The rise of ternary quantity and its alignment with laryngeal articulations

2.1 Sample and sources

The data on 27 varieties analysed for this study are taken from existing publications, my own field research (on Soikkola Ingrian), and personal communication with experts (on Nenets, West Nilotic).

The sample on alleged ternary quantity includes the following areal-genetic groupings (the number of the varieties considered here is indicated in parentheses; see their full list in Supplementary Material 1; SM1):

1. Uralic (5 Finnic and 1 Saami) and Indo-European (1 Celtic, and 2 West Germanic) varieties of North-Western Europe;

2. Uralic (2 Nenets) and Yeniseian (2) varieties in North-Western Siberia;

3. West-Nilotic languages of East and Central Africa (4);

4. languages of eight families in North America (5) and in Mexico (5).

Available synchronic and diachronic data on many varieties are scarce, and their prosodic systems are often complex; thus the place of ternary quantity within these systems is not always clear. Particular data subsets considered in Sections 3 and 4 are outlined in Section 3.

2.2 Basic notions and transcription

In most varieties of the sample, ternary quantity is related to metrical stress.^[1] The relevant prosodic domain of quantity and stress is typically the metrical foot (usually mono- and/or disyllabic, trochaic or iambic, rarer trisyllabic). The ternary quantity contrast appears in stressed syllables in almost all cases (excluding Southeastern Finnish dialects, cf. Supplementary Material 2; SM2).

Several languages of the sample feature lexicalised pitch. In most cases, however, lexicalised pitch is stress-dependent (cf. Hyman 2006), so I treat it as a phonetic cue of stress (referred to below as pitch-stress) rather than as lexical tone. The only clear (and well-studied) exceptions are West Nilotic languages, which synchronically feature lexical tone in the absence of any metrical prominence. In West Nilotic, the domain of both ternary quantity and tone is the syllable.

In all varieties, the domain of lexical prosody (foot or syllable) has a quite clearly defined structure. Prosodic weight is often relevant for prosodic contrasts and processes, including the distinction of light and heavy syllables. Light syllables include (C)V and sometimes word-final unstressed (C)VC structures, and heavy syllables include all other structural types.

All phonetic transcriptions are transformed into IPA for better comparability. The three degrees of quantity are coded as Q1, Q2, Q3 – from the shortest to the longest. In cases of phonologically relevant prosodic weight, Q3 is possible only in structurally heavy syllables. Phonetic studies also show that phonetic overlength in Q3 is a stress feature.

1. In stressed syllables, the durational ratios between the three degrees of quantity are typically 1 : 2 : 2.5–3.5, i.e. phonetically [short : long : overlong], reflected below as Q1 [ˈV/C] – Q2 [ˈVː/Cː] – Q3 [ˈVːˑ/Cːˑ].

2. In unstressed syllables (Southwestern Finnish, see SM2) or in stressless tonal languages (West Nilotic), the contrast of durational ratios is roughly 1 : 1.5 : 2, i.e. [short : lengthened : long] (phonetic transcriptions with such cases are not included in the main text, but see SM2).

4.1 The laryngeal articulator

A proper account for the activity of the lower vocal tract is still missing from typology (Seifart et al. 2018: e329–e330). However, recent years have seen the rise of the Laryngeal Articulator Model (e.g. Esling 2005; Moisik 2013), together with attempts to implement it in phonology (Moisik et al. 2021).

In the terminology of this model, adopted also here, ‘larynx’ equals ‘pharynx’ equals ‘lower vocal tract’. The lower vocal tract is an active articulator rather than a passive source of vocal fold vibration. All kinds of pharyngeal, epiglottal, and glottal sounds and phonations are produced by the complex aryepiglottic constrictor mechanism together with associated tongue retraction and larynx raising (Esling et al. 2019: 5).

In what follows, all such articulations are labelled as laryngeal articulations/actions. If a language description specifies a more precise mechanism of phonation, this can be mentioned. However, impressionistic descriptions might often be imprecise, because auditory identification of most laryngeal categories has not yet been widely practiced.

Pitch is set as a separate category, because of partially different properties as compared to secondary laryngeal articulations (e.g. the diagnostics outlined in the end of Section 4.3.2 cannot be straightforwardly applied to pitch).

