Plosives and affricates

The four-way laryngeal contrast in plosives and affricates is maintained word-initially and word-medially and is neutralized to voiceless word-finally. The neutralized consonants are generally unaspirated, while showing some degree of aspiration utterance-finally or in careful speech (which can be regarded as a phonetic effect). Neutralization appears to be blocked when voiced plosives become prenasalized (see the section on Nasalization). As in Old Indo-Aryan (Whitney Reference Whitney1945), aspirated and breathy voiced segments do not occur more than once within a word. Breathy voiced consonants are somewhat unstable. The breathiness can be realized on adjacent vowels or lost completely; it may also appear sporadically with lexically plain voiced plosives/affricates (Mørch & Heegaard Reference Mørch and Heegaard1997: 9–10; Heegård & Mørch Reference Heegård, Mørch and Saxena2004: 31–32, 50). In our data, the contrast is relatively well preserved in isolated forms; some loss of breathiness can be observed in the passage (see e.g. /ɡʱoˈˈĩ/ ‘quotative particle’ pronounced as [ɡoˈˈĩ]).

Duration differences among the laryngeal classes can be seen in Figure 2, which plots average release durationsFootnote ³ for word-initial plosives and affricates produced by our two speakers. In plosives, average release durations are 25 ms for plain voiceless, 110 ms for voiceless aspirated, 19 ms for voiced, and 103 ms for breathy voiced. This shows an average difference of 84 ms between the plain and aspirated/breathy categories. The contrast is somewhat smaller for the retroflex place. As would be expected, affricate releases are longer, but still maintain a sizable 67 ms duration contrast between plain and aspirated/breathy categories. On average, releases in our data are 103 ms for plain voiceless, 162 ms for voiceless aspirated, 74 ms for voiced, and 148 ms for breathy voiced affricates (the category that is limited to /ʥʱ/). These results show that the laryngeal contrasts are indeed well-differentiated by duration alone (see also Prosody on f0 differences).

Figure 2 Average release duration for initial (a) plosives and (b) affricates by place (labial, dental, retroflex, alveolopalatal and velar) and laryngeal features (plain and aspirated/breathy categories, voiceless and voiced), based on a total of 440 tokens; note that the breathy voiced dental and retroflex affricates are absent from the inventory.

Aspirated plosives are optionally realized in Kalasha as fricatives: /pʰ/ as [f] or [ɸ], /t̪ʰ/ as [θ], and /kʰ/ as [x] (or [kx]). At least [f] and [x] occur primarily in lexical loans (Trail & Cooper Reference Trail and Cooper1985, Heegård & Mørch Reference Heegård, Mørch and Saxena2004), and therefore can be considered marginal phonemes. In addition, the voiced bilabial /b/ is occasionally lenited to [w] (Mørch & Heegaard Reference Mørch and Heegaard1997: 47). All these realizations are present in the recorded passage.

Kalasha plosives are produced at the bilabial, dental, retroflex, and velar places of articulation. All affricates are sibilant coronals, occurring at the alveolar, retroflex, and alveolopalatal regions. This three-way contrast is typical of languages in the Hindu-Kush region (see Palula: Liljegren & Haider Reference Liljegren and Haider2009; Khowar: Liljegren & Khan Reference Liljegren and Khan2017). All the place contrasts occur word-initially, medially, and finally. Retroflex plosives are typically retracted post-alveolars, which can vary in the degree of retraction and apicality. Affricates, on the other hand, tend to be more retroflex – produced with the tongue tip curled back (Trail & Cooper Reference Trail and Cooper1985). This was confirmed by Mørch & Heegaard (Reference Mørch and Heegaard1997: 51–52) using static palatography.

Figure 3 FFT (black) and LPC (red) spectra (made with a 46 ms window centred on the peak of noise intensity) of the frication noise in tsat ‘size’, c’ar dyek ‘ripen’, and car ‘grass’ (single tokens) produced by both speakers.

