2.1. Metaphony

Metaphony covers a wide range of assimilatory phenomena that raise vowels in Romance languages (Calabrese Reference Calabrese1985, Reference Calabrese1998). Some consider it a morphophonological process or even deny that it is a phonological process at all (Pöchtrager & Kaye Reference Pöchtrager and Kaye2014). Others, however, situate it firmly in phonology (D’Alessandro & van Oostendorp Reference D’Alessandro and van Oostendorp2016), even when the triggers are opaque or invisible (analysed as floating; Mascaró & Torres-Tamarit Reference Mascaró and Torres-Tamarit2022).

A comparatively simple example of metaphony is shown in (2), where mid vowels are raised by following high vowels.

In the majority view that it is a phonological process, the aspects distinguishing metaphony from general vowel harmony vary between authors, as do the representational choices underlying the process. Some authors use [ATR] – although this is neither phonetically nor phonologically identical to the familiar African [ATR], they argue that there is phonetic grounding for this choice (Calabrese Reference Calabrese2011: 2). For example, Calabrese (Reference Calabrese1998) finds metaphony distinctive in that [] targets are either diphthongised or tensed to [] or raised, or both, depending on the variety in question.

More traditionally, Walker (Reference Walker2005) primarily uses the feature [], and the distinctive property of metaphony is the stressed target vowel and its unstressed trigger. Her model is based on positional licensing and is driven by the motivation for the harmonic feature to sit in a prominent, that is stressed, position. She ties this together with stress-conditioned harmony in Servigliano (Le Marche) and stress-vowel delimited harmony in Ascrea (Lazio; Walker Reference Walker2005: 918).

Canalis et al. (Reference Canalis, Jiménez, Lloret and Renwick2024: 873) define metaphony as an ‘umbrella term’ for a group of related vowel-height-assimilatory processes such that: (a) the trigger is a word-final unstressed high vowel (typically a morphological marker); (b) the target is a vowel carrying word stress; and (c) stressed vowels, typically mid vowels, are raised. This formally treats metaphony as a type of vowel harmony with some coincidental shared characteristics. It then becomes a matter of judgement whether specific cases of vowel harmony are metaphony or not, such as when stress does not determine the target, as in Camuno (Cresci Reference Cresci2014: 63). Since this otherwise involves the same sets of alternations such an exclusion seems unsatisfactory.Footnote ²

Mascaró (Reference Mascaró2016) limits his definition of metaphony as ‘stress dependent harmony’, either in the triggers or the targets.

Accepting the insight of this previous work, there is a more fundamental distinction between metaphony as a formal process and vowel harmony in general. This is particularly evident if we adopt an ET (Harris & Lindsey Reference Harris, Lindsey, Durand and Katamba1995) approach to vocalic representation. ET is a non-articulatory, privative system of features. Currently, it is most distinctive of Government/Strict CV approaches to phonology (Backley Reference Backley2011), but it has important precursors in Dependency Phonology (Anderson & Jones Reference Anderson and Jones1974; Anderson & Ewen Reference Anderson and Ewen1987), Natural Phonology (Donegan Reference Donegan1978) and Particle Theory (Schane Reference Schane1984).

Given our adoption of ET, we group vowel harmonies based on the vowel quality of the alternations (triggers and targets). This leads us to a more restrictive and formal definition of metaphony, characterised by the loss of unlicensed features. This contrasts it sharply with other types of harmony via feature spreading. Those include umlaut, which has much in common with metaphony: it involves raising, and it typically proceeds from an unstressed high vowel trigger to a stressed vowel target, and it is often a morphological marker. However, the phonological mechanism underlying it is still one of feature spreading/sharing. Indeed, there are decades-old analyses which treat umlaut as the phonological exponent of a syntactic head in the form of a floating feature (see Hansson & Wiese Reference Hansson and Wiese2024 for an overview).

