PITCH AND

[This file is accompanied by five zip files that contain the sound recordings on which this analysis is based. These include the codas, the plosive onsets, the and onsets, the nasal onsets, and the approximant onsets]

The issue of suprasegmental phonological features such as tonality and stress are contentious in many Tibeto-Burman languages of the Eastern Himalayas. Possible realisations, distinctiveness and perceptions of these features may differ among native speakers, between native speakers and researchers, and among the researchers describing the language at different moments in time. Chinese linguists tend to describe every language as tonal, whereas Indian linguists generally do not mark tonal contrasts, even in languages that are phonologically the same whether spoken at the Tibetan or the Indian side of the border, as the descriptions of Puroik by Sun Hongkai et al. (1991) and Remsangpuia (2008) show. Other examples are the various descriptions of tone in Boro (see for an analysis Burling and Joseph 2010), in Lhasa Tibetan (see for a summary Mazaudon 2014: 590, 609), in rGyalronic languages (e.g. the analysis for Jiǎomùzú and the overview of other rGyalronic varieties in Prins (2011: 51-54)) and in Tshangla. As Morey (2014) mentioned, in many languages of the Eastern Himalayas, pitch and contour (change in pitch) are not the only defining features of tone. Rather, tone consists of a bundle of features that may include phonation (glottalisation and creakiness, breathiness, tenseness and laxness), duration and intensity. Not only Morey’s Tangsa, but also Watter’s Kham (), and other languages of the Eastern Himalayas display exactly this kind of complex interaction between different phonological features. Lowes (2009: ) describes the situation for Kurtoep where voiceless and voiceless aspirated obstruents are followed by the high tone, and voiced aspirated obstruents are followed by the low tone.

TONE IN TSHANGLA

The absence or presence of tone in Tshangla has been an issue that has divided linguists for a considerable time now. Depending on the origin and background of the speaker and the researcher, Tshangla is described as a language that has no tone, no distinctive tone, distinctive tone only on some onsets, or a language with distinctive high versus low register tone on all syllables.

Perhaps the most interesting but unfortunately underrepresented feature of Das Gupta’s 1968 description of Dirang Tshangla is the assertion (1968:14) that ‘tone plays a significant part. As has been observed already, k and g, c and j, t and d, and p and b owe their distinction to their tonal values as well. lengths are also affected by tones’. He then lists the following examples: khu [rising] dog vs. khu [level] rice, sha [rising] meat vs. sha [level] tooth, ra [rising] brass vs. ra [level] paddy, shi [rising] excreta vs. shi [level] bamboo and nga [rising] fish vs. nga [level] full. Unfortunately, in the remainder of his work no further mention is made of lexical tone. Own research on Dirang Tshangla has thus far not resulted in conclusive evidence for minimal pairs that confirm the distinctiveness of tone in the Dirang Tshangla varieties. The data from Metok presented in Zhāng (1986), Anon. (1991), Sun and colleagues (1981) show a two-tonal (high level versus low rising) contrast on all syllables. In fact, the main difference in inventory between standard Bhutan Tshangla and Pemakö Tshangla stems from the fact that Zhāng does not distinguish the voiced from their unvoiced counterparts. Partially in line with what Das Gupta earlier asserted, Zhāng argues that the voiceless initial consonants are only present in syllables with a high tone, and that voiced initial consonants only occur in syllables with a low tone (Zhāng 1988:4). According to Zhāng, there is a distinct two-tone contrast, namely high-level or high-falling tone, which he represents by the high-level tone ˥ (55) and the low-level or low-rising tone which he represents as the low-rising tone ˨ (13) (Zhāng 1988:10-11). Thus, the voiceless and /p/, /pʻ/, /t/, /tʻ/, /k/, /kʻ/, /ʔ/, /ts/, /tsʻ /, /tʂ/, /tʂʻ/, /ʨ/, /ʨʻ/, /s/, /ɬ/, /ʂ/ and /h/ and the complex onsets /pr/ and /pʻr/ are only present in the high-tone syllables, whereas the voiced plosives and fricatives /b/, /d/, /g/, /dz/, /dʐ/, /dʑ/ and /z/ and the complex onsets /br/ and /mr/ are only present in low-tone syllables. Zhāng does not present minimal pairs to confirm this theory. In fact, speakers themselves consider tone as non-distinctive, at least in the Didong dialect (Zhāng 1988:10).

