This Illustration focuses on the variety of the Xumi language (旭米, /LPʂuketɕɜ/ or /EPʂu-hĩ ketɕɜ/) that is spoken in the upper reaches of the Shuiluo river (水洛河) in Shuiluo Township (水洛乡) (hereafter Upper Xumi).Footnote 1 The township is located in Muli Tibetan Autonomous County (木里藏族自治县, smi li rang skyong rdzong in Written Tibetan, hereafter, WT), in the South-West of Sichuan Province (四川省) in the People's Republic of China.
Upper Xumi is one of the two dialects of the Xumi language, the other one being Lower Xumi. That dialect is spoken in the lower and middle reaches of the Shuiluo river, as we discussed in our phonetic and phonological sketch of Lower Xumi (Chirkova & Chen Reference Chirkova and Chen2013b). The present overview of Upper Xumi complements our overview of Lower Xumi.
The two dialects of Xumi are closely related to each other, but they are distinctly different, due to their contact with different languages. Upper Xumi is essentially influenced by the local dialect of Tibetan (Gami Tibetan), as well as, to a lesser extent, by Pumi. By contrast, Lower Xumi is mostly influenced by the local dialects of the Pumi, Mosuo, and Naxi languages. The observed diversity is highly remarkable given:
(a) the small geographical area occupied by the group (there is a mere 15.25 km beeline between the northernmost village of Lanman (兰满村) in the upper reaches of Shuiluo river and the southernmost village of Mianbang (免邦村) in the lower reaches of Shuiluo river);
(b) the small number of speakers (less than 2,000 people);
(c) the continuing contact and exchange between villages in the upper, middle, and lower reaches of Shuiluo river.
The comparative data on the two varieties are of interest for studies on language change in progress, and they may provide insight into the mechanisms at work in the implementation of language change and the diffusion of linguistic innovations.
The present sketch provides an overview of Upper Xumi with an emphasis on those features that distinguish it from Lower Xumi. It also contains an instrumental study of the voiceless–voiced contrast in nasals and laterals (nasal and oral airflow and electroglottographic measures), made possible by the fact that we could work with our principal Upper Xumi language consultant in a phonetics laboratory.
The description is based on the first author's fieldwork. The word list and the recorded passage provided with this paper were recited by a fifty-nine year old male native speaker of Xumi, who was born and raised in Lanman village. To facilitate comparison with Lower Xumi, the present description uses (whenever available) cognates in both varieties as illustrative examples. (In the sound archive, sound files with Lower Xumi forms are marked with ‘LX’.)
Consonants
The consonant inventory of Upper Xumi comprises of 44 consonants, as listed below. (Low frequency phonemes are given in parentheses. See section ‘Prosodic organization’ for tonal variation on monosyllabic words.) The appendix provides the distributional patterns of consonants and vowels that include permissible sequences in syllables as attested in our corpus of c. 2,800 words. Upper Xumi is characterized by a reduced segmental inventory (44 consonants compared with 50 in Lower Xumi), different phonotactic constraints, and the presence of prenasalized clusters.
There is a general three-way contrast in stops and affricates: voiceless aspirated, voiceless unaspirated, and voiced. Velars and uvular stops are in near complementary distribution. Uvular stops are found before non-high vowels, and velar stops are found elsewhere. The two series contrast before /e ɜ u
/, e.g. /Rkhe/ ‘tax’ (WT khral?) vs. /Rqhe/ ‘lime’, /ʜkhɜ/ ‘foot’ vs. /ʜqhɜ/ ‘feces’, /ʜku/ ‘to be able’ vs. /ʜqu/ ‘hearth’, /RPɲɜ-k/ ‘to warm oneself’ vs. /ʜq/ ‘fate, life’. Unlike Lower Xumi, Upper Xumi does not have a voiced uvular stop. Words with the voiced uvular stop in Lower Xumi correspond to words with the cluster /Nɡ/ in Upper Xumi (see section on prenasalized clusters below).