2.2. Element Theory

ET recognises that the traditional vowel quadrilateral is perceptually more of a triangle (Lindsey Reference Lindsey, Lindsey and Nevins2017), as illustrated in Figure 1. The corners of the vowel triangle are characterised by distinctive spectral shapes (Harris & Lindsey Reference Harris, Lindsey, Durand and Katamba1995) shown in Figure 2.Footnote ³ Each shape has an ‘autonomous interpretation’ , but detailed acoustics is left to phonetic interpretation. (See Zahorian & Jagharghi Reference Zahorian and Jagharghi1993 for experimental evidence for spectral shape as a better cue than formants for vowel quality.) Thus each element has an associated prototypical speech sound, while features may be abstract, articulatory or acoustic, and acoustic features such as describe broad characteristics, not actual sounds.

Figure 1

Vowel quadrilateral vs. triangle.

Figure 2

Corner vowel spectra.

Unlike standard feature theory where mid vowels are all [−high, −low] and distinguished by other articulatory features, in ET (and all the precursors listed above), mid vowels are an amalgam of the corner spectral shapes, a combination of these primitive elements: |A.I| = and |A.U| = , where |A.I| means |A| + |I|. Further subdivisions can be made by headedness, a stronger presence of one of the elements, marked by underlining; for example, |A.I| = , |A.I| = . This is illustrated in Figure 3: in the left chart, we see the gradual deepening of the curve (roughly, decreasing $\mathit {F}_1$ ) and rightwards movement of the peak (increasing $\mathit {F}_2$ ) as we interpolate between and (the peak for is off-chart). The right chart shows the interpolation between and .

Figure 3

Vowel spectra.

Figure 3 Long description

The figure consists of two side-by-side panels containing spectral envelopes.

In the left panel, four curves are shown. The curve labeled ‘a’ is a dotted line with a prominent peak in the upper-left quadrant. The curve labeled epsilon is a dashed line with a peak in the center-right. The curve labeled ‘e’ is a dash-dot line descending to a trough at center-left then with a sharp peak at right. The curve labeled ‘i’ is a solid line forming a deep U-shaped trough at the bottom of the graph.

In the right panel, five curves are shown. At the top, a dotted line labeled ‘a’ shows a broad, high-amplitude peak. Below it, a solid line labeled alpha descends into a shallow trough. Further down, a dash-dot line labeled open o and a dashed line labeled ‘o’ follow similar descending paths with minor fluctuations. At the bottom, a solid line labeled ‘u’ forms a deep, wide U-shaped curve similar to the ‘i’ curve in the first panel.

The phonetic interpretation of these element combinations is relatively free and language-specific, allowing, for example, the light-diphthong interpretation of certain vowels (perhaps only in computed forms; Kula Reference Kula2008). Segments represented by an empty set may also have a language-specific phonetic interpretation, such as neutral vowels or schwas (Kaye Reference Kaye1990; Charette Reference Charette1991). Some examples of such interpretations are shown in (3).

The final elaboration of ET we employ is the floating element, or variable element (van der Hulst Reference van der Hulst, Petrosino, Cerrone and Hulst2018). A floating element, which we mark by grey shading as in ||, is present in the underlying representation, but does not surface unless licensed by a phonological rule, for example, by a similar full element in an adjacent segment. Floating elements allow an apparently negative process (suppression of an |A| or [+low]) to be viewed as a positive process (licensing of the |A| in the unsuppressed case), as we demonstrate shortly.

2.3. Metaphony and umlaut in ET

Although the term metaphony was originally coined (as French métaphonie) as a translation of German Umlaut,Footnote ⁴ which ostensibly also involves raising of vowels in similar contexts, for example, / ‘man/manly’, / ‘cow sg/pl’ (Lowenstamm Reference Lowenstamm2012), the formal mechanisms which describe umlaut and metaphony are very different in ET.Footnote ⁵

Its own particularities aside, umlaut is straightforward vowel harmony by feature spreading, or in ET by spreading of element |I|, which has the effect of fronting and raising target vowels.

