“Panda” or “Bear, cat”: Mandarin-speaking preschoolers use duration and pitch to distinguish compounds and lists

Abstract

Compounds (e.g., jellybeans) and list forms (e.g., jelly, beans) can be distinguished by the presence or absence of boundaries, marked by durational and pitch cues. Studies have shown that 5-year-olds learning English have acquired both cues for distinguishing compounds and lists. However, it is not clear how and when this ability is acquired by children speaking tonal languages, such as Mandarin. This study examined whether Mandarin-speaking preschoolers can use durational and pitch cues to distinguish compounds and lists and whether their productions are adult-like. Thirty-one 4-year-olds, 34 5-year-olds, 29 6-year-olds, and 43 adults participated in an elicited production experiment. Results showed that similar to English-speaking preschoolers, Mandarin-speaking preschoolers can use durational cues to mark boundaries, triggering appropriate pitch changes for distinguishing compounds and lists, though these were not fully adult-like, even in the oldest age group.

摘要

复合词(如“jellybeans”)与其对应的列表形式(如“jelly, beans”)可通过边界存在与否进行区分, 在韵律上使用时长和音高线索进行标记。以往的研究表明, 以英语为母语的 5 岁儿童已能利用这两种韵律线索区分复合词与列表结构。然而, 对于普通话等声调语言为母语的儿童何时获得这种能力, 学界尚未明确。本研究考察了汉语普通话学龄前儿童能否运用时长与音高线索区分复合词与列表结构, 以及其产出是否达到成人水平。实验招募 31 名 4 岁、34 名 5 岁、29 名 6 岁儿童及 43 名成人参与产出任务。结果显示, 与英语儿童相似, 普通话学龄前儿童能够使用时长线索标记边界, 并通过引发适当的音高变化来区分复合词与列表结构, 但即使最大年龄组的儿童, 其表现仍未完全达到成人水平。.

1. Introduction

Prosody can be used to chunk utterances into meaningful units by marking boundaries, a phenomenon known as prosodic phrasing (Keating, 2004; Ladd, 1986; Liberman & Prince, 1977; Selkirk, 1984). For example, English speakers usually insert a boundary within a list (e.g., jelly, beans) to differentiate it from a compound (e.g., jellybeans), using prosodic cues such as duration and pitch (Ladd, 1986; Wynne et al., 2018; Yuen et al., 2021). Previous studies have found that English-speaking preschoolers can use durational and pitch cues for phrasing by age 5 (Vogel & Raimy, 2002; Wells et al., 2004; Yoshida, 2007; Yuen et al., 2021). However, little is known about how prosodic phrasing is acquired by preschoolers speaking tonal languages, such as Mandarin, where pitch is used for both lexical meaning and prosodic phrasing simultaneously. Although compounding structures in English and Mandarin differ (Zhang et al., 2012), studies in Mandarin can provide cross-linguistic evidence regarding potential acoustic cues that children can use for boundary marking. Thus, the current study aims to investigate Mandarin-speaking preschoolers’ use of prosodic cues to contrast compounds and lists in their productions.

2. Prosodic phrasing in English

Different linguistic units, such as compounds and lists, can be disambiguated by the existence of a prosodic boundary, inserted in lists but not in compounds (Nespor & Vogel, 1986; Wells et al., 2004; Yuen et al., 2021). Durational cues, such as preboundary lengthening and pauses, mark the existence of a boundary. For example, the duration of jelly is longer in the list “jelly, beans …” when preceding a word boundary compared to that in the compound “jellybeans …” without a boundary. A pause is also inserted between jelly and beans in the list but not in the compound (De Pijper & Sanderman, 1994; Katz et al., 1996; Kentner & Féry, 2013; Krivokapić, 2007; Price et al., 1991; Turk & Shattuck-Hufnagel, 2007; Wagner, 2005; Yuen et al., 2021). Both preboundary lengthening and pauses are robust cues for marking boundaries and have been found across many (non-tonal) languages (Dutch: Cambier-Langeveld, 2000; French: Michelas & D’Imperio, 2012; German: Huttenlauch et al., 2021; Petrone et al., 2017).

