Dinka is a Western Nilotic language within the Nilo-Saharan family. There are over two million speakers (Gordon Reference Gordon2005). The Dinkas live along the banks of the White Nile and its tributaries in Southern Sudan. There are also communities elsewhere within Sudan, in particular in the capital Khartoum, and abroad. Cows play a key role in the Dinka economy, and also in the sociocultural system. Four major dialect areas are commonly distinguished: Padang, Rek, Agar, and Bor. The variety of Dinka represented here – Luanyjang or, as in Roettger & Roettger (Reference Roettger and Roettger1989), Luac – is part of the Rek dialect group. The term Luanyjang ‘Luac Dinka’ refers in the first place to a section within the Dinka ethnic group, and by extension also to their variety of the Dinka language. The main Luanyjang town is Wuncuei, located about 170 kilometers east of the city of Wau. Luanyjang has around 15000 speakers according to Ethnologue (Gordon Reference Gordon2005), although the current number is likely to be much higher.
Dinka is of particular interest because of its rich suprasegmental system, which includes independent distinctions of tone, length, and voice quality. A key publication on Dinka phonology is the description of Agar Dinka by Andersen (Reference Andersen1987). The Agar dialect is geographically adjacent to Luanyjang. Other contributions on the Dinka sound system include Malou (Reference Malou1988), Gilley (Reference Gilley2003), Edmondson & Esling (Reference Edmondson and Esling2006), Remijsen & Gilley (Reference Remijsen and Gilley2008), and Remijsen & Ladd (Reference Remijsen and Ladd2008); Remijsen & Macmartin (Reference Remijsen and Macmartin2008) is a collection of sound materials on Luanyjang Dinka, with the primary aim of documenting the rich inventory of suprasegmentals. In the following sections, we present a brief description of the sound system of Luanyjang Dinka. While this description may inform ongoing work on the development of Dinka orthography, it is not a proposal to modify the existing orthography. The passage transcribed in this Illustration is ‘The North Wind and the Sun’, recorded by the second author. A version of this article annotated with embedded sound files for all the Dinka examples is available on the journal website, as supplementary material to this article.
Syllable structure and word structure
It is a defining characteristic of the Dinka language that inflected stems are mostly monosyllabic, like the corresponding lexical roots (Tucker Reference Tucker and Perrot1981: 311, Andersen Reference Andersen1990: 6). That is, instead of being marked by affixes, nominal and verbal inflections are encoded predominantly by segmental changes in the nucleus and the coda, and by changes in vowel length, tone and voice quality. Consider, for example, the singular–plural pairs w
al–wàl ‘plant:s/p’, where number is marked by vowel length, voice quality, and tone, and 
il–jὲɛl ‘thistle:s/p’, where we find the same morphological distinction marked by breaking and lowering of the vowel, in addition to tone.Footnote 1, Footnote 2 The stem-internal nature of inflectional marking can also be seen from the glosses of content words throughout this paper: most content words are morphologically complex.
The segmental template of monosyllabic content-word stems – both with and without inflection – is presented in (1). The same template is proposed in Andersen (Reference Andersen1993) for the Agar dialect.
(1) C (w) (j) V (V) (V) C
As seen from (1), syllables tend to be closed, and consonant clusters are limited to the onset, where one or two semivowels may follow an initial consonant. Open monosyllables are highly uncommon among content words. That is why we treat forms like càa ‘milk:s’ and ɲàa ‘daughter/girl:s’ as exceptions, instead of expanding the syllable template.Footnote 3 By contrast, many of the function morphemes are open syllables. Phonotactic constraints on consonants and the three-level vowel length distinction will be discussed in detail in the ‘Consonants’ and ‘Vowels’ sections below.
As for polysyllabic words, three categories can be distinguished. First, there are native monomorphemic nouns that consist of two syllables. The composition of such forms is highly constrained: the first syllable is invariably /a/, and the second syllable conforms to the template in (1) – examples are presented in table 1 (left). As noted by Storch (Reference Storch2005: 168), these forms are the only polysyllabic nouns in Dinka. Native verb stems are strictly monosyllabic. The remaining two categories of polysyllabic words are loan words and native compounds – cf. table 1 (middle and right, respectively).
Polysyllabic words in Luanyjang Dinka.
Consonants
Luanyjang Dinka has 20 consonant phonemes. Voiced and voiceless plosives, and nasals, come in five places of articulation. There are no fricative phonemes.
(2)
While the characteristic realisation of these phonemes is as implied by their IPA transcription, they may be produced differently in certain contexts. First, voiceless plosives may undergo weakening, either to their voiced counterparts or to a homorganic fricative or approximant. Examples are provided in (3) and (4).
(3)
woman.s agrs-carry_onhead.zero sorghum
‘The woman carries dura on her head.’
(4)
‘He had wrapped himself up in a coat.’
