Consonants

Below we illustrate near-minimal pairs for consonants in the frame /CəC(C)(C)/. Because /ə/ surfaces as [а] after /ħ/, this consonant is shown with a following [а] instead. As the surface representations demonstrate, the vowel /ə/ has many allophones. The allophony is governed by neighboring consonants and discussed further in the section Vowels. For all of the words below, the suffix /-k҆/ is the indefinite article. Some of these forms may be loans from standard Abkhaz in Mr. Pysipa’s speech, and may differ slightly from the corresponding forms in other varieties of Abkhaz (Chirikba, p.c., Hewitt p.c.). Cwyzhy has a larger consonant inventory than Abzhywa due to additional coronal contrasts. For example, Abzhywa does not contrast /sʷ ʃʷ/ as Cwyzhy does (Hewitt Reference Hewitt1979). The larger coronal inventory is shared with Bzyp (Chirikba Reference Chirikba2003). Chirikba (Reference Chirikba and Inal-Ipa2014: 298) states that Cwyzhy also possesses pharyngealized voiceless uvular fricatives /χ҆/ and /χʷ҆/, but Mr. Pysipa’s variety lacks these. Apart from this the Cwyzhy consonants presented here are identical to those Chirikba (Reference Chirikba and Howard1996a) describes for Sadz dialects.

Realization of consonants

In word-initial position the voiced stops to our ears sound slightly imploded in Mr. Pysipa’s speech, and the non-ejective voiceless stops and affricates are heavily aspirated. In word-final position, voiced stops may be devoiced and aspirated, so that /а-kʰalbad/ ақаᴫбад ‘sock’ can surface as [akʰaɫˈbatʰ]. Some final stops, particularly ejectives, may have a relatively weak final release (noted for Sadz dialects by Chirikba Reference Chirikba and Howard1996a: 67). The labialized anterior fricatives and affricates, /sʷ zʷ tsʰʷ tsʷ҆ dzʷ/, are slightly palatal, as seen in the transcriptions above. Mr. Pysipa’s /ɾ/ has a degree of pharyngealization, and may also surface as an approximant, especially when geminated. The fricatives which we transcribe /ʃ ʃʲ/ for Cwyzhy Abkhaz correspond to /ʂ ʃ/ respectively in the literary Abzhywa dialect. Unlike Abzhywa, Cwyzhy does not have any appreciable retroflexion in either of these consonants.

Abkhaz has long consonant clusters, which we illustrate in the spectrograms and waveforms below with CCC clusters consisting only of obstruents. They are representative of a general pattern in the language where each consonant is released separately, and consonants are not typically elided in these circumstances. However, in fast speech, there may sometimes be no acoustic evidence for particular consonants. Examples can be seen in the narrow transcriptions in the list of minimal pairs above. The absence of acoustic evidence does not imply that no articulatory movements for the relevant consonants are being performed.

The waveforms in Figures 2 and 3 show that compared to aspirated stops, ejectives typically have a louder and shorter release burst.

Figure 2

Waveform and spectrogram of [tsʰəpʰχʲk҆] ‘a spark’.

Figure 3

Waveform and spectrogram of [məskʰʲk҆] ‘a musk, good-smelling soap’.

Labialization

Labialization has three different realizations in Cwyzhy:

1. i. The labialized coronal stops, /tʰʷ dʷ tʷ҆/, are coarticulated with a labial stop, as in /tʷ҆ə-k҆/ Təьɪκ ‘a slave’, realized as [҆ək҆].

2. ii. The labialized coronal fricatives and affricates, /sʷ zʷ ʃʷ ʒʷ tsʰʷ dzʷ tsʷ҆/, and /ħʷ/, have a front rounded secondary articulation, as in /zʷə-k҆/ Жəьɪκ ‘an old thing’, realized as [ʑᶣək҆].