Lexically relevant laryngeal articulations are the focus of this study. The broad phonetic notion of “laryngeal articulations” covers three main types of both synchronic and diachronic phonological units:

1. laryngeal and pharyngeal consonants (e.g. h, ʔ), and other types of consonants in case of the diachronic sources of synchronic laryngeals (see e.g. Yeniseian);

2. secondary laryngeal articulations of consonants (e.g. pre-aspiration in Saami or Scottish Gaelic) and voice quality in vowels (in West Nilotic);

3. word-prosodic units – lexicalised prosodies based on laryngeal articulations (e.g. hiatus in Scottish Gaelic; cf. also Kuznetsova 2018, 2022).

Finally, an important distinction related to lexicalised segmental laryngeal articulations is the contrast of lenis and fortis consonants. This contrast has been distinguished for many varieties studied here, but its phonetic manifestation varies both across languages and across consonantal types. In this study, I apply these terms in cases when more than one acoustic cue based either on duration or on laryngeal articulations distinguishes the members of the correlation (following a similar definition of “weak” and “strong” consonants by Trubetzkoy 1939/1969: 148). The variability of the primary cue may depend either on the prosodic position of a consonant or on its quality (stop, fricative, sonorant, etc.). For example, Leer (1985: 83–87) describes fortis consonants in Alutiiq (Eskimo-Aleut) through a complex bundle of features: as voiceless, tense, longer in duration than lenis (but shorter than geminates), truncating the preceding vowel, pre-occluded, with a preceding hiatus, a reduction of air pressure, or with a schwa-like transition. For the contrasts based on duration only, I use the terms “short” and “long”, while for those based only on voicing, I use the terms “voiced” and “voiceless”.

Ternary quantity and laryngeal articulations show different alignment under two main types of lexical prosody, stress and tone. In what follows, I discuss and exemplify the main types of alignment, starting with tonal languages (Section 4.2) and proceeding with stress languages (Section 4.3).

4.2 Symmetric “orthogonal” alignment: West Nilotic tonal languages

In stressless West Nilotic languages (Dinka, Nuer, Reel, Shilluk), lexical tones, vowel length, and voice quality are synchronically mutually independent parameters. These features show symmetrical orthogonal alignment: most combinations of their values are possible. Such combinations, including also the parameters of vowel quality and consonantal lenition, alternate to express most grammatical meanings, cf. a partial matrix of verbal derivation (columns) and inflection (rows) for tḛem/H ‘to cut’ in Agar Dinka in Table 1 (Andersen 1992–1994: 29). The domain of length, tone, and voice quality is the syllable (Remijsen and Ladd 2008); tone and voice quality are marked once per syllable in Table 1

Modern West Nilotic languages vary by the systems of quantity, tone, and voice quality. There are languages with zero, binary, and ternary quantity contrasts (Andersen 1990). Agar Dinka has three tones (high, mid, low), while tone number reaches nine in Shilluk (Remijsen et al. 2019). Agar Dinka has a binary voice quality contrast (breathy vs. creaky), while in Bor Dinka, the voice quality distinctions include breathy, modal, creaky, and hollow (faucalised) (Edmondson and Esling 2006).

Proto-West-Nilotic apparently had only high and low tone (Storch 2005), a binary quantity contrast (Andersen 1990; Monich and Baerman 2022), and a [±ATR] contrast (Andersen 1990; Esling et al. 2019: 176–177). The subsequent enrichment of quantity, tonal, vowel and voice quality contrasts happened through the loss of one or two monosyllabic suffixes following the root.

In particular, the loss of suffixes triggered compensatory lengthening in stems. Original short (Q1) and long (Q2) vowels lengthened to Q2 and Q3 respectively. In other cases, original length was preserved, or Q2 vowels even shortened to Q1. These processes of compensatory lengthening and shortening have created ternary quantity (Andersen 1990; Monich and Baerman 2022). Similar compensatory lengthening of vowels before occasionally eliding glides or vowels is still an ongoing process in Shilluk and Nuer (Reid 2009, 2019).

The binary breathy/creaky voice contrast, as in Agar Dinka, likely originates from [±ATR] (see Section 5.2). In some cases, now vanished [+ATR] suffix vowels harmonised stem [-ATR] vowels, which has produced stem alternations in vowel and voice quality (Andersen 1990, 2017). Tonal enrichment apparently also partially stemmed from the loss of suffixes (Remijsen, pers. comm.).

4.3 Asymmetric alignment in stress languages: key notions

4.4 Limited alignment in a binary pattern

We start from the cases of relatively limited alignment between ternary quantity and laryngeal articulations. Here, the alignment only concerns either certain groups of sounds (typically obstruents) or consonants at foot boundaries, i.e. outside the syllable nucleus, where overlengthening occurs.