As in other languages, Kalasha retroflex plosives are acoustically more salient in post-vocalic position than in prevocalic position (see Dave Reference Dave1977 for Gujarati; Hussain et al. Reference Hussain, Proctor, Harvey and Demuth2017 for Punjabi). The following words illustrate the dental–retroflex contrast in plosives post-vocalically: /bʱot̪/ bhot ‘evening meal’ vs. /pʰoʈ/ phot’ ‘chaff, wood shavings’; /bʱut̪/ bhut ‘evil spirit’ vs. /muʈ/ mut’ ‘tree’.

In affricates, the place contrast is distinguished primarily by a combination of noise spectra and formant transitions with adjacent vowels. Spectra for sample tokens are shown in Figure 3. They indicate that the anterior affricate /ts/ is characterized by a greater concentration of high intensity noise at higher frequencies (centre of gravity (COG) is 7300–8200 Hz). In contrast, the two posterior affricates, /ʈʂ/ and /ʨ/, have a concentration of high-intensity noise at much lower frequencies (COG of 3300–3700 Hz). Both also show two main peaks, with the lower frequency one being more intense. The two posterior affricates differ from each other in the lower-frequency slope towards the peak, which is steeper for /ʨ/ than /ʈʂ/. The spectral differences among the affricates are further enhanced by CV transitions, with F2 being the highest for /ʨ/ and lowest for /ts/. F3 is lowered by /ʈʂ/. (For a more extensive acoustic investigation of the Kalasha affricates, the reader is referred to Kochetov & Arsenault, published online 17 January Reference Kochetov and Arsenault2020)

Voiced affricates can vary freely with fricatives for some speakers [ɖʐaʈʂ ∼ ʐaʈʂ]j’ac’ ‘spirit beings’ (Trail & Cooper Reference Trail and Cooper1985), but they are realized consistently as affricates in our data. Other notable variation includes the realization of voiced plosives /b/ and /ɡ/ as [m] and [ŋ] after a nasal vowel or before a nasal consonant. In utterance-final position, voiceless velar /k/ is sometimes unreleased or realized as [ʔ] or glottalization on the vowel (see Trail & Cooper Reference Trail and Cooper1985).

Fricatives

The main class of fricatives in Kalasha are sibilants, which, like affricates, occur at the alveolar, retroflex, and alveolopalatal places. As with affricates, this three-way place distinction is typical of languages in the Hindu-Kush region (Ramanujan & Masica Reference Ramanujan, Masica and Sebeok1969, Bashir Reference Bashir, Cardona and Jain2003, Arsenault Reference Arsenault2017). Voicing is contrastive in sibilant fricatives, but only voiceless fricatives occur word-finally. Non-sibilant fricatives [ɸ f θ x] occur as variants of aspirated plosives (see the discussion of plosives above).

Like affricates, the place of fricatives is distinguished by spectral shapes and vowel transitions. As shown in Figure 4, anterior /s/ has a higher-frequency concentration of energy, while the energy of posterior /ʂ/ and /ɕ/ is concentrated at lower and mid frequencies. Compared to the alveolopalatal affricate discussed above, the alveolopalatal fricative /ɕ/ shows an even sharper drop in energy at the lowest frequencies and a more defined primary peak. Like its affricate counterpart, it substantially raises F2 in CV transitions (Mørch & Heegaard Reference Mørch and Heegaard1997: 66). The retroflex fricative lowers F3 (Mørch Reference Mørch1995) in a similar way to the retroflex affricate.