According to Zhāng, nine consonants can occur in both high-tone as well as low-tone syllables: [m], [n], [ŋ], [ȵ], [w], [l], [r], [j] and [ɕ] because of the change from [ʑ] to [ɕ]. He then continues to list 13 (near-) minimal pairs out of his total of 2128 lexemes. It is important to differentiate between native Tshangla lexemes and nativised borrowings and changed pronunciation as a result of contact with other languages. Thus, ŋa five might be realized as [ŋa˥] in Pemakö Tshangla under the influence of Tibetan high-register [ʼnga] (lnga), and jaŋ wealth, prosperity is a borrowing from Tibetan [ʼyang] (g.yang) through religious concepts such as jaŋ kʰukpe to conduct a ritual to accumulate wealth. Furthermore, in translating from one language to the other one has to be careful not to project semantic meanings onto the target language.

For Bhutan Tshangla it was reported by authors such as Hoshi (1987:6), Egli-Roduner (1987), Wangdi (2005:56), van Driem (2001b) and Lowes (2006:94) that Tshangla does not have contrastive tone. This lack of tone distinction is even carried over by Tshangla speakers into other languages they might speak, including Dzongkha and Tibetan. In both languages, native Tshangla speakers do not pronounce the low-high initial tone contrast. This is a very typical feature by which Tshangla speakers of Dzongkha are easily distinguished from native Dzongkha speakers, or even Dzongkha speakers with another Central Bodish or even East Bodish language background, who are able to hear and produce tone differences. Andvik (1999:32, 2003) described Tshangla as contrasting high versus low tone following sonorant consonants for a small subset of words with some dialects currently replacing a contrast in voicing of initial obstruents with a contrast of high/low tone on the following vowel. Grollmann, finally, in the description of Bjokapakha Tshangla stated that “the differentiation of minimal pairs on the basis of register tones in Bjokapakha is, if present at all, only in the process of being built up and by no means an elaborated strategy”, with “not enough homophone lexemes to assume systematic tone differences in Bjokapakha”.

As proposed in Bodt and Gyatso (2012) and Bodt (2012), ongoing research on the various Tshangla dialects in Bhutan and elsewhere is pointing to both the existence of distinctive tone in some varieties, the recent loss in other varieties, and ongoing tonogenesis in again other varieties. In Bodt (2012: 200) I argued that the various Tshangla varieties display a shift from distinctive syllable-final consonant clusters (Dirang Tshangla), to distinctive tone (Pemakö Tshangla), to distinctive morphophonological features and homonymy (Bhutan Tshangla). But there appear to be several independent and sometimes contradicting pathways to the development of distinctive lexical tone, including a loss of voicing of plosives, a loss of coda consonant clusters and the influence of coda glottal stops, and the introduction of tonal distinction through loanwords from language where lexical tone or register tone is distinctive, in particular Tibetan and Dzongkha. Tone in Tshangla, though, remains elusive, and unless detailed descriptions of the many Tshangla varieties become available will probably remain elusive in the near future.

TONE IN DUHUMBI

In order to analyse the absence or presence of tone in Duhumbi, an approach was taken as also advocated by various authors such as Morey (2014) and Mazaudon (2014). First, tonal contrasts were identified, then, the phonetic details of these contrasts were explored. The identification of tonal contrasts was through a combination of identifying minimal and near-minimal pairs. These minimal and near-minimal pairs were then discussed in detail with the main consultant, in which the core phrases became khengdeq [kʰʲɛŋdḛʔ] ‘it pulls’ and khengbang [kʰɛŋbaŋ] ‘it doesn’t pull’ or, in frequent code-shifting to Dirang Tshangla, the cognates jangkla [ʥaŋkla] and jangkmala [ʥaŋkmala] or in a curious creole of both as jangteq [ʥaŋtḛʔ] and jangpang [ʥaŋpaŋ]. ‘Pulling’ was the way in which the informant expressed what she described as a need to ‘pull up’ a consonant from somewhere, rather than simply letting it flow from the point of articulation outward. In the end, perhaps, simply accepting this extremely clever laywoman’s view of the issue turned out to quite neatly pin-point at the core of the suprasegmental phonological feature of tone.