3. iii. All other labialized segments have a back rounded secondary articulation, as in /qʷ҆əd-k҆/ ҟəьɪдκ ‘a bean’, realized as [qʷ҆ɵdk҆].

A fourth type of labialization, labiodentalization, is found in some varieties (see Catford Reference Catford, André and René1972, Reference Chirikba, Yuri and YuryChirikba, to appear), but is absent from Mr. Pysipa’s speech.

Labial–coronal double articulations as shown in (i) are extremely rare cross-linguistically, and have previously been hypothesized not to exist (Maddieson Reference Maddieson1983, Reference Maddieson1987). Ladefoged & Maddieson (Reference Ladefoged and Ian1996: 343–348), however, argue that true labial–coronal sounds occur either allophonically or phonemically in a handful of languages. Concerning the Caucasian cases, including Abkhaz, they say that ‘it might be more justifiable to consider this gesture as phonetically a secondary articulation’ (Ladefoged & Maddieson Reference Ladefoged and Ian1996: 344), a conclusion that we do not think follows from the descriptions cited. Catford is explicit that these stops involve ‘complete labial closure’ (Catford Reference Catford, André and René1972: 681), and the closure is also described as ‘complete’ by Hewitt (Reference Hewitt1979: 256). As the degree of stricture is the same at labial and coronal places of articulation, we feel that the term ‘double articulation’ is justified.

Catford does mention that the closure is between the inner surfaces of the lips (endolabio–endolabial in Catford’s (Reference Catford1977) terminology), unlike for plain labial stops, where the contact is typically between the outer surfaces (exolabio–exolabial). These descriptions match our impression of the labial–dental stops in Cwyzhy, from video footage of Mr. Pysipa and his mother. We illustrate the contrast between Mr. Pysipa’s [p҆] and [҆] in Figure 4.

Figure 4

Left: lip closure for [p҆] in [ˈdaːjzaːp҆] ‘he came’. Right: lip closure for [҆] in [iːuːˈzəmɪtʃʰʲħaɾa҆ə] ‘as hard as he could’.

The closure of the lips on the left appears slightly less dark. As the closure for [p҆] is between the outer surfaces, the typical dark vermilion zone of the lips is obscured. When the inner surfaces form the closure, as on the right, the vermilion zone is more visible, leading to a darker appearance. However, this visual evidence is also consistent with an incomplete closure for [҆]. Articulatory work is necessary to shed additional light on exactly how [p҆] and [҆] are distinguished.

Lateral allophony

Andersson (Reference Andersson2017) reports on an acoustic study of lateral allophony in Georgian and Cwyzhy, on which the current section builds. Variation in /l/ has not previously been reported for Abkhaz, but impressionistically [l] and [ɫ] both occur. We therefore decided to study how the backness of intervocalic /l/ is affected by the backness of preceding and following vowels. The data are taken from previous Cwyzhy recordings (Vaux & Pysipa Reference Vaux and Zihni2020). While the recordings do contain full sentences, most of the data consists of single words elicited orthographically from English and Turkish translations. The style is therefore most closely comparable to wordlist speech. Nevertheless, it is important to note that the data are less controlled than single words recorded in a carrier sentence, for example. Care must therefore be taken when interpreting the data, which we have not been able to control for factors such as utterance length and phrasal position.

We extracted 54 tokens of the lateral, with as many preceding and following vowel qualities as possible. Only intervocalic tokens were studied. Because of the small vowel inventory of Abkhaz, as well as the nature of vowel coloring (see the Vowels section), it is relatively rare to find /l/ in certain contexts. For example, the sequence [iːliː] requires an underlying sequence of five segments /əjləj/. Because of this it was necessary to include some repeated tokens of the same word. The 54 tokens come from a set of 33 words.

Praat (version 6.0.29, Boersma & Weenink Reference Boersma and David2017) was used to analyse the recordings. The segmentation criteria employed for the lateral are given in (7). The list draws heavily on Skarnitzl’s (Reference Skarnitzl, Anna and Robert2009) work on Czech, which Andersson (Reference Andersson2017) found to be applicable to Georgian and Cwyzhy.

