DUHUMBI AND GLOTTAL CONSTRICTION

The phonologic position of the glottal stop /ʔ/ in Duhumbi is elusive. The glottal stop and glottal constriction are clearly phonetic features among all speakers of all varieties of the . However, there is no convincing evidence that it occurs as a distinctive phoneme. Rather, the glottal stop co-occurs as a glottal closure on onset /a, e, i, /, as a sub-phonemic feature combined with creaky and rising pitch on certain open vowels /a, e, o, u/, or as of the , mainly the velar /k/, in coda position in certain phonotactic environments. In the following description, the first section describes the occurrence of the glottal stop in syllable- initial position. The next section describes the occurrence of the glottal stop as allophone of the (velar) plosive in coda position. Debucallisation of the syllable-final (velar) plosive can be shown to result in sandhi forms depending on morphophonotactic position of the velar plosive. Because of this diachronic process of lenition, minimal pairs that appear to show the opposition between the plosive /k/ and the glottal stop /ʔ/ are shown to solely depend on the morphophonotactic position. The final section describes the distinctive Duhumbi vowels /aq, eq, oq, uq/.

GLOTTALISED VOCALIC ONSETS

In Duhumbi, there are not many instances of onset vowels in unique roots, mainly limited to prefixes, interjections and Bodish and Tshangla loans, with only a limited number of occurrences in native lexemes. Excluding the prefixes, the most common onset is /u/, followed by /a/ and /o/. There are only a handful of instances of onset vowels /i/, all in native lexemes, and /e/, mainly in loans. Three of the most widely used prefixes have a vocalic onset, namely the kinship prefix a- (marginally o-/u-); the adjective prefix o- and the adjective prefix u-. Hence, onset vowels are common in lexemes containing the basic question morpheme a, e.g. abe ‘what’; adi ‘how’; apʰi ‘how much’; and the basic demonstrative o this and its derivatives, e.g. 0ʨʰi ‘this here’, okʰo ‘there’, ose ‘like this/that’. They also occur as prefixes on nouns such as okʰow ‘liquid’; ojow ‘fat’; utɕʰaŋ ‘lid’; udam ‘piece’; uguŋ ‘spirit, shadow’; ulap ‘leaf’; uli ‘seed’; uɕu ‘fruit’; utʰuŋ ‘stem’; ujuŋ ‘handle’, where they often correspond with a- in other Western Kho-Bwa . Other occurrences vocalic onsets in native lexemes include ak {ta} ‘get stuck’; an {da} ‘select’; ar {da} ‘become dry’; at {da} ‘kill’; aw {da} ‘itch’; aj {da} ‘see’; en {da} ‘be spilt’; i {da} ‘die’; is {ta} ‘recognise’; ir {da} ‘ride, mount (a horse)’; in {da} ‘say’; ikku ‘hiccup’; om ‘three’; om {ta} ‘keep waiting’; oŋ {da} ‘go’; oŋ {ta} ‘fit’; orba ‘snore’; oj ‘husband; uk ‘hide ; ur ‘arrow body’; ur {ta} ‘mix’. Onsets in vowel position are also found in Bodish and Tshangla loans, such as akpu crow, cf. TT akpu, TSB akpakʰa; alak branch, cf. BL alak; am mango, cf. Tshangla amse > Hin. आम aːm; ani nun, cf. Tib. a-ni; ara liquor, cf. TSB ara; emriŋ swastika, cf. Tib. g.yuṅ-druṅ, TSD emriŋ; eri raw silk, cf. TSD eri ~ elin; er Kaleej pheasant, cf. TT er, BL er; ortoŋ internal part of throat, cf. TSB ɔrtɔŋ throat. Finally, we widely find onset vowels commonly in interjections and pauses, such as ah ‘ehm’; akej ‘oops’; ana ‘right, ok, well’; atsow ‘ouch’; aj ‘ok’; eh ‘oh’; oŋ ‘yes, right’ and owej ‘hey’. Except for the close /u/, all the other vowels, when occurring in onset position, whether in open or in closed syllables, are preceded by a non-phonemic glottal stop or glottal constriction, which is usually and most clearly manifested in creaky voice of the vowel. This can be shown in the spectrograms and waveforms for the respective tokens and minimal pair ar {da} [ʔa̰ɹ ~ ˀa̰ɹ] ‘to dry’,