Fricatives are pronounced at six different places of articulation: alveolar, retroflex, alveolopalatal, velar, uvular, and glottal. Voiceless velar and glottal fricatives are in complementary distribution, the former only occur with oral vowels, whereas the latter only occur with nasal vowels, e.g. /ʜxɔ/ ‘cooked rice, food’, /ʜh/ ‘to stretch’. /ɦ/ has the most restricted distribution of all fricatives. It co-occurs only with the nasal vowels // and /
/; examples include: /ʜɦ/ ‘to dare’ (compare with /ʜ/ [ʜʔ] ‘self (first person subject)’), /Rɦ/ ‘self (third person subject)’ (compare with /ʜ/ [ʜʔ] ‘sheep’). Also unlike Lower Xumi, Upper Xumi does not have a voiced velar fricative, but it has the voiceless uvular fricative /χ/ as an independent phoneme, which can occur word-initially, e.g. /LPχɜ-χɐ/ ‘to itch’ (compare Lower Xumi /LPqhɐ-χɐ/), /ʜχɐ/ ‘difficulty’ (WT dkaˈ).
Xumi nasals and laterals contrast (i) alveolar and alveolopalatal places of articulation and (ii) voiced and voiceless counterparts.
Alveolar and alveolopalatal laterals and nasals are minimally distinguished before /u/; in addition, alveolar and alveolopalatal nasals are also distinguished before /ɐ/. Examples include: /LPɲu-ɲu/ ‘breast’ vs. /LPnu-tʂhu/ ‘bean curd’, /Rlu/ ‘again’ vs. /Rʎu/ ‘come (imp)’, /RPnɐʁɐ/ ‘twenty’ vs. /LPɲɐʁɔ/ ‘shadow’. For nasals, the contrast is not found before the vowels /i ɛ ɜ/, and for laterals, before /ɛ/. In the above cases, we use the symbols for the alveolopalatal nasals and laterals, respectively. Example include: /ʜɲi/ ‘you, thou’, /ʜɲe/ ‘snivel, snot’, /ʜɲɛ/ ‘milk’, /ʜɲɜ/ ‘to have a craving (for something), to be hungry’; /Rʎɛ/ ‘predestined affinity’.
Bilabial and alveolar nasals show a clear voiced vs. voiceless contrast as confirmed with our electroglottography (EGG) data (see Figures 1 and 2). Voiceless nasals are infrequent. They mostly occur in loanwords from Tibetan, but they are also attested in the native vocabulary, e.g. /RP
jɛtsũ/ [RPjætsũ] ‘tail’, /R
/ ‘fur, animal hair’.Footnote 4 Upper Xumi has two voiceless nasals (/ , /), which is two voiceless nasals less than in Lower Xumi (/
/). This is surprising, given that Upper Xumi is in closer contact with a local Tibetan dialect than Lower Xumi and that that local Tibetan dialect has four voiceless nasals (/
/, Chirkova, in press). Moreover, /EP
ɐ
mu/ (WT lha rnga mo), the word for ‘camel’, is distinct in Upper Xumi not only from the corresponding form in Lower Xumi (/EP ɑm
/ ‘camel’), but also from that in Gami Tibetan, /ʜŋam/ (WT rnga mong), suggesting distinct donor languages for this form.
Figure 1 Waveforms, spectrograms, EGG, NAF, and OAF for the pairs /ʜmjɛ/ ‘bamboo’ vs. /ʜ/ ‘medicine’, /ʀᴘnɐʁɐ/ ‘twenty’ vs. /ʜ
/ ‘incantation’, and /ʀnɔ/ ‘whole, complete’ vs. /ʀ/ ‘fur, animal hair’.