1. (7) Segmentation criteria for the lateral

   1. a. weak F2 (relative to surrounding vowels)

   2. b. lower intensity in higher frequency ranges (2 kHz and above)

   3. c. amplitude dip in the waveform

   4. d. simplification of the shape of the waveform

   5. e. antiformant between 2 kHz and 3 kHz

   6. f. visible release of tongue tip contact

In some cases, these criteria did not clearly identify the beginning and end points of the lateral. In such cases, a conservative segmentation was chosen, giving the lateral a shorter duration rather than a longer one. The surrounding vowels were segmented using standard segmentation criteria (Peterson & Lehiste Reference Peterson and Ilse1960).

The measure chosen for backness was F2 at the midpoint of the lateral (see Recasens, Fontdevila, & Dolors Pallarés Reference Recasens, Fontdevila and Pallarès1995, Andrade Reference Andrade, Ohala, Yoko, Manjari, Daniel and Bailey1999, Carter & Local Reference Carter and John2007, Yuan & Liberman Reference Yuan and Mark2009). For the surrounding vowels, F2 at the midpoint was again used. Measurements were made automatically in Praat. Each token was inspected manually to confirm that Praat’s formant tracking was accurate. Where it was not, the formant tracking settings were adjusted to obtain an accurate automatic measurement, following the methodology used by Frisch & Wodzinski (Reference Frisch and Sylvie2016).

It is clear that there are some tokens with relatively clear [l] (brighter shade), especially when preceding and following vowels are front (top right of Figure 5). Similarly, when both preceding and following vowels are back (bottom left of Figure 5), a more velarized [ɫ] (darker shade) appears. Example waveforms and spectrograms illustrating clear and velarized laterals are shown in Figures 6 and 7.

Figure 5

F2 (Bark) of the lateral (color gradient) by F2 (Bark) of preceding (x-axis) and following (y-axis) vowels. Figure made in R (R Core Team 2019) with a modified version of the function CGSPlot from Brinda (Reference Brinda2020).

Figure 6

Spectrogram of a velarized lateral between central vowels in [aːˈɡəɫaɾa] ‘to stop’. The duration of the [əɫa] sequence is 266 ms.

Figure 7

Spectrogram of a clear, non-velarized lateral between front vowels in [aˈnejliːk҆aː] ‘the message’. The duration of the [eːliː] sequence is 302 ms.

Vowels

Varieties of Abkhaz and other Northwest Caucasian languages are usually analysed with two underlying vowel phonemes, viz. /a/ and /ə/ (see Spruit Reference Spruit1986: 81–82 for discussion), and Cwyzhy is no different in this respect (see Chirikba Reference Chirikba and Howard1996a on Sadz dialects generally). If unaffected by neighboring consonants, these two vowels surface as [a] and [ə] (see below for more detailed phonetic data).

However, each vowel typically acquires the additional vocalic features [high], [back], [round] from the features of the immediately preceding and following consonants, if they are present. This gives rise to a much larger inventory of vowels on the surface. Some works do not transcribe these allophones (e.g. Spruit Reference Spruit1986), but where details are given (e.g. Hewitt Reference Hewitt2010), they generally agree with what we present for Cwyzhy below. Very similar patterns of allophony can also be found in other Northwest Caucasian languages (e.g. Kabardian; Kuipers Reference Kuipers1960, Allen Reference Allen1965, Anderson Reference Anderson, Alan and Hooper1978, Choi Reference Choi1991, Colarusso Reference Colarusso1992, Wood Reference Wood1994).