4.5 Extended alignment in a binary pattern

In the varieties analysed in Section 4.5, the system is still binary, but an interaction between durational lengthening and laryngeal articulations is more immediate and more extended throughout the system. Laryngeal activity here is more closely related to stress rather than to single consonants and it occurs in the prosodic centre of the foot rather than at its margins.

4.6 Alignment in multiple patterns

Varieties in Section 4.6 present three different types of alignment in multiple patterns (from four to six quantity-laryngeal(-pitch) foot templates).

4.6.1 Livonian: synergistic alignment

Livonian represents an innovative multiple Finnic pattern distinguishing between five quantity-stød foot templates. Two slightly different descriptions of these templates (referred to as “accents” in both) are combined in Figure 1.

Figure 1:

Duration-stød templates in Livonian: labels of the “accents” by Viitso are in bold black, by Kuleshov – in bold grey (“F” = strong, “L” = weak). Vertical arrows show main paradigmatic alternations between templates, illustrated by examples.

The two systems differ in the grouping (shown by) of the [ˈVTː(C)(V)] type together with either [ˈVRː(C)(V)] on phonetic grounds (lack of stød; Viitso 1974, 1975) into the “grave accent”, or with [ˈVːˀC(C)V] on the basis of alternations in paradigms (Kuleshov 2012) into the “F1 accent”. Both systems in any case agree in their classification of certain foot templates as strong (called “strong” or “F”; grey-shaded in Figure 1) and others as weak (“weak” / “L”).

Strong templates, like in Saami, show lengthening or other types of coda strengthening in the stressed syllable (Viitso 2007), including the consonantal Q3 in [ˈkilːˑə], [ˈkitːˑə], or stød, as in [ˈpiˀdːə] or [ˈk^uoˀigə]. Like in Saami, Livonian Q3 and stød are restricted to strong templates, so their alignment is synergistic. The two features are in complementary distribution across foot structures. Stød is possible in heavy stressed syllables with a long vowel/diphthong/triphthong, or with a short vowel followed by a voiced geminate/cluster.

In Livonian, coda strengthening in strong templates happened along with post-stress vowel reduction or loss. By contrast, weak templates manifest the lack of post-stress vowel reduction, e.g. [ˈkiːla], or even the lengthening of this vowel: [ˈkitˑaː], [ˈk^uoigiːd̥] etc.

Notably, certain structures with either a weak (“L2 / acute accent”) or a strong (“3 / broken”) template have lengthened their stressed vowel, e.g. [ˈkuːoja], [ˈläːˀdə] (cf. Finnish k oi , l ä hteä). This late process happened on top of an already existing system of strong and weak templates. The resulting distribution is peculiar because this is the only case in the sample where we synchronically observe stressed vowel lengthening in weak templates.

In sum, the Livonian system represents the “gradation” quite different from a common Finnic type (Viitso 2007: 55). It can be described through a combination of two parameters: (a) presence versus absence of stød and (b) the locus of the durational peak in the foot: in the 1st syllable (σ₁) nucleus or coda, or, in the “weak/unmarked” pattern, in the 2nd syllable (σ2) nucleus (Figure 1).

The diachronic processes of the consonantal ternary quantity formation are nevertheless similar to other Finnic languages (Viitso 2007, 2008), with an addition of stød rules. Livonian stød corresponds to a former syllable boundary (Kiparsky 2017; Wiik 1989; Table 7-d, f), which in some cases can ultimately originate from a loss of the consonant h (Table 7-g). The mechanisms of the ternary quantity and stød development can be summarised as follows:

1. lenition of original short obstruents to voiced (Table 7-a);

2. compensatory shortening of the stressed heavy syllable before a closed syllable in Q2 (Table 7-b);^[5]

3. “anti-compensatory” lengthening of the stressed light syllable before a long vowel or a diphthong (subsequently shortened to ə; Table 7-f):

   1. a long voiceless stop into Q3 (Table 7-c);

   2. a voiced stop or a sonorant through stød (Table 7-d);

4. compensatory strengthening (with a loss of the post-stress vowel) in:

   1. the stressed heavy syllable through lengthening (Table 7-e);

   2. the stressed light syllable through stød (Table 7-f);

5. the development of stød with the loss of *h in some mono- and disyllables (Table 7-g).

