Phonotacticon: a cross-linguistic phonotactic database

4.1 Phonemic inventory

The phonemic inventory part of each lect’s profile contains the segmental phonemes of the lect. Since Phonotacticon is a phonological database and not a phonetic database, it only lists phonemes as members of the phonemic inventory, excluding its possible allophones.

One challenge of transcribing phonemes using the International Phonetic Alphabet (IPA) is the overspecification of articulatory features. A phoneme is “a combination of certain (simultaneous and/or successive) features, leaving other features unspecified” (Chao 1934: 373). The IPA, on the other hand, is an alphabet representing the articulatory possibilities of human speech. As such, every IPA consonant symbol specifies the articulatory place, manner, and phonation of the consonant it represents. The problem arises when a phonologist has to transcribe the nasal phonemes of a lect using the symbols <m n ŋ>, which represent voiced nasal consonants, even when the lect do not specify the feature [+voice] in their nasal phonemes. Thus, using <m n ŋ> to transcribe nasal phonemes that do not specify voicedness is technically an overspecification. Transcribing nasal consonants as <m n ŋ> may be based on the fact that they are articulated as voiced in most environments, but that would be transcribing their (most common) phonetic surface forms and not the combinations of their phonological features, which is the definition of a phoneme. As van der Hulst (2017) puts it, “IPA symbols are mere shorthand for feature representations” (p. 41) and not equivalent to the feature representations. Nevertheless, for pragmatic purposes, every phoneme is defined as an IPA symbol in Phonotacticon.

Most of the time, a phoneme is described in the consulted literature as having a single underlying form that can be transcribed as an IPA symbol. But rarely, a phoneme is described as consisting of several allophones, without a single underlying form. In such cases, we judge which allophone represents the underlying form based on which one appears in the most unmarked environments. If there is a form in isolation, then that form is chosen as the representing segment. An example is Japanese moraic nasal /N/, which may occur as [n], [mː], [nː] or others depending on phonotactic context (Iwasaki 2013). /N/ occurs as [nː] when it does not precede any segment (e.g. san さん [sɑ̃nː] ‘three’), so we have transcribed it as /nː/.

If there is absolutely no reason to prefer one allophone over another, then we choose the one that appears first in the cited literature. For example, if a phoneme is described as <s/ʃ>, without specifying whether /s/ or /ʃ/ represents the underlying form, and there is no reason to believe one is more unmarked than the other, then it is transcribed in Phonotacticon as <s>.

Archiphonemes – phonemes that have other phonemes as their allophones – are generally treated as equivalent to their allophonic phonemes. Tuvinian [tuvi1240] archiphoneme /I/ can be realized as /i/, /y/, /ɯ/, or /u/, all of which are phonemic in Tuvinian, based on vowel harmony: /àt^h-I/ > [àtɯ] ‘his horse’; /k^hyç-I/ > [k^hyʝy] ‘his strength’ (Anderson and Harrison 1999: 4). In this case, /I/ is treated as equivalent to the phonemes /i y ɯ u/, without being coded as a separate phoneme.

Another problem to be addressed is the xenophone, a phoneme that only occurs in loanwords (coined by Eklund and Lindström 1998). The main complication is that the status of a xenophone can vary from fully nativized to extremely marginal and it can sometimes be difficult to judge whether a xenophone is truly a part of a given lect’s phonology. Some xenophones are indistinctively part of a lect’s phonology, such as German /ʒ/, while some xenophones are distinctively foreign, such as German nasal vowels, which remain unstable and are often replaced by native phonemes (Wiese 2000: 12). Thus, whether a xenophone forms a part of the phonology of a lect is essentially a gray area, leaving us with the question of which xenophones to record in the database and which ones to leave out.

In Phonotacticon, we have included the xenophones as part of the phonemic inventory (and consequently, part of the onset, nucleus, or coda forms) if they are considered to be an integral part of a lect’s phonology by the consulted literature. This is mostly inferred from how general a statement is regarding the status of xenophone within a lect’s phonology. For instance, if a grammar simply writes “X is a phoneme of this lect” or lists it in within the phonemic inventory table, then we take that to mean that that grammar considers X to be an integral part of the phonemic inventory. On the other hand, if the grammar writes a statement in the lines of “in addition to the above-listed phonemes, X only occurs in some loanwords”, then we assume that the grammar does not consider it to be an integral part of the phonology. As ambiguous as this strategy of tone-reading can be, it is arguably an appropriate approach to the status of xenophones which is by nature ambiguous. Furthermore, we have excluded xenophones that occur only in certain varieties of the lect and/or are freely variable with native phonemes, such as the German nasal vowels.