However, in the dialects of Italy, the trigger of metaphony can be either or , and either of these can raise either or both of and , depending on the pattern. In ET, metaphony is analysed as the suppression of underlying features/elements where they are not licensed (Maiden Reference Maiden1991; Carvalho & Russo Reference Carvalho, Russo, Hathout and Montermini2007a,Reference Carvalho, Russo and Trotterb; Russo Reference Russo2007, Reference Russo2014; D’Alessandro & van Oostendorp Reference D’Alessandro and van Oostendorp2016; van der Hulst Reference van der Hulst, Petrosino, Cerrone and Hulst2018). For example, for the data from (2), the surface element content is shown in (4) for the first vowel.

From these forms, one can analyse metaphony as A-licensing: |A| is underlyingly present in all forms, but only appears when licensed by an |A| in the final vowel.Footnote ⁶ Formally, the underlying form has floating ||, which needs a licence to attach to a skeletal slot and be realised. The rule is shown diagrammatically in (5a), together with its application in (5b) and non-application in (5c). The notations are standard: the top row is the skeleton of C and V slots; the second row notes the segments associated with each slot; and the final row shows the elements associated. Solid association lines are underlying, while dashed associations are the result of phonological rules. Arrows indicate licensing relations.

This models the contrastive presence of |A| in these nuclei, but also the alternation with its non-appearance in the metaphonic forms.

The Foggia example shows a simple example of the mechanism. Metaphonic systems show a dazzling array of complexity: interaction with morphology, opacity of the triggers, and possible interpretation with floating features (Mascaró & Torres-Tamarit Reference Mascaró and Torres-Tamarit2022), or the role of stress in defining the locality of targets of metaphony (Walker Reference Walker2011), or the non-application of metaphony to either lexical exceptions or where the process is restricted to close-mid vowels (e.g., Grado; Walker Reference Walker2005). Space precludes going into these in any depth.

Directly relevant to us is Reference CalabreseCalabrese’s (Reference Calabrese2011) claim that a weakness of the ET approach to metaphony, such as Maiden’s (Reference Maiden1991) A-demotion account, is the diphthongal output of the metaphony rules. Maiden (Reference Maiden1991) proposes a special ‘resolution’. More recent ET developments improve on the understanding of this phonology-to-phonetics mapping; and moreover, they open a space for the diphthongal outcome of metaphony, while providing an elegant account of its cross-linguistic variation and the trickiest aspect of its analysis, its chain-shifting nature.

In the Calvello examples in (6), we see a contrast between close and open mid vowels, which behave differently in metaphony. The outcome is also hard to model because of the distorted chain-shift → , → [u].

In ET, the open-mid vowels are usually interpreted as A-headed: = |A.I|. However, as with the close-mid, we posit that in Calvello even the headed A is floating: |.I|. The licensing rule is extended so that an unlicensed headed floating element still surfaces unheaded, and the head is transferred to the other element in the segment, as shown in (7) together with the application of the rule. (7b) and (7c) are the same as in Foggia, while in (7d), a floating headed || is licensed by the following |A|, and in (7e), a floating headed || ‘self-licenses’ unheaded |A| while also heading the other element |U|.

Systems like Calvello are quite phonetically transparent in reflecting the phonological shapes. The high-vowel element as the head is interpreted as the onset of a light diphthong.

However, ET is not rigidly grounded, and has ‘slack’ in the phonetic implementation of linguistic structure (Scheer Reference Scheer2019). This is accepted even in more standard theories. For instance, Calabrese (Reference Calabrese2011) acknowledges that the exact implementation of the diphthongisation is variable: and (among others). Also, phonetic contours need not be phonological contours: light diphthongs and affricates are often single segments, as in Calvello or English .

This slack in the phonetic implementation of phonological structures allows interpretative space for language variation, even when the phonological pattern is otherwise uniform. Thus a single phonological structure, such as |U.A|, may be interpreted differently in different languages, for example, as or .