PLOSIVE CODAS AND PITCH CONTRAST

The first general observation that can be made is that syllables ending in rhymes with a plosive consonant /k, t, p/ as well as the constricted vowel rhymes /-aq, -eq, -oq, -uq/ are almost unequivocally realised with a rising-falling pitch by all speakers. This observation implies that the rising-falling pitch pattern is a co-articulated, inherent feature of closed rhymes ending on the plosive consonants /k, t, p/ or a glottal stop/glottal constricted vowel. Characteristically, the onset of the syllable does not influence the pitch pattern: a rising-falling pitch pattern has been attested following all the permitted onsets and onset clusters, and there are no minimal pairs in which both onset and rhyme are the same and only the pitch pattern is distinct. The spectrograms and waveforms Figure 1 to Figure 6 represent examples of all permitted plosive coda and constricted rhymes {-ak, -at, -ap; -uk, -ut, -up; -ik, -it, -ip; ek, -et, -ep; -ok, -ot, -op, -aq, -oq, eq} with a wide variety of permitted native consonants to show that the feature of rising-falling pitch is indeed related to the rhyme of the syllable. Note that various degrees of creakiness of the vowel can be observed in the case of rhymes -ak/-ok/-ek and to a lesser degree -ik/-at/-ot/-et and -ap/-op/-ep.

Figure 1. Waveforms and spectrograms of wat [wa̰t̚] ‘clothes’, ngak [ŋa̰k̚] ‘speech’ and yap [jap̚] ‘attic’.

Figure 2. Waveforms and spectrograms of zek {da} [zʲɛk̰ ]̚ ‘to bite’, phet {da} [pʰɛt̰ ]̚ ‘to arrive’, khep {da} [kʰɛp̚] ‘to cover’.

Figure 3. Waveforms and spectrograms of mik [mḭʔ] ‘arrow’, tit {da} [tit]̚ ‘to pull down skin’ and khip {da} [kʰip̚] ‘to cry’.

Figure 4. Waveforms and spectrograms of lok [lɔ̰ʔ] ‘friend’, chot {da} [ʨʰɔ̰t]̚ ‘to make’, khop [kʰɔp̚ ] ‘nest’.

Figure 5. Waveforms and spectrograms of duk {ta} [duk] ‘to touch lightly’, hut [hut]̚ ‘hand, arm’ and rup [ɹup̚] ‘money’.

Figure 6. Waveforms and spectrograms of goq [gɔ̰ʔ] ‘1sg.agt’, naq [na̰ʔ] ‘2sg.gen’ and beq [bḛʔ] cop1.

Plosive codas rhymes, as well as constricted rhymes, have an inherent tendency of rising-falling pitch. Thus, in these phonotactic conditions, distinctiveness on basis of the pitch contour pattern cannot be considered.

A similar situation has been described from, for example Jiǎomùzú rGyalrong, for which Prins (2011: 51-54) considers the possibility that the process of glottal replacement of coda plosives with finally erosion of the coda is one of the pathways to the development of Jiǎomùzú tonal contrasts.

PLOSIVE ONSETS AND PITCH CONTRAST

Secondly, an analysis was made of the pitch patterns following plosive onsets. Three sets of (near-) minimal pairs of lexemes were selected in which the plosive, fricative or affricate onset differs only in one feature, namely the (i.e. the voiced/unvoiced/aspirated distinctions). These sets were identified to find out whether Duhumbi pitch contrasts are independent of onset or perhaps co-occurring with certain onsets only. To avoid possible interaction with the pitch pattern following rhymes described in the preceding section, only nasal rhymes were selected. In addition to the (near-) minimal pair sets of which the waveforms and spectrograms are presented here, other tokens were also identified and analysed. These are presented in Figure 7 to 9.

The Duhumbi pitch contour with plosive onsets turns out to be predictive. Unvoiced, unaspirated and aspirated plosive onsets have an audibly higher pitch onset. Syllables with unvoiced plosive onsets have a sharply falling pitch pattern, whereas syllables with an aspirated plosive onset have a rising-falling pitch pattern. Voiced plosive onsets have lower pitch onsets and rising-falling pitch patterns. The average onset pitch differences are considerable: for the particular speaker in these examples, a 30-50 Hz difference between unvoiced unaspirated and aspirated plosive onsets and voiced plosive onsets is common.

Figure 7. kang {da} [kaŋ] ‘to be filled’, khang {da} [kʰaŋ] ‘to lift completely; carry; become cooled off’, gang {da} [gaŋ] ‘to lift partially’

Figure 8. ten {da} [tɛn] ‘to hold back’, then {da} [tʰɛn] ‘to affect’, den {da} [dɛn] ‘to know; be alone; be right’.