Table 7:

Examples of the origins of consonantal ternary quantity and aligned types of contrasts in Livonian.

	Q1	a. [ˈpidaː] ‘keep:imp’	< *pidäk
	Q2	b. [ˈkitˑaː] ‘burn:imp’	< *küttäk
	Q3	c. [ˈkitːˑə] ‘burn:inf’	< *küttǟ < küttätäk
	Other types	d. [ˈpiˀdːə] ‘keep:inf’	< *pitǟ < pitätäk
		e. [ukːʃ] ‘door’ f. [käˀd] ‘hand:gen’ g. [vɔːˀ] ‘wax’	< *uksi < *käten < *vaha

5.1 Structural, genetic, and areal biases in the distribution and rise of ternary quantity and its alignment with laryngeal articulations

Ternary quantity, although rare, has been found in unrelated parts of the world. Most of these varieties have quite innovative word prosody. For many of them, areal contact has been suggested as a source of their phonological innovativeness (see SM2 for details).

Ternary vocalic length is far more common than consonantal (23 and five cases respectively for the 27 varieties considered in this study, cf. SM1). This could be linked to the fact that ternary quantity typically rises through stress-induced and compensatory processes. Compensatory lengthening and shortening are sometimes considered as directly induced by stress (White and Turk 2010; White 2014). Under this view, most if not all of the five main evolutionary mechanisms listed in Section 3 (paths 2–6) are likely the effects of stress (or of some kind of stress-like inequality in prominence, as in the history of West Nilotic). It is of no surprise then that stress-related lengthening more often affects vowels than consonants or that ternary quantity is mostly restricted to stressed syllables.

Consonantal ternary quantity is found only in Finnic and Saami (Uralic). The synergistic alignment of overlength with laryngeals is also mostly concentrated in these varieties. The emergence of both phenomena is closely linked to the specific “anti-compensatory” lengthening of consonants before longer segmental or syllabic matter. This process, not yet accounted for by Articulatory Phonology models, might be possibly linked to the tense pronunciation of the first stressed syllable in these languages (Kuznetsova 2023).

Ternary quantity is considered unstable. This has been explained by its poor perceptual salience and/or difficult articulation (Blevins 2004: 202). Ternary quantity might be especially unstable in unstressed syllables, like in Southeastern Finnish, where the phonetic distance between quantity degrees is very small (cf. SM2). Two paths of a likely further evolution of ternary quantity include either the restoration of binarity (usually a Q2/Q3 merger), or the development of other phonetic features which reinforce ternary quantity (Kuznetsova 2015; McRobbie-Utasi 2007; Remijsen and Gilley 2008).

The features reinforcing ternary quantity in stressed syllables often include an inverse ratio of the duration of certain foot segments to the duration of the sounds in three quantity degrees. For example, in North Low Saxon, Franconian, Lowland Mixe we find shorter vowel quantity degrees before (longer) fortis consonants but vocalic overlength before (shorter) lenis consonants. In Estonian and partly in Soikkola Ingrian, the length degrees in the first syllable are in an inverse relation to the post-stress vowel duration.

Additionally, we often observe correlating vowel quality contrasts (tense/lax, close/open, diphthongised/plain, cf. SM2), a difference in pitch and intensity curves, and antagonistic or synergistic laryngeal articulations. In particular, many of the durational and laryngeal articulations correlating with ternary quantity are part of the contrast of fortis and lenis consonants.

Most of those correlations reinforcing ternary quantity seem to be related to the general type of prosodic innovativeness of the stress languages analysed. In these languages, we observe extreme strengthening of the prominent syllable at the expense of extreme reduction and loss of non-prominent syllables. Metrically prominent syllables receive maximal prosodic enhancement by a whole array of mutually related phonetic features. Some of these cues may remain purely phonetic while others become lexicalised as independent phonological contrasts (see Sections 5.2–5.3).

The degree and/or the type of stress-induced strengthening in the prominent syllable is different across foot structures in the stress languages considered. We observe an interaction between durational lengthening and such factors as the number of syllables in the original structure, the presence of open/closed syllables in certain prosodic positions, the presence/absence of particular laryngeal articulations, the vowel quality type, and the type of pitch register and/or curve. These interactions have created from two to six distinctive quantity-laryngeal(-pitch) foot templates in these varieties.