Figure 2 Waveforms, spectrograms, and EGG of the minimal pairs /ʀlɜ/ ‘stutterer’ vs. /ʀɜ/ ‘soul, spirit’, and /ʜʎɛ/ ‘correct; right’ vs. /ʜ
ɛ/ ‘flavourless’
Table 1 provides the measured values of nasal airflow (NAF) and oral airflow (OAF) for the three (near) minimal pairs: /ʜmjɛ/ ‘bamboo’ vs. /ʜ/ ‘medicine’ (WT sman), /RPnɐʁɐ/ ‘twenty’ vs. /ʜɐ/ ‘incantation, curse’ (WT sngags?), and /Rnɔ/ ‘whole, complete’ vs. /R/ ‘fur, animal hair’. In the table, NAFmin = minimum NAF; NAFmax = maximum NAF; NAFmean = mean NAF value; the provided measures represent an average of two to three repetitions and are measured during the nasal closure only. Figure 1 provides waveforms, spectrograms, EGG, NAF, and OAF for the three pairs.
Table 1 Measured values of nasal airflow (NAF) and oral airflow (OAF) (in liter/second) for the (near) minimal pairs /ʜmjɛ/ ‘bamboo’ vs. /ʜ/ ‘medicine’ (WT sman), /ʀᴘnɐʁɐ/ ‘twenty’ vs. /ʜɐ/ ‘incantation, curse’ (WT sngags?), and /ʀnɔ/ ‘whole, complete’ vs. /R/ ‘fur, animal hair’.
A surprising finding of this pilot investigation is that all values are greater during the articulation of voiceless nasals. More precisely, while in the articulation of voiceless nasals the air escapes both through the nasal and oral cavity, there is almost no airflow through either nasal or oral cavity for the voiced nasals. Piezoelectric data, recorded for the same corpus, confirms that in voiced nasals, there is resonance recorded in the nose but no nasal (and no oral) airflow. A more detailed study (in preparation) will hopefully provide a more accurate description of this potentially new kind of nasals.
All examples of the voiceless laterals include frication and an abrupt release of the front closure that occurs slightly before the onset of voicing for the vowel (see Figure 2). The burst separates the segment into two parts with different spectral structure. The pre-burst part has the greatest energy centered around 3000 Hz and greater amplitude of frequencies above that peak. The post-burst part has the noise present more equally across the frequencies, with minor peaks at the frequencies corresponding to the formant peaks of the following vowel. In addition, the post-burst part resembles the spectral structure of the voiced lateral counterpart. The presence of the burst and the spectral structure of the post-burst part suggest that these segments have developed from clusters.
Syllabic consonant
As a syllabic consonant, // may occur with a zero-initial (e.g. /RPkhu-/ ‘to be in debt’), and after alveolar sibilants and retroflexes (attested in our corpus after /tstshdzsztʂh ʂ ʐ/). We take the position that this syllabic consonant is phonologically the same as the fricative vowel after alveolar sibilants and retroflexes (/sztstshdz ʃ ʒ tʃhtʂh/). In the latter case, // is realized as homorganic to the preceding consonant onset. Consider the following examples: /ʜts/ [ʜts
] ‘to use’, /ʜtsh/ [ʜtsh] ‘to cut with scissors’, /ʜdz/ [ʜdz] ‘wheat’, /RPmɐs/ [RPmɐs] ‘tomorrow’, /RPmɐz/ [RPmɐs] ‘cat’, /ʜtʂh/ [ʜtʂh
] ‘to sell’, /ʜʂ/ [ʜʂ] ‘fishing net’, /Rʐ / [Rʐ ] ‘to sleep’. Similar to the distribution of the syllabic consonant // in Lower Xumi, // in Upper Xumi contrasts with the high vowels (/iu/), as in /ʜtsi/ ‘lock’, /ʜts/ ‘to use’, /ʜtsu/ ‘to pluck (facial hair)’. We therefore analyse it as a separate phoneme in this language.