We extracted three tokens of each of the 13 words above from the materials in Vaux & Pysipa Reference Vaux and Zihni2020. These are subject to the same constraints and limitations as the lateral data from the same source that were discussed earlier. As some vowels appear only after back consonants, words were generally chosen which had the relevant vowel after a velar or uvular. Occasional exceptions had to be made, either due to restrictions based on coarticulation (see below), or because of a lack of relevant words in our materials. In almost all cases, we used three repetitions of the same word but in some cases morphologically related forms of a word were used to get three tokens. The vowels were segmented in Praat using standard segmentation criteria for vowels (Peterson & Lehiste Reference Peterson and Ilse1960). F1 and F2 were measured at the midpoint of each vowel in Praat. All tokens were manually inspected to verify the automatic measurements. Where errors were present, the formant tracking settings were adjusted so that an accurate automatic measurement could be made. Figure 8 shows the realization of the surface vowels of Cwyzhy.

Figure 8

F1–F2 plot (Bark scale) for Cwyzhy vowels.

The relationship between the two contrastive vowels, and the thirteen realizations that we have identified, is shown in Table 1.

Table 1

Variation in vowels conditioned by surrounding consonants.

Note that the allophony of Cwyzhy is variable, and as our phonetic transcriptions throughout the article suggest, for many of the given environments, the [a] and [ə] qualities may also appear.

In some cases the glides /j w ɥ/ appear to surface as long vowels when syllabic: [iː]/[ɪː], [uː]/[ʊː], and [yː]/[ʏː], respectively. For example, when /-j/ ‘and’ is attached to a consonant-final noun, the result is a long vowel:

1. (8) Apparent syllabic glides

   

However, these syllabic glides occur in precisely the environments where we would independently expect epenthetic [ə] (see Spruit Reference Spruit1986: 83). An alternative analysis, therefore, would derive these long vowels from the sequence [ə] + glide which arises from epenthesis. Vowel coloring by glides is independently motivated in this environment (see Table 1), and produces the correct surface forms with long vowels.

Some may feel that a two-vowel analysis is needlessly abstract. Why, for example, would we think that [s̩ˈtsʰoːt҆] ‘I am going’ is really derived from /s-tsʰa-ˈwa-jt҆/ (Chirikba Reference Chirikba and Howard1996a: 76)? It is worth pointing out that while many allophonic rules assume fairly abstract underlying forms, vowel coloring is productive across morpheme boundaries and is sometimes found even across word boundaries:

1. (9) Justification for vowel coloring

   

Some have attempted to reduce the vowel inventories of Northwest Caucasian languages further, to one vowel (Allen Reference Allen1965 on Abaza, with discussion of Kabardian and Abkhaz, Anderson Reference Anderson, Alan and Hooper1978 on Abaza and Kabardian), or even no vowels at all (Kuipers Reference Kuipers1960 on Kabardian). We use two vowel phonemes in light of minimal pairs for the two vowels, both of which contrast with the absence of a vowel:

1. (10) Minimal pairs for vowels

   

For further discussion of the contrast between /ə/ and the absence of a vowel, and other ways of accounting for it, see the section on stress, below. Some argue for a third vowel phoneme, long /aː/, in the related Circassian languages (Appelbaum & Gordon Reference Appelbaum, Matthew, Chundra, Shinae and Sandy2013: 14 and references therein). Cwyzhy also has a phonetic long [aː], and there are some phonological arguments for analysing it as a single phoneme rather than a sequence /aa/. While the sequences [aj] or [aw] do not appear because of vowel coloring, [aːj] and [aːw] appear freely:

1. (11) Differences between long and short /а/

   

Despite this, we believe that [aː] should be analysed as a sequence /aa/ at the phonological level. In some cases, the underlying geminate status is clear from morphological considerations. For example, in [aːˈɡəɫaɾa] aaгьɪᴫapa ‘to stop’, the first /a/ is a separate morpheme, which together with a suffix /ɾa/ marks the verb as nonfinite. Moreover, the second /a/ of an /aa/ sequence can bear stress independently of the first, as in /aˈalaɡəjt҆/ [aˈaɫaɡiːt҆] ааᴫагьɪит ‘… began.’ This form would begin with /j-/ ‘it, they’ in isolation, but when the subject of the verb immediately precedes it, this agreement marker is not present.