The voiceless laryngeal /h/ is highly variable. Between vowels or next to sonorants, it is realized as breathy voiced [ɦ] or lost altogether. There are also cases of /h/ shifting to another syllable or [ɦ] being produced in words where it is absent phonemically. The latter has been observed in pretonic intervocalic position, where [ɦ]-insertion is an apparent hiatus resolution strategy (Mørch & Heegaard Reference Mørch and Heegaard1997: 47–48; e.g. /ʂo˞ ˈek/ ʂo˞ ˈɦek]s’o’ek ‘to plug’ produced by SK). This highly labile behavior of /h/, combined with the variability of breathy voiced plosives/affricates (see above), suggests the possibility of analyzing aspiration/breathiness in Kalasha as a word-level prosody (Heegård & Mørch Reference Heegård, Mørch and Saxena2004: 31–32).

Figure 4 FFT (black) and LPC (red) spectra (made with a 46 ms window centred on the peak of noise intensity) of the frication noise in /sak/ sak ‘very great’, s’a ‘king’, and shak ‘vegetable’ (single tokens) produced by both speakers.

Nasals

There are fewer place contrasts for nasals than for plosives. Bilabial /m/ and dental/alveolar /n/ are well established, contrasting in a variety of contexts. The status of retroflex /ɳ/ is more ambiguous. It contrasts marginally with /n/ in word-final position as a result of (variable) consonant cluster simplification: [moɳ ∼ moɳɖr]mond’r ‘whey’ as opposed to [mon ∼ mond̪r]mondr ‘word’. However, some researchers report that it may occur independently in a small number of words (Mørch & Heegaard Reference Mørch and Heegaard1997: 51; Trail, p.c. 2009). An articulatory investigation is required to confirm these observations, as dental and retroflex nasals are difficult to distinguish reliably on purely acoustic and auditory grounds. Moreover, native speakers do not appear to have clear intuitions about retroflex nasals, apart from those arising through final cluster simplification. The process of cluster simplification can also result in word-final alveolopalatal and velar nasals, which are otherwise sub-phonemic, occurring only before homorganic affricates and plosives (see Mørch & Heegaard Reference Mørch and Heegaard1997).

Laterals

Kalasha has two lateral approximants. /ɫ̪/ is a velarized apico-dental, while /lʲ/ is a palatalized lamino-alveolar (Trail & Cooper Reference Trail and Cooper1985), as in /ɫ̪ap/ l’ap ‘quickly’ and /l^jʲaw/ law ‘harvest’. These differences of articulation were corroborated by Mørch & Heegaard (Reference Mørch and Heegaard1997: 52; see also Heegård & Mørch Reference Heegård, Mørch and Saxena2004) using static palatography. Acoustically, the laterals differ in F2 during the closure; F2 is lower for /ɫ̪/ and higher for /lʲ/, indicative of moderate velarization and palatalization. The velarized lateral induces some backing in adjacent vowels, while the palatalized lateral induces fronting. This can be seen in Figure 5, which provides an annotated spectrogram of the two laterals occurring word-initially and word-finally. Note that the place difference in F2 during the closure is about 700–800 Hz. Our larger-scale acoustic investigation of Kalasha laterals (produced by 14 speakers of different age groups and genders) revealed that /ɫ̪/ velarization was clearly present in younger speakers and largely absent in older speakers (Kochetov, Heegård Petersen & Arsenault Reference Kochetov, Heegård Petersen and Arsenault2020). This suggests that palatalization is the main distinctive feature of the contrast, while velarization has developed more recently, either as an enhancement strategy or due to the influence of Khowar.Footnote ⁴ Thus, the contrast can be alternatively classified as /l̪/ vs. /lʲ/.

Figure 5 Formant patterns (Hz) in words with contrastive laterals: l’ap ‘quickly’ vs. law ‘harvest’ (top), and wal’ ‘protection’ vs. wal ‘pity’ (bottom) by speaker SK.

