, Verdonschot et al. (2020) suggested that Hangul may encourage processing in syllables (not a body-coda fashion) given its visually salient syllable block. Therefore, it seems that the visual saliency of syllable block of Hangul could not explain the preference of the body-coda structure well either.

The two accounts that address the saliency of CV body in Korean may explain the selection of body as the primary phonological preparation unit. The first one is the prevalence of the CV structure in Korean. Consonant clusters in a syllable are not common in Korean, and re-syllabification also increases the frequency of the CV structure in spoken word production (e.g., a disyllabic word with CVC-VC structure (군인/gun.in/, solider) that only has one CV unit would be changed to CV-CVC (구닌/gu.nin/) that has two CV units). As a result, it may be more efficient to prepare spoken words in a body-coda fashion. Note that re-syllabification also increases the frequency of the CV structure in English for the same reason, but the phoneme instead of the CV body functions as the primary phonological preparation unit. This is probably due to the more complicated syllable structure in English than Korean (e.g., words with CCVC and CCCVC structures like glad and spread are common in English but are not legal in Korean), decreasing the prevalence of the CV structure in English, at least when comparing with Korean. Second, the emphasis on CV body knowledge in early literacy instruction during childhood may also contribute to the preference of body as the primary phonological preparation unit. However, caution is needed when taking instructional approach into consideration here, as such influence may not last long enough to adulthood, as suggested in Li and Wang (2017) — Chinese children’s primary phonological preparation unit switch from onset-rime to syllables once they no longer receive extensive training on Pinyin. Therefore, we speculated that the preference of body-coda unit may be more likely due to the phonological features of Korean. Early learning experience may help as well, but future research is needed to study whether its influence is long-lasting.

In conclusion, the findings of the present study suggested that native Korean speakers tend to plan spoken words in a body-coda fashion by default in concept-driven spoken word production. This raises a question concerning the determining factors for the primary phonological preparation unit in spoken word production in different languages. Although the tonal frame account explains cross-linguistic differences for languages such as English, Chinese, and Japanese, it does not seem to help explain the preference of body-coda structure in Korean. We propose that two major factors may jointly contribute to the body-coda preference in spoken Korean word production most — the strong statistical prevalence of the CV structure and instructional/learning approach in childhood. These factors seem to weigh more in determining speakers’ primary phonological preparation unit compared to other factors including phonological features like resyllabification and visual/orthographic features like salient syllable block.