A slight concession had to be made in the case of the bilabial plosive onset as a good recording of tokens of pang was not available at the moment of analysis.

Figure 9. paŋ{bu} [paŋbu] ‘chest’, pha [pʰa] ‘axe’, baŋ {ba} [baŋ] ‘to be not’

Similar pitch patterns for plosive onsets has been reported more widely for Bodish languages. Van Driem (1998: 57) reports from Dzongkha that both unvoiced, unaspirated plosives and unvoiced, aspirated plosives have a high register tone onset, with voiced and de-voiced plosive onsets having low register tone onsets. For Kurtoep, Lowes (2009: 43-46) reports a similar situation, except that Kurtoep has no devoiced plosives. Lowes indicates that as Kurtoep is undergoing a process of tonogenesis in which tone phonologises on the vowel (with low-falling tone following voiced plosives and high-falling tone following voiceless and aspirated plosives), the voicing distinction between the voiced and the unvoiced plosive onsets (as measured in a positive VOT even for voiced plosives which would be normally associated only by unvoiced plosives) disappears.

In Duhumbi, syllables with unvoiced unaspirated and unvoiced aspirated plosive onsets have a higher pitch onset than syllables with voiced unaspirated onsets. Syllables with unvoiced unaspirated onsets have a falling pitch pattern, and syllables with unvoiced aspirated and voiced unaspirated onsets have a rising-falling pitch pattern. Now, if there would be minimal pairs that are only distinctive in terms of the pitch onset, there would be distinctive register tone in Duhumbi, like in Dzongkha. If there would be minimal pairs only distinctive in terms of pitch contour on the syllable, there would be distinctive tone in Duhumbi. But. since at least in plosive onsets, the pitch onset and pitch pattern are predictive on basis of the onset, neither of these two situations is applicable. Pitch onset and pitch pattern in what needs to be considered the native lexeme ka {da} ‘to chew’, cf. Proto-Western Kho-Bwa *ka CHEW and the loan ka ‘order’, cf. Tib. bkaḥ ‘order’, are not significantly different within the same speaker.

There are other differences between the Bodish languages and Duhumbi too. Duhumbi does not show any tendency to devoice voiced plosives in combination with a change in the pitch contour on syllables with initially a voiced onset. The distinction between aspirated and voiced plosives is neatly maintained in all phonotactic conditions and by all speakers.

Now, there may thus be a high likelihood that the pitch pattern in Duhumbi may be a relatively recent phonological introduction or development ostensibly taking place under influence of Bodish contact languages. Unvoiced unaspirated plosives can almost all be traced to have Bodish or Tshangla origins in Duhumbi, attested both by cognates in the Bodish languages or Tshangla and, conversely, in the absence of cognates in the other Western Kho-Bwa languages. The Western Kho- Bwa languages display a similar distribution, with a general absence of unvoiced unaspirated obstruents, leading to the conclusion that at the Proto-Western Kho-Bwa level there was a distinction in obstruents based on Onset Timing rather than on aspiration and voicing. Just to mention the Bodish cognates of the Duhumbi lexemes with unaspirated unvoiced obstruents: kang {da} ‘to be filled’, cf. Tib. skaṅ-ba/bskaṅgs-pa ‘to fulfil, amend, atone, appease, correct, remedy, make up, repair, reconcile, mend’; kang ‘marrow’, cf. Tib. rkaṅ ‘marrow’; ten {da} ‘to hold back’, cf. Tib. rten-pa ‘to base on, by, keep, hold, adhere to, lean on, support, to fix upon’; pak {ta} ‘to select and move’, cf. Tib. spags-pa > spog-pa ‘to mix; go to another place; remove and to bring near by turns’ and spag-pa ‘to shift, transfer oneself’. Analogy with the Bodish source lexemes with high register tone could well be the primary factor for the high onset and falling pitch contour in the Duhumbi cognates. Note that the written Tibetan spellings of these lexemes, and in effect almost all the Tibetan cognates of Duhumbi unvoiced unaspirated plosive onsets, have a superscript consonant {ra, sa, la}. There are, in fact, only very few Tibetan words that in the written Tibetan spelling start with a simple initial unvoiced consonant {ka, ta, pa}.