Rhotics

The phoneme /r/ has a rather variable realization. It tends to be weakly trilled or fricated word-initially and after a plosive (Mørch & Heegaard Reference Mørch and Heegaard1997: 51). Similar, but devoiced, realizations occur utterance-finally. Intervocalically, /r/ is often a tap. Our data confirm these observations, while also pointing to occasional retroflex approximant-like realizations (see e.g. [pʰaɻ]phar ‘load’ by SK). Utterance-initially, /r/ may be preceded by a short vocoid (e.g. /ra/ ra ‘way’ as [^ᵊra]). Next to a nasalized vowel, /r/ is optionally realized as [ɾ͂] (e.g. the first instance of /ˈsirã/ sirã’ ‘wind’ in the passage, realized as ˈsiɾ͂ã]).

In addition to /r/, the Birir variety of Northern Kalasha has a retroflex flap /ɽ/. This phoneme does not occur in the Bumburet variety described here, which has retroflex vowels instead (see the discussion of vowels, below). (See also Mørch & Heegaard Reference Mørch and Heegaard1997: 63–64 and 80–85; Heegård & Mørch Reference Heegård, Mørch, Hansen, Hyllested and Jørgensen2017: 243 on rhotics in other varieties of Kalasha.)

Glides

There are two glides in Kalasha. The bilabial /w/ is often realized in onset position as labiodental [ʋ] or [v], especially before front vowels (Trail & Cooper Reference Trail and Cooper1985; Mørch & Heegaard Reference Mørch and Heegaard1997: 52–55). The palatal /j/ shows little variation.

Consonant clusters

Onset clusters consist of obstruent + liquid or /j/. Examples include /praɕ/ prash ‘soft’, /t̪rip/ trip ‘sickness’, /ʈrits/ t’rits ‘kind of bird’, /ˈkriʂna/ kris’na ‘black’, /blʲats/ blats ‘short’, and /kɫ̪uʨ/ kl’uc ‘quickly’. Sequences of dental plosive + /j/ are realized as (alveolo)palatal plosives (e.g. /d̪jek/ in dhar dyek ‘to lean’ as [ɟek]). Some initial clusters can be broken up with a short epenthetic vowel (e.g. [s^ᵊras(t̪)]srast ‘avalanche’; see Mørch & Heegaard Reference Mørch and Heegaard1997). Trail & Cooper’s (Reference Trail and Cooper1999) dictionary also includes sequences of sonorant + /h/, as for example in /mhalʲ/ mhal ‘curse’, /ɫ̪ho˞ʂʈ/ l’ho’s’t’ ‘charcoal’, and /rha/ rha ‘cedar’. In our data, however, these words tend to lack frication or breathiness. They are pronounced much like the single-onset sonorants in /malʲ/ mal ‘goods’, /ɫ̪om/ l’om ‘a measure’, and /ra/ ra ‘way’. The word /ˈlʲhojak/ lhoyak ‘smooth’, as produced by speaker SK, is an exception: /h/ is realized here as breathy voiced [ɦ] and preceded by a short epenthetic vowel ([lʲəˈɦojak], which makes it clearly different from the onset in /lʲots/ lots ‘lightweight’). This indicates that sonorant + /h/ sequences are only marginally present in this variety of Kalasha.

Final coda clusters include: homorganic nasal + plosive; fricative + plosive; dental plosive + /r/; and approximant /w, j/ + consonant. Examples include /ɖoɳɖ/ d’ond’ ‘double bride-price’, /prost̪/ prost (hik) ‘to bow’, /ɡoʂʈ/ gos’t’ ‘barn’, and /ʈʂʰet̪r/ c’hetr ‘field’. In many lexical items final clusters are simplified by leaving the second consonant unpronounced, as for example in [ɖoɳ ∼ ɖoɳɖ]d’ond’ ‘double bride-price’, [d̪raʂ ∼ d̪raʂk]dras’k ‘small amount’, and [s^ᵊras ∼ s^ᵊrast̪]srast ‘avalanche’.

