Toward a comprehensive framework for tonal analysis: Yangru tone in Southern Min

2.1 Zhangzhou speech

Zhangzhou is a prefecture-level city situated in the southern part of Fujian Province in mainland China at longitude 116°54′0″ to 118°08′0″ east and latitude 23°34′0″N to 25°15′0″ north. Its east is opposite to the Taiwan Strait, while its southwest borders Chaozhou, a prefectural city in Guangzhou Province. The city covers an area of approximately 12,600 km² and hosts a population of approximately 5.05 million in the 2020 census (Huang 2021, 2022a). The colloquial language spoken by native Zhangzhou people is Southern Min (ISO 639-3 [nan]; anglicized as Hokkien), which is mutually intelligible to Xiamen, Quanzhou, and Taiwan Southern Min but is entirely unintelligible to other Sinitic languages, such as Mandarin, Hakka, Cantonese, and Wu. The city has four districts (Xiangcheng, Longwen, Longhai, and Changtai) and seven counties (Hua’an, Nanjing, Pinghe, Zhao’an, Yunxiao, Zhangpu, and Dongshan). Certain regional variations can be observed in their sound systems (Ma 1994, Yang 2008). For example, this Yangru tone is transcribed as a short high contour [4] in Longhai, a mid-level contour [33] in Changtai, a low-rising contour [13] in Dongshan and Zhao’an, and as a convex contour of either [121] or [131] elsewhere (Yang 2008). This study restricts the research locality to Longwen and Xiangcheng districts, the innermost urban area of Zhangzhou City, which is considered to be historically, socially, culturally, and linguistically representative of Zhangzhou (Ma 1994, ZZG 1999). This selection can maximally ensure that derived scientific patterns are truly representative of Zhangzhou Southern Min as an independent language.

2.2 Corpus

The empirical data processed in this study were collected by the author from 21 native speakers, comprising 9 males and 12 females, in a formal field in the city of Zhangzhou in 2015. They were all born and grew up either in Longwen or Xiangcheng district and have been speaking Zhangzhou Southern Min as their mother tongue since childhood. Their parents and spouses are also native speakers of this language. None of them had physical issues to utter and perceive speech sounds. The average age for male speakers was 56.5 and 50 for female speakers at the time of recording. Language data recording was conducted in an acoustically soundproofing room in Zhangzhou Hotel through a professional cardioid condenser microphone at a sampling frequency of 44,100 Hz in Praat. The corpus incorporated 160 monosyllabic tokens, as illustrated in Table 1, and 588 disyllabic phrases, as illustrated in Table 2, in which this Yangru tone is denoted as tone 7. In this study, about 192 disyllabic tokens (12 samples × 8 combinations × 2 contexts) were elicited from individual speakers to examine how the Yangru tone changes realizations in linguistic contexts of phrase-initial and phrase-final, and how different tonal forms are related. What should be noted is that in Table 1, each tone is given its corresponding name in terms of the MC tonal category to make it diachronically traceable and synchronically comparable with other Sinitic languages. Tone 8 is given the name of Yangru (new) because it is a new tone emerging from those syllables that are conventionally transcribed with a glottal stop coda in Yangru tone, but the glottal stop coda is found to undergo deletion in synchronic speech (Huang 2018, 2023c). To maintain its origin, it is named under the category of Yangru but given a note to signify its difference from Yangru tone 7 which is related to syllables ending in obstruent codas and having a different F0 contour.

2.3 Acoustic processing

Two main acoustic correlates of tone, fundamental frequency (F0) and duration, were quantified using Praat. In alignment with most tonal studies in Sinitic languages (e.g., Rose 1986, 2016, Huang et al. 2016, Huang 2018, 2020, 2022a, 2023a), tonal duration was determined to incorporate all syllable segments, except onset consonants. Figure 1 illustrates how the tonally relevant duration was determined (Huang 2020). The duration onset was set at a glottal pulse, where the amplitude of air pressure fluctuation began to increase, the periodicity of speech wave vibration appeared regular, and the formant patterns of related tokens were clearly stable in spectrograms. In contrast, the offset was set at a point where the periodicity of speech wave vibration became irregular and formant patterns ceased to be visible (Huang 2020, 2022a). Raw F0 and duration values were extracted using a pre-designed script at ten equidistant sampling points within labeled tonal duration. About 33,600 F0 values (160 tokens × 21 speakers × 10 sampling points) and 3,360 durational values (160 tokens × 21 speakers) were measured for monosyllabic tones, and about 40,320 F0 values (192 tokens × 21 speakers × 10 sampling points) and 4,032 durational values (192 tokens × 21 speakers) were extracted to examine this tone in disyllabic phrases. In addition, the first three formant values, labeled F1, F2, and F3, were also extracted to examine how vowel quality and obstruent codas change their phonetics across different linguistic contexts, and how they interact with suprasegments to encode tonal morphemes in this tone.