English-learning children acquired these durational cues for boundary marking at around 4 to 5 years. For example, 5-year-olds can already produce preboundary lengthening and pauses to distinguish between compounds and lists in an adult-like manner, although they insert more pauses in compounds than adults (Dankovičová et al., 2004; Wells et al., 2004; Yoshida, 2007; Yuen et al., 2021. See Katz et al., 1996 for different findings). These findings show that English-speaking preschoolers can build mappings correctly between prosodic cues and phrasing. However, little is known about whether and when preschoolers speaking tonal languages acquire the prosodic cues required for boundary marking.

3. Prosodic phrasing in Mandarin

Like English, Mandarin also uses preboundary lengthening and pauses for boundary marking (Bu et al., 2022; Chen, 2022; Kuang et al., 2022; Shen, 1992; Wang et al., 2017). However, unlike English, Mandarin has four lexical tones with distinct pitch contours: a high-level tone (T1), a rising tone (T2), a low-dipping tone (T3), and a falling tone (T4). T1, T2, and T4 have simple contours while T3 has a complex contour (see Figure 1). These four lexical tones are associated with distinct lexical meanings (e.g., ma1 “mother,” ma2 “hemp,” ma3 “horse,” and ma4 “to scold”). The acoustic realization of lexical tones can be influenced by the existence of word boundaries (Kuang et al., 2022, 2023; Lin, 1999; Wang et al., 2017; Xu & Wang, 2009; Zhang, 2012). For example, the pitch range of dynamic tones is generally expanded when preceding a boundary (e.g., lists), whereas no pitch range expansion is expected when not preceding a boundary (e.g., compounds) (Xu et al., 2024; Zhang, 2012). For the level tone T1, instead of pitch range variation, its overall pitch height is higher when following a boundary, compared to the pitch preceding a boundary (Kuang et al., 2023; Wang et al., 2017). T3 is a special tone as it involves a complex phonological process, that is, tone sandhi,¹ in multisyllabic words (Chen, 2022; Chen & Yuan, 2007; Lai & Kuang, 2016; Shih, 1997; Speer et al., 1989; Zhang, 2022). T3 was not investigated further here since children only acquire this complex process after the preschool period, especially in multisyllabic words (Tang et al., 2019a).

Figure 1. Pitch contours (citation forms) of lexical tones in Mandarin.

To our knowledge, only a few studies have investigated the acquisition of prosodic cues for sentence-level boundary marking by Mandarin-speaking preschoolers. One study found that Mandarin-speaking 4- to 5-year-olds can use preboundary lengthening and pauses to mark sentence boundaries, albeit with longer duration and inconsistent insertion of pauses compared to adults (Liu et al., 2023). Another study found that children can also implement pitch range expansion when preceding a boundary between sentences (Yu et al., 2020). Around this age, children are also producing connected speech and tonal processes in disyllabic words (Tang et al., 2019a; Xu Rattanasone et al., 2018). Together, these findings suggest that by age 4 to 5, preschoolers can use both duration and pitch to mark sentence boundaries. However, it is yet unclear whether they can also use these cues to mark word boundaries. Word boundaries are critical to understanding speech as it differentiates a single item (i.e., compound) from multiple items (i.e., a list), essential for word recognition in connected speech (Beach et al., 1996; Frazier et al., 2006).

So far, there has been only one study exploring Mandarin-speaking preschoolers’ use of prosodic cues at the word level, focusing on the implementation of preboundary pitch range expansion (Xu et al., 2024). The results showed that Mandarin-speaking 6-year-olds can correctly implement preboundary pitch range expansion for T2 and T4, with adult-like productions. However, it remains unclear whether and when children can use durational cues to mark boundaries and when children’s use of durational and pitch cues for phrasing become adult-like. As the pitch modification for phrasing is superimposed on tonal implementations, Mandarin-speaking children might only be able to use pitch for prosodic phrasing after they have acquired lexical tones by age 3 (Hua & Dodd, 2000; Li & Thompson, 1977; Tang et al., 2019a; Xu Rattanasone et al., 2018). They might then only be able to use pitch in an adult-like way after their tonal productions become adult-like after age 5 (Xu Rattanasone et al., 2018). Furthermore, T1 is a level tone where pitch height is more critical than pitch range (Tupper et al., 2020; Zhang & Gu, 2023), which was not investigated by Xu et al. (2024). This study therefore investigates the productions by Mandarin-speaking 4- to 6-year-olds, examining both durational (i.e., preboundary lengthening and pause) and pitch cues (i.e., pitch range for T2/T4 and pitch height for T1).

