Syllable onsets in Southern Min: inventory, processes, and constraints

3.1 The size of onset inventory

In this language, only 15 onsets are phonemically contrastive to make lexical distinctions. They are summarized in Table 3 and illustrated in Table 4 with (near-) minimal pairs. The size of this onset inventory is smaller than that of most languages and/or dialects in China and many others worldwide. For example, among a sample of 70 Chinese dialects (Zee and Zee 2014), the number of onset phonemes ranges from 14 in Putian of Middle Min and Ningde of Northern Min to 29 in Xinhua of the Xiang language. Thus, it seems that only a few languages in China have a smaller onset inventory than this language. Likewise, in a cross-linguistic survey of 317 languages worldwide (Maddieson 1984), the East Papuan language Rotokas has the lowest number of 6, whereas the Khoisan language! Xũ has the greatest number of 46, which is primarily due to the large number of clicks and laryngeal contrasts exploited by both click and non-click consonants (Gordon 2016; Maddieson 1984). As can be predicted, the number of 15 onset consonants in Zhangzhou is supposed to be lower than many of the world’s languages. The reason why the onset inventory is small in this language is unclear, but it might be ascribed to a compensatory relationship between the number of vowels and onset consonants. For example, Zee and Zee (2016) report that among 70 dialects in China, the number of vowels ranges from a low of 3 in Yongding of the Hakka language to a high of 13 in Ningbo of the Wu language. Maddieson (1984) reports that the vast majority of languages have between five and seven vowels (64.7 %), with the modal number being five (30.9 %). In Zhangzhou, a total of 11 vowels (7 oral vowels and 4 nasal vowels) is predictable to be larger than in most of the world’s languages, and this might potentially reduce the number of consonants to function as onset phonemes in this language.

It should also be noted that conventional work all posited 15 onset phonemes for this language (FJG 1998; Gao 1999; Guo 2014; Lin 1992a, 1992b, 1992c; Ma 1994; Medhurst 1832; Schlegel 1886; Xie 1818; Yang 2008; Zhou 2006; ZZG 1999), with four voiced onsets transcribed differently from this study. They all documented a voiced alveolar affricate /dz/, but this onset was found to be a fricative /z/ in the empirical data across speakers. This can be seen in Figure 1, which shows the spectrograms of morphemes /zi33/ ‘character’ and /zu22/ ‘such as.’ No voiced bars or burst releases were identified before frication for either morpheme. This indicates that no oral constriction is created during their production. In other words, they are not articulated with an affricate manner of articulation, as transcribed as /dz/ in previous work. Instead, they are produced with a fricative manner of articulation, and this onset should thus be posited as /z/ to reflect the reality in the field.

Figure 1:

Spectrograms of /zi33/ ‘character’ (left) and /zu22/ ‘such as’ (right).

Prior work documented voiced onsets /b/, /l/, /g/ in their segmental inventory (e.g., FJG 1998; Gao 1999; Guo 2014; Lin 1992a; Lin 1992b; Lin 1992c; Ma 1994; Medhurst 1832; Schlegel 1886; Xie 1818; Yang 2008; Zhou 2006; ZZG 1999); however, these onsets are separately produced as implosives /ɓ/, /ɗ/, /ɠ/ in unmarked environments by native speakers regardless of genders and ages, and have several allophonic variants which will be discussed in the next section. Their implosive articulation can be demonstrated in the spectrograms. As shown in Figure 2, a voiced bar can be clearly identified before the formant patterns across all morphemes. This signifies a voiced manner of articulation and indicates that over the production of related onsets, a complete oral constriction is formed at a certain point in the speaker’s oral cavity, depending on their place of articulation, and the vocal folds vibrate in a regular fashion. In addition, the amplitudes of the air pressure fluctuations in the waveforms were maintained without significant change over the entire portion of the onsets, rather than dropping to zero before the release of oral constriction. This indicates a special ingressive glottalic airstream for implosive sounds. It occurs because speakers block off their glottis during articulation but continually suck in airflow from outside into their mouth until they achieve a maximum amount of air pressure, after which they open their glottis and release the constriction to let the air pressure from both lungs and mouth flow out of the vocal tract (Clements and Osu 2002; Cun 2010; Ladefoged and Maddieson 1996; Mori 2023). As such, it is empirically justifiable to propose implosives /ɓ, ɗ, ɠ/ as onset phonemes that falsify prior transcriptions but advance our knowledge of this language.

Figure 2:

Spectrograms and waveforms of Zhangzhou implosives.

3.2 Typologically common and uncommon characteristics

The onset system of this language can be characterized in several ways. It shows several universals in phonemic distribution but also presents several typologically uncommon features that contribute to our understanding of human language diversity. These onsets present a five-way contrast in terms of the place of articulation, which includes bilabial /p, pʰ, ɓ/, alveolar /t, tʰ, ɗ, s, z, ts, tsʰ/, velar /k, kʰ, ɠ/, pharyngeal /ħ/, and glottal /ʔ/. Seven of the 15 onsets were alveolars, accounting for 47 % of the total onsets. This indicates the high contribution of the front of the tongue and the alveolar place in speech production in this language. This reflects a universal tendency; for example, 97.5 % of the 317 languages in Maddieson ’s (1984) survey have alveolar sounds, followed by velars in 89.3 % and bilabials in 82.9 %. Likewise, these onsets present a three-way contrast in terms of the manner of articulation, including plosive /p, pʰ, ɓ, t, tʰ, ɗ, k, kʰ, ɠ, ʔ/, fricative /s, z, ħ/, and affricate /ts, tsʰ/. It can be seen that plosive sounds occupy as many as 67 % of the total onsets, implying that complete constriction is the dominant manner of articulation in speech distinction in this language. It also reflects the universality that languages most commonly contrast unaffricated plosives, particularly at three places of articulation (labial, alveolar, and velar) (e.g. Gordon 2016; Maddieson 1984; Nikolaev 2022).