Consonant clusters
Clusters with /w/
The approximant /w/ occurs in the second position in consonant clusters, where it may be realized as secondary labialization of the first position consonant. /w/ occurs after alveolars, alveolopalatals, retroflexes, velars, uvulars, as well as after /ŋ/ and /ɹ/. It may be followed by /ie ɛ ɜ ɐ / as well as, marginally, also by /ĩ/. If preceded by an alveolopalatal onset or followed by the front vowels /ie ɛ ĩ/, /w/ is realized as [ɥ], e.g. /ʜɡwi/ [ʜɡɥi] ‘bundle wrapped in cloth’, /Rdwe/ [Rdɥe] ‘to ask’, /Rdʑwɛ/ [Rdʑɥɛ] ‘bird’, /Rdʑwɜ/ [Rdʑɥɜ] ‘honey, sugar’, /EPɡwɛ-wɜ/ [EPɡɥɛ-wɜ] ‘to hunt’, /EPlɐ-ɡwɜ/ [EPlɐ-ɡɥɜ] ‘to hide something’, /Rɹwɜ/ ‘copper’, /Rɹwɐ/ ‘heavy’, /ʜqwɐ/ ‘to cry, to weep’.
Prenasalized clusters
Similar to its linguistic neighbour, Gami Tibetan (Chirkova, in press), Upper Xumi has prenasalized voiced stops and affricates clusters (/Nb Nd Ndz Ndʑ Ndʐ Nɡ/). The stop component of the cluster may only be voiced, and a sequence of a nasal and a stop must be homorganic.Footnote 5 Prenasalization in Upper Xumi is contrastive, as evidenced by the (near) minimal pairs below:
Footnote 6 Prenasalized clusters mostly occur in Tibetan loanwords (corresponding Gami sound files have been provided whenever available), and they are likely to be borrowed due to widespread Xumi–Gami Tibetan bilingualism in the upper reaches of the Shuiluo river. Examples include: /RPNdiwu/ ‘bullet’ (WT mdeˈu, Gami /ʜNdi/), /RNdʐɔ-Ndʐɔ/ ‘to be similar’ (WT ˈdra, Gami /ʜNdʐɔ/). (Compare the corresponding words in Lower Xumi: /RPdiwu/ ‘bullet’, /EPdʐe-hĩ/ ‘similar’, with the nominalizing marker /hĩ/.) Prenasalized clusters are also attested in Xumi native vocabulary. We note that such words (invariably with the cluster /Nɡ/) are cognates with Lower Xumi forms with the voiced uvular stop, as in ‘to stew’: Upper Xumi /RNɡu/, Lower Xumi /Rɢo/. As suggested by the external reviewer of this paper, the presence of Tibetan loanwords containing the cluster /Nɡ/ has possibly paved the way for the reinterpretation of the voiced uvular stop as a /Nɡ/ cluster in Upper Xumi. We note that such a development would be consistent with the correlation between stops and voicing, as discussed in Ohala (Reference Ohala and MacNeilage1983: 194–201). Given that the farther forward in the vocal tract a stop is articulated, the better able it is to accommodate voicing, /ɢ/ is the most difficult stop to voice. The prenasalized stop /Nɡ/ avoids the problems associated with maintaining voicing, while maintaining the original phonemic contrast between /ɡ/ and /ɢ/.
Consonant lenition
Similar to Lower Xumi, Upper Xumi has a set of productive lenition rules, which transform some intervocalic voiced and voiceless aspirated stops and affricates into spirants. However, compared to the lenition rules in Lower Xumi, those in Upper Xumi are reduced in number and solely yield independent full-fledged phonemes that can occur word-initially. The attested lenition rules are summarized as follows:
The noted changes are not only less numerous, but are also less regular than the lenition rules in Lower Xumi, and for some initials, also sporadic and marginal. This is the case for /Rdʑ
/ ‘to have’, where the initial /dʑ/ irregularly lenites to /ʐ/, as in [RPmu = ʐ
] ‘not have’.
Vowels
The Xumi vowel system comprises nine oral vowels and six nasal vowels (i.e. one oral vowel more than in Lower Xumi). Vowels exhibit a wide range of surface realizations (also among the language consultants that we worked with) with most vowels occurring in a rather narrow range of contexts. The number of hypothesized underlying vowels and the relation between the surface vowel realizations and the phonemic vowel categories have been established on the basis of cognates in both varieties. See also the vowel chart plotted on the relative F1/F2 formant values below. (F1 and F2 were measured at one quarter of the vowel duration (25%), at mid point (50%), and at three quarters of the vowel duration (75%); the mean formant value of a vowel was calculated by averaging over the three measurements).