Figure 1

Praat labeled monosyllabic example /kɛ1/ ‘home’ (WYF, male).

Acoustic signals are expected to be variable because they contain not only linguistic content but also a range of extralinguistic information such as speakers’ sociocultural background, vocal tract anatomy, and pragmatic intent (cf. Lobanov 1971, Rose 2016, Huang et al. 2016). This study conducted a series of normalization processes to reduce intra-and inter-speaker variation in the hope of deriving tonal representations of linguistic significance. An z-score approach was used to normalize raw F0 values, as presented in formula (1), whereas an absolute approach was employed to normalize raw duration values, as shown in formula (2) (Rose 2016, Huang et al. 2016, Huang 2018, 2020, 2022b). Lobanov’s (1971) approach was adopted to normalize formant frequency values as shown in formula (3).

2.4 Statistical testing

This study employed the technique of pairwise t-test by effect size (Levshina 2015, Huang 2022a, 2023a) to determine whether a particular acoustic variable (F0 or duration) of this tone differs significantly as a function of surrounding tones in a specific context. This test was conducted under the assumption that each paired difference is independently and identically normally distributed. Thus, an exhaustive comparison of all possible paired differences is required. For example, examining whether the F0 offset of this tone is affected by its following tones in disyllabic phrases requires testing 28 (=8 × 7/2) paired differences, as illustrated in Figure 2.

Figure 2

Pairwise t-test on F0 offset of Yangru tone across subsequent tones.

During the testing, Bonferroni correction was performed to control for a type I error and to ensure significance (Levshina 2015, Huang 2022a, 2023b). The corrected alpha was calculated by dividing the critical P value of 0.05 by the number of paired differences. For example, in the phrase-initial context, the corrected alpha value was 0.00186 (=0.05/28). If a calculated t-value was less than the corrected alpha of 0.00186, its related paired difference was considered statistically significant and vice versa. The pairwise t-testing results were visualized using hierarchical clustering algorithms to present a statistical relation among acoustic values in an objective way. Acoustic processing and statistical testing enabled this study to establish an analytical framework that is scientifically grounded and that can objectively interpret tonal behaviors and processes with a higher level of generalization.

3.1 Tonal F0

A number of documents have depicted Zhangzhou citation tones (e.g., Schlegel 1886, Dong 1959, Lin 1992, Ma 1994, ZZG 1999, Gao 1999, Huang 2018, 2020, 2022a, 2023a, b, c, d). All works prior to Huang’s (2018) study were based on auditory impressions and characterized Zhangzhou tones in terms of the MC tonal categories, with pitch values transcribed differently among scholars. In addition, they documented seven tones in this language, contradictory to Huang (2018, 2019, 2023d)’s proposal of an eight-tonal system, in which the eighth tone emerges from those syllables that are conventionally transcribed with a glottal stop coda in the MC Yangru tone, but the glottal coda is discovered undergoing deletion in the synchronic Zhangzhou speech. In addition, Huang (2018) asserts a multidimensional framework to characterize tonal behaviors and the totality of tonal contrasts in this language. This is proposed based on the fact that tones sharing a similar F0 contour can differ in other parameters, for example, tone 4 (Yinqu) and tone 6 (Yinru) are both realized as mid-falling [41] in the citation, but tone 6 is significantly shorter than tone 4. Likewise, tone 8 shares an identical low-level contour with tone 2 in the citation, but behaves differently in other contexts. It is beyond the scope of this study to provide a detailed introduction of how the eight-tonal system and the multidimensional framework are derived, but as an extension to understanding the nature of tonal property in this language, this eight-tone proposal is adopted in this study. Figure 3 plots normalized F0 contours of the eight lexical tones in citation, whereas Figure 4 visualizes the pairwise t-test result that examines how many F0 levels are contrastive in the citation context (Huang 2022a).

Figure 3

Z-score-normalized F0 pattern of individual Zhangzhou citation tones.

Figure 4

Pairwise t-test results for F0 levels of all Zhangzhou citation tones.

As can be seen, Zhangzhou citation tones vary considerably in F0 contour shapes. An inventory of one rising (tone 1), three levels (tones 2, 5, and 8), three falling (tones 3, 4, and 6), and a level with a falling tail (tone 7) can be derived from acoustic patterns. The pairwise t-test result in Figure 4 reveals five F0 heights that are statistically significantly different and they are ranked from the highest F0 height (5) to the lowest F0 height (1) using the hierarchical clustering algorithm. For example, tone 3 at the 10% sampling point (onset) and tone 1 at the 100% sampling point (offset) are both tested to be the highest. In contrast, tones 3, 4, 6, and 7 at the 100% sampling point (offset) are tested to be the lowest. Thus, the eight lexical tones can be represented using numerical notations, as summarized in Table 3, along with their corresponding names in the MC tonal categories, to make them diachronically traceable and synchronically comparable with other Sinitic languages. Examples of individual citation tones with (semi-) minimal pairs are provided in Table 1 in Section 2. Some additional significant characteristics can also be generalized with respect to the property of Zhangzhou tones in the citation context.