Phonemes that are used by only a portion of the whole speaker population of a given lect have been excluded as well: we have only included phonemes that are used by all or most speakers. An exception to this rule is that phonemes used by older generations but not by younger generations have been included, due to the fact that younger generations generally reflect the ongoing change of a lect and it is not appropriate to fully reflect an ongoing change as if it were already complete.

In cases where the source describes a sociolinguistic distinction between prescriptive “educated” speech and real-life “colloquial” speech, we have generally chosen the latter as better reflecting the phonology of a given lect.

When transcribing the phonemes based on a reference grammar, we relied first and foremost on the articulatory description of that phoneme rather than its orthographic transcription. If a phoneme is transcribed as <c> but described as “voiceless palatal affricate”, then we transcribed it as /c͡ç/ (which is the voiceless palatal affricate) rather than the verbatim /c/ (which is the voiceless palatal stop).

All transcribed phonemes are those found in the PanPhon database (Mortensen et al. 2016, as of 23 July 2020). In other words, phonemes that are not found in PanPhon are transcribed in a way that fits PanPhon. The reason for this alignment between Phonotacticon and PanPhon is to make all segments in Phonotacticon machine-readable via PanPhon and convertible into featural values, hence allowing immediate computational analysis. To align the two databases, some theoretical sacrifices have been made. In the case of diphthongs, PanPhon does not include diphthongs (or triphthongs) as independent segments, even though some grammars argue that a diphthong forms an independent phoneme in the described lect. Hence, even if a diphthong phoneme of a lect consists of two vowels that are not found as monophthongs in that lect, those two vowels nevertheless occur in Phonotacticon as individual phonemes, contrary to the grammar’s description. For example, if a grammar describes a lect as having /ɛ͡ɪ/ as a diphthong phoneme while not having /ɛ/ or /i/ as monophthong phonemes, we have still listed /ɛ/ and /i/ as phonemes instead of /ɛ͡ɪ/.

This approach is beneficial to the database, since it not only allows it to be compatible with PanPhon, but also because it avoids the highly controversial nature of the status of diphthongs as individual phonemes. For example, whether diphthongs in a given lect constitute individual phonemes or are combinations of two vowel phonemes is a matter of debate (Berg 1986; Pike 1947; Eliasson 2023) and is thus highly subject to theoretical bias. Another problem is that most reference grammars do not explicitly address this question. Very few grammars –in our experience –discuss whether a diphthong in a lect has phonemic status or not. Many descriptive grammars simply provide a list of monophthongs and diphthongs without specifying whether the diphthongs are sequences of phonemes or independent phonemes. By listing all diphthongs as combinations of monophthong phonemes, we can make all the vowel phonemes compatible with PanPhon and allow cross-linguistic analysis, albeit at the sacrifice of favoring one theoretical approach to diphthongs over another. Moreover, regardless of the phonemic status of diphthongs and triphthongs, they are still listed in the nucleus part of the database, so there is no sacrifice at the descriptive level.

Exceptionally, we have made the following changes to PanPhon:^[3]

1. The features [hitone] and [hireg] were excluded, since they only pertain to tones and not segments.

2. We included prenasalized and preaspirated segments, as these concepts are employed by quite a few grammars but absent in PanPhon. Their features are identical to the nasal and aspirated equivalents, except that prenasalized segments are assigned 0 value to the [nasal, sonorant] features and preaspirated segments are assigned 0 value to the [constricted glottis] feature. The prenasalized consonants are transcribed with <ⁿ> followed by a segment (<ⁿb>, <ⁿd>), whereas preaspirated consonants are transcribed as <h> followed by a tie bar and a segment (< h p ͡ >, < h t ͡ >).

3. We included the fortis (or tense ) counterpart of all consonants, transcribed by the segmented followed by a small plus sign, as this concept is employed in works on Korean (Lee 2021), Swiss German (Fleischer and Schmid 2006), or other lects but not present in PanPhon. The feature of each fortis consonant is identical to its non-fortis counterpart, except that its [tense] feature is 1 and not 0.