Supraglottal plosives can be further classified into three subcategories in terms of aspiration and voicing: voiceless unaspirated /p, t, k/, voiceless aspirated /pʰ, tʰ, kʰ/, and voiced /ɓ, ɗ, ɠ/. This is typologically uncommon. For example, according to Maddieson’s (1984) survey of 317 languages, less than 24.0 % of languages present a three-way laryngeal contrast on unaffricated oral stops, while over 51.1 % of languages present a two-way contrast between voiceless and voiced stops (Gordon 2016). In addition, this language is typologically uncommon because it employs a special ingressive glottalic airstream to construct implosive onsets of three places of articulation. This is because only 10 % (Ladefoged and Maddieson 1996) to 20 % (Clements and Osu 2002; Cun 2010) of the world’s languages are reported to have an implosive articulation at an underlying level. The three-way contrast among supraglottal plosives can be demonstrated by the acoustic parameter of voice onset time (VOT), which is the time between the release of an oral constriction for plosive production and the onset of vocal fold vibration (Abramson and Whalen 2017). This is shown in Figure 3. The VOT value is slightly above zero across voiceless unaspirated plosives /p, t, k/, reflecting that speakers vibrate their vocal folds immediately after the release of oral constriction. The VOT value is steeply positive from 0.067 to 0.083 s across voiceless aspirated plosives /pʰ, tʰ, kʰ/, reflecting that after an oral constriction is released, there is a period of articulation for aspiration, which could induce a delay in vocal fold vibration. In contrast, the VOT value is steeply negative between −0.077 and −0.064 s for voiced sounds /ɓ, ɗ, ɠ/, reflecting that speakers vibrate their vocal folds before releasing an oral constriction.

Figure 3:

VOT of Zhangzhou occlusive onsets.

An additional typologically uncommon feature of this onset system is that nasal consonants are not phonemic in this onset position. This is cross-linguistically unusual. In Maddieson (1984)’s survey, only seven of 317 languages (2.2 %) lacked phonemic nasal consonants. In Zee and Zee (2014)’s survey of 70 Chinese dialects, only one dialect lacked a labial nasal onset, 12 dialects lacked a velar nasal onset, and 14 dialects lacked an alveolar nasal onset. This language is among the very few to lack any of nasal onsets. It should be noted that while nasal onsets are not attested, the nasality feature is extensively used to construct nasal vowels, syllabic nasals, and nasal codas in this language. For example, four of seven oral vowels can be nasalized and used phonemically, such as /ti35/ ‘pig’ versus /tĩ35/ ‘sweet’, /kɛ35/ ‘home’ versus /kɛ̃35/ ‘net’, /kɐ/35/ ‘glue’ versus /kɐ̃35/ ‘prison’, /kɔ22/ ‘past’ versus /kɔ̃22/ ‘snore’. The bilabial syllabic nasal /m̩/ can carry five different pitch contours to differentiate five different words, such as /ʔm̩24/ ‘drink’, /ʔm̩22/ ‘flower bud’, /ʔm̩51/ ‘aunt’, /ʔm̩41/ ‘affirmative’, and /ʔm̩33/ ‘negative’. The velar syllabic nasal /ŋ̩/ can combine with twelve out of fifteen onsets to create monosyllabic morphemes, such as /pŋ̩51/ ‘list of names’, /pʰŋ̩51/ ‘ebb tide’, /tŋ̩35/ ‘wait’, /tʰŋ̩41/ ‘hot’, /ɗŋ̩22/ ‘small taro’, /sŋ̩51/ ‘play’, /tsŋ̩35/ ‘village’, /tsʰŋ̩22/ ‘bed’, /kŋ̩41/ ‘steel’, /kʰŋ̩41/ ‘hind’, /ħŋ̩35/ ‘prescription’, and /ʔŋ̩35/ ‘centre’. In the meanwhile, nasal codas of a three-way contrast in the place of articulation can be identified, such as /sim35/ ‘heart’, /sin35/ ‘new’, and /siŋ35/ ‘rise’.

Such an asymmetrical distribution of the nasality feature is also typologically rare enough. Because it is common for languages to possess nasal onsets only and lack nasal vowels and syllabic nasals in their phonemic inventory. In contrary, it is uncommon for languages to possess nasal vowels, syllabic nasals and nasal codas and do not have any nasal onset at the underlying level. However, this does not mean that the nasality feature does not perform any function on syllable onsets in this language, instead, it can affect their phonetic realizations and induce diverse outputs at the surface level. This will be discussed in detail in next section.

3.3 Distinctive features

As introduced above, this language possesses an onset system smaller than many of languages in the world. It presents several universals, such as the preference of the alveolar sounds and plosive manner of articulation, reflecting cross-linguistic tendencies in the construction of onset consonants in natural languages. It reveals several typologically unusual characteristics, such as the three-way contrast of supraglottal stops, the phonemic usage of implosive sounds, and the lack of nasal onsets, contributing to the typology of speech sounds and our understanding of the diversity of human languages. However, in spoken languages, speech sounds do not always act independently, instead, multiple sounds often participate in the same sound patterns (Mielke 2008). In other words, a group of sounds in an inventory may behave similarly because of specific properties, rather than because of the sounds themselves (Kennedy 2016; Mielke 2008). To capture the way that speech sounds function as a system, the concept of natural class is proposed in literature to refer to a group of sounds within a language that may either trigger or undergo the same phonological process, and the concept of distinctive features is proposed to group them into natural classes (cf. Arthur 1983; Chomsky and Halle 1968; Clements 2003; Hyman 1975; Keating 1991; Kennedy 2016; Lisker and Abbramson 1971; McCarthy 1988; Mielke 2008). Distinctive features have been widely assumed to be part of Universal Grammar since the mid-twentieth century and have been considered as the building blocks of phonological patterns (Kennedy 2016; Mielke 2008). They decode components of individual phonemes that help distinguish them from others, but also express phonemes formally as groups because they share one or more features.

In this language, a set of ten distinctive features are posited, as presented in Table 5. They cover three major categories, including features of the place of articulation ([labial], [coronal], [dorsal], [pharyngeal]), features of the manner of articulation ([voice], [spread glottis], [constricted glottis]), and laryngeal features ([continuous], [sibilant/strident], [delay release]). The binary features are used to specify onset phonemes, with a positive value [+] denoting the presence of a specific feature, and a negative value [−] indicating the absence of a feature.

As indicated in the table, these ten distinctive features form a geometry to characterize the components of individual onset phonemes in this language. For example, the bilabial implosive /ɓ/ can be denoted as [+labial, +voice, +continuant]. They can also effectively group a set of onsets as a natural class because of sharing one or more distinctive features. For example, the onsets /ɓ, ɗ, ɠ, z/ form a natural class because of the common distinctive feature [+voicing]. This feature geometry is important to serve as a base to explore how onset phonemes change their quality at the surface level; what specific features trigger their alternations, and how their alternations can be expressed using rules within the paradigm of generative phonology.

4.1 Palatal-conditioned variation

The high front vowel /i/ and palatal glide /j/ are active to trigger allophonic alternation on syllable onsets of this language. The trigger can be further specified as the [+palatal] feature that is shared by these two front vocoids, and it is further observed affecting the surface of onset phonemes in three different ways, depending on their place of articulation at an underlying level. This is generalized in Table 7.