Variation in the possible interpretation of headed structures is what makes other metaphonic systems so confounding. Take, for instance, the chain shift of Servigliano in (8).

This can be interpreted just as Calvello, the only difference being the phonetic implementation of the non-A-headed structures, as shown in (9). The from in verde is representationally distinct from the derived from of pedi. Likewise, we will see that the same phonetic surface vowel in ǃXóõ can appear as the result of two different elemental forms. (Although in ǃXóõ, there may indeed be a subtle phonetic difference.)

More complicated cases involve the nature of the A-loss, in what combinations it is allowed, what is the phonetic outcome of the derived structure and so on. The details are found in the specific studies.

To summarise, vowel harmony may be the result of the additive process of spreading, or the subtractive process of (non-)licensing, as schematised in (10).

The pattern from ǃXóõ that we re-analyse in this article is complicated precisely because it involves both types of vowel harmony, which together give the appearance of a single complex gradient pattern of vowel raising (Lionnet Reference Lionnet2018). We argue that part of this pattern is attributable to the same formal mechanism as metaphony in Romance.

3.1. ǃXóõ phonemes

We summarise the phonemic inventory of ǃXóõ, and refer the reader to Bradfield (Reference Bradfield2014) for a full review of evidence and analyses up to 2014, and to Naumann (Reference Naumann, Vossen and Haacke2016) for a detailed phonetic study of the Western ǃXóõ inventory. We shall use phonemic notation in ‹guillemets› when citing ǃXóõ forms, and reserve phoneme for explicitly phonemic discussion. We adopt Bradfield’s symbol for a generic click.

The published primary sources are Traill (Reference Traill1985) and Traill (Reference Traill1994). We have been given, most generously, access to some of the ongoing dictionary of the Taa Documentation Project by Christfried Naumann et al. at the DOBES archive, and we cite this as TDP.

Lemmas for content words mostly have the bimoraic form C₁V₁(C₂)V₂. Here, C₁ is a pulmonic consonant, a simplex click, complex click–obstruent cluster or even an obstruent–obstruent cluster.

Prima facie, V₁ is usually , with arising (with a few exceptions) from the raising process that is our topic. V₁ can be further specified for breathiness, pharyngealisation and glottalisation. Current accounts favour describing V₁ as underspecified vowels (or [−round], [+round]; originally proposed for ǀGui by Nakagawa Reference Nakagawa2006, Reference Nakagawa2010; extended across Khoisan by Güldemann & Nakagawa Reference Güldemann and Nakagawa2018); we follow this, with a slight modification.

C₂ is . V₂ is with possible nasalisation; or , in which case C₂ is empty.

Function words and affixes are typically CV, with C simplex.

ǃXóõ is not considered to have lexical stress, but impressionistically the first syllable is usually stronger than the second, and distributional analyses analyse the prosodic accent on the initial CV (Lionnet & Hyman Reference Lionnet and Hyman2018: 655–658). Thus the typical prosodic word is a heavy monosyllable CVV, a trochee CVCV or a dactyl CVCV-CV, a structure generally found in the Khoisan languages (Nakagawa et al. Reference Nakagawa, Witzlack-Makarevich, Auer, Fehn, Gerlach, Güldemann, Job, Lionnet, Naumann, Ono and Pratchett2023), and one typical of Romance metaphony. The language is tonal, with high or low tones on each mora in the current analysis, and complex grammatical tonal concord rules; we mark tone, but it plays no role in this article.