FRICATIVE AND AFFRICATE ONSETS AND PITCH CONTRAST

A similar approach to that in the previous paragraph was taken with the fricative and affricate onsets. First, minimal pairs were identified, and consequently the pitch patterns for these minimal pairs were analysed. The results were again consistent: voiced onsets have lower initial pitch and rising-falling pitch pattern, whereas voiceless and aspirated onsets are characterised by higher pitch onsets and a steeper decline. The results are presented in Figures 10 to 15.

This pattern can also be observed in the case of the unvoiced unaspirated and voiced unaspirated fricatives /h/, /s/ and /sh/ and /z/ and /zh/. The pitch distinction in onset – high in the case of the unvoiced unaspirated fricatives, low in the case of the voiced unaspirated fricatives - is maintained and the pitch pattern in the case of the unvoiced unaspirated fricatives is falling, and in the case of the voiced unaspirated fricatives rising-falling. This appears to be consistent among all lexemes. Note that although not very obvious in the spectrogram, the pitch onset of zang {da} is on average 35 Hz lower than that of sang {ta}.

Figure 10. Waveforms and spectrograms for sang {da} [saŋ] ‘to become light’ and zang {da} [zaŋ] ‘to clear’.

High onset and, albeit less pronounced, falling pitch is also observed in the fricative onset, as the spectrogram and waveform of haŋ {da} ‘to show’ in Figure 11 shows. The onset hw- shows a similar pattern. Figure 11. Waveforms and spectrograms for hang {da} [haŋ] ‘to show’ and hwang [hwaŋ] ‘thread’.

The pitch onset in the case of zhangq {ta} is again lower than in the case of shang {da}, the rising- falling pitch pattern of zhangq {ta} is distinct from the falling pattern of shang {da}.

Figure 12. Waveforms and spectrograms for shang {da} [ɕaŋ] ‘to warm up (weather)’, zhangq {ta} [ʑaŋʔ] ‘to construct; to rise up (H)’.

The waveform and spectrograms in Figure 13 indicate that a similar pitch onset and pattern remains even with another rhyme and is thus not influenced by the coda glottal stop of the complex coda cluster.

Figure 13. Waveform and spectrogram of shey {da} [ɕɛj] ‘to buy’ and zhey {da} [ʑɛj] ‘to laugh’. Figure 14. Waveforms and spectrograms for con [ʨɔn] ‘crack’, chon [ʨʰɔn] ‘boil’ and jong {da} [ʥɔŋ] ‘to jump; lift up’.

For the alveolar affricate series, a proper comparison was not possible as the voiced alveolar affricate /dz/ does not occur natively in Duhumbi. Even the tokens for the onsets /ts/ and /tsh/ have clear Bodish cognates, i.e. Tib. rtse ‘top’ and Tib. tshe ‘life’. Figure 15. Waveforms and spectrograms for tse [ʦe] ‘top’ and tshe [ʦʰe] ‘life’.

As the previous figures show, the basic distribution of the pitch patterns among the unaspirated unvoiced, aspirated unvoiced and unaspirated voiced affricate and fricative onsets in Duhumbi is similar to that of the plosive onsets. The sub-phonemic features of pitch onset and pitch pattern in these onsets is predictable on basis of the onset. The correspondences are summarised in Table X.X.

Table 1. Phonemic and sub-phonemic features in plosive, fricative and affricate onsets in Duhumbi.

k, t, p, ts, c, s, sh, h kh, th, ph, tsh, ch g, d, b, j, z, zh voice - - + aspiration - + - VOT ~0 + - pitch onset high high low pitch pattern falling rising-falling rising-falling

NASAL AND APPROXIMANT ONSETS AND PITCH CONTRAST

Analysis of the nasal and approximant onsets shows a clear non-distinct and level pitch contour over the entire syllable. This can be illustrated in Figure 16 for the nasal onsets and Figure 17 for the approximant onsets.

Figure 16. Pitch contour in nasal onsets: waveforms and spectrograms of ngan {da} [ŋan] ‘to become defunct’, nan {ta} [nan] ‘to add’, man {da} [man] ‘to achieve’, nyan {da} [ɲan] ‘to stretch’.

Among the approximants, the spectrograms of tokens with onset /w/, /y/, /r/ and /l/ show considerable variation among and between speakers, as well as variation in respect to rhymes. In general, however, speakers indicate that in native lexemes the pitch onset is pitch patterns are low and level. This is shown in Figure 17, with as only exception the token for ra [ɹa] ‘brass’. This is, however, a plausible loan, cf. Tib. ra-gan ~ rag ‘brass’, where the distinctive pitch pattern may be the result of the underlying velar coda.