4. Some segments that we judge to be missing as accidental gaps were added. For example, / t s ͡ ’^wː/ is absent in PanPhon, even though / t s ͡ ’ː/ and / t s ͡ ’^w/ are present. As such cases are clearly gaps created by mistake, we added such segments in with appropriate feature values.

As different phonological databases offer different featural parameters and values per segment, users may convert the IPA segments of Phonotacticon into featural values using databases other than PanPhon, such as PHOIBLE (Moran and McCloy 2019) or the Cross-Linguistic Transcription Systems (Rubehn et al. 2024), although they may not cover every segment in Phonotacticon.

In some cases, a source may specify only a certain class of segments as part of a permissible sequence of phonemes. For example, the source may indicate that a plosive plus a liquid may form an onset cluster, without specifying whether all logically possible combinations of plosive + liquid are permitted in the onset position. In such cases, we have used capital letters to describe the permitted sequence without specifying the segments: PL for plosive (P) plus liquid (L).

Table 3 shows the capital letters used to represent underspecified segments and how they are defined in terms of features and/or graphemes. <j, w, ɥ, ɰ> means any segment including any one of these graphemes in its IPA symbol. !<h, ɦ> means any segment not having these graphemes in its IPA symbol. Other than V, which stands for vowels, all the capital letters represent consonants or glides: N refers to nasal consonants and glides only, and excludes nasalized vowels.

Many grammars published in China that describe monosyllabic lects do not describe the lect’s phonemic inventory in terms of segmental phonemes but rather in terms of initials (shengmu 聲母)and finals (yunmu 韵母), which correspond to onsets and rhymes. When consulting such grammars, we have interpreted the description in terms of phonemes. For example, if a grammar of a lect describes it as having initials /p-, t-, k-/ and finals /-a, -i, -u, -an, -in, -un/, we have interpreted that as a phonemic inventory of /p, t, k, n, a, i, u/.

All geminates are considered to be consonant sequences and not independent phonemes unless the literature explains why they are independent phonemes.

4.2 Onset, nucleus, and coda forms

The onset, nucleus, and coda sections of Phonotacticon describe the possible onset, nucleus and coda forms of a given lect. They consist of phonemes listed in the phonemic inventory section, as singleton phonemes or a sequence of phonemes. An exception is the obligatory epenthetic phones, which may not be present in the phonemic inventory section but may be present in the onset, nucleus or coda sections. For example, Bantawa [bant1281] (Sino-Tibetan) does not have a glottal stop as a phoneme, but does have it as an epenthetic phone to fill in the obligatory onset slot (Doornenbal 2009). In this case, <ʔ> was transcribed in the onset section of Bantawa. Epenthetic phones that are only optionally inserted were not included. The null onset and the null coda are represented as <#> in the onset and the coda sections.

Some grammars list word-initial, word-medial, and word-final consonant clusters instead of consonant clusters in onset and coda position. In such cases, we have interpreted the data as follows:

1. Word-initial clusters are interpreted as onset clusters.

2. Word-final clusters are interpreted as coda clusters.

3. Word-medial clusters are interpreted as onset consonants, coda consonants, or the mixture of both. If the grammar does not state the syllable boundary that divides a word-medial cluster, we have located the syllable boundary according to the following principles:

   1. If a cluster occurs word-initially or word-finally, then we favored the interpretation that it also exists in a word-medial cluster. For example, if /lp/ occurs word-finally, then the medial cluster /lpt/ would be interpreted as /lp.t/, instead of /l.pt/, given that /pt/ does not occur word-initially.

   2. If a medial cluster does not contain sequences that appear as initial clusters or final clusters, then we favored the interpretation that reflects the sonority sequencing principle (Clements 1990). The sonority sequencing principle is here defined as the normative sequence of vowel > glide > liquid > nasal > obstruent in relation to the vicinity to the nucleus. For example, if /lp/ does not occur word-finally and /pt/ does not occur word-initially, then the medial cluster /lpt/ would be interpreted as /lp.t/ rather than /l.pt/, because /Vlp/ reflects the sonority sequencing principle (vowel – liquid – obstruent), whereas /ptV/ does not (obstruent – obstruent – vowel). Not reflecting the sonority sequencing principle is preferable to violating it: For example, /mmp/ would be interpreted as /mm.p/, since /Vmm/ does not reflect but does not violate the sequencing principle (vowel - nasal - nasal), whereas /mpV/ violates it (nasal – obstruent – vowel).