There are five basic clicks: . Clicks may carry manners such as aspiration, nasality, ejectivity, etc. Clicks also occur followed by glottal stop, glottal fricative or uvular stop, perhaps with ejectivity, aspiration or affrication. For example, in the East ǃXóõ dialect of Traill (Reference Traill1985), the lemma for ‘gnaw meat off a bone’ is in modern cross-Khoisan orthography): the initial dental click is prevoiced, and the posterior uvular closure is held and then released ejectively with affrication, followed by a creaky vowel. There has been a long debate about whether these sounds are clusters or single phonemes: see Bradfield (Reference Bradfield2014) and Bennett (Reference Bennett and Sands2020) for reviews. Most Khoisanists now prefer the cluster analysis; we adopt a compromise, having two root nodes attached to a single skeletal slot (cf. Bradfield Reference Bradfield2014, Reference Bradfield2018; Polgárdi Reference Polgárdi2014).

Pulmonic C₁ also ranges widely, with stops in five places of articulation , with voicing, aspiration and/or ejectivity, and (for coronal and labial stops) combined with the same uvulars that can follow clicks; also fricatives . For simplicity, the dental will just be written .

3.2. The back vowel constraint

Khoisan languages all have some form of phonetic ‘back vowel constraint’ (BVC; Traill Reference Traill1985; Miller-Ockhuizen Reference Miller-Ockhuizen, Fujimura, Joseph and Palek2000; Nakagawa Reference Nakagawa2006), informally expressed as: a back consonant is not followed by a front vowel . The definition of ‘back’ varies somewhat between languages, as does the strength of the constraint, from a tendency to a strict rule. In ǃXóõ as described by Traill, the back consonants are the clicks and the velar and uvular non-clicks. For example, the form is licit, but * is not.

3.3. ‘A-raising’

The phenomenon we here discuss is traditionally known as ‘A-raising’. It was introduced by Traill (Reference Traill1985) in the process of promoting the BVC from a phonetic constraint to a phonological constraint in ǃXóõ. He observes that not only is licit, but * is not found; moreover, V₁ = is only found before the same V₂. Traill (Reference Traill1985: 90–91) thus proposes that V₁ is always underlyingly after any click, and when it surfaces as , it does so by assimilation of to V₂.

Contrariwise, Miller (Reference Miller2011) on Juǀ’hoansi (a Kx’a language, so unrelated, but with a similar phenomenon) argues that instead underlying is lowered to before low V₂. Nakagawa (Reference Nakagawa2010) on Gǀui (of a third unrelated family, Khoe–Kwadi) suggests the intermediate view that V₁ is one of two archiphonemes , specified only for rounding, and filling in its other features from V₂. As remarked, this analysis is popular now among Khoisanists, and we adopt it, except that we use rather than , as we analyse both back and front V₁ to have a default low height rather than a completely unspecified height.

Regardless of the analysis, what makes this process particularly interesting in ǃXóõ is the detail of when the assimilation or in-filling happens. The description in Traill (Reference Traill1994) is as in (11):Footnote ⁷

In a more detailed account, Traill (Reference Traill1985) describes the raised as varying over . We use for all of these; the first author’s observations from audio data suggest the variation is fairly free.

Traill does not discuss the case of V₂ = , but in the dictionary entries themselves he notes becoming in raising contexts.

In Traill’s surviving recordings, the in (11c) is hard to distinguish from the raising described in (11a), and is predicted to be the same by the analysis of Bradfield (Reference Bradfield2014). We will not rely on it, but it turns out that our analysis can account for it as an automatic side-effect; see example (22).

Lionnet (Reference Lionnet2018) undertook a meticulous examination of Traill (Reference Traill1994), and observes that some words with explicit uvulars (namely ) are marked with full raising, contrary to (11b). However, since he wrote, the long-awaited revised dictionary (Traill Reference Traill2018) has been posthumously published, and therein the first three words have been reassigned to the corresponding non-uvular segments . Moreover, the first author’s examination of audio data for relevant words from TDP supports the original statement. The problem of ‘Nama person’ remains; this word also appears as in TDP, but unfortunately we do not have the audio. We stand by (11b) as stated, and suspect that may actually have been partially raised .