Figure 17. Pitch contour in approximant onsets: waveforms and spectrograms of wa {da} [waː] ‘to move’, ja {da} [jaː] ‘to reach’, la {da} [laː] ‘to fetch’, ra [ɹaː] ‘brass’.

This is more or less confirmed by the following Figure 18, although here the token wam ‘house’ shows a divergent pitch pattern, which is not present in token recorded from other speakers.

Figure 18. Pitch contour in approximant onsets: waveforms and spectrograms of wam [wam] ‘house’, jaŋ {da} [jaŋ] ‘to want, desire, feel like’, lam {da} [lam] ‘to search; to be cold’, ram {da} [ɹam] ‘to tear’.

Despite the observation that nasal and approximant onsets in general do not have any distinctive pitch contrasts, speakers identified a number of lexemes with nasal /ng, n, m/ or approximant onsets /l, w, y/ and with open or nasal rhymes where a rising-falling pitch contour could be observed. Table 2 presents the sets of minimal pairs that were identified. Those lexemes that were described as ‘pulling’ are indicated by a circumflex accent on the vowel, indicating a rising-falling pitch. As can be seen, the minimal pairs are limited to lexemes with the nasal onsets /m, n, ng/, approximants /w, l, y, r/. This picture is partially comparable to the situation in Bodish languages, including, for example, Kurtoep. Lowes (2009: 36) describes a high versus low tone contrast following the sonorants /m, n, ng, ny, r, l, w, y/ and the palatal fricative /ç/ [ɕ]. Absent in the Duhumbi inventory are the nasals /ny/ and rhotic /r/. More about these later on. Another curiosity appears to be the fact that minimal pairs are limited to lexemes with the vowel /a/. Other occurrences do exist, but these lack contrastive minimal pairs.

Table 2. Possible minimal pairs for tone in Duhumbi

onset set # lexeme cognates in contact languages m- 1a mâ [maː] ‘wound’ Tib. rma; TSD ma; TSB maha ma {da} [maː] ‘to finish’ - ma [maː] ‘thing; warp’ PWKB prefix *ma-/mə- (§) 1b man {da} [man] ‘to - achieve’ mân [man] ‘medicine’ Tib. sman ‘medicine’; TSD [man]; Khispi pen [pɛ] 1c mâng [maŋ] Tib. dmaṅs ‘public; TSB [maŋ] ‘public, community’ ‘community’ mang [maŋ] ‘garlic’ - n- 3a nâ [naː] ‘oath’ mnaḥ ‘oath’; TSB [na] ‘oath’ na {da} [naː] ‘to be sick; Tib. na-ba ‘to be ill’; TSD [nan {bo}] ‘to be sick’; [na {lo}] ‘to to function’ function’ 3b nang {da} [naŋ] ‘to bark’ - nâng {da} [naŋ] ‘to give’ Tib. gnaṅ-ba ‘to give (H)’; TSD [naŋ {bo} ‘to give (H)’ nângda [naŋdaː] snaṅ-dag ‘worry, doubt’ ‘problem’ l- 2a la [laː] ‘mountain pass’ Tib. la ‘pass’; TSD [la] ‘mountain pass’; but Sartang and Sherdukpen lu ‘mountain, mountain pass’ suggests PWKB *la ‘mountain pass’ lâ [laː] ‘large leaf’ Tib. lo-ma; lo-ḥdab; BP [marra lâɕiŋ] ‘tree with leaves used to wrap butter’; TSD [laga] ‘large leaf’ la {da} [laː] ‘to fetch’ - y- 4a yang [jaŋ] ‘height Tib. yaṅ-rtse ‘high point’; TSD [jaŋ] ‘height’ (animate)’ yâng {ta} [jaŋ] ‘to - bloom’ yang {da} [jaŋ] ‘to wish, - to want’ yang [jaŋ] ‘even’ Tib. yaṅ ‘even’ yang [jaŋ] ‘green’ Tib. ljaṅ-khu/ku/gu ‘green’; TSD jɛŋlu ‘green’ 4b yap [jap] ‘attic; tent’ TSD [jap] ‘attic’ yap {da} [jap] ‘to TSB [jap {pʰe}] ‘to scatter’ scatter; to winnow’ yap [yap] ‘father (H)’ Tib. yab father (H); TSD [jap] ‘father (H)’ w- 5 wâng ‘blessing’ dbaṅ empowerment; TSD [waŋ] ‘blessing’ wang {da} ‘to be bright’ wâng {ta} ‘to occupy’ dbaṅ ba control, obtain, dominate