(a) Alveolar plosives /t/, /tʰ/ and /ɗ/, regardless of their manners of articulation, are found to be dentalized and become [t̪], [t̪ʰ] and [ɗ̪], respectively, in this palatal environment. This is posited because over the articulation, native speakers do not raise their tongue to the alveolar ridge, but rather raising their tongue against the back of upper incisor. For example, the morpheme /ti35/ ‘pig’ is produced as [t̪i35] by native speakers, and the morpheme /tʰi51/ ‘tear’ is uttered as [t̪ʰi51] in real-world utterances. As can be seen, the alternation between /t, tʰ, ɗ/ and [t̪, t̪ʰ, ɗ̪] involves shifting the location of primary constriction from the alveolar ridge to the back of upper incisor. As such, this allophonic process can be expressed as dentalization, which can be characterized using the distinctive feature [-distributed] (Kochetov 2011). Likewise, the three alveolar plosives that participate in this dentalization process can be grouped as a natural class using the distinctive features [+alveolar; -strident]. Correspondingly, this palatal feature-induced dentalization can be formally expressed using Rule (1) as shown below.

(b) Alveolar affricates /ts/ and /tsʰ/ and fricatives /s/ and /z/ are found to surface as alveolar-palatal sounds [tɕ], [tɕʰ], [ɕ], and [ʝ], respectively, in this palatal context. For example, the morpheme /tsi22/ ‘potato’ is pronounced [tɕi22], and the morpheme /zi/ ‘press’ is produced as [ʑi22] in utterances. This can be justified by speakers’ articulatory gesture. They are observed moving their tongue backward to the hard palate to create constriction, rather than maintaining their tongue around the alveolar ridge. As can be seen this alternation between /ts, tsʰ, s, z/ and [tɕ, tɕʰ, ɕ, ʝ] involves shifting the primary place of constriction from the alveolar ridge to the hard palate. As such, this process can be considered as alveolar palatalization and can be interpreted using the feature [-anterior]. The four alveolar affricates and fricatives can also be grouped as a natural class using the distinctive features [+alveolar; +continuous; +anterior]. Correspondingly, the whole alternation can be expressed using Rule (2), as shown below.

(c) Onset phonemes that have a place of articulation after the alveolar ridge, like /k/, /kʰ/, /ɠ/, /ħ/, and /ʔ/, are found to undergo palatalization and surface as [k^j], [kʰ^j], [ɠ^j], [ħ^j], and [ʔ^j], respectively, before the high front vowel /i/, and palatal glide /j/. Because native speakers move their tongue forward to the hard palate, but do not arrive the hard palate. For example, the morpheme /ki22/ ‘flag’ is pronounced as [kʲi22], and the morpheme /ħi22/ ‘fish’ is uttered as [ħ^ji22] ‘fish’ by native speakers. As can be seen, this alternation between /k, kʰ, ɠ, ħ, ʔ/ and [k^j, kʰ^j, ɠ^j, ħ^j, ʔ^j] does not involve shifting the primary place of articulation of related onset consonants, as found in other processes on alveolar onsets, but rather, they acquire a secondary articulation of palatalization because of the [+palatal] feature from their subsequent sounds. As such, the alternation can be characterized as palatalization using the feature [+palatal], and the five phonemes can be grouped as a natural class using the features [−bilabial; −alveolar]. Corresponding, the whole alternation can be expressed using Rule (3), as shown below.

As discussed above, all onset phonemes, except labial plosives /p, pʰ, ɓ/, are easily affected by the palatal feature from their subsequent sounds and change their forms at the surface level. Labial onsets are exceptional to an alternation in this context, and this may be ascribed to their different articulator at lips, while the palatal-inducing process dominantly affects the tongue, as generalized by Gordon (2016), based on the 100-language WALS sample. In addition, it can be observed that, the alternations with alveolar onsets exclusively involve shifting the location of the primary constriction to either before the alveolar ridge as in dentalization for alveolar plosives /t, tʰ, ɗ / or after the alveolar ridge as in the alveolar palatalization for alveolar affricates and fricatives /ts, tsʰ, s, z/. In contrary, the alternations with onsets that are non-labial and non-alveolar exclusively involve superimpositions of a secondary gesture of palatalization. As such, it can be generalized that, the impact of the palatal feature from vocalic sounds on syllable onsets is asymmetrical in this language, and this is dominantly determined by the place of articulation of target onsets at the underlying level. There appears to be a tendency that alveolar sounds are more likely to be affected by the palatal nature of surrounding sounds and are more likely to shift their primary place of constriction because of the impact.

4.2 Labial-conditioned variation

The high back round vowel /u/ and the labial-velar glide /w/ appear to be the most active to trigger allophonic variations on onsets of Zhangzhou. The trigger can be further specified as [+labial], which is discovered impacting syllable onsets of this language in six different ways, depending on the nature of target onsets. This is generalized in Table 8.

(a) All onset phonemes surface a labial feature before the vowel /u/ or the labial glide /w/. This is because native speakers are found rounding their lips over their articulation of any onset consonant in this labial environment, such as the morpheme /ku35/ ‘beetle’ is pronounced as [kʷu35], the morpheme /kʰwɐ35/ ‘boast’ is produced as [kʰʷwɐ35], and the morpheme /ʔu33/ ‘have’ is uttered as [ʔʷu33] by speakers. This phenomenon is articulatorily understandable and is phonologically predictable, because it is a natural process for onset consonants to acquire a secondary articulation of labialization due to the labial nature of their immediately subsequent sounds. As such, the alternation can be characterized as labialization and expressed using the Rule (4), as shown below. It should be noted that, the three labial plosives are already characteristic of the [+labial] feature themselves. It is thus redundant to impose a labialization effect on their surface form.

(b) The labial implosive /ɓ/ and the velar implosive /ɠ/ are found to separately become a labial voiced fricative [β] and a labialized velar voiced fricative [ɣʷ] in this labial context. Because no complete constriction can be observed from speakers’ articulatory gesture, and the airflow from their lungs can be continuously perceived over the production of related tokens. Such as the morpheme /ɓu33/ ‘frost’ is pronounced as [βu33], and the morpheme /ɠu22/ ‘cow’ is produced as [ɣʷu22] by speakers. As can be seen, the alternation between /ɓ, ɠ/ and voiced [β, ɣʷ] involves shifting the airstream mechanism from glottalic ingressive to pulmonic egressive and changing the manner of articulation from implosive to fricative. As such, this process can be considered as lenition, which is also known as spirantization or fricativization in literature (Gordon 2016; Kennedy 2016). In addition, the voiced velar onset acquires the labial feature as a secondary feature from its immediately subsequent sound. In other words, there are two different processes of lenition and labialization occurring on non-alveolar implosive onsets. As such, the lenition process can be characterized using the distinctive features [+continuous; +strident], and the labial and velar implosives /ɓ, ɠ/ can be grouped as a natural class using the features [−alveolar; −strident; +voiced], while the whole alternation can be formally expressed using Rule (5), shown below.