1 er [ʔɛ̰ɹ ~ ˀɛ̰ɹ] ‘blood pheasant’; ir {da} [ʔḭɹ ~ ˀḭɹ] ‘to mount and ride (a horse)’; orba [ʔɔ̰ɹbaː ~ ˀɔ̰ɹbaː] ‘a snore’; but ur {ta} [uɹ] ‘to mix (flour in boiling water to a dough); to scratch (scalp and hair); to hold and shake (someone by clothes or arms)’. As can be seen in the respective spectrograms and waveforms (Figures 1-5), all vocalic onsets except the vowel /u/ show a more or less pronounced glottal onset and in all cases creaky voice that is not phonotactically conditioned by the following rhotic but the result of the preceding glottal onset.

2

Figures 1-5. Spectrograms and waveforms of pre-glottalised vocalic onsets

The creakiness of vocalic onsets /a, e, i, o/ is a highly pronounced phonetic feature, with impact on all the surrounding vowels. Hence, the choice of a carrier sentence (goq ___ innyi) with a glottal constriction preceding and following the tokens was one of the major methodological flaws of the PhD research, as this influenced the quality of tokens with vocalic onsets and open rhymes tremendously.

The situation in Duhumbi has some parallels in other languages, and from a historical-comparative point of view perhaps, most noteably with Tibetan. In written Tibetan, the onset ཨུ་ is almost exclusively found in loans and nativised loans from Sanskrit (such as ut-pal ‘blue poppy Meconposis sp.’, cf. Sanskrit ut-pa-la ‘blue lotus’), Chinese (such as u-yon ‘committee member’, cf. Chinese 委员 wěiyuán ‘committee member’) and Mongol (such as u-rge ‘the top of a tent pole that protrudes above the canvas’, cf. Monglian Өрх [ɵrx] ‘top of the tent’). There are hardly any native Tibetan lexemes that have the onset ཨུ་, if any at all; however, the same can be said of the vocalic onset that are written in Tibetan as ཨི་, ཨོ་ and ཨེ་: all of these occur almost exclusively in Sanskrit, Chinese or Mongolian loans. In fact, the only common onset is ཨ་, which is, for example, common in kinship terms. On the other hand, the vocalic onset འུ་ is much more common, and occurs in what may be presumed to be native lexemes, such as ḥu-lag ‘compulsory post or labour service’, ḥu-su ‘coriander’, ḥug-pa ‘owl’, ḥud ‘swaggering, boasting’, ḥub-pa ‘to collect, to sweep together’ and ḥur ‘busy activity, rush’. Zhang (1987: 41–6) analysed the Tibetan letter ཨ that Hill (2010: 115) transcribes as q as representing a glottal stop in Old Tibetan. Hill (2010: 115-116), however, argues against this analysis and observes that q functions in the Tibetan script as a null consonant marking a vocalic onset. He does, however, remark that, perhaps, in Old Tibetan all vowel-initial words might have been articulated with a sub-phonemic glottal stop, similar to German. As for the Tibetan grapheme འ་, Hill (2005), transcribing it with , argues on basis of evidence from Tibetan dialects that when preceding a vowel onset and the labial --, it represents a voiced velar [ɣ].

3

Contemporary evidence from Duhumbi indicates that in onset position and particularly monosyllabically, the point and of the vowel /u/ trigger a more open, voiced or perhaps breathy onset closer to a onset [ɦu], whereas all the other vowels have a clear glottal reinforcement in the form of an unvoiced glottal plosive preceding the vowel onset which results in creakiness of the vowel, i.e. [ʔa̰, ʔɛ̰, ʔḭ, ʔɔ̰]. However, the voiced glottal fricative and the unvoiced glottal plosive and creakiness of the vowel have only a sub-phonemic status. The glottal onsets of the vowels /a, e, i, o/ are represented in the , but not in the phonemic transcription, used for transcribing the Duhumbi texts.