In (3), the phonemes /k/ and /c/ are realised as [ɰ] and [j], respectively. In (4), the coda /
/ gets realised as [
] when followed by a vowel. Second, the plosives may be produced without a release before a phrase boundary – a case in point is the final /p/ in example (3). The other consonants may also be underarticulated in the same context – that is, the nasals, /l/, and /r/ can be very short before a phrase boundary, to the point that their presence is to be inferred mainly from their effect on the formant transitions in the preceding vowel. Finally, the palatal /c/ – as in p
ic ‘stirring stick.s’ – has fricative [ɕ ç] and affricate [cɕ cç] variants (see Edmondson, Silva & Willis, to appear). The variation between these allophones and [c] is essentially free, even in the absence of contexts that induce weakening.
Syllable structure and phonotactic constraints on consonants
Any consonant can appear in a simple onset. By contrast, when the onset is complex, its composition is severely constrained. The main limitation is that the initial onset consonant can only be followed by one or two semivowels – cf. the template in (1). Also, /ɰ/ does not appear in complex onsets at all, and onset /
/ tends to be followed by /j/. The only exceptions to this generalisation that we know of are 
òm ‘head.s’ and 
iim ‘head.p’.
As for the coda, there are two restrictions. First, voicing in plosives is not distinctive in this slot. On the basis of their phonetic realisation in citation form and in phrase-final position, we represent coda plosives as underlyingly voiceless. Underlyingly, because unvoiced codas may surface voiced intervocalically, as in (4). Second, the semivowel /ɰ/ does not appear in the coda slot, just as it is not found in complex onsets. We will return to this phoneme below.
In the analysis proposed here, vocalic sequences – in a phonetic sense – within the syllable are not treated as diphthong nuclei. Instead, we distinguish in them a nucleus vowel preceded and/or followed by semivowel consonants. For example, we transcribe ‘canoe.s’ as rj
ɛj rather than riɛi, even though, superficially, phonetic evidence does not rule out the latter option. There are several arguments that support the ‘consonantal’ analysis. First, if we were to interpret coda /j/ and /w/ as part of a diphthong nucleus instead, then we would predict that such sequences can be followed by a consonant, just like other vowels can. To the best of our knowledge, no such native forms exist. If we treat coda /j/ and /w/ as consonants, the exception disappears – they cannot be followed by a consonant because they themselves occupy the slot of the coda consonant. As for prevocalic /j/ and /w/, they have been analysed in Agar as part of diphthong nuclei in Andersen (Reference Andersen1987), and as consonants in Andersen (Reference Andersen1993). Under the diphthong analysis, the composition of the onset is reduced to a single consonant, but the number of vowel phonemes is greatly increased. Here again, as for the coda, we favour a consonantal interpretation. One argument is of a morphophonological nature. As seen from (5a), number can be marked on nouns by a change in vowel length, usually in combination with a change in tone. When forms with a prevocalic glide are involved, this prevocalic glide is not lengthened (see (5b, c)), suggesting that it is not part of the vowel. A second argument, relating to the alignment of tone patterns, will be presented in the section on tone, below.
(5)
As seen in (6), sequences of semivowels only arise as a result of morphological processes that involve vowel breaking in forms that already have a semivowel.
(6)
The phoneme /ɰ/ has three allophones: [ɰ], [ɦ], and [zero]. [ɰ] is found before back vowels, and [ɦ] before front vowels. Finally, /ɰ/ is not realised at all in non-emphatic realisations of function words like ɰὲɛn ‘1s’ and ɰòok ‘1p’. /ɰ/ is the phoneme most constrained in its distribution. First, there is a phonotactic constraint: /ɰ/ only appears in non-complex onsets. Because of this, it could be reinterpreted as a phonetic characteristic of syllables that do not have an onset. This is the analysis proposed by Duerksen (Reference Duerksen1989) on the basis of data from speakers of the Padang and Bor dialects, who do not realise /ɰ/ consistently. This analysis does not work in Luanyjang Dinka. In this dialect, /ɰ/ is systematically realised in the onset of monosyllabic words, but not at the start of words beginning with /a/. This means that the presence of /ɰ/ is not predictable: it is part of the lexical representation of particular morphemes.
Vowels
Vowel quality
Luanyjang Dinka has seven vowel phonemes: /i e ɛ a ɔ o u/. These vowel phonemes combine fairly freely with the various suprasegmental distinctions – voice quality, vowel length, and tone. There are two restrictions. First, the high back vowel is invariably breathy. Second, the vowel /ɛ/ does not occur in the shortest level of vowel length. Morphological evidence suggests that short /ɛ/ has merged with short /a/. As seen in (7a), a long stem vowel /e/ lowers to /ɛ/ in the second plural inflection of the verb. But when the stem vowel is a short /e/, as in (7b), it changes to /a/, presumably for lack of the short /ɛ/.Footnote 4
(7)
The phonetic quality of the seven vowels is represented schematically in figure 1. The peripheral side of the ellipses represents the phonetic quality when it is realised saliently, i.e. when the vowel length is mid or long (cf. the section on vowel length below). As seen from figure 1, the distance between the high vowels /i u/ and the high-mid vowels /e o/ is smaller than that between the latter and the low-mid vowels /ɛ ɔ/. This means that the phonetic quality of /ɛ/ is typically between [ɛ] and [æ], and that of /ɔ/ between [ɔ] and [ɒ]. The centralised realisation of short vowels – represented by the central side of the ellipses in figure 1 – will be discussed below in the section on vowel length.