(c) The alveolar implosive /ɗ/ is observed to surface as a labialized approximant [lʷ] before the labial segments /u/ or /w/. For example, the morpheme /ɗu51/ ‘female’ is produced as [lʷu5], and the morpheme /ɗwɐ22/ ‘spicy’ is uttered as [lʷwɐ22] by speakers. This occurs because speakers move the central part of their tongue, rather than the tongue tip, nearly approaching but not touching the alveolar ridge to form a constriction that is slightly wider than the constriction needed for fricative sounds. Meanwhile, the airflow can be continually perceived passing through the constriction, while speakers’ lips are rounded over the production. As can be seen, the alternation between /ɗ/ and [lʷ] involves shifting the airstream mechanism from glottalic ingressive to pulmonic egressive, altering the manner of articulation from plosive to approximant, changing the active articulator from the tongue tip to the central part of the tongue, and acquiring a secondary articulation of labialization from subsequent sounds. The allophonic alternation of this alveolar implosive in this labial context can thus be considered as a compounding effect of lenition, laminalization, and labialization, and can be formally expressed using Rule (6), as shown below.

(d) The voiceless alveolar plosives /t/ and /tʰ/ are observed to separately surface as [t̻ʷ] and [t̻ʰʷ] in this labial context. This is because native speakers raise up their tongue blade, rather than their tongue tip, to form an oral obstruction at the alveolar ridge, while their lips are rounded over the production. For example, the morpheme /tu35/ ‘pile, heap’ is pronounced as [t̻ʷu35], while the morpheme [tʰu41] ‘to shovel’ is produced as [t̻ʰʷu41] by speakers. As can be seen, the alternation between /t, tʰ/ and [t̻ʷ, t̻ʰʷ] mainly involves shifting the active articulator from the tongue tip to the tongue blade and acquiring a secondary articulation of labialization. The alternation can thus be considered as a compounding effect of laminalization and labialization, which can be characterized using the features [+laminal; +labial]. The two alveolar plosives /t/ and /tʰ/ can also be grouped as a natural class of the [+alveolar; −voiced; −continuous] features. Correspondingly, their allophonic alternation in this labial context can be formally expressed using Rule (7), as shown below.

(e) The alveolar fricatives /s/ and /z/ and affricates /ts/ and /tsʰ/ are found to surface as palatal-alveolar sounds [ʃʷ], [ʒʷ], [tʃʷ], and [tʃʰʷ], respectively, in this labial setting. This is because speakers are observed to move their tongue body from the alveolar ridge towards the hard palate to form a constriction, while rounding their lips. For example, they produce the morpheme /su22/ ‘word’ as [ʃʷu22] and the morpheme /zu22/ ‘such as’ as [ʒʷu22]. As can be seen, the alternation between /s, z, ts, tsʰ/ and [ʃʷ, ʒʷ, tʃʷ, tʃʰʷ] involves shifting the location of primary constriction from the alveolar ridge to palatal-alveolar and acquiring a secondary articulation of labialization. The alternation can thus be considered as a compounding effect of alveolar-palatalization and labialization and can be interpreted as [−anterior; +labial]. The four alveolar onsets /s, z, ts, tsʰ / can be grouped as a natural class for sharing distinctive features [+alveolar; +distributed; +strident]. As such, their allophonic alternation can be expressed using Rule (8), as shown below.

(f) The pharyngeal fricative /ħ/ is found to surface as a labialized glottal fricative [hʷ] before the labial segment /u/ or /w/. This is because speakers reported they made a partial constriction at the glottis, rather than in the pharyngeal cavity, while rounding their lips over the production of related tokens. For example, they produce the morpheme /ħu22/ ‘help’ as [hʷu22], and the morpheme /ħwɐ35/ ‘flower’ as [hʷwɐ35]. As can be seen, the alternation between /ħ/ and [hʷ] involves shifting the location of primary constriction from the pharyngeal cavity to the glottis and acquiring a secondary articulation of labialization. The allophonic alternation of this onset can thus be considered as a compounding effect of glottalization and labialization, and can be expressed using Rule (9), as shown below.

As discussed above, the allophonic alternations of syllable onsets in this language are much more complex in the labial context than in the palatal context. It can be observed that the labial-inducing impacts on onset phonemes are diverse, which include (a) changing the airstream mechanism, such as from the glottalic ingressive to pulmonic egressive as found in implosive onsets; (b) changing the manner of articulation, such as from implosive to fricative as found in /ɓ, ɠ/, or from implosive to approximant as found in /ɗ/; (c) shifting the active articulator, such as from the tongue tip to the central part of the tongue as found in /ɗ/, and from the tongue tip to the tongue blade as found in /t, tʰ/; (d) shifting the location of primary constriction, such as from alveolar ridge to palatal-alveolar as found in /s, z, ts, tsʰ/, and from the pharyngeal cavity to the glottis as found in /ħ/; and (e) acquiring a secondary articulation of labialization as found in all onsets except the labial plosives, because they have already had the labial feature on their own. Likewise, the types of allophonic processes are also diverse, which include labialization, lenition, laminalization, alveolar-palatalization, and glottalization. In addition, the allophonic processes mostly do not operate on their own, instead, they compound with other process (es) to impact the surface forms of target onsets. For example, the alternation between the alveolar implosive /ɗ/ and the labialized approximant [lʷ] involves lenition, laminalization, and labialization. In general, except the voiceless labial plosives, the two voiceless velar plosives /k, kʰ/ are the least affected in the labial context, because their alternation only involves acquiring a secondary articulation of labialization.

4.3 Nasality-conditioned variation

The nasal vowels /Ṽ/and syllabic nasals /N̩/ can also induce allophonic alternations on onset phonemes of this language. The trigger can be specified as the [+nasal] feature as shared by these two types of nuclei. The alternations mainly manifest in three different ways, depending on the nature of target onsets. This is generalized in Table 9, in which the symbol * indicates no empirical data were tested in the field.