Footnote 7Footnote 8Footnote 9 Upper Xumi nasal vowels include /ĩ
ũ
/, as in /ʜhĩ/ ‘man, person’, /ʜh
/ ‘to cut, to slice (meat)’, /ʜhũ/ ‘to blow’, /ʜh
/ ‘vegetable’, /RPlɐ-h
/ ‘have disappeared’. Marginally, Upper Xumi also has the vowel /
/, which is only attested in one word in our corpus, namely /LPm
dɐ/ ‘on the roof, upstairs’ (compare /LPmĩdɐ/ ‘pitiful’).
Vowel realization
/i/ and /e/ overlap with some initials (see below for discussion).
Front mid vowels contrast two degrees of vowel height: /e/ vs. /ɛ/, as in the following examples (where these vowels are also contrasted with /i/): /Rji/ ‘to exist (of animate beings)’, /Rje/ ‘to lick’, /Rjɛ/ ‘vegetable oil’; /RPɐdʑi/ ‘alcohol, spirits’, /ʜdʑe/ ‘happy’, /ʜdʑɛ/ ‘rich’.
The front vowel /ɛ/ seems to be limited to occur only with [j], e.g. /ʜbjɛ/ ‘hoof’, /ʜdjɛ/ ‘fox’, /Rɡjɛ/ ‘eggplant’, /ʜmjɛ/ ‘bamboo’. The combination /jɛ/ is realized as [jæ] for some speakers, e.g. /LPmjɛ-tshu/ [LPmjæ-tshu] ‘downstairs’.
/ʉ/ has a restricted distribution. It only occurs after bilabials (e.g. /Rbʉ/ ‘to rub (with hands), to twist’), and velars, where it is realized as [ɯ] (e.g. /ʜkʉ/ [ʜkɯ] ‘to bake, to roast, to toast’).
/o/ has many different realizations, for it represents a change in progress, that is, raising to /u/ (see below). After alveolars, it is realized as [o] or [ɵ] (e.g. /ʜtho/ ‘manner, way (to do things)’, /ʜto/ [ʜtɵ] ‘to build’). After alveopalatals, it is realized as [əʊ] (e.g. /ʜtɕho/ [ʜtɕhəʊ] ‘to insert, to stick in’, /ʜɕo/ [ʜɕəʊ] ‘meat’). Occasionally, it can also be realized as [ɤ] (e.g. /ʜtso/ [ʜtsɤ] ‘crown of a head’).
/ɔ/ frequently appears in Tibetan loanwords (where it corresponds to WT a), e.g. /EPthuwɔ/ ‘hammer’ (WT tho ba), /ʜbɔ/ ‘mask’ (WT ˈba), /ʜdʐɔ/ ‘enemy’ (WT dgra), /RPlɐtɕhɔ/ ‘instrument, utensil’ (WT lag cha), /EPmɐzɔ/ ‘peacock’ (WT rma bya). In the native vocabulary, /ɔ/ contrasts with /ɐ/, e.g. /ʜqɔ/ ‘mountain range’ vs. /ʜqɐ/ ‘to move’, /ʜʁɔ/ ‘strength’ vs. /RPnɐʁɐ/ ‘twenty’.
Comparison with Lower Xumi
The main differences between the vowel systems of Lower and Upper Xumi can be summarized as follows. Compared to their counterparts in the vowel system of Lower Xumi, the vowels of Upper Xumi evidence a chain shift that consists of the raising of most vowels. The chain shift is possibly triggered by the addition to the system of the phoneme /ɔ/. Hence, Lower Xumi /eo ɐ ɑ 
/ generally correspond to Upper Xumi /eu ɜ ɔ ĩ ũ /, respectively. For the oral mid vowels /eo/, the change is in progress, yielding a number of intermediate forms (see above on the realization of these vowels) and resulting in near merger for some words, as discussed below.