Finally, (11a) does not list the alveolars as raising consonants. Like Lionnet (Reference Lionnet2018), we shall add these.Footnote ⁸

Taking these observations into account, we restate ‘A-raising’ for the purposes of this article in (12), where (and henceforth) we write C⁺ when C is either a click involving or a dental/alveolar pulmonic, and C $^{-}$ otherwise. (Lionnet Reference Lionnet2018 writes C_[+], C_[−].) We write C^q to mean C has an explicit uvular, as have, and C for C with no explicit uvular.

We shall see that even this description is probably not entirely accurate, according to Traill’s own data – because he did not consider the archiphonemic idea, he did not consider that should never appear as underlying V₁.

4.1. Skeletal tier

We adopt a framework that distinguishes consonants and vowels by C or V slots at the skeletal tier. We are not here concerned with arguments about Strict CV phonology (Lowenstamm Reference Lowenstamm, Durand and Laks1996; Scheer Reference Scheer2004), but will interpret complex onsets ( etc.) as attached to a single C slot, and mark C₂ in the skeleton even when empty.

Each slot has one or two place nodes and a manner node, each containing elements. We are generally not concerned with the manner nodes, so we elide them from our representations.

4.2. Elements for consonants

We adopt a headed version of ET with few elements, mostly following Backley (Reference Backley2011). Sounds can contain multiple sets of place elements, used for example in English long vowels and diphthongs; we use the same technique for click consonants. We assign elements for places of articulation of simple consonants as in (13).

Here, |U| is found in both labials and velars, but is headed in labials. Similarly, |I| is found headed in palatals, but also occurs in coronals that pattern with palatals. ǃXóõ is a little unusual in that its dental/alveolar stops pattern with palatals.Footnote ⁹ Additional evidence for |I| in dentals is that the particle has a concordial form that should be , but is often realised as , by assimilation to |I|. Velars per Backley (Reference Backley2011) are just |U|, but for reasons that will be clear from analysis to follow (see also §6.1.2), in ǃXóõ they contain |A|; uvulars are distinguished by heading the |A|. (See Bellem Reference Bellem2008 for other approaches to uvulars in ET.)

If C₁ always occupies one C-slot, we must represent clicks, click–obstruent clusters (e.g., ) and obstruent–obstruent clusters (e.g., ) in a single slot. Adapting ideas from Bradfield (Reference Bradfield2018) and Polgárdi (Reference Polgárdi2014), we assume that clicks and clusters have two place nodes, notated as, for example, |A;I|. Clicks are distinguished by velarity |U| in the first slot. The clusters have the click or first obstruent elements in the first slot, and the second obstruent elements in the second. The plain click first nodes contain either |I| for the raising clicks or |A| for the non-raising clicks, with headedness employed as for pulmonics.

The simplex clicks have |U| in the second place node, corresponding to the velarity of the posterior closure release, as in (14).

The clicks with uvular accompaniments have |U.A| in the second node, corresponding to the uvular obstruent.

The clicks have an empty second place node: for example, is |I.U; $\varnothing $ |. The rationale is that these clicks do not have an audible posterior release transition: the click is followed in by a short silence before the glottal release, and in by crescendo aspiration (and, according to Ladefoged & Traill Reference Ladefoged and Traill1984, even has ingressive nasal airflow to ensure there is no air pressure to make an audible release).

Elements for manner, shown in (15), are mostly standard in the style of Backley (Reference Backley2011). The voiced versions have an additional |L|.

Thus, for example, has place nodes |I.U;U.A| and manner node |ʔ.H|.

4.3. Elements for full vowels

The basic five vowels that occur in V₂ position and in monomoraic words are represented standardly as |I|, |A.I|, |A|, |A.U|, |U|, with the use of headed elements allowing lower allophones to be represented unheaded. Manner for breathy and creaky vowels is also standard: add |H| and |ʔ|.

ǃXóõ also has pharyngealised vowels, and ‘strident’ (epiglottalised) vowels. The latter are arguably phonologically breathy pharyngealised. There is no agreed representation in ET; since we operate with distinct place and manner nodes, we can stipulate that |A| in a manner node represents pharyngealisation (for more on A and pharyngealisation, see Bellem Reference Bellem2008 and Faust Reference Faust2017).