Waveforms and spectrograms of these contrastive lexemes show one main distinctive feature between the lexemes that are said to ‘pull’ and those that are not, namely, a much higher pitch onset in these lexemes. As a secondary feature, and probably related to the first, in most of these lexemes the pitch pattern of the ‘pulling’ lexemes is much more steeply falling. A third feature is the fact that the average length of the onset is shorter in the ‘pulling’ lexemes as compared to the non-pulling lexemes. This is shown in the Figures 19 till 23.

Figure 19. Pitch contour in nasal onsets: waveforms and spectrograms of ma [maː] ‘thing; warp’; ma {da} [maː] ‘to fijinish’ and ma [maː] ‘wound’.

Figure 20. Pitch contour in nasal onsets: waveforms and spectrograms of man {da} [man] ‘to get’ and man [man] ‘medicine’.

Figure 21. Pitch contour in approximant onsets: waveforms and spectrograms of mang [maŋ] ‘garlic’ and mang [maŋ] ‘community’.

Figure 22. Pitch contour in approximant onsets: waveforms and spectrograms of na {da} [naː] ‘to be ill; to function’ and na [naː] ‘oath’. Figure 23. Pitch contour in approximant onsets: waveforms and spectrograms of nang {da} [naŋdaː] ‘to bark’; nang {da} [naŋdaː] ‘to give (H)’ and nangda [na̰ŋdaː] ‘problem’.

It is perhaps significant to note that the Khispi equivalent of Duhumbi yang [jaŋ] ‘height (animate)’ is zhang [ʑaŋ], adding additional evidence to the fact that Duhumbi yang might be a more recent Tshangla loan rather than an old retention. Similarly, instead of Duhumbi yap [jap] ‘attic; tent’, Khispi has zhak ~ zhap [ʑak ~ ʑap] ‘attic’, and Bodish loan gul ~ gur [gul ~ guɹ] ‘tent’, cf. Tib. gur ‘tent’.

Strictly speaking, according to these results, a distinctive level tone versus high rising-falling tonal contrast on the vowel /a/ for onsets starting with the nasals /m, n/ and the approximants /y, l/ (but not for w or r) needs to be posited for Duhumbi. However, as Table 2 indicates, most of these lexemes have Bodish cognates in which the Written Tibetan orthography with a prefixed d-, g- or ma- or a superscript r- or s- as well as a Tibetan and Brokpa pronunciation with a high tone or a high register onset, depending on the analysis. This suggests that a tonal difference in these lexemes is not a native Duhumbi phenomenon, but rather a borrowed phonological feature.

Though not an absent feature, tone in Duhumbi carries a low functional load and is in the process of developing itself under influence of contact languages, rather than a native distinctive feature.

NOTE ON THE ERRORS IN THE APPROACH

The analysis of pitch patterns brought a significant error in the approach to light that was unfortunately too late to amend. A carrying sentence used to elicit lexemes in a phonetically equal context should be carefully chosen. Not only should it make sense to the speaker, but it should also be a phonetically neutral environment. For the case of Duhumbi that means, for example, that it is favourable if the lexeme under investigation is flanked by syllables with unvoiced, unaspirated stops at both sides, introducing a pause between the lexeme under question and the previous and next syllable without breaking the flow of the carrying sentence. Syllables with, for example, high or palatals are not a good choice because they inherently have a higher pitch, tainting the result. Similarly, the flow from and to glides and approximants and other vowels is often too smooth if the lexeme in question also has a coda or onset vowel, making it difficult to distinguish between the end of one syllable and the onset of the next. Significantly, syllables with glottal stops or creaky voiced onsets or coda often carry the creaky voice over in the lexeme under investigation. Creaky voiced syllables, especially obstruents, often show an ‘undefined’ pitch pattern. As such, the carrying sentence goq ___ innyi [gɔ̰ ___ ʔḭnɲi] I said ____, whilst semantically perfect for each and every lexeme, could phonetically probably not have been a worse choice, with creaky voiced vowels, and that also one high vowel, flanking the lexeme under investigation.