Importantly, the distribution of quantity degrees, laryngeal articulations, and pitch curves is always asymmetrical to some degree in the stress languages. In particular, two main types of alignment between quantity and laryngeal articulations within foot templates have come to light. Phonetic overlength (Q3) is invariably linked to the “strong” templates (those based on the diachronic processes of strengthening or the lack of weakening). The lexicalised laryngeal articulations interacting with quantity occur either in the strong templates, thus creating a synergistic alignment with overlength, or in the “weak” templates (diachronically based on weakening or the lack of strengthening), thus showing an antagonistic alignment with Q3 (see Section 5.4 for chronological considerations).

By contrast, a group of stressless tonal West Nilotic languages manifests a symmetrical orthogonal alignment between quantity degrees, laryngeal articulations, and pitch. Most values of the three categories combine, creating numerous quantity-laryngeal-pitch foot templates. The lack of metrical stress is apparently part of the story here. Stress is more prone than tone to creating a “conspiracy” between durational, laryngeal, and pitch cues. Indeed, we do not observe anything close to the extremely numerous West Nilotic foot templates in the stress languages, where the combinations of quantity, laryngeal, and pitch values are far more restricted.

A complementary distribution of asymmetrical alignment between duration and laryngeal articulations in the stress languages versus orthogonal alignment in the tonal languages might, therefore, stem from a deep prosodic contrast between stress versus stressless languages. A possible articulatory basis of this difference is further discussed in the next section.

5.2 Potential role of the laryngeal articulator in languages with ternary quantity

The tonal and stress varieties analysed manifest differences also in the types of laryngeal articulations associated with quantity and pitch contrasts.

West Nilotic voice quality (register) contrast, orthogonal to both length and pitch, is a syllable-length larynx postural setting (cf. Esling et al. 2019: 2). The two registers of Agar Dinka – breathy and creaky/modal – correspond to two fundamental states of larynx: unconstricted/lax versus constricted/tense (Esling et al. 2019: 78–79). As noted in Section 4.2, this contrast has been linked to the original contrast in [+ATR] versus [-ATR]. Bor Dinka, with its four registers, “exploit ‘cardinal corners’ of laryngeal contrast: phonatory quality, sphincteric tightening, and larynx height” (Esling et al. 2019; cf. also Edmondson and Esling 2006). Esling and colleagues have also discussed a link of voice quality to other features. Larynx lowering versus raising is seen as a property connecting consonant voicing, pitch height bias, phonatory and vowel quality (Esling et al. 2019: 175–176; cf. also Denning 1989). A possible relation between voice quality and duration, however, is only mentioned in passing (Edmondson and Esling 2006: 188).

For the stress languages with ternary quantity, detailed articulatory studies on larynx activity are still missing. However, existing experimental and impressionistic observations and phonological descriptions show that the lexicalised laryngeal articulations in these languages are generally of a different type, as compared to West Nilotic registers. Rather than a longer-term larynx postural setting, here we observe a momentary “ballistic” articulatory gesture associated with the stressed syllable, typically that of laryngeal/pharyngeal constriction.^[8]

The fact that the gesture is momentary likely explains the high variability in its acoustic output, observed for many of the languages analysed. This variability is seen in the phonetic experiments on Livonian stød (Tuisk 2015), Yeniseian laryngeal and pharyngeal articulations (Werner 1969: 88), laryngeal features in Otomanguean languages (Avelino Becerra et al. 2016: 172), in North Franconian laryngealisation under Accent 1 (Kehrein 2017: 169), in realisations of Scottish Gaelic hiatus (Mandić 2021), and in Nenets word-internal glottal stops (Janhunen 1986: 43; Lyublinskaya 2014). Momentary laryngeal constriction can provide a wide range of output, e.g. a glottal stop or a pharyngeal, stronger or weaker vowel glottalisation or pharyngealisation, a pitch and intensity dip, or even no visible acoustic cues (cf. with similarly variable realisation of Danish stød, a glottal prosody, in Grønnum and Basbøll 2007: 200).

Earlier accounts of some of the varieties analysed have suggested that such a laryngeal gesture could be the original source of the variable synchronic acoustic output which does not necessarily always feature overt laryngeals. For example, Harms (1975) and Sammallahti (1977: 63, 1998, 2012) proposed an additional “sub-laryngeal / sub-glottal pulse” on the first post-stress consonant before an open syllable (discernible on spectrograms as an increase in amplitude) as a single source of the strong foot template in Saami, which manifests synergistic/enhancing laryngeal articulations, durational lengthening and epenthetic vowels (cf. Section 4.5.1). Harms (1975: 435) links this alleged extra “ballistic” increase in sub-laryngeal pressure followed by its momentary loss also to a peculiar realisation of the Saami voiced geminates in the strong template: oabbá [ˈŏăbˑpaː], as compared to the voiceless geminates in the weak template: oappá [ˈoapːaː] (Table 5-g). Voiced geminates start as voiced, then become voiceless during closure, and resume voicing at the onset of the next syllable, i.e. are pronounced roughly as [bpb].