Our analysis entails that the sound is sometimes a phonetic realisation of an empty V-slot and alternates with , following Reference LionnetLionnet’s (Reference Lionnet2018) observation of when has as an allophone. However, in other forms and contexts, is an allophone of a V-slot containing |A.I|, and it alternates with and even . Hence, we stipulate that the empty vowel | $\varnothing $ | is realised as .

The moraic nasals occupy V₂ position and (according to Traill) both trigger raising after C⁺ ₁. is dental, and so should be |I.L.ʔ|. However, Traill (Reference Traill1985) records that unlike the other dentals, it has an apical rather than laminal articulation, and auditorily it has ‘A-resonance’ with an -like quality rather than -like. We suggest that it is rather |I.A.L.ʔ|. is simply |U.L.ʔ|.

4.4. Elements for V₁

As described in §3.1, a popular and attractive analysis of V₁ in Khoisan is as an underspecified vowel carrying only a rounding value, or in ET terms presence or absence of |U|. Since the height of V₁ in bimoraic words is predictable from context, ascribing any other features than rounding to its underlying form is necessarily stipulative. However, is by far the most common vowel in the language, with around half the lexicon having V₁ = . We choose on sprachgefühlig grounds to specify with floating || to acknowledge this, and so allow our metaphonic analysis of ‘back’ consonants enabling the underlying quality. According we specify as |.U|.

5.1. Rules

The first rule, in (16), licenses the floating || in V₁ if there is |A| in either C₁ or V₂. (There is no C₂ containing |A|.) This rule is parallel to the formal description of metaphony in §2, with the addition of A-licensing from C₁. As in many instances of metaphony, V₁ corresponds to the head of the foot, and its |A| can be licensed by the foot dependent, if we accept the idea of ǃXóõ as a basically trochaic language (§3.1).

The second rule, in (17), describes spreading of (headed or unheaded) |I| from V₂, that is, the ‘raising’. The licensing element may be headed, and then heads its licensee. However, in our analysis, this only applies when there is no intervening C₂.Footnote ¹⁰ This rule parallels our definition of umlaut in (10a).

The final rule, in (18), deals with the particular case of palatal C₂, which allows raising even when V₂ is .Footnote ¹¹ It is best seen as a ‘plateau effect’, where C₁ and C₂ form an assimilatory domain for the element I, affecting the interpolated V₁. This rule only applies when the rightwards element is headed.Footnote ¹²

Plateau effects are seen in both tonal and segmental phonological processes, though they are commoner in tonal ones (Hyman Reference Hyman, Goldsmith, Riggle and Yu2011: 500–501). The ban on HLH and H $\varnothing $ H is a constraint found in several languages (Cahill Reference Cahill2007, reported in Hyman Reference Hyman, Goldsmith, Riggle and Yu2011), so much so that Yip (Reference Yip2002: 137) gives it a constraint *Trough.

Plateau effects are also reported in vowel harmony. Hyman (Reference Hyman, Goldsmith, Riggle and Yu2011: 218–219) highlights Yaka, which lowers exclusively when bounded by mid vowels on both sides: e/o…i…e/o → e/o…e…e/o. And in Woleaian (Howard Reference Howard1972; Nevins Reference Nevins2010), an underspecified vowel only raises to when surrounded by a [+high] donor on each side; for example, . This example shows a plateau effect for the same feature |I| as we propose for ǃXóõ.

5.2. Examples

We present examples discussed by Lionnet (Reference Lionnet2018) and given in (1), which are all drawn from Traill’s dictionary or text. The underlying form is on the left, and the form computed by the rules in §5.1 on the right. Unassociated floating elements are not realised in the surface form. As a notational convention, a double line represents a bundle of association lines from the slot to all the elements listed, while a single line links only to the specific element it is aimed at. Recall that the notation |A.U;U| represents a complex consonant with two place nodes.