(a) Implosive onsets /ɓ/, /ɗ/, and /ɠ/ are observed to surface as their corresponding homorganic nasal plosives [m], [n], and [ŋ], respectively, in a nasal context. Because speakers are found to release the airflow through their nose, rather than through their mouth, when they produce these onsets before nasal vowels. For example, they pronounce the morpheme /ɓĩ33/ ‘noodle’ as [mĩ33], and the morpheme /ɗɔ̃33/ ‘two’ as [nɔ̃33]. It should be also noted that implosive onsets are not attested to occur before syllabic nasals in this language. As can be seen, the alternation between /ɓ, ɗ, ɠ/ and [m, n, ŋ] involves shifting the manner of articulation from oral to nasal. It thus can be considered as an effect of nasalization. The three implosive onsets can be grouped as a natural class, using the features [+voiced; -strident] and the nasalization process can be formally expressed using Rule (10).

(b) The pharyngeal fricative /ħ/ is observed to surface as a prenasalized breathy onset with three different forms that include [^mɦ], [ⁿɦ], [^ŋɦ], depending on the nature of its subsequent nasal segments. This occurs because speakers reported they moved a constriction from the pharynx to the larynx where they continuously generated airflow from lungs and let them freely pass through their vocal tract. They also reported before the articulation of this onset, their velum had already been lowered as that the airflow could also rush out through their nose. For example, they produce the morpheme /ħɛ̃41/ ‘rest’ as [ⁿɦɛ̃41], the morpheme /ħm̩41/ ‘beat’ as [^mɦm41], and the morpheme /ɦŋ̩35/ ‘recipe’ as [^ŋɦŋ̩35]. As can be seen, the alternation between the pharyngeal fricative /ħ/ to three forms of prenasalized breathy onset [^mɦ, ⁿɦ, ^ŋɦ] involves shifting the location of primary constriction from pharynx to larynx, creating voicing for the airflow, and acquiring an articulatory gesture of nasalization before onsets. The alternation can thus be considered as a compounding effect of glottalization, voicing, and pre-nasalization. The three surface forms differ in the place where a constriction is created in the oral cavity for the articulation of subsequent nasal sounds. Specifically, the form [^mɦ] occurs before the labial syllabic nasal, the form [ⁿɦ] occurs before nasal vowel, while the form [^ŋɦ] occurs before the velar syllabic nasal. As such, the allophonic alternations of this onset can be formally expressed using Rule (11).

(c) All other onsets that are non-voiced and non-pharyngal are found to surface a nasal feature in this nasal context. This is because before the articulation of onset consonants, speakers are observed to lower their velum and get an articulatory gesture ready for a nasal production. For example, they produce the morpheme /tɛ̃33/ ‘press’ as [ⁿtɛ̃33], the morpheme /pĩ 35/ ‘weave’ as [^mpĩ33], and the morpheme /kɐ̃51/ ‘dare’ as [^ŋkɐ̃51]. As can be seen, the alternation involves acquiring an additional articulatory gesture of nasalization before onset articulation and assimilating the place of nasal articulation to that of the target onset. For example, the alveolar plosive /t/ surfaces as [ⁿt], the labial plosive /p/ surfaces as [^mp], and the velar plosive /k/ surfaces as [^ŋk]. As such, the alternation can be considered as a prenasalization as an impact of the nasal feature from subsequent nasal vowels and syllabic nasals. It can be expressed formally using Rule (12).

As discussed in this section, the allophonic alternations of syllable onsets are incredibly active in this language. Twelve different types of alternation processes are generalized from empirical data and are formally expressed using rules within the paradigm of generative phonology. They manifest varying impacts of featural properties of triggers from the vocalic segments, but also reflect varying reactions of onset phonemes in different phonological environments. In general, the labial-inducing alternations are much more complex and diverse that account for half of the total alternation types and involve changing the airstream mechanism, manner of articulation, active articulator, the location of primary constriction, and acquiring a secondary articulation. In most cases, the individual process compounds with other process(es) to impact the surface of onsets. The nasal-inducing alternations are less complex that involve changing the manner of articulation from oral to nasal, shifting the location of primary constriction, and generating pre-nasalization. However, the nasal-inducing alternations can cause certain onsets to have more than one surface form, for example, the pharyngal fricative onset /ħ/ has three surface forms [^mɦ], [ⁿɦ], and [^ŋɦ], so does the glottal stop with three variants [^mʔ], [ⁿʔ], and [^ŋʔ]. The palatal-inducing alternations seem the least complex that mainly involve shifting the location of primary constriction on alveolar onsets or acquiring a secondary articulation of palatalization on velar onsets, while labial onsets are not affected in the palatal context.

As for the onset phonemes, they suffer varying impacts from their immediately subsequent sounds. Alveolar and velar onsets, except the voiced fricative /z/, have four surface forms; the pharyngeal fricative /ħ/ and the glottal stop /ʔ/ both have as many as six variants at the output; the labial implosive /ɓ/ has three forms while the other two voiceless labial plosives only have two surface forms. This suggests that the voiceless labial stops are the least affected by their subsequent sounds, while onsets with a place of articulation outside the oral cavity are the most valuable to the featural properties of their subsequent sounds. In addition, voiced onsets are more likely to change the manner of articulation as an impact of their following segments, while voiceless onsets have different reactions. In general, voiceless alveolar onsets are more likely to shift the location of primary constriction, change an active articulator, and/or acquire an additional secondary feature, while voiceless velar onsets largely involve acquiring a secondary feature of articulation. Voiceless labial onsets are the least easy to be affected but if yes, they only involve acquire an additional secondary feature. As indicated, the featural impacts of vocalic segments appear to be most severe on alveolar sounds that involve the engagement of the front of the tongue in this language. This tends to be a cross-linguistic tendency, as Gordon (2016) observed among 100-language WALS sample. It may be ascribed to the fact that the front of tongue executes faster gestures than both dorsum and lips, as such it is more prone to be overlapped by surrounding sounds (Gordon 2016).

5.1 Usage frequency of Zhangzhou onsets

Tracing back to the Eastern Han Dynasty (25–220 AD), Chinese scholars had devised the method of Fanqian to document the pronunciation of Chinese Characters, which were monosyllabic, through using two other characters (Dong 2023). The first character represents the initial of the character being referred, which corresponds to a syllable onset in modern linguistics, while the second character represents its final, which refers to the constituent that covers all syllable components except onset. In other words, within the Sinitic convention, individual syllables have long been considered to incorporate two parts of Initial and Final (Dong 2023; Duanmu 1990, 1999; Huang 2021; Li and Yao 2008; Třísková 2011; Zhang 2006). This fanqie approach has thereafter been prevalent as an effective tool to indicate speech sounds of Chinese languages over history. During the Six Dynasties period (220–589 AD), a series of rhyme dictionaries were compiled using this method, such as Lu Fayan’s Qieyun, which are regarded as the earliest and most direct reference to the study of the phonological system of the Early Middle Chinese and above (Dong 2023). In these dictionaries, each rhyme can be seen as a range of rhyming characters for poetry and prose compositions, and characters sharing the same pronunciation are listed together. As such, rhyme dictionaries contain all possible characters being practically used. For example, there are about 12,000 characters in Lu Fayan’s rhyme book Qieyun. They also contain the maximum types of initials and finals, as well as their combinations to formulate monosyllabic characters.