GLOTTALISATION OF THE CODA PLOSIVES

Among Duhumbi speakers there are various realisations of the syllable-final velar plosive /k/ following the open and open-mid back vowels /a, o, e, i/. Glottal reinforcement of these rhymes is extremely common, not only in Duhumbi, but also in other languages, such as Tshangla (Bodt 2014). Whereas the rhyme -ak/-ok/-ek/-ik may be realised pre-glottalised and released [-ˀk] or pre- glottalised and unreleased [-ˀk̚], it is also commonly realised as glottal stop combined with creaky phonation on the vowel [a̰ʔ], [ɔ̰ʔ] [ḛʔ], [ḭʔ] or simply as a creaky vowel [a̰], [ɔ̰], [ḛ], [ḭ].

A coda preceded by vowel /a/ in a monosyllabic word or as the coda ultima of a polysyllabic word in isolation is elided in Khispi with subsequent lengthening of the vowel to [aː] but commonly realised as a glottal stop with maintenance of the short vowel length in Duhumbi [aʔ], i.e. *[-ak] > Duh. [-aʔ] / Khis. [-aː]. As the waveform and spectrograms (Figures 6-8) show, the exact realisation of the syllable-final velar plosive, even when spoken in a single sequence by a single speaker, varies between an unreleased velar plosive [bʲɛk̚] with little or no creaky voice, and an increase in the degree of creakiness in the vowel of [bʲɛ̰ʔ].

4

Figure 6. Waveform and spectrogram of bek {da} [bʲɛk̰ ]̚ ‘ripen, mature (vi)’.

Figure 7. Waveform and spectrogram of bek {da} [bʲɛʔ]̰ ‘ripen, mature (vi)’.

5

Figure 8. Waveform and spectrogram of bek {da} [bʲɛʔ]̰ ‘ripen, mature (vi)’.

Phonotactic conditions play a major role in the extent to which glottalisation and glottal replacement of coda plosives occurs. Released and unreleased realisations are more commonly attested when a root is followed by an unvoiced suffix, glottaly constricted realisations are more often observed when followed by a syllable with a voiced onset. An example are the verb roots, such as the minimal pairs dek {da} ‘shoot’ and dek {ta} ‘be stunted’. The fact that a verb root falls in the soft-stemmed Da-class or the hard-stemmed Ta-class is a historically conditioned one, determined by the presence of a syllable-final /s/ in the verb stem. In other words, it is not the exact realisation of the syllable-final velar plosive that determines the conjugational class of the verb, but rather the conjugational class of the verb and its corresponding prefix determine the most common realisation of the velar plosive in the stem, i.e. either as an (un-)released velar plosive or as a glottal constriction most commonly characterised by creaky voice on the preceding vowel. When a coda velar plosive in the stem is followed by a voiced morpheme such as -da ADV, there is a syllable-final glottal constriction which is not completely realised but is apparent from a creaky voiced vowel: dek {da} [dʲɛ̰ˀda] shoot-ADV; ɕak {deɁ} [ɕa̰ˀdḛˀ] be bright-PRES; ʨʰok {da} [ʨʰɔk̚da ~ ʨʰɔ̰ʔda] peck, poke, strike-ADV; and hek {da} [hʲɛk̚da ~ hʲɛ̰ʔda] become red-ADV. When a coda velar plosive in the stem is followed by an unvoiced morpheme such as -ta ADV, the syllable-final plosive is audibly released, there is no creaky voice, and there is a silence between the final plosive and the onset of the next syllable: dek {ta} [dʲɛktaʰ] be stunted-ADV; ɕak-teɁ [ɕaktḛˀ] cut in half-PRES, ʨʰok {ta} [ʨʰɔkta] allow-ADV and hek-ta [hʲɛkta] swirl-ADV. Beyond word boundaries, similar morphosyntactic rules can be observed. A coda velar stop in the ultima of a word is realised as a glottal stop when the first syllable of the next word has a voiced onset, and as an unreleased velar plosive if the onset of the next word in unvoiced. Within the same morphological word, a coda velar stop in the penult or antepenult of a polysyllabic word is realised as a glottal stop /ʔ/ if the onset of the ultima is a voiced consonant, and as an unreleased velar plosive /k̚/ if the onset of the ultima is an unvoiced consonant.