Nasalization

Following Trail & Cooper (Reference Trail and Cooper1985, Reference Trail and Cooper1999), vowels that occur before a homorganic nasal + plosive sequence are analyzed as nasalized vowels followed by a plosive; the homorganic nasal is treated as non-phonemic. Thus, a broad transcription such as /o͂ɡ/ for õg ‘thigh’, shown above, typically corresponds to [o͂^ŋ ɡ] or [o͂ŋɡ] in narrow transcription (see Masica Reference Masica1991 for discussion of this issue in other Indo-Aryan languages). (Note that final devoicing does not apply here, being blocked by consonant nasalization.Footnote ⁵) In some cases, the nasalization spreads through the following plosive resulting in realizations such as [ba͂ŋ] for /ba͂ɡ/ (in bãg dyek ‘to poke’). In polysyllabic words, the precise locus of nasalization can vary, as nasalization optionally spreads to vowels in adjacent syllables (e.g. [ˈwe͂.a ∼ ˈwe.a͂ ∼ ˈwe͂.a͂]wẽa ‘far upstream’; see Mørch & Heegaard Reference Mørch and Heegaard1997: 43–44).

Retroflexion

Retroflexion is the most intriguing aspect of the Kalasha vowel system. Kalasha may be the only known language with a fully symmetrical set of retroflex (or rhotacized) and non-retroflex vowel phonemes.Footnote ⁶ The precise articulatory mechanism involved in the production of these vowels is not clear. Trail & Cooper (Reference Trail and Cooper1985) suggest that the vowels are produced by ‘slightly curling’ the tip of the tongue, but Heegård & Mørch (Reference Heegård, Mørch and Saxena2004) found variation among their informants, who produced the vowels either by curling the tongue tip or by raising and bunching the middle of the tongue (see Ladefoged & Maddieson Reference Ladefoged and Maddieson1996). Despite the variation, Heegård & Mørch favour the term ‘retroflex’ over ‘rhotacized’, arguing that retroflexion of the tongue tip appears to be the most common strategy for producing the vowels. Whether this is correct or not (see Hussain & Mielke Reference Hussain and Mielke2018 for some ongoing articulatory work), the term ‘retroflex’ seems appropriate given the phonological patterning of the sounds (as part of a larger system of consonants and vowels) and their historical sources.

Historically, the retroflex vowels evolved from plain vowels primarily under the influence of a following intervocalic retroflex plosive or nasal, or Old Indo-Aryan (OIA) /r/ in its various phonetic forms. The consonants were eroded through lenition, leaving retroflexion on the vowels (and nasalization, in the case of [ɳ]); e.g. /aˈʐa˞i/ az’a’i ‘apricot’ < OIA /aʂaɖʱiːja/; /ˈg^ɦũi˞/ ghũ’i ‘goat-hair coat’ < OIA /ɡoːɳiː/ (Mørch & Heegaard Reference Mørch and Heegaard1997: 77–86; Trail & Cooper Reference Trail and Cooper1999; see also Morgenstierne Reference Morgenstierne1973: 191; Heegård & Mørch Reference Heegård, Mørch and Saxena2004: 67–72). A retroflex flap [ɽ] is still preserved before /i/ in the Birir (Biriu) variety of Kalasha, and sporadically in others (compare Birir /aˈʐaɽi/ ‘apricot’ and /ˈɡu͂ɽi/ ‘goat-hair coat’ with their Bumburet cognates, listed above; Heegård & Mørch Reference Heegård, Mørch and Saxena2004, Reference Heegård, Mørch, Hansen, Hyllested and Jørgensen2017; see also Di Carlo Reference Di Carlo2010).Footnote ⁷

Acoustically, retroflex vowels are more centralized than their plain counterparts, and are most clearly characterized by a lower F3 (see Heegård & Mørch Reference Heegård, Mørch and Saxena2004). This is evident in Figure 6, which plots the formants F1 and F2 (the top plots), and F2 and F3 (the bottom plots) for non-nasalized vowels produced by our two male speakers.Footnote ⁸ Note that the retroflex vowels produced by SK are more centralized (in F1 and F3) than those for TKK; both speakers, however, show similar differences in F3 (see also Kochetov, Arsenault & Heegård Petersen Reference Kochetov, Arsenault, Heegård Petersen, Calhoun, Escudero, Tabain and Warren2019).