4. The current study

This study addressed two research questions: First, we want to know whether Mandarin-speaking preschoolers can implement durational cues for word boundaries to disambiguate compounds (N1 + N1) and lists (N1, N2), triggering changes of pitch cues (preboundary pitch range expansion and pitch height changes). If so, we then also want to know whether preschoolers show adult-like prosodic productions. The following hypotheses were formulated to test these research questions.

4.1. Hypothesis 1 (H1): Preboundary lengthening

As preboundary lengthening is a robust cue for boundary marking in Mandarin and has been acquired by 5-year-olds for sentence-level boundary marking (Bu et al., 2022; Chen, 2022; Kuang et al., 2022; Shen, 1992; Wang et al., 2017), we expect that (a) Mandarin-speaking preschoolers should be able to use preboundary lengthening to distinguish compounds and lists (i.e., produce longer syllable durations for N1 in lists than compounds), but (b) might produce longer durations in both conditions compared to adults due to a slower speaking rate (Nip & Green, 2013; Yang et al., 2021).

4.2. Hypothesis 2 (H2): Pause

Based on Liu et al (2023) and Yuen et al. (2021), where Mandarin- and English-speaking 4- to 5-year-olds were found to insert pauses for boundary marking, we expected that Mandarin-speaking preschoolers would (a) insert pauses to disambiguate compounds and lists, with longer pause durations in lists than in compounds, but (b) might insert more pauses in compounds than adults and produce longer pauses.

4.3. Hypothesis 3 (H3): Pitch

Based on Xu et al. (2024), where 6-year-olds can expand the pitch range for marking word boundaries, and Tang et al. (2019a) and Xu Rattanasone et al. (2018) who provide acoustic evidence for the realization of tonal processes in disyllabic words by younger children, we expected that (a) Mandarin-speaking preschoolers should be able to use an expanded pitch range (for T2 and T4) and a higher pitch (for T1) in lists compared to compounds, though (b) their productions might not be like adults.

5. Methods

5.1. Participants

A total of 94 typically developing Mandarin-speaking preschoolers participated in this study, with 11 preschoolers subsequently excluded due to incomplete data (four 4-year-olds and seven 5-year-olds). The remaining 83 child participants were then divided into three age groups according to chronological age (see Table 1). Child participants were recruited from kindergartens in Hebei and Shenyang Province, where Standard Chinese is used in daily teaching activities. In China, children above the age of 6 can enrol in elementary school, but many are still in kindergarten as they are born later in the year and do not meet the age needed to enter elementary school when enrolment begins. Additionally, 43 adult native speakers of Mandarin (born in Beijing) were recruited from local universities as controls. None of the participants had any speech, hearing, or cognitive disorders. All received a gift for their participation. This study was conducted in accordance with the ethics protocol approved by Macquarie University’s Human Ethics Panel. Consent was provided by adult participants and by the principal of the kindergarten for child participants.

Table 1. Demographics of participants in each age group

5.2. Stimuli

A total of 16 picturable noun-noun items (N1-N2) were selected to form compounds (N1 + N2) and their related list forms (N1, N2). Of these, four items were used in the practice trials and twelve items in the test trials (see Supplementary Material 1, hereinafter, Supplementary Material S1). All items were selected from spontaneous speech produced by 1- to 6-year-olds in the following corpora: the Chang Corpus (Chang, 1998), the Tong Corpus (Deng & Yip, 2018), the Li/Zhou PeerTalk Corpus (Li & Zhou, 2008), the Zhou Corpus (Zhou, 2001), and the Zhou Narratives Corpus (Li & Zhou, 2011), all available as part of the CHILDES database (MacWhinney, 2000). These corpora consisted of spontaneous speech produced by preschoolers aged one to six. As T3 is acquired later and involves phonological process (cf. Tang et al., 2019a), only the lexical tones T1, T2, and T4 were used as N1.