According to Harms, in the strong Saami template, not only the syllable peak is more prominent, but also the syllable boundary is more marked: “the stressed syllable is clearly set off from the following syllable by a distinct syllable boundary”, which is phonetically expressed by various means including also epenthetic vowels (Harms 1975: 437). This is parallel to observations on Yupik, where the foot in the strong template is prosodically “set apart” by strengthening both in its nucleus and at its boundaries (cf. Section 4.4.1 and SM2), and, as discussed below, on Estonian Q3, as well as to svarabhakti in the “strong” pattern of Scottish Gaelic (Section 4.5.2).

The variable synchronic output in the strong template in Saami, however, as noted by a reviewer, can also be considered as a direct transformation of an original durational cue rather than of a laryngeal gesture. Indeed, quantity contrasts are widespread across Finno-Saamic languages, while overt lexicalised laryngeal features are observed most prominently only in Saami and Livonian. Therefore, only lexicalised quantity is reconstructed for common Finno-Saamic and for Proto-Finnic. It is an open question, though, whether a durational cue or a laryngeal gesture proposed by Sammallahti and Harms should be reconstructed for Proto-Saami.

A “ballistic” laryngeal gesture might be proposed as the source of variable synchronic output also for some languages with antagonistic/blocking laryngeals. Hiatus versus its absence (accompanied by vowel epenthesis in some structures) is usually seen as the source of the contrast between the strong and the weak templates in Scottish Gaelic (Section 4.5.2). Synchronically, hiatus is realised through a laryngeal action, from “a break in the tension of the glottal chords” between vowels (Borgstrøm 1940, 1941) to an overt glottal stop, or through a falling pitch contour, or shorter duration, or a combination of some of these features (see phonetic details e.g. in Ladefoged et al. 1998; Mandić 2021; Nance 2015; Ternes 2006). There are no articulatory studies on hiatus, though; nor is it clear which realisation is diachronically primary. Given such a varied output and the pitch fall in the hiatus pattern, a laryngeal action associated with the syllable boundary might potentially be the source of synchronic variability (rather than for example the pitch cue). If that were the case, then such laryngeal articulation could have blocked stressed vowels from strengthening through (over-)lengthening (unlike in Saami) thus creating the “weak” pattern.

A unified account through a difference in laryngeal settings has been proposed also for North Franconian dialects (likewise featuring antagonistic/blocking laryngeals, cf. SM2). Engelmann (1910: 383–385) impressionistically described the difference between Accent 1 and 2 (the weak and the strong template) in Moselle-Franconian of Vianden as follows. Under Accent 1, syllables end with a sudden loss of expiration, and under Accent 2, they fade slowly. Engelmann also observed laryngoscopically an overt glottal closure after a vowel or a coda sonorant in Accent 1, especially under strong sentence-final stress (cf. SM2). Like Sammallahti for Saami, Engelmann holds an underlying laryngeal gesture in Franconian Accent 1 versus its absence in Accent 2 responsible for all observed acoustic differences in vocalic and consonantal duration between the two accents, as well as for the voicing contrast of consonants (while the pitch curve appears the same under both accents in this variety).

Franconian Accent 1, however, has only in part emerged out of original disyllables with a voiced intervocalic consonant, for which a syllable boundary marking with a laryngeal gesture can be proposed, cf. the “strong” template *[ˈhaʋz(ə)] ‘house:dat’ > Q3 [huːˑz̥] in North Low Saxon (Prehn 2012: 57, 288), but the “weak” template under a similar process in Franconian: *[ˈhaʋz(ə)] > [¹huːs] in Cologne ∼ [¹haʋs] in Arzbach (Gussenhoven and Peters 2004: 280; Köhnlein 2016: 92). This path of the rise of Accent 1 in Franconian appears secondary to its original rise out of long non-high vowels, where the intrinsic durational and F0 properties of vowels might have been at play as the primary cues (see SM2), so the role of the laryngeal articulator is not entirely clear here.