1. Tones that share a similar F0 contour shape can differ considerably in F0 contour height. For example, both tone 2 and tone 5 show a leveling trend in the citation context, but tone 5 has a statistically significant higher contour height. Likewise, tone 3 and tone 4 both present a falling contour, but the contour height of tone 3 is statistically significantly higher than that of tone 4. This reflects the phonemic function of F0 contour shape and F0 contour height to distinguish lexical items in this language.

2. Tones can neutralize F0 contrasts in the citation context. For example, tones 4 and 6 both present a mid-high falling contour that is noted as [41], whereas tones 2 and 8 both present a low-level trend that is noted as [22]. The reason why they are separately posited as phonemically contrastive is because of their different realizations in other parameters, including duration, phonation, vowel quality, syllable coda type, and/or linguistic context (Huang 2018, 2020, 2023b, c, d). For example, tone 4 has a significantly longer duration than tone 6 in both citation and sandhi (non-right-most; non-phrase-final) positions. Tone 2 presents a mid-level contour, whereas tone 8 shows a mid-falling trend in the non-phrase-final sandhi position.

3. Yang-registered tones have an F0 range lower than Yin-registered tones. For example, Yinping tone 1 presents a mid-rising contour, whereas Yangping tone 2 shows a low-level F0. Likewise, Yinqu tone 4 shows a mid-high falling contour, whereas Yangqu tone 5 appears as mid-level F0. This asymmetrical distribution can be ascribed to the diachronic constraint on the selection of tonal register with respect to a voicing status of onset consonants. Yang-registered tones are historically assigned to voiced onsets, while Yin-registered tones are assigned to voiceless onsets. As acknowledged in modern phonetic theory, voiced onsets have a depressing effect, whereas voiceless onsets have a raising effect on subsequent F0 contours (cf. Hombert 1978, Yip 2002, Abramson and Erickson 1978); thus, it is articulatorily understandable for Yang-registered tones to be pitched in a lower range in this language.

4. The Yangru tone, labeled as tone 7, neither presents a low convex contour [121] (Ma 1994, Yang 2008, FJG 1998, Guo 2014, ZZG 1999), nor a low rising contour (Gao 1999, Dong 1959). Instead, its acoustic pattern shows a low-level trend with a fall tail in this context, which can be noted in [221]. What deserves special attention is that this tone shares a similar low-levelling trend with tones 2 and 8 over the first half of the duration; however, in the second half, its F0 contour is depressed with a shorter duration and a lower F0 offset. The depression is considered to result from an effect of laryngealization that is induced by the non-realization of obstruent codas in utterance-final contexts (Huang 2018, 2023d), which will be discussed in Section 6.

3.2 Tonal duration

Zhangzhou tones also present variations in their duration. Figure 5 plots normalized duration patterns of individual citation tones, whereas Figure 6 visualizes the pairwise t-test result that examines the number of contrastive durations in citation (Huang 2018, 2022b). In total, four levels are tested to be statistically significantly different, which are ranked from the longest (1) to the shortest (4) and are summarized in Table 4. As indicated, Yinru tone 6, which is diachronically related to syllables with voiceless onsets and obstruent codas, is tested to be the shortest, while the three-level tones (tones 2, 5, and 8) are among the longest. Interestingly, the duration of Yangru tone 7 is not the shortest, as conventionally documented (Ma 1994, Yang 2008, Guo 2014); instead, it is longer than tones 3, 4, and 6. This falsifies the default convention in the studies of Sinitic tones which considers Ru tones that are related to stop-ending syllables to be the shortest (e.g., Ma 1994, Yang 2008, Guo 2014, Li and Yao 2008, Yang 2005). In addition, Yangru tone 7 is shorter than tones 2 and 8, which share a similar low-level F0 contour. This can also be considered an effect of laryngealization induced by the non-realization of obstruent codas in an utterance-final position.

Figure 5

Normalized duration of Zhangzhou citation tones.

Figure 6

Pairwise t-testing result on duration of Zhangzhou citation tones.