Compounds and their related list forms were embedded into a carrier sentence “Zhe4-li3 you3…” (Here have …), followed by a filler noun (N3), constructing a two-item list in the compound condition (N1 + N2, N3) and a three-item list condition (N1, N2, N3). All items were presented on the screen using coloured clipart pictures.

5.3. Procedure

At the beginning of the test, a picture-naming task was conducted to familiarize all participants with the target words and their corresponding pictorial representations. Each picture depicting a set of items was presented on the screen one at a time, and the participants were asked to name all items shown in each picture (see example in Figure 2). The experimenter would provide the correct label if the participants did not produce any or the correct label. After the picture-naming task, four practice trials (two compound items and two list items) were presented to familiarize participants with the carrier sentence Zhe4-li3 you3… “Here have …” (without any conjunctions between N1 and N2). In instances where participants did insert a conjunction, for example, he “and,” during the training phase, they were provided further training.

Figure 2. Examples of compounds (left) and lists (right) in the testing phase.

During the testing phase, either two pictures eliciting compounds or three pictures eliciting lists were shown on the screen, and the experimenter asked “Can you tell me what these are?” Participants were instructed to answer the question using the provided carrier sentence to name objects from left to right. For example, if given the compound picture in Figure 2, participants were expected to say Zhe4-li3 you3 xiong2-mao1 he2 xi1-gua1 “Here have panda and watermelon,” and Zhe4-li3 you3 xiong2, mao1 he2 xi1-gua1 “Here have bear, cat, and watermelon” when given the list form. At this stage, if participants inserted conjunctions between N1 and N2, they were returned to the training phase until they stopped using conjunctions. Otherwise, the testing was discontinued after five attempts of the training (with 11 preschoolers not completing the study).

Compounds and lists were presented in a pseudo-randomized sequence, in which pictures of compounds (e.g., xiong2-mao1 “panda”) would never occur right after their corresponding list forms (e.g., xiong2, mao1 “bear, cat”) and vice versa. Two test versions were generated and counterbalanced across participants to avoid any list effects with no more than three compounds or lists shown successively. The productions were audio-recorded using a Marantz PM661 solid-state recorder and an AKG G520L head-worn microphone at a sampling rate of 44100 Hz. Each participant produced 24 target sentences, resulting in 1032 tokens from adults and 2256 tokens from preschoolers (744 tokens from 4-year-olds, 816 tokens from 5-year-olds, and 696 tokens from 6-year-olds).

5.4. Annotation and measurements

All recordings were annotated in Praat (Boersma & Weenink, 2024). We adopted the following criteria to define the onset: (a) when the consonant was a plosive or an affricate, the onset was defined as the occurrence of a release burst in the spectrum and the waveform; (b) for fricatives, the onset of a syllable was determined by the beginning of high energy noise in the spectrum and a clear frication noise on the waveform; (c) for laterals, nasals, or glides, the onset was identified by the beginning of higher formants in the spectrum and periodic waveform. Since pitch information in lexical tones is mainly located in the rhyme part of the syllable in Mandarin, the onset of the rhyme was defined as the beginning of the second formant (F2) and the beginning of the periodic waveform. All syllables were either open syllables or had coda nasals. Hence, the criterion for the offset was the end of high formants and the end of periodic waveform. In compounds, for syllables where the onset of the second syllable was a vowel or a nasal, the offset of the first syllable was defined as the transition of F2 and the point of change in the periodic waveform.

Acoustic parameters were extracted using a Praat script by Xiong (2017), including the syllable duration of N1, pause duration between N1 and N2, and pitch range and pitch height over N1. A criterion for the threshold of the presence or absence of a pause was based on Xu (1986), who reported that closure duration was 47 ms and 62 ms for aspirated/unaspirated plosives and 39 ms and 50 ms for aspirated/unaspirated affricates. Thus, the silent interval between N1 and N2 that exceeded the reference closure duration was coded as a pause. As acoustic parameters of plosives and affricates are stabilized by age 4, we adopted the same criterion to determine the presence or absence of a pause in preschoolers’ production (Peng & Chen, 2020). Pause duration was calculated by subtracting the temporal offset of N1 from the onset of N2. Pitch tracks of annotated rhyme parts were checked and manually revised to avoid pitch doubling or halving errors. We also analysed normalized syllable and pause duration to account for speaking rate differences between children and adults, yielding results consistent with the raw data. For the sake of readability and organization of this manuscript, all normalized durational data are included in Supplementary Materials. Duration was normalized for each speaker using the following formula (1). Original F0 values were extracted at 10 equidistant points in Hz by the default algorithm in Praat. To better match human perception, F0 values from observed Hertz were further converted to semitones with a reference of 40 Hz. Pitch range was then derived by subtracting the lowest pitch from the highest pitch. The pitch height of T1 was normalized for each speaker to eliminate any individual differences in pitch height, using the following formula (2).