   3. If a medial cluster contains both an initial cluster and a final cluster, or if a medial cluster does not contain sequences that appear as onset or coda, and if multiple possible interpretations reflect the sonority sequencing principle, then we resorted to the maximal onset principle (Kahn 1976), favoring complex onsets over complex codas. For example, if /pl/ is an initial cluster and /lp/ is a final cluster, /lpl/ would be interpreted as /l.pl/, instead of /lp.l/.

   4. For triconsonantal or longer medial clusters, we applied the maximal onset principle within the length of the initial cluster. For example, for a medial cluster /lpml/, we can divide it into /l.pml/ if a three-consonant cluster is attested word-initially. But if only two-consonant clusters are attested as onset, we can only divide it into /lp.ml/.

   5. Some works (such as Riad 2013) only list word-initial and word-final clusters and do not list word-medial clusters. In such cases, we interpreted the word-initial and word-final clusters as the same as onset and coda clusters.

As a reviewer pointed out, this approach of interpreting word-based data into syllabic information, rather than directly listing word-based data into the database, may have compromised the rawness and theory-neutrality of the database. But we argue that such interpretation is an essential part of the database, whose goal is to provide the syllabic phonotactic information of each lect, based on the (often incomplete) data provided by the literature. Additionally, we would like to point out that many grammars do directly provide syllabic information. It is only when a grammar did not do this that we had to resort to this interpretation algorithm.

In some cases, a given set of phonemes may be described as permitted in a given position of a sequence. For example, a source may indicate that /p t k s/ may precede /l r w j/ to form a biconsonantal onset cluster, without specifying whether all the 4 × 4 = 16 logically possible combinations are actually attested. In such cases, we have used square brackets to denote any one of the phonemes within this bracket: [ptks][lrwj] to mean any one of /p t k s/ followed by any one of /l r w j/.

If a consonant is described as occurring word-initially or as an onset, then we assumed that it can occur alone as a single onset. Technically, this may not always be the case, as a consonant may occur word-initially in the onset position as the initial part of a cluster and not on its own (for example, /s/ occurring in /spV/ only and not in /sV/). But unless stated otherwise, we assumed that its occurrence in word-initial or onset position implies its occurrence as a single onset. The same rule applies for word-final and/or coda consonants.

Often, a grammar does not mention whether an onset is obligatory in a syllable. If we detected at least one syllable without an onset, then we judged that that lect does not oblige an onset.

If the literature does not mention syllabic consonants, then we assumed that the syllable requires at least one vowel.

4.3 Tonemes

Tones are transcribed in capital letters (H, M, L, F, R, or any combination of these) or Chao letters (1 to 5 or any combination of these). For example, a high rising tone may be transcribed as HR in capital letters or 35 in Chao letters. If a grammar employs Chao letters, then the Chao letters are transcribed verbatim in Phonotacticon. If a grammar uses other means of description, then the tones are transcribed in capital letters. If a lect has no tones, then the absence of tones is marked with <->.

As a rule, the tones are transcribed in terms of pitch (level or contour) unless a toneme is not distinguishable by pitch only. A toneme often has acoustic cues other than pitch, such as length and phonation. Only when two tonemes are only distinguished by non-pitch cues have we transcribed the non-pitch information in Phonotacticon: <ˀ> for creaky voice, <C> for checked tones, and <^h> for aspiration. For example, Burmese tones are transcribed as L (low), Hˀ (high creaky), and H^h (high aspirated) (based on Jenny and Hnin Tun 2016).

In some cases, a tone may be described as having more than one allotones, rather than one single underlying form. In those cases, the allotones are transcribed and separated by slashes. For example, the three tones of Asho Chin [asho1236] are transcribed as <55, 44, 22/11> (based on Zakaria 2018).

Many grammars of atonal lects do not specifically mention the absence of tone. If the cited literature does not mention tone, then we have assumed that the lect has no tone.

4.4 Bibliographical sources

The database includes the bibliographical information of the source consulted for each lect’s profile. The sources are either the “long grammars” as defined by Glottolog 4.4 or any other source we deem relevant and accessible. As most reference grammars are very long, we did not read each grammar in its entirety (including syntax, morphology and other sections), but rather focused on its phonology section. The consulted phonological data were then transcribed manually into Phonotacticon.