6

A similar development may be observed in the alveolar and bilabial coda plosive, such as -at/-ot/- et/-ap/-op/-ep, though commonly not as pronounced as in the case of the velar plosive. Glottal reinforcement of the coda bilabial plosive has thus far not been attested.

Finally, it must be noted that glottalisation of coda plosive /k/ and constriction of the rhyme leading to a creaky vowel, especially in rhymes /-ak/ and /-ok/, is also observed in monosyllabic words and in verb roots without prefix that may be modified by a suffix. Examples include ʨʰak [ʨʰak̚ ~ ʨʰa̰ʔ] ‘taro’; tʰak [tʰak̚ ~ tʰa̰ʔ] ‘rope’; gak {ta} [gak̚ ~ ga̰ʔ] ‘to block, to obstruct’; zok [zɔk̚ ~ zɔ̰ʔ] ‘lie’; lok [lɔk̚ ~ lɔ̰ʔ] ‘friend’; and pʰok {da} [pʰɔk̚ ~ pʰɔ̰ʔ] ‘to pull out’. Even in rhymes /-ik/ and /-uk/ and in rhymes with the other coda plosives, Duhumbi speakers will often not release the coda stop, such as, for example, in buk [buk̚] ‘breath’; ʦʰik [ʦʰik̚] ‘joint; knode; word’; pʰet {da} [pʰɛt]̚ ‘to arrive’; pʰap [pʰap̚] ‘yeast’; and dut {da} [dut]̚ ‘to forge; to smoke out’. This is to some extent a Bodish contact language influence, as in most Bodish languages the glottalisation of the coda velar stop is in advanced stages and the Bodish contact languages of Duhumbi, mainly Brokpa, Tibetan, Tawang Monket and Chocangaca, are no exception to that. Compared to the other Western Kho-Bwa varieties, Duhumbi is conservative in its retention of rhymes with syllable final plosives, with the Sartang and Sherdukpen varieties having no retentions of rhymes /-up, -op, -ip, -ep, -et, -ot/.

THE CONSTRICTED VOWELS AND THE GLOTTAL STOP

Duhumbi has several minimal and near minimal pairs of morphemes and lexemes that can be distinguished on basis of a package of several phonological features. The main distinction appears to be between ‘constricted’ and modal vowels. The ‘constriction’ of the vowel can be attributed to a glottal closure, be it partial or complete, at the end of the syllable. Thus, Duhumbi distinguishes a long, close-mid, near-front unrounded vowel /e/ [eː], a long, open, central unrounded vowel /a/ [aː], a long, close, back, rounded vowel /u/ [uː] and a long, open-mid, back rounded vowel /o/ [ɔː] from a distinctive ‘constricted’, short, creaky, close-mid, near-front unrounded vowel /eq/ with a rising pitch1 [ḛ́ˀ], a ‘constricted’ creaky, short, open, central unrounded vowel /aq/ with rising pitch [á̰ˀ], a ‘constricted’ creaky, short, close, back, rounded vowel /uq/ [ṵ́ˀ] and a ‘constricted’ creaky, short, open-mid, back rounded vowel /oq/ with rising pitch [ɔ̰́ˀ]. A similar distinction is not found in the /i/, with the only allophone of vowel /i/ attested as glottalised in onset position, [ʔḭ ~ ˀḭ]. The reason for this is probably phonotactic, with the vowels /a, e, o, u/ articulated in the lower back part of the oral cavity, closer to the point of articulation of the glottal stop.