Ivanov (1971, 1975) also suggested a laryngeal cue (a transformation of some obstruents into laryngealisation/pharyngealisation under certain prosodic conditions, cf. Section 4.6.3) as the source of Yeniseian (Ket and Yugh) “tones”. Werner (1979, 1996), however, argued that a distinction between two pitch melodies on disyllabic words is older, otherwise some facts remain unexplained. A similar discussion exists e.g. on Danish stød, a lexicalised glottal prosody. Franconian Accent 1 is phonologically comparable to Swedish and Norwegian Accent 1 and to stød in Danish, while Franconian Accent 2 is comparable to Scandinavian Accent 2 and to lack of stød (Peters 2006). It is likewise contentious whether the Scandinavian pitch-based Accent 1 or the stød is older, the predominant view giving primacy to the pitch cue (Iosad 2016).

For Yeniseian, we can in any case say that the lexicalised durational cue is clearly more recent than both the laryngeal and the pitch cue. However, this holds only for glottalisation (the antagonistic laryngeal articulation in Yeniseian). Pharyngealisation (the synergistic laryngeal articulation), according to Werner (cf. Section 4.6.3), has developed later out of laryngealisation, when the durational cue had already been formed from vowel contractions.

From the diachronic point of view, therefore, duration, pitch, and laryngeal articulations in the languages with ternary quantity can apparently all serve as diachronically primary cues and can manifest transformations into each other. Vowel quality might also have a role in some cases, as in Franconian. Below I propose a possible articulatory model which might account for this.

5.3 Possible articulatory account for the rise of quantity-laryngeal-pitch templates in languages with ternary quantity

As said, the Laryngeal Articulator Model does not yet explicitly link duration to other articulatory events in the lower vocal tract. However, as this study also shows, durational events are closely interrelated with other types of laryngeal activity (e.g. laryngealisation, pharyngealisation, pitch and intensity curves), as well as with some events higher up in the vocal tract (e.g. vowel quality and its mutations; cf. SM2). This interaction is strong in stress languages and especially strong in longer types of stressed syllabic nuclei (Q2 and Q3), for the possible reasons explicated below.

If we follow an “approach to speech-motor control based on … phonology-extrinsic timing mechanisms and symbolic phonological representations” (Turk and Shattuck-Hufnagel 2020: 313), recently re-introduced to Articulatory Phonology, we may say that abstract phonological quantity controls the length of time that laryngeal mechanisms accompany oral postures over the course of their articulation.

A progressively longer duration of postures in greater phonological quantity degrees means progressively higher phonetic variability of sounds, because variability in both speech and non-speech movements grows linearly with an increase in duration of an interval (Turk and Shattuck-Hufnagel 2020: 90–95). Greater variability means a larger pool of phonetic realisations and, therefore, more diverse potential sources for sound change (Ohala 1989). Therefore, for long and especially overlong syllabic nuclei, we might expect a higher variability also in laryngeal articulations. For example, in longer sounds we can expect more important changes in the degree of larynx constriction than in shorter sounds. These bigger changes in the states of the larynx trigger, among other things, more prominent pitch and intensity falls in longer sounds, as well as some specific vowel quality mutations (e.g. vowel diphthongisation), observed less frequently or not at all in shorter sounds. Higher articulatory variability in longer sounds causes more phonetic variability, not only in pitch and intensity curves, but also a higher probability of stronger overt voice quality effects (such as laryngealisation, pharyngealisation, aspiration) accompanying durational events than in shorter sounds.

In the progression of a sound change, pitch and voice quality acoustic cues might lexicalise and take over durational cues in primacy, as indeed can be suggested for several varieties in this sample. However, a reverse process, where e.g. durational cues take over the laryngeal ones, or laryngeal cues take over the pitch cues, is also plausible and suggested for some varieties (see SM3). As noted above, stress, i.e. strong metrical prominence, is typically realised through a conspiracy of duration-laryngeal-pitch cues, with their variable primacy across languages and structures. With extra durational lengthening in languages with ternary quantity, the probability of some of these cues becoming prominent enough to be phonologised additionally increases.

As an example of a pool of subtle duration-laryngeal-pitch variability in extra-long syllabic nuclei, let us consider Standard Estonian. Estonian synchronically features complex lexicalised duration-based contrasts and only very limited and late lexicalised laryngeal contrasts interacting with quantity (Section 4.4.1). However, some experimental and impressionistic accounts suggest that subtle articulatory laryngeal activity, synergistic to the distinction between the strong (Q3) and the weak (Q1–Q2) template, might be present more broadly.