4.1 Tonal F0

Tones may change their realization categorically in connected speech, inducing a phenomenon of tone sandhi. Tonal citation form is considered to be the form when its monosyllabic morpheme is produced in isolation, in the case of Sinitic languages, while tonal sandhi form is referred to as the form when its monosyllabic morpheme alters its realization, particularly at both phonetic and phonological levels, in multisyllabic constructions. Tonal languages where relevant can be classified into be either right-dominant or left-dominant, depending on where tonal form in the citation is preserved and where tonal form undergoes a categorical change in connected speech (e.g., Shih 1986, Ballard 1988, Rose 2004, 2016, Chen 2000, Zhang 2007, Huang 2018, 2020, 2022b, Shen and Rose 2016). Specifically, in left-dominant languages, tones at the leftmost position are expected to maintain their citation forms but extend the realization over the whole domain. In contrast, in right-dominant languages, tones at the right-most position are assumed to maintain their citation forms while at a non-right-most position undergo categorical change. Zhangzhou Southern Min presents a typical right-dominant tone sandhi system (Huang 2018, 2020, 2022c). This can be straightforwardly seen in Figure 7 that plots the F0 contours of tone 3 in three contexts. It presents a high-falling contour in the citation and preserves this high-falling contour at the rightmost position of disyllabic constructions, but this tone changes the contour to rising at the non-right-most position, justifying a right-dominant tone sandhi system in this language.

Figure 7

Illustration of right-dominant tone sandhi in Zhangzhou.

Driven by the forcing effect of right dominance of a sandhi system, this Yangru tone is expected to change its realizations categorically at a non-right-most position, a non-phrase-final position in particular. To explore how this tone changes its realization, Figure 8 plots the normalized F0 patterns of this tone across eight lexical tones, whereas Figure 9 visualizes the pairwise t-testing result that examines whether its F0 may be affected by subsequent tones and to what extent it may be affected (Huang 2018).

Figure 8

Normalized F0 patterns of Yangru tone across phrase-final tones.

Figure 9

Pairwise t-test result on F0 offset of Yangru tone across phrase-final tones.

As shown, this Yangru tone consistently presents a decreasing trend between 0.5 and −0.5 s.d. across eight phrase-final tones, which can be noted as [32]. The change from a low-level trend with a falling tail [221] in citation to a mid-falling contour [32] at the non-phrase-final position justifies the effect of the right-dominant tone sandhi system occurring on this tone. In addition, the test result reveals no statistically significant difference in the F0 offset realization of this tone as a function of eight tones at the phrase-final position. This suggests that its F0 realization is phonologically inert to the category of following tones, because regardless of whether its following tone is rising, level, or falling, its F0 contour is consistently the same.

In addition, the domain where a tone sandhi process is expected to occur in this language is found to be aligned with the boundary of syntactic phrases XP, which also include multisyllabic words (Huang 2018). If tones occur at a non-phrase-final position, they alter their realization phonetically and phonologically, but maintain their citation forms at the phrase-final position. The tonal alternation within the general environment of syntactic phrases of this language is referred to as the general tone sandhi process (Huang 2018, 2022c). Figure 8 that examines the F0 realization of Yangru tone as a function of eight subsequent tones is plotted based on the empirical data of disyllabic phrases or words, as illustrated in Table 2. In addition, the general tone sandhi process appears to be irrelevant to the internal structure of syntactic phrases XP. Because regardless of whether the phrase is AP, NP, or VP, tones at a non-phrase-final sandhi position have the same output in F0. For example, the phrases /tit32.tit221/ ‘very straight’ (AP), /sit32.ɗik221/ ‘strength’ (NP), and /ħɐk32.sik221/ ‘to study’ (VP) have the same Yangru tone 7 + Yangru tone 7 combination but different internal structures at the underlying level. However, at the surface level, the F0 contours of the first Yangru tone 7 at phrase-initial morphemes consistently present a mid-falling trend, which is noted as [32], justifying its realizations is irrelevant to the category of syntactic phrases. What should also be noted is that this study restricts the discussion on the general tone sandhi process within the context of syntax phrases to demonstrate the forcing effect of a right-dominant tone sandhi system. There are also some other special tone sandhi patterns that are also right-dominant, but occur in very specific morphological processes, like infixation (Huang 2023e) and suffixation (Huang 2024b, c). For example, tonal morphemes before the suffix ʔɐ51, which can function as a nominal marker, diminutive maker, verbalizer, or nominalizer among others, alter their pitch contours to either rising or high level depending on their dominant contour shape in citation. Specifically, if tones have a dominant falling contour in citation, before this suffix ʔɐ51, they change their contour to rising; whereas if tones have a dominant-none falling contour, before this suffix ʔɐ51, their contour becomes rising (Huang 2024b). For this Yangru tone 7, since it possesses a dominant level contour with a falling tail, its related morphemes are found to bear a rising contour over the suffixation with ʔɐ51, such as /ɠjɔk35.ʔɐ51/ ‘jade’, and/ sik35.ʔɐ51/ ‘wing’. It is out of the scope to discuss special tone sandhi processes in this study, but they can be referred to in my other work in morphology (Huang 2023e, 2024b, c).