Near-minimal pairs showing the distinctiveness of the constricted versus the modal vowels in open syllables are largely, but not exclusively, restricted to stressed headwords and unstressed particles or suffixes. Table 1. Distinctiveness of modal versus constricted vowel phonemes

/a/ vs. /aq/ -ba [-baː] NOM vs. -baq [-bá̰ˀ] INF -pa [-paː] NOM vs. -paq [-pá̰ˀ] INF ga [gaː] ‘1SG; saddle’ vs. gaq [gá̰ˀ] 1SG.GEN wa [waː] ‘move’ vs. waq [wá̰ˀ] 3PL.GEN

1 As in Old Greek, this rising pitch is indicated in the phonetic transcriptions with an acute accent on the vowel [V́].

7

na {da} [naː] ‘to be sick’ vs. naq [ná̰ˀ] 2SG.GEN ta [taː] ‘year of the horse’ and -ta [-taː] EMPH vs. dungtaq [duŋtá̰ˀ] ‘frog’, shiptaq [ɕiptá̰ˀ] ‘horse’ /e/ vs. /eq/ be [beː ~ bʲɛː] ‘down’ vs. beq [bḛ́ˀ] COP12 tʰe [tʰeː ~ tʰʲɛː] ‘the other side’ vs. tʰeq [tʰḛ́ˀ] ICF depa [deːpaː] ‘past; faith (R)3’ vs. -deq [-dḛ́ˀ] PRES teʨak [teːʨak] ‘whip’, tewa [teːwaː] ‘host’, ‘original’ vs. -teq [-tḛ́ˀ] PRES, loteq [lɔːtḛ́ˀ] ‘caretaker’ and kʰapteq [kʰaptḛ́ˀ] ‘counter’ gepu [geːpu] ‘king’ vs. sargeq [sargḛ́ˀ] ‘eight’ /o/ vs. /oq/ lo [lɔː] ‘musk deer’ vs. -loq [-lɔ̰́ˀ] ABL go [gɔː] ‘beginning; before’ vs. goq [gɔ̰́ˀ] 1SG.AGT ʥo [ʥɔː ~ gjɔː] ‘over there’ vs. pʰiʥoq [pʰiʥɔ̰́ˀ] ‘thorn’ /u/ vs/ /uq/ lu [luː] ‘song’ vs. duluq [dulṵ́ˀ] ‘goitre’

The figures 9-14 show the waveforms and spectrograms of several of the minimal pairs in Table 1. The waveforms show the distinctive creakiness of the vowel, as observable by irregularly spaced pitch periods and reduced acoustic intensity in the waveform. In the spectrogram, the same is characterised by increased distance between vertical striations and a usually very abruptly rising pitch. As can be seen, in some cases the glottal stop is clearly present, whereas in other cases, the glottal closure is much less pronounced, more of a glottal restriction than an actual stop. Variation among speakers, variation within the same speaker and variation based on phonotactic environment are all very high. But whereas a syllable-final glottal closure is not always realised by all Duhumbi speakers, and the glottal closure is often absent in running speech, they always maintain at least a shorter, creaky vowel. On the other end, Khispi speakers do not realise a glottal closure nor a short creaky vowel at all, and for Khispi speakers all of the minimal and near-minimal pairs in Table 1 are, in fact, homophonous. Duhumbi speakers are quick to state that a failure to make the distinction between the constricted and modal vowels identifies another speaker as a non- mother tongue speaker of Duhumbi; be it a Tawang Monket speaker from Laphyak or Samtu, a Khispi, or an outsider.

2 And also the third person indefinite pronoun bi in the genitive case: beʔ [bḛʔ]. However, as was elicitated in [CHUK14102015Cnorec(298)es], this phonetic form occurs mainly in rapid speech, with slow and careful speech maintaining a more regular form bijaʔ [bija̰ʔ ~ bja̰ʔ]. 3 Cf. Tib. ḥdas-pa ‘past’ and dad-pa ‘faith, reverence, respect, confidence’.

8

Figures 9-14. Spectrograms and waveforms of the modal versus constructed vowels.