The Lower Xumi /ʉ/ generally merges with /u/ in Upper Xumi, except after bilabial and velar initials, where it continues to be contrastive with /u/, e.g. /ʜbʉ/ ‘bowels, intestines’ vs. /ʜbu/ ‘Pumi people’, /Rbʉ/ ‘to rub (with hands), to twist’ vs. /Rbu/ ‘crops’, /ʜkʉ/ to bake, to roast, to toast’ vs. /ʜku/ ‘to be able’. In all other cases, we observe merger with /u/. For example, the words ‘to count’ and ‘to wipe’, which are contrastive in Lower Xumi (/ʜsu/ and /ʜsʉ/, respectively), are homophonous in Upper Xumi, /ʜsu/ ‘to count’ and /ʜsu/ ‘to wipe’.
The clearest examples of vowel raising involve the (Lower Xumi) low vowels /ɐ ɑ
/, which rise into the empty mid vowel space. Thus, Lower Xumi /ɐ/ generally corresponds to Upper Xumi /ɜ/, Lower Xumi /ɑ/ generally corresponds to Upper Xumi /ɔ/, Lower Xumi /
/ generally corresponds to Upper Xumi /
/, and Lower Xumi /
/ generally corresponds to Upper Xumi /
/; this is illustrated by the following examples:
The situation is more complex for the (Lower Xumi) oral mid open vowels /eo/, which by raising towards (Upper Xumi) /eu/ cause the overcrowding of the high vowel space, leading in some cases to merger or near merger. The minimal pair ‘now’ and ‘cloth’ (Lower Xumi /ʜɹi/ ‘now’, /ʜɹe/ ‘cloth’) is a case in point. A minimal pair test investigating speakers’ intuitions about their pronunciation of these words (see Labov Reference Labov1994: 353–354) suggests complete merger for most Upper Xumi speakers that we consulted: they judged the two words to be homophonous (compare the realization of these words by the speaker DDTR, /ʜɹi/ ‘now’ and /ʜɹi/ ‘cloth’). A more careful pronunciation of these words, as in word elicitation and discussion with our principal consultant, yielded a distinction, which our main consultant claimed to be that of tone. He pronounces the word for ‘now’ with a continued fall after the abrupt f0 rise in the initial part of the syllable, which gives the perception of a falling tone (i.e. [ɹi451] ‘now’). Conversely, he pronounces the word for ‘cloth’ with a high-level pitch contour after the abrupt rise, which gives the perception of a high level tone (i.e. [ɹi⁴⁵⁵] ‘cloth’). We note, however, that our principal language consultant was inconsistent in his judgements, and that this type of variation in the realization of the high tone is non-contrastive in this variety (see below on tone).
Consider another example, involving the back vowels /u/ and /o/, as in the minimal pair /ʜqhu/ ‘year’ and /ʜqho/ ‘bowl’, Lower Xumi /ʜqhʉ/ and /ʜqho/, respectively. While our principal language consultant maintains a distinction, the difference is acoustically very minor. Video recording reveals a difference in the relative degree of lip rounding (see the video clips ‘Year’ and ‘Bowl’ in supplementary materials). Clearly, broader surveys and commutation tests (Labov Reference Labov1994: 356–357) are needed to provide a complete account of this merger in progress.
Syllable structure
The canonical syllable of Upper Xumi may contain up to two optional elements in the following two types of linear structures: (i) (C1)(C2)V, and (ii) (C2)(C3)V, where C1 stands for nasal (before voiced stops and affricates), C2 can be any consonant, and C3 can only be -w-; V stands for vowel (or the syllable consonant /
/), and parentheses indicate optional constituents. A non-phonemic glottal stop can be inserted at the left edge of a vowel-initial stressed syllable (e.g. /ʜ
/ [ʜʔ
] ‘sheep’, /ʜũ/ [ʜʔũ] ‘collar’), and at the left edge of a vowel-initial root (e.g. /RPlɐ-
/ [RPlɐ-ʔ
] ‘to be drunk’, /EPmjɛ-ũ/ [EPmjæ-ʔũ] ‘to swallow down’).