As discussed, this language is right-dominant in its tone sandhi system. In the domain of general syntactic phrases, the tone sandhi process is only sensitive to the boundary of syntactic phrases, while is phonologically inert to the category of subsequent tones, and is syntactically irrelevant to the internal structure of related phrases. These properties render this language to be typologically different from many other tonal languages. For example, the well-known third tone sandhi in Mandarin is entirely phonologically evoked because it only occurs in the third tone when it occurs before another third tone. Tone sandhi in Wenzhou and Shanghai are sensitive to the internal structure of syntactic phrases (cf. Ballard 1988, Chen 2000, Zhang 2007, Duanmu 2005). For example, in Shanghainese, the modifier-noun phrase (NP) shows a left-dominant tone sandhi pattern, whereas the verb-object phrase (VP) exhibits a right-dominant sandhi pattern (Duanmu 2005). In Zhangzhou, the general tone sandhi process is only sensitive to the boundary of syntactic phrase, the non-right-most (non-phrase-final) position.

4.2 Tonal duration

As an important parameter of tonal properties, this tone also changes its duration at a non-right-most (non-phrase-final) sandhi position. Figure 10 plots the normalized duration of Yangru tone as a function of eight tones at the phrase-final position, and Figure 11 shows the pairwise t-test result that examines whether the duration may be affected by subsequent tones in a statistically significant way. As can be seen, there is no statistically significant difference in its durational realization across subsequent tones. In other words, the change of tonal categories at the right-dominant position does not affect its duration at both phonological and phonetic levels.

Figure 10

Normalized duration of phrase-initial tone 7 across phrase-final tones.

Figure 11

Pairwise t-test result on duration of Yangru tone across phrase-final tones.

To depict the duration status of this tone with respect to all tones in the sandhi context, Figure 12 shows the normalized duration of all tones, and Figure 13 shows the pairwise t-testing result that examines how many duration levels are contrastive at the sandhi position. What should be noted that tone 5 is tested to have two variants in this context with one being statistically significantly shorter before tone 3 and tone 6 (Huang 2018, 2022b). All other tones only have one variant.

Figure 12

Normalized duration of phrase-initial tones.

Figure 13

Pairwise t-testing result on duration of Zhangzhou phrase-initial tones.

As can be seen, only two duration levels are tested to be statistically contrastive at the sandhi position. The two Ru tones of Yinru tone 6 and Yangru tone 7, also known as stopped tones, are grouped together because they both present an extra-short duration. All other non-Ru tones are clustered into one single category because they present a medium duration. Thus, the duration appears to be asymmetrically distributed with respect to the type of syllable codas. Because within the Sinitic Convention, Ru tones are assigned to syllables ending in obstruent segments, whereas non-Ru tones are assigned to syllables ending in sonorant segments that include zero codas, nasal consonants, and glides. As such, it is understandable that tones are realized asymmetrically in duration, either as extra-short or medium, depending on the syllable ending type. In addition, this Yangru tone changes its duration from long in citation to extra-short in sandhi, further reflecting the forcing effect of the right-dominant sandhi system in this language. The extra-short property also reflects the interaction between segments and suprasegment in this language, which is discussed in detail in Section 6.