Figure 6 F1/F2 (top, Hz) and F2/F3 (bottom, Hz) plots showing means (indicated by IPA sound symbols) and 2 standard deviations (indicated by ellipses) for oral vowels produced by the speakers SK and TKK (based on 157 and 140 tokens, respectively); the vowels occur in words with no other retroflex vowels, retroflex consonants or /r/.

As with nasalization, the locus of retroflexion can vary in polysyllabic words, and sources differ as to which vowels are transcribed as retroflex (e.g. [ˈku.a˞k ∼ ˈku˞.ak ∼ ˈku˞.a˞k]kua’k ‘child’; see also the words ‘to plug’ and ‘to warn’ in the table above). Moreover, retroflexion tends to ‘spread’ from retroflex vowels to plain vowels in adjacent syllables, often permeating intervening consonants and crossing morpheme boundaries (Mørch & Heegaard Reference Mørch and Heegaard1997: 44; Arsenault & Kochetov, Reference Arsenault and Kochetovin press), as, for example, in [t̪raˈpu˞i ∼ t̪raˈpui ∼ t̪raˈpui]trapu’i ‘eye disease’. In our data, retroflex harmony of this kind is bidirectional, but predominantly progressive, and tends to spread through non-coronal consonants (other than /j/).

Allophonic variation

Vowels show some moderate variation depending of the place of articulation of adjacent consonants, and most notably next to alveolopalatals and labials. As can be seen in Figure 7, the front mid vowel /e/ appears to be most variable, showing some differences in height, [e ∼ ɛ] (see Trail & Cooper Reference Trail and Cooper1985; Mørch & Heegaard Reference Mørch and Heegaard1997: 41). All vowels are considerably fronted and show an [j]-like offglide/onglide next to alveolopalatals. As a result, back vowels in this context can be narrowly transcribed as [ʉ]/[u̟], [o̟], and [Æ] (Mørch & Heegaard Reference Mørch and Heegaard1997: 41). In addition, all vowels show some lowering in F3 (not shown in the figure) next to retroflex consonants and /r/. The degree of this retroflexion appears to be greater in the context of retroflex fricatives. At least some of these cases can be considered categorical assimilation resulting in a neutralization of the retroflex contrast for vowels in that context.Footnote ⁹

Figure 7 F1/F2 (Hz) plots showing means (indicated by IPA sound symbols) and 2 standard deviation ellipses for oral stressed vowels in CVC syllables by selected places of articulation of the preceding and the following consonant, as produced by SK (based on 334 and 267 tokens, respectively); the consonant places are ‘lab(ial)’, ‘pal(atal)’, and ‘vel(ar)’ (with other places, as well as the consonants /r j w/ excluded).

Vowel length

Vowel length is closely tied to prosody in Kalasha (discussed below) and is not generally considered phonemic. However, some researchers have suggested that there may be limited contrast. Bashir (Reference Bashir1988: 35) notes that vowel length and stress may be independent variables in Kalasha, although she does not elaborate. Heegård & Mørch (Reference Heegård, Mørch and Saxena2004) suggest that length may be contrastive, at least for /a/ in open stressed non-final syllables (see Heegård Reference Heegård, Niemi, Odlin and Heikkinen1998), but they leave the question open for future, more focused studies.

Sequences of vowels

Sequences of two vowels are common and can include sequences of identical vowels. Vowel sequences are always syllabified as two syllables (e.g. ˈʂo˞/.ek/ s’o’ek ‘to plug’, /ˈɡri.ik/ in ku’ griik ‘to get singed’) (and variably separated by [ɦ], as noted above). Hence, they are regarded as true sequences as opposed to diphthongs or long vowels.