This study adopts the Sinitic convention of syllabification and rhyme dictionary to examine the phonotactics of syllable onsets in Southern Min. Each syllable is divided into two parts of Initial that corresponds to onset and Final that covers all other syllable components. This is hierarchically represented in Figure 4. In total, 15 syllable onsets (initials) are contrastive, as described in Section 3. Likewise, 61 finals can be generalized from the empirical data, as summarized in Table 10 with examples. They can be four types: V, GV, VX, and GVX, in which V covers oral vowels, nasal vowels, and syllabic nasals, while X covers nasal codas, obstruent codas and glide codas. The 15 onsets and 61 finals will serve as an important base to explore how the onset phonemes of this language are distributed in relation to different finals (or final types), and how their combinations may be constrained at the syllable level.

Figure 4:

Syllable modelling in this study.

The phonotactic exploration in this study is data driven. It is built upon 61 rhyme tables that Huang (2019) adopts the conventional notion of rhyme dictionary to tabulate combinations of individual onsets, finals, and tones to generate syllables in this language. Table 11 illustrates the rhyme table under the final /ɐ/, while Table 12 illustrate the rhyme table under the final /ɐ̃w/. The top row indicates the final being addressed; the second rows list the eight tones in sequence while the leftmost column lists the 15 onset phonemes. Each slot in the table represents a combination of an onset, a final, and a tone to generate a syllable in this language. If a syllable is empirically attested, a represented character is put in that correspondingly represents one or more monosyllabic morpheme(s). For example, the syllable /kɐ35/ in Table 11 is attested because it can generate lexical morphemes like ‘glue’ in Chinese character 胶 ( jiāo ). Likewise, the syllable /ɓɐ̃w33/ in Table 12 can generate a lexical morpheme like ‘appearance’ in Chinese character 貌 ( mào ). In contrast, if a slot is left empty, indicating that its related syllable is theoretically possible to occur, but it is not attested in the field. For example, the syllable /kɐ22/ is not attested, because it is not related to any of morphemes in this language.

As can be seen in these two tables, the arrangement of rhyme table is an effective way to present what syllables are attested within a particular final, and what syllables are not attested but they are theoretically possible to occur. A whole set of 61 rhyme tables thus delivers a full inventory of syllables that are empirically used by the speech community of this language. It is also an effective way to present the occurrence of individual phonemes at the syllable level and show their combinability with other syllable components. If a syllable fails to obtain any lexical morpheme, as shown in an empty slot, it indicates the combination of a specific onset, final and tone is subject to phonotactic constraint(s). In contrast, if a syllable is able to obtain lexical morphemes, it indicates this onset is combinable with the related final and tone to make lexical items, and there is no constraint governing their sequencing. As such, the usage frequency of individual onset phonemes can be quantified by calculating the number of syllables that can be obtained under the target onsets. Likewise, the phonotactic constraints on syllable onsets can also be interpreted by calculating the number of syllables that are theoretically permissible but are not obtained in the field.

Assuming this language has an inventory of 15 onsets, 61 finals, and 8 tones, each onset is theoretically possible to generate 488 syllables (=61 finals × 8 tones), and a total of 7,320 syllables (=15 onsets × 61 finals × 8 tones) are logically possible to be used by the speech community. However, the usage frequency of onset phonemes in real-world situations is significantly lower than the theoretically expected. This can be seen in Table 13. It shows the number of empirically attested syllables under individual onset phonemes, as shown in the column of syllable, and the discrepancy between empirically obtained syllable number and the theoretically expected number, as shown in the column of gap. They are calculated based on the 61 rhyme tables that Huang (2019) constructed. In addition, Figure 5 visualises the results.

Figure 5:

Usage frequency of onset phonemes in Zhangzhou.

As can be seen, individual onset phonemes show great variations in their usage frequency in this language. The alveolar unaspirated affricate /ts/ appears to be the most frequently used with 189 syllables obtained, followed by the glottal stop with 188 syllables obtained in the field. The voiced alveolar fricative /z/ has the lowest usage frequency that only generates 33 syllables. However, in comparison with the number of 488 that each onset is theoretically expected to obtain, there is a great discrepancy. For example, the number of syllables that the onset /ts/ generates is 299 fewer than the expected number, while the number of syllables that the onset /z/ generates is exactly 455 fewer than the theoretical assumption. The great divergency between the expectation and the reality implies that the usage frequency of onset phonemes and their combinability with finals and tones have been substantially constrained at the syllable level. As a matter of fact, only 2105 syllables are attested in the field, and as many as 5215 syllables show gaps in the rhyme tables. This signifies that more than 71% of theoretically possible syllables fail to be employed by native speakers in communication. The rest of this section will be devoted to exploring how the combinability of onsets with other syllable components are constrained and what factors trigger the constraints.

5.2 Onset-final occurrence constraints

Phonotactic constraints occur on the alignments of onset phonemes and finals in Southern Min. Table 14 calculates the number of syllables that are obtained in relation to individual onset-final combinations. For example, in the pɐ combination, 3 syllables are obtained which bear three different tones, while in the zɐ combination, no tokens are tested in the field. Based on the data shown in this table, several generalizations can be made as to how the onset-final occurrence is constrained to formulate syllables in this language.

One of the most significant factors that affect the usage frequency and combinability of onsets can be ascribed to the nasality feature from nuclei in Southern Min. As discussed in previous sections, the nasality feature is not employed phonemically to construct nasal onsets in this language, but the existence of nasal vowels and syllabic nasals can induce diverse phonological processes and change the surface forms of onset phonemes at the output. Similarly, the nasality feature from the nucleus system can also affect the usage frequency of onset phonemes and their combinability with other components at the syllable level. This can be seen in Table 14, no tokens are found to contain syllables like *ɓN, *ɠN, and *zN in the field, and only one lexical item has a voiced onset before a syllabic nasal, which is /ɗŋ22/ ‘small taro’, but it is not extensively used in the community. As such, it can be generalized that there is a phonotactic constraint on the alignment of the voicing feature of onset consonants with the nasality feature from syllabic nasals. This is because voiced onsets dominantly cannot occur before syllabic nasals, whereas voiceless onsets can occur before syllabic nasals. For example, all voiceless onsets can combine with the syllabic velar nasal /ŋ̩/ to formulate monosyllabic morphemes, such as /pŋ̩51/ ‘list of names’, /pʰŋ̩51/ ‘ebb tide’, /tŋ̩35/ ‘wait’, /tʰŋ̩41/ ‘hot’, /sŋ̩51/ ‘play’, /tsŋ̩35/ ‘village’, /tsʰŋ̩22/ ‘bed’, /kŋ̩41/ ‘steel’, /kʰŋ̩41/ ‘hind’, /ħŋ̩35/ ‘prescription’, and /ʔŋ̩35/ ‘centre’. The labial syllabic nasal /m̩/ can only combine with two onsets /ħ/ and /ʔ/ which are both voiceless, such as /ħm̩41/ ‘hit with a stick’ and /ʔm̩35/ ‘drink’.