(1) $$ \mathrm{Normalized}\ \mathrm{duration}=\frac{\mathrm{Observed}\ \mathrm{duration}-{\mathrm{Mean}\ \mathrm{duration}}_{\mathrm{Individual}}}{\mathrm{Standard}\ \mathrm{deviation}\;{\mathrm{of}\ \mathrm{duration}}_{\mathrm{Individual}}} $$

(2) $$ \mathrm{Normalized}\ \mathrm{pitch}\ \mathrm{height}=\frac{\mathrm{Observed}\ \mathrm{pitch}-{\mathrm{Mean}\ \mathrm{pitch}}_{\mathrm{Individual}}}{\mathrm{Standard}\ \mathrm{deviation}\;{\mathrm{of}\ \mathrm{pitch}}_{\mathrm{Individual}}} $$

5.5. Statistical analysis

A total of 984 tokens from adults and 1992 tokens from preschoolers (648 tokens from 4-year-olds, 648 tokens from 5-year-olds, and 696 tokens from 6-year-olds) were included for further statistical analysis. An additional 48 tokens from adults and 264 tokens from preschoolers (96 tokens from 4-year-olds and 168 tokens from 5-year-olds) were excluded due to the insertion of the conjunction hai3-you3 “and” between N1 and N2, the presence of creaky voice, and/or the presence of environmental noise. All data were processed and analysed in R (R Core Team, 2023). All models were fitted using the lme4 package (Bates et al., 2015). The syllable duration, pause duration, and pitch range were evaluated by linear mixed-effect models, and the occurrence of pause was evaluated by the generalized linear-mixed effect model with binomial distribution. To determine the model of best fit for each analysis, models with maximal random structures allowed by the data (without singularity or convergence issues) were selected based on a likelihood ratio test (Pinheiro & Bates, 2009). Statistical significance in all models was evaluated using the ANOVA function from the car package (Fox & Weisberg, 2019), providing omnibus main effects and interactions among factors by conducting F-tests to estimate p-values. Pairwise comparisons were performed using emmeans package with Tukey’s honestly significant difference (HSD) adjustment (Lenth, 2024).

6. Results

The analyses were reported in three sections, with results on preboundary lengthening presented first, followed by pause insertion and finally pitch range expansion and height.

6.1. Preboundary lengthening

This section tested H1a: whether Mandarin-speaking preschoolers produce longer syllable duration in lists than in compounds and use preboundary lengthening to contrast these two structures, and H1b: whether preschoolers’ productions are adult-like. A linear mixed-effect model was fitted with syllable duration as the dependent variable and Condition (Compound and List) and Age group (4-, 5-, and 6-year-olds and adults) as fixed effects. Subject and Item were entered as random intercepts and slopes. Figure 3 shows the mean duration of N1 in Compound and List for each age group.

Figure 3. Duration of N1 in Compound (N1 + N2) and List (N1, N2) with +/− 1SE (standard error) for each Age group.

The results (see Table 2) revealed significant main effects for Condition and Age and a significant interaction for “Condition $ \times $ Age.” See Supplementary Material S2 for parametric values of the model and Supplementary Material S3 for the results of the post hoc analysis. Results of normalized syllable duration can be found in Supplementary Material S4. Together, these results suggest that while all groups produced longer syllable duration in List (M = 579.25, SD = 175.52) than Compound (M = 377.43, SD = 118.18), all preschoolers (4-year-olds: M = 529.01, SD = 193.30; 5-year-olds: M = 556.21, SD = 197.87; 6-year-olds: M = 499.93, SD = 173.15) produced longer duration in List and Compound than adults (M = 381.23, SD = 109.59). The only difference between preschoolers was that 5-year-olds had a longer syllable duration in Compound than 6-year-olds.