Lehiste (1960) and Harms (1962: 8, 11–12) impressionistically described Estonian Q3 through an additional energy pulse. Lehiste (1962) found a clear decrease of subglottal pressure at the end of the Estonian Q3 syllable coda. She interpreted this decrease as a phonetic marker a clear syllabic boundary in Q3, unlike in Q1–Q2 syllables. Like in Saami or Yupik, both the syllable peak and the syllable boundary seem to be strengthening in the strong template here. Harms characterised Q3 as a “free ballistic unchecked movement”^[9] versus a shorter and more controlled gesture in Q2. He suggested that longer duration and a specific pitch curve in Q3 might be secondary consequences of this gesture.

Following Harms, Eek (1986; Eek 1990; Eek and Meister 1997) also described Estonian Q3 as a localised “ballistic” articulatory energy effort at the end of the stressed syllable rime leading to the uneven distribution of articulatory energy throughout the foot and creating such effects as overlength in the first syllable, a shortened vowel in the second syllable, a sharp fall of pitch and energy curve, and long mid vowel raising (most obvious in South Estonian). The “controlled” pattern (Q1–Q2), in turn, is characterised by more even and controlled distribution of articulatory energy, which leads to a smoother pitch and intensity curve and a more even distribution of duration through the foot.

Palatographic, cineradiographic, and acoustic studies by Liiv (1961, 1962a, 1962b) and Eek (1973) indeed suggest higher tenseness of articulators in Q3 vowels and consonants. The Q3 vowels had the tensest and the clearest articulation (with less coarticulation and looser contact with the following consonant), more velar/pharyngeal/back and tongue-retracted and tongue-raised realisations, and in general stronger pharyngeal constriction and more stable position of the larynx. The larynx in Q3 was lower than in Q1–Q2, which might be explained by a deeper pitch fall in the end of a sentence-final Q3 vowel (John Esling, pers. comm.).

Liiv (1961) explicitly argued in favour of classifying the Q1–Q3 vowels by the level and type of pharyngeal constriction rather than by their oral properties. He noted that while Q1 vowels differed from Q2/Q3 by duration the most, the acoustic differences between Q2/Q3 vowels were rather more qualitative.

Moreover, in the perception experiments by Eek and Lehiste (cf. Eek and Meister 1997: 77), a modification of just durational and pitch parameters led to a very low recognition of Q3, which indicated that something essential might be missing. Also, Türk et al. (2017: 22) could not find any clear articulatory “signature” in the oral vocal tract for the Estonian quantity system. This “signature” might be potentially sought for lower in the throat, although, as said, the durational acoustic cue is clearly primary both diachronically and synchronically for the Estonian quantity system.

5.4 Chronological relations between quantity and laryngeal articulations in languages with ternary quantity

Finally, let us address a possible broader correlation between the relative chronology of durational and laryngeal cues and the type of their alignment (antagonistic/blocking vs. synergistic/enhancing) in the stress languages (see column “Diachr. Q/L” in SM3). In those (relatively few) cases when we observe synergistic alignment (Saami and Finnic languages, Yupik, pharyngealisation in Yeniseian), we find a lexicalised quantity contrast but no lexicalised laryngeal articulations interacting with quantity in the proto-language. The only exception here is the synergistic glottal stop in Lowland Mixe, reconstructed both as a segment and as prosodic glottalisation for Proto-Oaxacan-Mixe.

Instead, in cases of the antagonistic laryngeals, which are more widespread and diverse across languages, we rather find such lexicalised laryngeal articulations interacting with quantity which are either diachronically prior to the lexicalised quantity (North Low Saxon, Franconian, Nenets), or roughly equally old (Nenets, Scottish Gaelic, laryngealisation in Yeniseian, glottal fricative in Lowland Mixe).

It is possible, therefore, that the synchronic antagonistic alignment reflects the diachronic path in which older or otherwise more independent lexicalised laryngeal articulations have blocked newer phonetic extra lengthening. On the contrary, the synergistic alignment instead seems to reflect the path where the lexicalised durational cue, under further lengthening, triggered additional strengthening and lexicalisation of some subtle laryngeal articulations which had previously accompanied phonological quantity distinctions within a pool of phonetic realisations (as discussed for Estonian and Saami). The role and the exact place of pitch, as well as of vowel quality, in this picture needs to be further studied on a broader sample of languages.