There shows less restriction on the alignment of onset phonemes with nasal vowels. Because implosive onsets can occur before nasal vowels at an underlying level, such as syllables like ɓṼ (/ɓĩ33/ ‘noodle’), ɗṼ (/ɗɔ̃33/ ‘two’), and ɠṼ (/ɠɔ̃51/ ‘five’) are all attested in the field, but it should also be noted that, these implosive onsets surface as their homorganic nasal stops /m, n, ŋ/ at the output. The voiced fricative /z/ cannot occur before nasal vowels, as no lexical items contain syllables like *zṼ, *zGṼ, *zṼG, *zGṼG. However, although implosives are allowed to combine with nasal vowels, the number of syllables that they can generate is not significant. As can be seen in the table, only 32 syllables are obtained before the Ṽ type, 26 before the GṼ type, 10 before the ṼG type, while only 1 syllable is attested before the GṼG type.

There seems no restriction on the alignment of onset phonemes with the nasality feature form syllable codas. This can be seen in the Table 14. Not only voiceless onsets, but also woiced onsets can occur before finals with nasal codas, for examples, syllables like ɓVN (/ɓɐn51/ ‘mosquito’), ɓGVN (/ɓjɐn51/ ‘waive’), ɗVN (/ɗɐm22/ ‘male’), ɗGVN (ɗjɐm22/ ‘sticky’), ɠVN (/ɠɐn51/ ‘eye’), ɠGVN (/ɠjɐn41/ ‘attractive’), zVN (/zim51/ ‘to tolerate’), and zGVN (/zjɐm51/ ‘infect’) are all attested to construct lexical items in this language. As such, it can be generalized that the phonotactic constraint on onset phonemes with the nasality feature from other syllable components dominantly operates on their alignments with nasal nuclei and does not affect their alignments with nasal codas.

Another significant factor that affects the usage frequency of onset phonemes in this language can be ascribed to the phonotactic constraint on the occurrence of the labial feature within syllables. As can be seen in the table, all labial onsets /p, pʰ, ɓ/, regardless of their manner of articulation, cannot combine with finals of a labial coda. This is because no lexical items can be attested having syllables like *Pɐm; *Pim; *Pɔm; *Pjɐm; *Pɐp; *Pip; *Pɔp; *Pjɐp, which involve as many as 24 combinations (=3 labial onsets × 8 finals). In contrast, these labial onsets can occur before finals with a non-labial coda, as can be seen in morphemes like /pɐn22/ ‘bottle’, /pin22/ ‘side’, /pɐŋ22/ ‘bedroom’, /pjɐn41/ ‘build’, /pɐt221/ ‘another’, /pik41/ ‘force’, and /pjɐt221/ ‘shoot’. Likewise, non-labial onsets can occur before finals with a labial coda, as can be seen in morphemes like /sim35/ ‘heart’, /sɐm35/ ‘shirt’, /sɔm35/ ‘frost’, /sjɐm22/ ‘hug’, /sɔp42/ ‘tie up’, /sip41/ ‘humid’, and /tsjɐp41/ ‘connect’. It thus can be generalized that the labial feature is subject to an OCP (Obligatory Contour Principle) constraint that disallows it to co-occur at both onset and coda positions. Because of this restriction, a considerable number of syllables that are theoretically possible to contain labial onsets and labial codas are not able to be attested in the field. Such a labial restriction is also reported in other Sinitic languages, such as Wu (Zhang 2006) and Cantonese (Kirby and Yu 2007).

Additionally significant factors that can create phonotactic constraints on onset phonemes in this language come from the final type and the complexity of finals. The leftmost column in Table 14 lists the types of finals in accordance with the nature of their nuclei and coda, and the occurrence of prevocalic glides. As can be seen, onset phonemes tend to generate less syllables with a nasal final type (Ṽ and N) than with an oral final type (V). This can be seen that a total of 402 syllables are obtained under the V type, while only134 syllables are obtained under the Ṽ type and 42 syllables under the N type. Likewise, onset phonemes generate less syllables with finals ending in glide and obstruent codas (VG and VP) than with finals ending in nasal codas (VN). This can also be seen that a total of 421 syllables are obtained under the VN type, while only 162 syllables are obtained under the VP type and 109 under the VG type.

In addition, the complexity of finals can also impact the usage frequency of onset phonemes in this language. Voiceless onsets largely cannot combine with the GṼG final to form syllables. This is because no lexical items are found to contain syllables like *pGṼG, *pʰGṼG, *tGṼG, *tʰGṼG, *kGṼG, *kʰGṼG, *tsGṼG, and *tsʰGṼG. Only the voiceless alveolar fricative /s/, pharyngeal fricative /ħ/ and the glottal stop /ʔ/ can occur before this final, which separately generate 2, 1, and 2 syllables, such as /swɐ̃j22/ ‘mango’, /swɐ̃j22/ ‘sound of spinning movement’, /ħwɐ̃j22/ ‘chubby status’, and /ʔwɐ̃j41/ ‘wrench’. Implosive onsets are able to occur before the GṼG final, such as ɓGṼG (/ɓjɐ̃w22/ ‘copy; describe’), ɗGṼG (/ɗjɐ̃w22/ ‘cat’), and ɠGṼG (/ɠjɐ̃w35/ ‘itchy’), but only 7 syllables can be obtained. As a total, only 12 syllables are able to contain this final. As can be seen onset phonemes have the lowest combinability with this final and this can be ascribed to the complexity of this final that contain nasal vowels and prevocalic and postvocalic glides.