Xumi is phonologically monosyllabic with a strong tendency towards disyllabicity in its lexicon. Trisyllabic and quadrisyllabic words are mostly composite, e.g. /LPth
-sĩp
/ ‘pine tree’ (WT thang) (from the word /LPsĩp
/ ‘tree’), /RP
ɜwu-dʑwɜ/ ‘peach’ (from the bound root /dʑwɜ/ ‘fruit’, as in the word /LPdʑwɜ-dʑwɜ/ ‘fruit’), /LPɲɜmibuxu/ ‘sunflower’ (from /LPɲɜmi/ ‘sun’, /RPbuxu/ ‘flower’). At the same time, Upper Xumi also has a handful monomorphemic trisyllabic words (both native and loanwords), e.g. /LPbɜz
ki/ ‘pillar’, /EPl
mutɕhi/ ‘elephant’ (WT glang po che).
Prosodic organization
Similar to Lower Xumi, Upper Xumi has a two-way tonal contrast on monosyllabic words and a three-way contrast of tonal melodies on polysyllabic words.
Tone and tonal melodies on lexical words
On monosyllabic roots, we observe a two-way tonal contrast: (i) rising (R) (e.g. /Rje/ ‘to lick’, /Rjɛ/ ‘vegetable oil’, /Rlɐ/ ‘suspension bridge’, /Rwɐ/ ‘cow’), and (ii) high (H) (e.g. /ʜje/ ‘tobaccoʼ, /ʜjɛ/ ‘to buyʼ, /ʜlɐ/ ‘tiger’, /ʜwɐ/ ‘tooth’). The pitch contours of the two tones differ slightly from the corresponding tones in Lower Xumi, as in Figure 3. Similar to Lower Xumi, non-contrastive variation abounds in the actual realization of the two lexical tones in Upper Xumi. In particular, the high tone may be realized as a falling tone. For example, the word /ʜjɜ/ ‘tent’ may be realized as [jɜ⁴⁵⁵] or as [jɜ451].
Figure 3 Pitch contours for the pairs /ʀje/ ‘to lick’ vs. /ʜje/ ‘tobaccoʼ, and /ʀwɐ/ ‘cow’ vs. /ʜwɐ/ ‘tooth’ in Lower and Upper Xumi.
In polysyllabic monomorphemic words, we observe a three-way contrast of tonal melodies. Here, again, we adopt the prosodic system developed for Lizu (Chirkova & Chen Reference Chirkova and Chen2013a) but it is important to note that just like Lower Xumi, Upper Xumi does not show a consistent correlation between duration and melody as observed in Lizu. Rather, the three patterns are more consistently signaled via their melodic difference. These tonal melodies are:
(i) Equally-Prominent Contour (EP). There is no salient rise or fall over any of the syllables. Rather, it seems to be high-level pitch contours throughout the two syllables. This pattern is mostly attested in monomorphemic words and in loanwords, e.g. /EPmɐmi/ ‘soldier’ (WT dmag mi), /EPtɕɐzɐ/ ‘airplane’ (WT lchags bya).
(ii) Left-Prominent Contour (LP). The high f0 peak is realized before the end of the first syllable, where the pitch starts to fall already and continues to fall in the second syllable, e.g. /LPmɐɲe/ ‘fire tongs’, /LPdʑɜɕi/ ‘chopsticks’, /LPbemi/ ‘axe’.
(iii) Right-Prominent Contour (RP). The high f0 peak is realized over the last syllable of the word, e.g. /RPmɐɲi/ ‘mani pile’ (WT ma ni, pile of stones with the Mani Mantra of Avalokiteshvara), /RPtɕɐzɐ/ ‘weeding hoe’ (WT lcags gzar), /RPdʑɜɕi/ ‘to stand’, /RPbe-mi/ ‘sow’.