5.1 Tonal F0

The conventional criterion or default principle that determines the dominance of tone sandhi system relies exclusively on where the tonal citation form is preserved (cf., Zhang 2007, Ballard 1988, Rose 1996). In other words, the tonal citation form is considered to be intactly preserved without any change at the dominant position. Specifically, in a left-dominant sandhi system, tones at the leftmost position are assumed to maintain their citation forms, whereas in a right-dominant system, tones at the rightmost position preserve their citation forms. However, in the reality of connected-speech production, the vocal apparatus of human beings operates in a continuous fashion, which could cause considerable overlapping articulatory gestures and give rise to diverse outputs at the surface level (Han and Kim 1974, Shen 1990, Palmer 1969, Xu 1994, Gandour et al. 1994, Scholz and Chen 2014, Zhang and Liu 2011, Huang and Hyslop 2022, Abramson 1975). For example, vowels are easily nasalized if they occur after nasal consonants, such as in Mandarin, the morpheme /ma55/ ‘mother’ is often perceived as [mã55]. Consonants can also be easily nasalized if they occur before nasalized vowels, such as in Zhangzhou, three implosive onsets /ɓ/, /ɗ/, and /ɠ/ are realized as their homorganic nasal plosives [m], [n], and [ŋ], respectively, before nasalized vowels [Ṽ], such as /ɓĩ33/ ‘noodle’ → [mĩ33]/_ [Ṽ]; /ɗĩ51/ ‘dye’ → [nĩ51]/_ [Ṽ]; /ɠɔ̃51/ ‘five’ → [ŋɔ̃51]_ [Ṽ] (Huang 2022d, Huang and Hyslop 2022). This phenomenon is known as co-articulation in the literature and well-attested in the domain of segments (e.g., Han and Kim 1974, Shen 1990, Palmer 1969, Xu 1994, Gandour et al. 1994, Scholz and Chen 2014, Zhang and Liu 2011, Abramson 1975). Although less studied, tones may also be easily affected by their surrounding tones at the surface level, and this has been attested in some tonal languages in Asia. For example, Mandarin tones can change their F0 height in connected speech (Shen 1990). Thai tones can change their F0 height and F0 slope because of the co-articulation effect (Palmer 1969). Vietnamese tones can change their F0 slope and height because of the influence of surrounding tones (Han and Kim 1974). Given this cross-linguistic tendency of speech co-articulation, it appears questionable to consider that tones maintain their citation form without any change at the dominant position and use it as a default principle to determine the dominance of the tone sandhi system. Therefore, it is important to explore how tones are realized in the dominant position and to test whether their realizations may be affected by surrounding tones at the phonetic level. To address this issue, Figure 14 visualizes the normalized F0 pattern of this Yangru tone as a function of eight preceding tones. Figure 15 shows the pairwise t-test result that examines whether its F0 may be affected by preceding tones in a statistically significant way (Huang 2018).

Figure 14

Normalized F0 patterns of Yangru tone across phrase-initial tones.

Figure 15

Pairwise t-test result on F0 onset of Yangru tone across phrase-final tones.

As shown in Figure 14, this Yangru tone consistently presents a falling trend between 0 and −2 s.d. across eight tones at the phrase-initial position, regardless of whether the preceding tone is rising, level, or falling. In other words, a change in the tonal category before this tone does not cause any categorical change in its F0 contour shape. As such, this tone can also be seen as being phonologically inert to the category of surrounding tones at the dominant position. However, the pairwise t-testing result reveals two variants that are significantly different in the F0 onset height. The onset is tested to be marginally higher after tones 1, 2, 3, and 6 that have their F0 offset either at or above the midpoint. Contrastively, its onset is tested to be lower after tones 4, 5, 7, and 8 that all have an offset below the midpoint. As such, the two variants are predictable and are primarily conditioned by the offset height of preceding tones. Thus, they can be denoted as [311] and [211] at this right-dominant (non-sandhi; phrase-final) position. Comparing with a low-level with a fall tail, noted as [221], in the citation context, the leveling plateaus of the two variants is lower that also can be ascribed to the F0 declining effect at an utterance-final position, which has been well-attested in the world’s natural languages (e.g., Lieberman 1967, Pierrehumbert 1989, Maeda 1976, Cohen et al. 1982, Ladd 1984, Yuan and Liberman 2010, Rose 2014).

As indicated, the F0 realization of this tone is not exactly the same as a function of its preceding tones. This reflects the cross-linguistic trend of co-articulation and provides a scientifically attested pattern to the typology of co-articulation that occurs in the realm of tone. Meanwhile, the F0 realization is not exactly the same as its citation form, indicating that the citation form is not intactly preserved without change at the dominant position. As such, the conventional criterion or default principle to specify the dominancy of the tone sandhi system does not accurately capture the phonetic reality of tonal behaviors in real-world utterances, which could create misleading information on our understanding of the nature of tone in connected speech. Thus, this exploration has significantly practical implication to supersede our conventional impression and enlighten a theoretical discussion on how the tone sandhi dominancy should be better defined from a modern linguistic perspective.

5.2 Tonal duration

Similar to F0, the parameter of duration also presents variation at the dominant position. Figure 16 plots the normalized duration of this Yangru tone across eight preceding tones. Figure 17 shows the pairwise t-testing result that examines whether the duration may be statistically significantly affected by preceding tones. As seen, two phonetic variants are tested to be significantly different, with one longer after tone 7 than after the other tones. Thus, the variation in duration can be considered to be tonally relevant in this context.

Figure 16

Normalized duration of phrase-final tone 7 across phrase-initial tones.

Figure 17

Pairwise t-test result on the duration of Yangru tone across preceding tones.

To better understand the duration of this tone with respect to all tones in this context, Figure 18 plots the normalized duration of all Zhangzhou tones at the phrase-final position of disyllabic constructions. Figure 19 visualizes the pairwise t-test result that examines how many duration levels are contrastive in this setting. What needs special attention is that tone 5 is also discovered to have two variants at the dominant position, with one variant significantly longer after tone 8 (Huang 2022c).