StressFootnote 3

Word-level stress is contrastive in Abkhaz, as in the minimal pair in (12):

1. (12) Minimal pair for stress

   

Stress in Abkhaz has not previously been studied acoustically, although intensity is claimed to be one important cue (Arshba Reference Arshba1979: 7). The spectrograms in Figure 9 show a near-minimal pair for stress, where the stressed syllable has higher pitch and intensity. Stressed vowels also appear to be slightly longer. Future work studying stress more systematically is needed to test how representative these patterns are of the language.

Figure 9

Initial stress on [ˈala] ‘the eye’ (left), and final stress on [aˈlːa] ‘the dog’ (right). Pitch (100–175 Hz) is shown in blue, and intensity (50–75 dB) in yellow.

Segments in Abkhaz are underlyingly specified as either accented (marked with an underscore) or unaccented (Kathman Reference Kathman, GermÁn, Benjamin and Hee-Rahk1992: 211).Footnote 4 The stress is then determined in most cases by Dybo’s Rule (Dybo Reference Dybo1977). In Spruit’s (Reference Spruit1986) formulation, Dybo’s Rule says: Assign word stress to the leftmost accented segment not immediately followed by another accented segment. If a string contains no such segments (either because all segments are accented, or because none are), assign final stress.

For example, in [ˈaχʷːa] ‘the ash(es)’ above, the root /χʷːa/ contains no lexical accents. The prefix /a-/ is accented, so it is the leftmost accented segment not followed by an unaccented segment. By Dybo’s Rule, the prefix is therefore stressed. By contrast, the vowel and consonant in the root /χʷːa/ ‘peak’ bear lexical accents. Since the word /a-χʷːa/ ‘the peak’ contains only accented segments, the word has final stress: [aˈχʷ׃a].

Lexical accent and stress are relevant to the status of underlying /ə/ in Abkhaz. Consider the minimal pair given for the /ə/–Ø contrast in (10) above, repeated here for convenience:

1. (13) /ə/–Ø

   

Note that the second word, unlike the word for ‘brown’, has stress on the schwa rather than the [a]. On an analysis that takes lexical accent into account, this is not a coincidence. For Spruit (Reference Spruit1986), both words end underlyingly in /z/. The difference between the two lies in lexical accent: [aˈzə] ‘for it’ has two accents, /a-z/, and the accented /z/ triggers schwa epenthesis. [az] ‘brown’ has its only accent on the prefix, /a-z/, so there is no epenthesis. In our Cwyzhy data, all /ə/–Ø minimal pairs differ in stress in this way, which might make it possible to analyse the language with only one underlying vowel. Whether or not such an analysis is viable depends on the status of unstressed schwa. Yanagisawa (Reference Yanagisawa2005) argues that unstressed schwas are also predictable from general phonotactic constraints, while Vaux & Samuels (Reference Vaux, Bridget, Diane and Lee2018) argue that they are not fully predictable in Cwyzhy.

Transcription of recorded passage

We provide phonemic, phonetic, and orthographic transcriptions of ‘The North Wind and the Sun’ in the Cwyzhy dialect, translated from the Turkish version (Zimmer & Orgun Reference Zimmer and Orhan1992) by Mr. Pysipa. ‘The North Wind and the Sun’ has also been translated into the literary Abzhywa dialect by Zaira Khiba (published by Hewitt Reference Hewitt1978). In the phonemic transcription, as elsewhere in this article, we treat /ə/ as underlying even in cases where it could be analysed as epenthetic.

As is common in Abkhaz storytelling, this story uses evidential endings on verbs. We have not translated the evidential meaning here, retaining the traditional English translation of the story.

Phonemic transcription

Phonetic transcription

Orthographic version