Tone patterns in compounds
If the tone of the leftmost monosyllabic root is high, the resulting compound has the left-prominent pattern, as illustrated with the following sequence:
/ʜsĩ/ ‘wood, tree’ + /ʜqho/ ‘bowl’ = /LPsĩ-χo/ ‘wooden bowl’
/ʜsĩ/ ‘wood, tree’ + /Rkhʉ/ ‘root’ = /LPsĩ khʉ/ ‘root of a tree’
/ʜsĩ/ ‘wood’ + /RPdʑ
h/ ‘house’ = /LPsĩ dʑh/ ‘wooden house’
Conversely, if the tone of the leftmost monosyllabic root is rising, the resulting tonal melody is right prominent, as the following illustrate:
/Rʂ
/ ‘iron’ + /ʜqho/ ‘bowl’ = /RPʂ-χo/ ‘iron bowl’/Rɹ
/ ‘horse’ + /Rkhwɜ/ ‘shed’ = /RPɹkhwɜ/ ‘stable, horse shed’/Rɹ
/ ‘horse’ + /RPjɛtsũ/ ‘tail’ = /RPɹjɛtsũ/ ‘horse tail’
Transcription of the recorded passage
Our principal language consultant was recorded at the Laboratoire de Phonétique et Phonologie (LPP) of the Centre National de la Recherche Scientifique (CNRS). The original recording (made with a Digidesign 003 Rack souncard, Sound Studio 3.6 software for iMac, and an AKG C520L microphone) has been made available to the JIPA along with this analysis. In the transcription, only lexical items are marked for tone, whereas function words are not. Xumi function words and discourse particles (e.g. the genitive particle /ji/, the topic marker /ʐ / in the appended text) are never pronounced in isolation. Their surface tone realization depends on the tone of the preceding (host) lexical word (similar to tone sandhi in compounds).
Semi-narrow phonetic transcription
North Wind and the Sun
Interlinear morphemic glossing
Abbreviations used in the gloss below follow the Leipzig Glossing Rules (LGR, http://www.eva.mpg.de/lingua/resources/glossing-rules.php). Non-standard abbreviations (those not included in the LGR) are: agt = agentive, anm = animate, itsf = intensifying.
Acknowledgements
We would like to thank our principal language consultants, Mr. Lurong Duoding 鲁绒多丁, Mr. Mulian Lanbu 木联兰布, and Mr. Duoding Tshi'er 阿果多丁次尔, for their patience and assistance. We are grateful to Jonathan Evans and the anonymous reviewers of this journal for helpful comments and suggestions. Thanks are also due to the directors of the Laboratoire de Phonétique et Phonologie (LPP) of the Centre National de la Recherche Scientifique (CNRS) Jacqueline Vaissière and Annie Rialland for making the phonetics laboratory of the LPP available to us through the project LabEx-EFL (Empirical Foundations of Linguistics); to Angélique Amelot (LPP, CNRS) for helping us with recordings; to Jos Pacilly (Leiden University) for helping us with Praat scripts; and to Wang Dehe 王德和 for helping us with videorecordings. We gratefully acknowledge the financial support of the Agence Nationale de la Recherche (France) as part of the research project ‘What defines Qiang-ness? Towards a phylogenetic assessment of the Southern Qiangic languages of Muli’ (ANR-07-JCJC-0063) to Katia Chirkova and Tanja Kocjančič Antolík. Yiya Chen is currently supported by the Netherlands Organisation for Scientific Research (NWO-VIDI 016084338) and the European Research Council (ERC-Starting Independent Researcher Grant, 206198).
Appendix. C–V and CC–V combinations in Upper Xumi
C–V combinations
The table is based on a vocabulary list of c. 2,800 words. Marginal phonemes are restricted to two or less occurrences. Chinese loanwords and place names (such as Lanman) are in grey.
CC–V combinations