Figure 18

Normalized duration of phrase-final tones.

Figure 19

Pairwise t-testing result on duration of all tones at the dominant position.

As shown, as far as the entire tonal system is concerned, three duration levels are tested to be statistically contrastive at the dominant position, which are noted as extra-long (tones 1 and 5), long (tones 2, 3, 7, and 8), and medium (tones 4, 6, and 7) (Huang 2022c). This Yangru tone has a markedly long duration that occurs after tone 7 and an unmarked medium duration that occurs elsewhere. This indicates that the tonal coarticulation effect can also affect duration. In addition, the unmarked form of a medium duration at the right-dominant position is different from the citation form of a long duration. This could also be ascribed to the final-position effect, which not only can decrease tonal F0 but also shorten tonal duration. This also justifies the fact that the tonal citation form is not intactly preserved without change at the dominant position, further questioning the conventional default principle on the specification of the dominance of the tone sandhi system in languages where relevant.

6.1 Changing phonetics of vowel quality

Seven oral vowels /i, e, ɛ, ɐ, ɔ, ɵ, u/ are phonemically distinctive in this language, which cover a three-way contrast of high /i, u/, mid /e, ɵ/ and low /ɛ, ɐ, ɔ/ in terms of tongue height (Huang 2018, 2020, 2021). In this Yangru tone, the two high vowels /i/ and /u/ alternate between monophthongs and diphthongs across different contexts. Specifically, they are separately realized as diphthongs [iɛ̯] and [uɤ̯] in an utterance-final context but as monophthongs [i] and [u] in a non-utterance-final context. The changing vowel quality in this tone can be justified by using scientifically quantified patterns. Figure 20 plots the normalized formant patterns of high vowels /i/ (left) and /u/ (right) in citation based on monosyllabic samples like /sit7/ ‘solid’ and /hut7/ ‘buddha’. Figure 21 plots the normalized formant patterns of high vowel /i/ (left) and /u/ (right) in the phrase-initial context based on disyllabic samples like /zit7.tʰɐw2/ ‘sun’ and /ħut7.kɐw4/ ‘Buddhism’. Each figure is derived by normalizing 3,404 values (=3 tokens × 2 high vowels × 1 tone × 9 sampling points × 3 formants × 21 speakers).

Figure 20

Normalized formant patterns of high vowels in Yangru tone in citation.

Figure 21

Normalized formant patterns of high vowels in Yangru tone in the phrase-initial context.

As indicated in Figure 20, the normalized formant patterns of vowels /i/ and /u/ both change in the citation context. For the vowel /i/, the F1 curve increases from −1.22 to −1.08 s.d., whereas the F2 curve decreases from 0.60 to 0.25 s.d. The increasing F1 curve but decreasing F2 curve suggests that native speakers moved their tongues toward a lower but more central position. It reveals a diphthong articulation of this high vowel /i/ in this tone and in this context, which can be transcribed as [iɛ̯]. For the high vowel /u/, the F1 curve increases from −1.14 to −0.95 s.d., while the F2 curve increases from −0.080 to −0.036 s.d. The increasing F1 and F2 curves indicate that native speakers moved their tongue toward a lower but fronting position. It also reveals a diphthong articulation of the vowel /u/, which can be transcribed as [uɤ̯]. Therefore, it is scientifically plausible to propose high-vowel diphthongization in this setting.

In Figure 21, the normalized formant patterns of vowels /i/ and /u/ are both relatively steady, without significant changes in the phrase-initial (non-utterance-final) context. In the high vowel /i/, its F1 curve is leveled at approximately −1.15 s.d., while its F2 curve is leveled around 0.45 s.d. but presents a slight falling trend toward 0.41 s.d. during the last 5% portion. This slight decline reflects a formant transition for the production of subsequent obstruent codas in this tone. Thus, the high vowel /i/ can be considered a monophthong [i]. Likewise, in the high vowel /u/, its F1 curve levels around −1.10 s.d., while the F2 curve levels around −053 s.d. Such steady formant patterns indicate that native speakers retain their tongues in the same position throughout articulation. This justifies a monophthong articulation of the high vowel /u/ in this tone and in this context. It is scientifically plausible to propose that high-vowels are realized as monophthongs in this setting.

As shown, tonal morphemes can change the phonetics of high vowels in this tone according to different linguistic contexts. The reason why high vowels undergo diphthongization in the utterance-final context but remain monophthong articulation in the non-utterance-final context is closely related to the phonetics of obstruent codas that can also change across contexts. This is discussed in detail below.