Table 2. Statistical results of linear mixed-effect model in syllable duration. F-values, degree of freedom (df), and p-values are provided

Note: Significant results are in bold. R code for the model: lmer(Duration ~ Condition * Age group + (Condition | Subject) + (Condition | Item)). *** p < .001.

In sum, Mandarin-speaking preschoolers can use preboundary lengthening to contrast compounds and lists at age 4, though their productions are still not fully adult-like even by the age of 6, with a longer syllable duration in both compounds and lists than adults.

6.2. Pause

This section addressed H2a: whether Mandarin-speaking preschoolers can insert pauses in lists to distinguish from compounds, and H2b: whether pause insertion and duration might be adult-like. Two mixed-effect models were fitted first to examine the occurrence of pause and a second model to examine pause duration. Both models had Condition (Compound and List) and Age group (4-, 5-, and 6-year-olds and adults) as fixed effects and random intercept for Subjects in pause occurrence and random intercept and slope for Subject in pause duration.

The occurrence of pause is shown in Figure 4. The first model (results see Table 3) detected a significant main effect of Condition and Age group, but no significant interaction was found. See Supplementary Material S5 for estimated parameters of the model and Supplementary Material S6 for results on the post hoc analysis of Age group. The results showed that all groups were more likely to insert pauses in List (M = 95.23, SD = 10.42) than in Compound (M = 41.20, SD = 13.74), while all preschoolers (4-year-olds: M = 45.99, SD = 15.04; 5-year-olds: M = 45.06, SD = 12.92; 6-year-olds: M = 46.26, SD = 12.32) were all more likely to insert pauses in Compound than adults (M = 31.91, SD = 9.30).

Figure 4. The occurrence of pause in Compound and List by preschoolers (4-, 5-, and 6-year-olds) and adults, with +/−1SE (standard error).

Table 3. Statistical results of pause occurrence. $ {\unicode{x03C7}}^2 $ , degree of freedom (df), and p-values are provided

Note: Significant results are in bold. R Model: glmer(Pause occurrence ~ Condition × Age group + (1 | Subject), family = binomial). *** p < .001.

Figure 5 displays the pause duration in Compound and List in each Age group. The results (see Table 4) from the second model revealed a significant main effect of Condition and Age group, as well as a significant Condition $ \times $ Age group interaction. See Supplementary Material S7 for the estimated parameters of the model and Supplementary Material S8 for results on the post hoc analysis. Results of normalized pause duration can be found in Supplementary Material S9. Together, these results suggest that all groups produced a longer pause duration in List (M = 567.97, SD = 358.33) than in Compound (M = 118.04, SD = 71.03). Additionally, preschoolers produced (4-year-olds: M = 121.51, SD = 63.51; 5-year-olds: M = 152.90, SD = 108.08; 6-year-olds: M = 113.84, SD = 50.20) similar pause duration in Compound compared to adults (M = 86.62, SD = 19.05). In List, however, pause duration became more adult-like with age, where 4-year-olds (M = 688.46, SD = 419.86) and 5-year-olds (M = 657.88, SD = 468.00) produced longer pause duration compared to 6-year-olds and adults (M = 474.55, SD = 183.13), while 6-year-olds (M = 515.09, SD = 331.75) produced adult-like pause duration.

Figure 5. Pause duration in Compound and List by preschoolers (4-, 5-, and 6-year-olds) and adults, with +/−1 SE (standard error).

Table 4. Statistical results of pause duration. F-values, degree of freedom (df), and p-values are provided

Note: Significant results are in bold. R Model: lmer(Duration ~ Condition * Age group + (Condition | Subject)). ** p < .01, *** p < .001.

In sum, the results showed that Mandarin-speaking preschoolers can use pauses to contrast compounds and lists at age 4, but their productions were not fully adult-like even at age 6, with more pauses in compounds as compared to those of adults. However, there was a developmental trend for pause duration, decreasing with age and becoming adult-like by age 6.