5.3 Onset-tone occurrence constraints

Phonotactic constraints also occur on the alignment of onset and tone in this language. Table 15 calculates the number of syllables that are obtained with respect to individual onset-tone combinations. For example, in the p-T1 slot, it presents 22 syllables that are attested to construct morphemes in this language. The letters I, II, III, and IV separately correspond to Ping, Shang, Qu, and Ru tone in terms of the Middle Chinese tonal category; while the letters a and b represent the Yin and Yang registers, respectively, which are determined by the voicing status of syllable onsets in the historical time. As can be seen, several significant characteristics can be generalized to account for the onset-tone occurrence constraints.

The type of syllable codas that tones are assigned to can substantially constrain the usage frequency of onset phonemes in this language. It can be seen in the table that onsets are consistently less frequently used in Tones 6, 7, and 8. For example, onsets in Tone 1 can generate as many as 392 syllables, whereas in Tone 6 can only generate 128 syllables. In Tone 3, onsets can generate 352 syllables, but in Tone 7, only 117 syllables are obtained. Likewise, in Tone 4, onsets can generate 360 syllables, but the number is reduced to 98 in Tone 8. As has been shown, the number of empirically attested syllables is consistently above 300 across Tone 1 to Tone 5, but it is dropped to around 100 among Tones 6, 7, and 8. This reduction can be ascribed to the diachronic constraint on assigning syllable codas in relation to tones. Tone 6 and Tone 7 both belong to Ru tones in the MC tonal categories, known as stopped tones, which diachronically request their syllables to end in obstruent codas (Huang 2018, 2021). Tone 8 is a new tone that emerges from those syllables that are conventionally transcribed with a glottal stop coda in the MC Yangru tone, but the glottal stop coda is found to have undergone deletion in the synchronic speech, because of this, its related syllables have become open, and a new tone is evoked (Huang 2018). On the contrary, all other tones (1–5) belong to non-Ru categories, known as unstopped tones, which request their syllables to end in sonorant segments, including zero codas, glides, and nasal stops. As can be generalized, the strict restriction on syllable coda type in relation to tones can negatively impact the usage frequency of onset phonemes and their combinability with tones to generate syllables in conversations.

The type of tonal registers can substantially impact the usage frequency of onset phonemes and the onset-tonal combinability in this language. It can be seen in the table that the number of syllables that are obtained in Yin registered tones is consistently greater than in Yang-registered tones. For example, a total of 392 syllables are obtained in the Yinping Tone 1, higher than that of 342 in the Yangping Tone 2. Likewise, a total of 360 syllables are attested in the Yinqu Tone 4, higher than that of 315 syllables in the Yangqu Tone 5. Similarly, in the Yinru Tone 6, 128 syllables are obtained, higher than that of 118 syllables in the Yangru Tone 7. The asymmetrical distributions of onset-tonal combinability in the generation of syllables can be ascribed to the diachronic constraint of assigning tonal registers with respect to the voicing status of syllable onsets. Yin-registered tones are historically assigned to syllables with voiceless onsets, while Yang-registered tones are assigned to syllables with voiced onsets. From a synchronic perspective of phonology, voiced onsets are more marked than voiceless onsets because they request an additional articulation of vocal fold vibration. As such, it is reasonable to obtain less syllables in Yang-registered tones than in Yin-registered tones. Also, it can be generalized that the type of tonal registers can impact the onset-tonal combinability in the generation of syllables for communication purposes.

The nature of onset phonemes themselves can also constrain the onset-tonal combinability and impact their usage frequency. As can be seen in the table, a total of 474 syllables are obtained with alveolar stops /t, tʰ, ɗ / across all tones, which is significantly higher than that of 383 with velar stops /k, kʰ, ɠ/ and that of 350 with labial stops /p, pʰ, ɓ/. This reflects that the feature of the place of articulation can affect the usage frequency of onset phonemes in generating linguistic entities. It also reflects the higher usage frequency of the alveolar place of articulation in speech communications. In addition, it can be observed that, a total of 482 syllables are obtained with voiceless unaspirated stops /p, t, k/ across all tones, higher than that of 377 with voiced stops /ɓ, ɗ, ɠ/ and that of 348 with voiceless aspirated stops /pʰ, tʰ, kʰ/. This reflects that the manner of articulation can also impact the onset-tonal combinations to generate syllables in this language. It also reflects that the higher usage frequency of voiceless unaspirated stops over their aspirated and voiced counterparts. In addition, it shows the feature of aspiration appears to be more marked than the feature of voicing because the former generate less syllables.

As discussed in this section, the usage frequency of onset phonemes and their combinability with other syllable components are subject to severe constraints in this language. This leads to more than 71% of theoretically permissible syllables to fail to be attested and used by native speakers. Only 2105 syllables are obtained in the field. Built upon a large corpus of 61 rhyme tables, this study quantifies the usage frequency of individual onset phonemes with respect to finals and tones and explores what factors have constrained their combinability in the generation of linguistic entities for communication purposes. It shows that the asymmetrical distribution of the nasality feature, the strict restriction of the occurrence of the labial feature, the type of finals and their complexity can all pose phonotactic constraints on the onset-final combinations. For example, voiced onsets dominantly cannot occur before syllabic nasals. Only a small number of syllables can have voiced onsets before nasal vowels. Labial onsets cannot co-occur with labial codas within the same syllables. Onsets are less frequent to combine with finals with marked features, such as nasal vowels, syllabic nasals, obstruent codas, and glide codas than combining with finals with ordinary features, such as oral vowels or nasal codas. Onset phonemes can also reduce their usage frequency when combining with finals with complex internal structures. For example, only 7 syllables are attested to contain the GṼG final type.

This exploration also reveals phonotactic constraints on onset-tonal combinations to generate syllables. It shows that the diachronic requirements on assigning tones with respect to syllable codas and onsets can negatively impact the usage frequency of onset phonemes and the number of empirically attested syllables. Specifically, onsets are less frequently used in tones that are assigned to syllables ending in obstruent codas, and in tones that are assigned to syllables with voiced onsets in the historical time. This is directly reflected in the number of syllables that is attested to be around 100 in stopped Tones 6 and 7 relating to obstruent-ending syllables, and in Tone 8 emerging from the synchronic deletion of glottal stop codas, while the number is above 300 in other tones relating to sonorant-ending syllables. Likewise, the number of syllables in Yin registered tones that are historically related to voiceless onsets is consistently greater than in Yang-registered tones that are related to voiced onsets. The nature of onset phonemes themselves can also influence their combinability with tones. In general, onsets with an alveolar place of articulation are more frequently used than onsets with velar and labial places of articulation. Similarly, onsets with a voiceless and unaspirated manner of articulation are more often used than onsets with other manners of articulation. Onsets with the feature of aspiration are less frequently used than onsets with a voiced feature.