6.2 Changing phonetics of obstruent codas

In East and Southeast Asian languages, tones are often classified into either stopped or unstopped categories, depending on the type of syllable ending to which they are assigned. For example, in the Early MC around 500 CE, four tones named Ping ‘level’, Shang ‘Rising’, Qu ‘Departing’, and Ru ‘entering’ had been documented by Yue Shen (441-513) and Yong Zhou (n.a) in Nanshi, an official historical of the Southern Dynasty, and in Qieyun, a rhyme dictionary edited by Fayan Lu in 601 CE (Baxter and Sagart 2014, Sagart 1989, Pulleyblank 1978, Pan and Zhang 2015, Li and Yao 2008, Yang 2005). The term ‘Ru’ equivalent to ‘stopped’ was created to refer to tones that are assigned to syllables ending in obstruent codas. The four MC tones are preserved in the present-day Zhangzhou Southern Min. Thus, this Yangru tone is only aligned to stop-ending syllables, and obstruent codas of a three-way contrast can be identified at the underlying level, as illustrated in Table 5. However, despite performing a phonemic function, the realization of obstruent codas is highly sensitive to linguistic context and presents a dynamic profile at the surface level (Huang 2023d). Specifically, they are not realized in an utterance-final context but are realized in a non-utterance-final context. The changing phonetics of obstruent codas are also justifiable from a scientific perspective. Figure 22 plots the normalized formant patterns of monosyllabic morphemes that have the same nucleus /ɐ/ and Yangru tone but differ in obstruent codas, like /tsɐp7/ ‘ten’, /ɓɐt7/ ‘dense’, and /ɗɐk7/ ‘six’. Figure 23 plots the normalized formant patterns of phrase-initial morphemes that also have the same nucleus /ɐ/ and Yangru tone but differ in obstruent codas, like /tsɐp7.ɗɐk7/ ‘sixteen’, /pɐt7.zit7/ ‘another day’, and /ɗɐk7.tsɐp7/ ‘sixty’. Each figure is derived by Lobanov-normalizing 5,103 formant values (=3 tokens × 9 sampling points × 3 formants × 3 obstruent codas × 21 speakers).

Figure 22

Normalized formant patterns for obstruent coda examination in citation.

Figure 23

Normalized formant patterns for obstruent coda examination in phrase-initial morphemes.

In Figure 22 about the utterance-final (citation) context, the normalized formant patterns are consistently stable without significant change over major portions of three rhymes /ɐp7/, /ɐt7/, and /ɐk7/ that differ in obstruent codas. A slight fluctuation occurs at the beginning of the F1 and F2 curves, which may be ascribed to the perturbation effect of onset production. From an articulatory perspective, steady patterns indicate that native speakers do not move their active articulators (e.g., lower lip, tongue tip) toward passive articulators (e.g., upper lip, alveolar ridge, hard palate) to form a complete constriction in the vocal tract (Huang 2023d). In other words, their articulators, especially tongues, are maintained in the same position over articulation. Therefore, the patterns reveal two important facts: (a) the related vowel is a monophthong and (b) no obstruent coda is articulated following the vowel. This justifies the assertion that obstruent codas are not realized in an utterance-final context.

In Figure 23, regarding the non-utterance-final context, the three formant patterns change differently over the last 10% of the duration. In the rhyme /ɐp7/, the F2 curve drops by approximately 0.5 s.d. indicating a bilabial place of articulation. In the rhyme /ɐt7/, the F2 curve increases by about 0.3 s.d. indicating an alveolar place of articulation. In the rhyme /ɐk7/, the F2 curve increases by more than 1 s.d., whereas the F3 curve abruptly drops by approximately 0.3 s.d. The two curves tend to converge at a point around 0.5 s.d. showing a typical characteristic of velar pinch that indicates a velar place of articulation. As shown, the three formant patterns change substantially toward different locus frequencies, which reflect different places in the oral cavity where the airstream is momentarily blocked for the articulation of different obstruent codas (cf. Laver 1994, Bickford and Floyd 2006, Gick et al. 2013, Huang 2023d). As such, it is scientifically justifiable to propose that obstruent codas are realized in this tone and in a non-utterance-final context.

As discussed above, tonal morphemes can change the segmental properties of both vowel quality and obstruent coda in different linguistic contexts in this tone. This indicates that segmental and suprasegmental properties are intrinsically correlated with each other to shape a dynamic profile of tonal morphemes in this language. Specifically, in the non-utterance-final context, obstruent codas are realized, which can shorten the duration of this tone and motivate monophthong articulation of high vowels. In contrast, in the utterance-final context, obstruent codas are not realized, which can lengthen the duration of this tone as a compensatory effect while motivating a diphthongization process on high vowels. Such intrinsic correlations between segments and suprasegments are not arbitrary but reflect a continuous motion of the vocal apparatus of human beings in speech production.