The accessibility issue includes language barriers as well. In most cases, including when the sources were written in French, German, Japanese, or Chinese, this was not a concern, as we could read those lects. In some cases when we could not read the lect a source was written in very well, such as Russian or Finnish, we read it with the aid of machine translation.

4.5 Note

In cases where further clarification is needed regarding how we retrieved the information from the cited source, we left a brief note in plain words in addition to the phonotactic profile.

4.6 Missing information

All the sample lects have, at the very least, its phonemic inventory transcribed. In some cases, the only accessible source did not provide any information on onsets/nuclei/codas. In case where this information was missing, we filled the slot with the symbol <?>. A total of 59 out of the 516 sample lects have no information about their onsets/nuclei/codas.

5.1 Syllable length

In this section, we visualize the distribution of syllable length in Eurasia. By syllable length we mean the number of segments (phonemes or epenthetic phones) that fill in the one of these three slots. For example, English permits up to three consonants in its onset position (/strit/ street, /splæʃ/ splash), three vowels in its nucleus position (/fa i ə/ fire, /aʊə/ hour), and four consonants in its coda position (/teksts/ texts, /ɡlimpst/ glimpsed) (Gut 2009). English, and European lects in general, allow longer onsets, nuclei and codas compared to other lects in the world. Hokkaido Ainu, for instance, allows only one segment in each of the three positions, its maximal syllable being CVC (Tamura 2000: 21).

To our knowledge, Maddieson (2013b) is the only work so far to have provided a typological overview of syllable length. Maddieson divides 486 lects worldwide into three categories based on their syllabic complexity: simple (maximal syllable is CV), moderately complex (maximal syllable is CCVC where the onset CC is stop + glide or stop + liquid), and complex (onsets can be clusters other than stop + glide/liquid and codas can be complex). He reports that ca. 56 % of the sample lects have a moderately complex syllable structure, ca. 30.9 % have a complex syllable structure, and ca. 12.5 % have a simple syllable structure. His data show that within Eurasia, East and Southeast Asian lects tend to allow moderately complex syllable structures, whereas complex syllable structures dominate elsewhere.

Maddieson’s overview, based on a ternary division based on syllable length, while by itself helpful, calls for a further analysis with finer resolution. The following figures provide such an analysis based on gradient values of onset, nucleus, and coda lengths.

Figure 2 shows the maximal length of an onset in the sample lects, in terms of the number of the phonemes allowed. Caucasus and Eastern Himalayas are zoomed in, due to the high concentration of lects. What is the most evident is that Eurasia is largely divided into three areas: North and Northeast Asia generally only permit singleton onsets, with the notable exception of the Qinghai-Gansu linguistic area (Dwyer 2013; Janhunen 2006; Xu 2017; Zhou 2020); South and Southeast Asia generally permit up to bisegmental onsets; and Europe generally permits up to triconsonantal onsets. The Middle East seems to be the most diverse without a dominant upper limit.

Figure 2:

Maximal length of an onset in each lect.

As the onset is optional in some lects, the minimal length of onset can be either zero or one segment in a given lect. Figure 3 shows whether an onset is obligatory in each lect. We see that the obligatory onset is mostly present in the Mainland Southeast Asian linguistic area (Enfield 2018; Sidwell and Jenny 2021; Vittrant and Watkins 2019) and the Middle East. All sample lects that mandate an onset in a syllable use the glottal stop [ʔ] as the filler segment to fill in the gap of a syllable that would otherwise lack an onset. The glottal stop may or may not be a phoneme in such lects.

Figure 3:

Obligatory onset.

Note that even the lects that do not have an obligatory onset filler may have a non-obligatory filler. English, for example, can insert /ʔ/ in the word-initial position, but it is certainly not obligatory (occurring about 50 % of the time in British English, according to Fuchs 2015). Furthermore, the glottal stop is normally not inserted in word-medial onsets (e.g. A. I. [(ʔ)ɛi.ɑɪ] and not *[(ʔ)ɛi.ʔɑɪ]).