6.3. Pitch

This section addressed H3a: whether Mandarin-speaking preschoolers use preboundary pitch range expansion preceding a boundary to disambiguate compounds and lists, as well as pitch height differences in T1, and H3b: whether their productions are adult-like. Two linear mixed effect models were constructed first to examine the use of preboundary pitch range expansion and a second model to investigate pitch height in T1. Both models had Condition (Compound and List) and Age group (4-, 5-, and 6-year-olds, and adults) as fixed effects. By-Subject and by-Item intercepts as well as the slope for Condition were entered as random effects. Tones (T1, T2, and T4) were also entered in the first model as a fixed effect. The mean pitch range of each tone in different age groups is shown in Figure 6.

Figure 6. Pitch range of N1 in Compound and List in T1/T2/T4 by Age group, with +/− 1 SE (standard error).

The results (see Table 5) revealed a significant main effect of Condition and Tone as well as a significant Condition $ \times $ Tone and Tone $ \times $ Age group interaction. No other significant main effects and interactions were detected. See Supplementary Material S10 for the parametric values of the model. For the results of the post hoc analysis, see Table 6 and Supplementary Material S11. These results showed that the pitch range of the two contour tones was significantly more expanded in List (T2 (M = 7.04, SD = 3.13) & T4 (M = 10.81, SD = 4.38)) than in Compound (T2 (M = 5.58, SD = 2.61) & T4 (M = 6.52, SD = 2.62)), whereas no significant pitch range difference was found over the level T1 in Compound (M = 2.49, SD = 1.15) compared to T1 in List (M = 2.45, SD = 1.29). Additionally, only T2 produced by 4-year-olds (M = 7.92, SD = 3.92) showed a significantly broader pitch range than that by adults (M = 4.89, SD = 1.63). No other significant pitch range differences were found in other groups and lexical tones. This suggests that Mandarin-speaking 5- and 6-year-olds’ productions are adult-like.

Table 5. Statistical results of pitch range of N1. F-values, degree of freedom (df), and p-values are provided

Note: Significant results are in bold. R Model: lmer(Pitch range ~ Condition * Tone * Age group + (1 | Subject) + (Condition | Item)). *** p < .001.

Table 6. Results of pairwise comparisons of Compound and List in different lexical tones

Note: Significant results are in bold. *** p < .001.

The mean pitch height in T1 can be found in Figure 7. See Supplementary Material S12 for the parametric values of the model. No significant main effects and interactions were detected. The results suggest that all preschoolers and adults did not produce T1 with significantly different pitch height in Compound versus List. There was also no significant pitch height difference among the groups (Table 7).

Figure 7. Pitch height in Compound and List in T1 by Age group, with +/− 1 SE (standard error).

Table 7. Statistical results of pitch height in T1. F-values, degree of freedom (df), and p-values are provided

Note: R Model: lmer(Normalized pitch height ~ Condition * Age group + (Condition | Subject) + (Condition | Item)).

Taken together, the results indicated that Mandarin-speaking preschoolers can use preboundary pitch range expansion in T2 and T4 with no difference in pitch height on T1 for these two structures, like adults. Furthermore, their productions of pitch range and pitch height were in general adult-like at age 4, though the pitch range on T2 produced by 4-year-olds was broader than adults.

7. Discussion

This study examined whether Mandarin-speaking preschoolers (4- to 6-year-olds) can produce prosodic cues, including preboundary lengthening, pause, preboundary pitch range expansion, and pitch height, to disambiguate compounds and lists, and if so, whether their productions are adult-like. Our findings support our hypotheses that Mandarin-speaking preschoolers, like their non-tonal language-speaking peers, can employ preboundary lengthening and insert pauses to disambiguate compounds and lists, but even 6-year-olds produce longer syllable duration and insert more pauses than adults. Additionally, our findings support our predictions regarding pitch range, showing that Mandarin-speaking preschoolers have mastered preboundary pitch range expansion for different tonal categories, disambiguating compounds and lists, where the older preschoolers were adult-like. Finally, neither preschoolers nor adults produced a pitch height difference for T1 between compounds and lists.

Our durational results extend previous findings from English (Yuen et al., 2021), suggesting that 4-year-olds speaking a tonal language (Mandarin) have also already acquired the use of durational cues such as preboundary lengthening and pause to contrast compounds and lists, producing adult-like patterns for word boundary marking. However, Mandarin-speaking preschoolers tended to produce longer syllable durations than adults, indicating that their productions are not yet fully adult-like.