No rules operate in (19), so V₁ surfaces with an empty place node, resulting in :

In (20), I-spreading operates, but the || remains unlicensed, resulting in a fully raised :

In (21), I-spreading and A-licensing both operate, giving |A.I| for V₁:

(22) is similar, but here || is licensed by the initial back click; the heading of the licensor transfers to the licensee, giving |A.I|, a more peripheral , accounting for the from Traill’s rule in (11c). This also covers the case of front clicks with uvular accompaniments.

In (23), the initial back click licenses |A|, but there is no I-spreading, so is realised as its default form :

5.3. Moraic nasals

Recall that ǃXóõ has moraic, tone-bearing nasals , which can take V₂ position (and in fact also V₁ in a couple of very common grammatical words). In §4.3, we argued for ascribing the elements |A.I| to , making it phonologically a blend of dental and alveolar. We can now reveal the real motivation: acts as a raising vowel like , but does not produce full raising as does. If it has |A.I|, then the following example from Traill (Reference Traill1985) is correctly predicted:

In (24), the |A| of the moraic nasal licenses the floating || in V₁ by (16), but also the |I| spreads left by (17), so V₁ surfaces as |A.I|, with a realisation. There is also unshown A-licensing from C₁, so the |A| in is not actually required; but in words such as ‘white bauhinia shrub’, it is.

Moraic also induces partial raising according to (11). However, Lionnet (Reference Lionnet2018) observes that in such words as ‘sour taste’, Traill (Reference Traill1994) does not notate raising. The first author’s inspection of data from TDP shows no sign of raising in words. We predict partial raising, as the vowel remains empty; the simplest modification to block this would be to posit that moraic , unusually, also contains |A|, for similar reasons as . (Lionnet tentatively suggests that moraic nasals raise by some other mechanism.)

5.4. Pharyngealised V₁

Traill’s rule in (11) says that A-raising does not apply to pharyngealised V₁ at all, and we have not so far addressed this. There is a natural rule that a manner |A| (if that is how we represent pharyngealisation) licenses || in the place node, which blocks full raising, and indeed strengthens it by spreading to |A|, which reduces further any raising from I-spreading – doubly linked elements tend to show special strength in phonology (Honeybone Reference Honeybone, Carr, Durand and Ewen2005).

There is very limited audio data from Traill. However, in the data from TDP, it does appear that words display some raising, to near .

We therefore suggest that there is no need to block partial raising for , and that the reduced raising can be left to phonetic implementation.

6.1. Apparent exceptions

6.2. /A/ with back rounded V2

So far, we have addressed only the realisation of V₁ = when it is followed by unrounded central/front . What happens when it is followed by back rounded ?

The element assignments and rules (16) and (17) give that after C⁺, is before and before , while after C $^{-}$ it is in both cases.

Reference TraillTraill’s (Reference Traill1994) description says nothing about , but does say that is raised and rounded to mid central rounded before – and gives an example with a back click, ‘salt pan’ → . In Traill’s published recordings, there are only two examples of such words: , which appears to be spoken with , as we predict; and ‘lie horizontally [inflected]’, which indeed appears with something like , though the rounding is light.

In TDP, words appear with a range of realisations. ‘salt pan’ is perhaps , and is perhaps similar (the words are too quickly spoken to be well analysed by spectrogram). The most strikingly affected word is (unsurprisingly) , which appears as or even fronter. On the other hand, has roughly with little rounding, while starts .

TDP has two examples of words, namely and . The centralisation of is slight, a little more in the latter.

Thus, our rules predict raising more or less in accord with the data. However, there is also some spreading back of |U| in CVV words. This could be modelled in phonology by positing that |U| (unheaded) is realised as a rounded schwa (as Backley Reference Backley2011 does for the reduced of more conservative English RP); or it could be left to phonetics.