Figure 4 shows the maximal length of nucleus in each of the Eurasian lects. We see that South, Southwest, and Central Asia tend to not allow complex nuclei, whereas in other areas, diphthongs or even triphthongs are common. Note that lects that only permit monosegmental nuclei may also have phonetic diphthongs if glides appear in their onset or coda position. For instance, according to Bauer and Benedict (1997: 57), Cantonese diphthongs are not analyzed as vocalic sequences within a nucleus but rather as a vocalic nucleus followed by a consonantal coda, based on the short duration of the offglides [j] and [w]. Adding to this argument, we can also argue for the nucleus-external hypothesis based on phonological grounds: if the Cantonese diphthongs were nucleus-internal, then it would be difficult to explain why they are not followed by a coda (i.e. *[VVC]). Given the fact that Cantonese only allows one segment as a coda, the impossibility of an offglide and the coda consonant coexisting favors the explanation that an offglide is a coda itself.

Figure 4:

Maximal length of a nucleus in each lect.

Figure 5 shows the maximal length of a coda in each lect. The distribution is very similar to the distribution of maximal onset length shown in Figure 2: European lects allow multiple (as long as six) codas, Southwest and South Asian lects allow up to two, and East Asian lects allow only one. The main difference between onset length and coda length distributions is that Southeast Asian lects do not allow complex codas and that several lects in Southwest China do not allow any coda at all. In sum, we observe a general correlation between onset length and coda length in the Eurasian macroarea, which both tend to increase westwards.

Figure 5:

Maximal length of a coda in each lect.

Figure 6 shows the average maximal length of onset, nucleus, and coda per each family. We see that generally, language families in western Eurasia, such as Indo-European and Afro-Asiatic, allow more segments per syllable than language families in eastern Eurasia, such as Tungusic and Sino-Tibetan.

Figure 6:

Average maximal length of onset/nucleus/coda by family.

To confirm our visual observation that the maximal lengths of onset and coda tend to be longer in western Eurasia compared to eastern Eurasia, we tested whether the maximal onset and coda lengths are correlated with the longitude of the Eurasian lects. First, it is necessary to test the spatial autocorrelation, as geographically neighboring lects may have similar phonotactic patterns. We identified the geographical neighbors of each lect, defined by lects whose coordinates are within 1,500 km distance of each other. This distance threshold leaves no sample lect without any neighbor. We then created a weight matrix and assigned the value of 1 to each neighboring lect pairs and the value of 0 to each non-neighboring lect pairs. We also needed to test the genealogical autocorrelation, since lects belonging to the same family may have similar onset or coda lengths. Similarly to the spatial weight matrix, we created a genealogical weight matrix where the lect pairs belonging to the same family were assigned the value of 1. Based on these two weight matrices (spatial and genealogical), we performed the Moran’s I test to test the spatial and genealogical autocorrelations on onset and coda lengths, which confirms that both onset length and coda length are areally and genealogically clustered (p < 0.001).

We then added these two weight matrices together. Thus, lects that are geographical neighbors and also belong to the same family are given the score of 2; lects that are geographical neighbors but belong to different families are given 1; lects that belong to the same family but are not geographical neighbors are given 1; and the rest are given 0. As a reviewer pointed out, this is under the assumption that spatial autocorrelation and genealogical autocorrelation are of equal weight. The rows of the converged weight matrix were then standardized so that the values of each row add up to 1.

Based on this spatio-genealogical weight matrix, we then ran the Lagrange multiplier diagnostics on the simple linear regressions between longitude and onset/coda lengths to decide whether the spatial lag model or the spatial error model is adequate to perform the spatial regression. The results show that both models are significant. We chose the spatial lag model from a theoretical perspective, since it is more likely that the onset and coda lengths of a lect are influenced by the onset and coda lengths of neighboring lects (spatial lag model) than that there are unobserved language-external factors causing onset and coda lengths of neighboring lects to be similar (spatial error model). Finally, based on the spatial lag model, we performed spatial regression to test the correlation between longitude of the lects and their onset/coda length, also weighing in the spatio-genealogical weight matrix. The results show that both onset and coda lengths are correlated with longitude (p < 0.01). The likelihood-ratio tests in both regressions show that the rho is significant (p < 0.001), confirming that the spatial lag model was suitable. This confirms our visual observation that the maximal onset length and the maximal coda length grow as one goes westwards in Eurasia, even when controlling for spatial and genealogical autocorrelations.

5.2 Syllabic consonants

In all the sample lects, and perhaps universally, the minimal nucleus length is one segment, as a syllable by definition requires at least one segment to form its nucleus. Some lects, however, do not require a vowel in nucleus position, as they allow consonants to form the nucleus. Consonants that form the nucleus are known as syllabic consonants.