In addition to preboundary lengthening, Mandarin-speaking preschoolers can also insert pauses to disambiguate compounds and lists although the duration of the pauses is not adult-like until 6 years. These observations are in line with the performance of English-speaking preschoolers (Yuen et al., 2021). In addition to developing articulatory functions, a longer pause by preschoolers in lists might also be due to the development of working memory and planning skills. Liu et al. (2023) showed a negative correlation between pause frequency and verbal working memory by Mandarin-speaking 4- to 5-year-olds. Preschoolers need more time to select the correct words and plan speech due to limited cognitive resources in working memory, sometimes resulting in speech disfluencies. Therefore, Mandarin-speaking preschoolers could be producing longer pauses than adults due to a slower overall speaking rate and longer speech planning time. Similar to Yoshida (2007) and Yuen et al. (2021), Mandarin-speaking preschoolers also inserted more pauses in compounds than adults, suggesting that they are less consistent in producing compounds as a single prosodic word.

Despite the potential challenge of simultaneously mapping pitch to both lexical and postlexical meanings, our results suggest that Mandarin-speaking preschoolers can implement appropriate preboundary pitch range expansion for dynamic tones (T2 and T4) to distinguish compounds and lists. However, neither children nor adults produce different pitch height for level T1 to distinguish compounds and lists (see below for further discussion on this point). Child productions are also adult-like with no significant group differences (except for T2 by 4-year-olds). This suggests that the non-adult-like lexical tone productions do not appear to impede preschoolers’ ability to implement preboundary pitch range expansion to convey postlexical meaning. Taken together with Tang et al.’s (2019a) report that Mandarin-speaking 3-year-olds have acquired different tonal categories (even though the acoustic realizations are not yet fully adult-like), our findings indicate that acquiring tonal categories early might be sufficient for preschoolers to simultaneously map pitch to word-level and sentence-level meanings. Hence, our results suggest that Mandarin-speaking preschoolers can map pitch to multiple levels of linguistic structure.

Our results regarding pitch range expansion are also consistent with previous studies where no obvious changes were found for the level tone T1, but pitch range expansion was detected in dynamic tones before word boundaries (Lin, 1999; Shi & Wang, 2014; Wang et al., 2017; Xu et al., 2024). Given that no pitch changes were found with the level tone, this suggests that pitch is not a robust cue across lexical tones for word boundary marking in Mandarin. This then has implications for the role of pitch and boundary marking for other tonal languages.

7.1. Limitations

Although our results suggest that Mandarin-speaking 4-year-olds can already use durational cues and pitch range expansion to distinguish compounds from lists, they employ a longer syllable duration and insert more pauses in compounds than adults. Future studies should include school-aged children to determine when their productions become adult-like.

While our results lend support to the claim that boundary cues trigger pitch range expansion in dynamic tones (Xu et al., 2024), we could not test this in the current study. Previous studies have shown cue-weighting among these prosodic cues for boundary marking, where pause insertion is more robust than preboundary lengthening (e.g., Zhang, 2012). Other studies have also pointed to the high correlation between prosodic cues and boundary strength in adults, such as longer pauses occurring with higher boundary strength (Cho, 2016; Frota, 2000; Horne et al., 1995; Kentner et al., 2023). These issues are worth re-examining in light of our findings.

8. Conclusion

This study is the first attempt to investigate whether Mandarin-speaking preschoolers can produce prosodic cues for word boundary marking to contrast compounds and lists and, if so, whether their productions are acoustically like those of adults. The results show that similar to their non-tonal language-speaking peers, Mandarin-speaking preschoolers can use durational cues by age 4, though they are not adult-like, producing longer durations and inserting more pauses in compounds before age 6. We also found that pitch information does not appear to be a reliable cue for marking boundaries in Mandarin, raising questions about the multiple roles of pitch in other tonal languages. Finally, this study provides a baseline for examining the acquisition of prosodic cues by atypical populations, such as preschoolers with cochlear implants who do not receive adequate pitch information due to device use, showing delays in acquiring lexical tones (Tang et al., 2019b). It also raises questions regarding whether Mandarin-speaking preschoolers can perceive these prosodic cues to guide their identification and understanding of compounds versus lists.