9

The nominaliser -ba/-pa and the infinitive -baʔ/-paʔ also display significantly distinct behaviour in morphosyntactic constructions, such as for example the distinction between in-ba-kʰo say-NOM-LOC [inbaːkʰɔː] ‘when saying, when was/were saying’ and in-baʔ-kʰo say-INF-LOC [imbkʰɔː] ‘in order to say’. Perhaps, the Sartang and Sherdukpen distinction between the non-past forms with a bilabial nasal such as Rahung, Khoitam and Khoina <-mɔ> and the past forms with a bilabial plosive such as Rahung <-bɔ> and Khoina and Khoitam <-pʰɔ> developed from cognate constructions at an earlier moment in time. Similar to Duhumbi, and in fact, in a much wider distribution, Sartang and Sherdukpen also make a distinction between short, constricted (or, in the words of Jacquesson 2015:23-24 ‘checked’) vowels and unchecked vowels. Most of the instances where Sartang and Sherdukpen have a shorter, constricted vowel, often accompanied by creaky voice, we can find a coda consonant in Duhumbi (e.g. Duhumbi at {da} ‘to kill vs. Rupa ɔ ‘to kill’; Duhumbi kʰej ‘stick’ vs. Rupa kʰɔ ‘stick’; Duhumbi zip-li ‘quiver’ vs. Rupa zi ‘quiver’, Duhumbi is {ta} ‘to recognise’ vs. Rupa ɛ ̃ ‘to recognise’).

The glottalised vowels are represented in the transcription of Duhumbi by the regular IPA symbol [ʔ] and in the Roman orthography by the letter /q/. For the ʼUcen orthography, however, an orthographic innovation was necessitated. While in modern Tibetan varieties a phonemic glottal stop does exist, this glottal stop is the result of the elision of coda /k, p, s/. A similar process can be observed in Duhumbi. However, given a few exceptions, there is no comparative evidence that supports an underlying coda plosive for the majority of the constricted syllables in Table 1. These exceptions are the additive focus marker -tʰḛʔ add (> reconstructed Proto-Western Kho-Bwa *tʰek) and the numeral ‘eight’ sargɛʔ̰ (> reconstructed Proto-Western Kho-Bwa *sar-gjat EIGHT). Therefore, after much deliberation, it was decided that the Duhumbi constricted syllables would be represented in ʼUcen Duhumbi by a so-called halanta under the syllable ga ག྄. The halanta (Sanskrit virāma) is used in many Brahmic scripts to surpress the inherent vowel /a/ of each consonant. The function in Duhumbi is thus completely unrelated to the original function, but the halanta is hardly used in written Tibetan and has no other function in the Duhumbi ʼUcen orthography, so confusion may be limited. Also, this orthographic solution maintains the connection with a (with the syllable ga representing the velar plosive /k/ in coda position) that may also become a glottal stop and affect the preceding vowel in the same way as for the constricted rhymes. The main disadvantage is that the halanta is relatively inconspicuous and might (especially in written ʼUcen Duhumbi) be confused for the ra-ta in a rhotic onset cluster གྲ་ or a subscript vowel /u/ གུ་.

Finally, glottal constriction and vowel creakiness are not limited to the Kameng area of Arunachal Pradesh. Earlier, Liu Sun reported it from Anong (Sun: 381-387). Calling the feature tenseness or laryngealisation, he observes that it is a phonetic feature correlating with the pitch level: high tone co-orrurs with fully laryngealised vowels, and mid and mid-falling tone with slightly laryngealised vowels. He concludes that tenseness is not a distinctive feature of vowels, and neither an acoustic cue to tone in the same way as pitch. Laryngealised/tense vowels may be accompanied by a glottal stop. This situation closely resembles the situation in Duhumbi, where constricted vowels may also be accompanied by a glottal stop and are furthermore characterised by a rising pitch contour.

10

REFERENCES Zhang liang Sheng, 1987. ‘The consonant system of middle-old Tibetan and the tonogenesis of Tibetan’, Ph.D. thesis, University of California, Berkeley. Hill Nathan W. 2010. An overview of old Tibetan synchronic phonology. Transactions of the Philological Society Volume 108:2 (2010) 110–125.

11