Figure 7 shows the distribution of lects that allow a syllabic consonant as its nucleus (blue circles) and those that do not (red crosses). We observe that syllabic consonants are generally permitted at the two extremes of Eurasia: In East and Southeast Asia and (to a much lesser degree) in Europe. Although not shown in the visualization, the phonotactic patterns of the syllabic consonants in these two areas also tend to differ. In East and Southeast Asia, syllabic nasals tend to occur as monosegmental syllables, such as Yue Chinese m ⁴ 唔 [m̩²¹] ‘not’, and syllabic fricatives tend to occur only after homoorganic fricatives, such as Mandarin Chinese sì 四 [sz̩⁵¹] ‘four’. In European lects, however, syllabic consonants have relatively less phonotactic restriction and can occur after a wider range of onsets, such as English button [bʌ.ʔn̩] or German Vogel [fo.ɡl̩] ‘bird’.

Figure 7:

Syllabic consonants.

The permitted syllabic consonants are mostly nasals and sometimes liquids or fricatives. This is an unsurprising result confirming that more sonorant segments tend to appear in the nucleus position.

5.3 Number of singleton codas

In Section 5.1, we saw that the maximal coda length varies across Eurasia. Codas are often limited not only quantitatively, but also qualitatively, as many lects only allow a subset of their phonemes to appear in the coda position. Although many lects also ban certain phonemes from the onset position as well, restriction in the coda position tends to be much stronger. For example, Mandarin Chinese only allows /n ŋ/ as codas, while allowing all consonant phonemes but /z ŋ/ as onsets.

Figure 8 visualizes the types of singleton consonants that can appear as coda, i.e. the types of mono-consonantal codas. (The sample lects are limited to those that have full information of singleton codas, i.e. excluding those whose singleton codas are underspecified as <C> in the database.) It shows that in the lects of East Asia and Southeast Asia, the coda is limited not only in terms of length but also in terms of the number of permitted consonants. Typologically, nasals, plosives and glides are the most common consonants as coda, whereas liquids, fricatives and affricates are less common.

Figure 8:

Number of singleton codas.

5.4. Number of tones

Maddieson’s (2013c) survey of 526 lects worldwide reveals that 220 of them are tonal. Among these tonal lects, 132 have a “simple tone system” with only two tones. The remaining 88 have a “complex tone system” with three or more tones. His data show that tonal lects are most heavily present in Sub-Saharan Africa and Mainland Southeast Asia. Complex tone systems (with three or more tones) are the majority in Mainland Southeast Asia, unlike in Sub-Saharan Africa, New Guinea, or the Americas, where simple tone systems are numerous as well.

Figure 9 shows the number of tones per Eurasian lect. The largest number of distinctive tonemes is ten, for example in Cao Miao (Wu 2015), and the lowest number is one, for example in Swedish, where the tonal distinction is privative, i.e., between the lexical tone and its absence (Riad 2013). It is easily observable that tones are a strongly areal phenomenon, concurring with Maddieson (2013c). Most tonal lects are distributed in Mainland Southeast Asia and China (with the notable exceptions of the Qinghai-Gansu linguistic area, Cambodia, and southern Vietnam). Within this area, the Guangxi province has the highest number of tones, the maximal number being ten. Elsewhere, tones are only sparsely present, lects having at most two tones. From this uneven distribution, we can see that tonogenesis (the emergence of tones) is highly prone to areal pressure, even though it can happen in non-tonal environments (e.g., in Swedish).

Figure 9:

The number of tones per lect.

It is worth noting that Korean, while depicted as atonal in Figure 9, retains the tones inherited from Middle Korean in certain varieties (notably the Southeast variety), while the Seoul variety is currently going through tonogenesis (Kang and Han 2013). In the light of the distribution of tones in East Asia, we can hypothesize that Korean tonogenesis may be motivated by areal pressure from Sinitic and Japonic.

5.5 Summary

In this section, we have seen how a number of phonological patterns vary across Eurasia. Crucially, different phonological patterns show different areal distributions: The distribution of tones (Section 5.4), for example, is not identical to the distribution of syllabic consonants (Section 5.2). It is therefore helpful to shed light on each one of the phonological patterns to understand their diverse areal shapes.