Schematic representations of the vowel height and advancement of the seven vowel phonemes. Based on acoustic measurements in Remijsen & Gilley (Reference Remijsen and Gilley2008).
Near-minimal sets supporting the distinction between four levels of vowel height, both for the front and for the back vowels, are presented in (8).
(8)
Voice quality
With the exception of /u/, the vowels appear in two voice qualities – modal and breathy. The breathy quality of the close back vowel is readily perceived by non-native speakers. As for the modal vowels, these often have a brassy or creaky quality. Exploratory measurements suggest that breathy vowels in Luanyjang Dinka are characterised by lower F1 values. The same finding is reported for Dinka in general by Malou (Reference Malou1988). In addition, there is relatively more energy at higher frequencies than at lower frequencies in the modal voice quality than in the breathy voice quality. This can be seen from the spectrum representations of /i/ in a minimal pair for voice quality in figure 2. The second harmonic (H2) has more energy than the first harmonic (H1) in panel (a) – impressionistically the vowel sounds creaky/brassy. When the same vowel is produced with breathy voice, H2 has about the same intensity as H1 (panel (b)). Similarly, the intensity of the harmonic boosted most by F2 (A2) is considerably weaker than H1 under breathy voice (panel (b)), but relatively less so under modal voice (panel (a)). A detailed acoustic investigation of voice quality in Luanyjang is in progress.
Spectrum representations calculated over a 30 ms window centered on the temporal mid point of the vowel /i/ in (a) modal-voiced k
iir ‘big river.s’, and (b) breathy-voiced k
iir ‘kind of tree.s’.
Near-minimal examples of each vowel phoneme except /u/ in each of the two voice qualities are presented in (9). Impressionistically, the mid vowels – /e/ vs. /ɛ/ and /o/ vs. /ɔ/ – are harder to tell apart when the vowel is breathy than when the vowel is modal, in particular when the vowel is not long (/VVV/).
(9)
Vowel length
Luanyjang Dinka vowels can be short, mid, or long. Near-minimal sets – identical but for tone – are presented in (10). In sentence-medial context, short mid and long vowels have average durations of about 70, 100, and 150 milliseconds, respectively (Remijsen & Gilley Reference Remijsen and Gilley2008). Vowel height and the presence of a following utterance boundary also weigh in at the phonetic level.
(10)
The short vowels additionally stand out from their mid and long counterparts in terms of their quality: i.e. when a vowel is short, its quality is audibly centralised. The quality of each vowel phoneme under centralisation is represented by the central side of the ellipses in figure 1. In addition to the examples in (10), consider the centralisation of /e/ in ᵑéᵑ ‘jawbone.s’ and cè ‘excrement.p’. The vowel /ɔ/ is an exception to this trend – its quality remains constant across the three levels of vowel length. In summary, the limited time domain over which they are realised appears to affect the quality of the short vowels. It should not come as a surprise, then, that the vowel system is reduced here, from seven to six phonemes – /ɛ/ does not occur as a short vowel. Further details on this centralisation phenomenon can be found in Remijsen & Gilley (Reference Remijsen and Gilley2008).
As seen from the examples in (10), and also from those in (5), the vowel length distinction has both a lexical and a morphological function; that is there are short stems and long stems, and each of these can appear in a short grade and in long grade. In the example in (10d), short and long lexical stems are represented by {tet} ‘pick’ and {teet} ‘divulge’, respectively, and morphologically short and long grades are represented by the inflections for negation and past tense, respectively. The phonologically short vowels, now, are found in the short grade of lexically short stems – e.g. tét ‘pick.neg’ in (10d). The phonologically long vowels are found in the long grade of lexically long stems – e.g. téeet ‘divulge.pst’ in (10d). The mid level of vowel length is ambiguous between the long grade of a short stem and the short grade of a long stem. In this way, in the example in (10d), we find the mid level of vowel length both in the long grade of a short stem (těet ‘pick.pst’), and also in the short grade of a long stem (tèet ‘divulge.neg’).Footnote 5
In certain nouns, the morphological grade distinction maps onto the short and long levels of phonological vowel length, i.e. skipping the mid level. Two sets of nouns can be distinguished in this context. First, there are noun stems ending in /r/. Vowel duration in the short grade of CVr stems – examples in the left column of (11a) – is as short as the short grade of short stems ending in another consonant – leftmost column in (10) above.
(11)
Vowel duration in the long grade of CVr stems – examples in right column of (11a) – is as long as in the long grade of long stems ending in another consonant – rightmost column in (10). But while the vowels of CVr stems are short in the short grade and long in the long grade, there is no evidence for the third, intermediate pattern of vowel duration in such stems. Second, the same pattern is found in a small number of nouns ending in a consonant other than /r/. Examples are presented in (11b). Further details on the quantity system of Dinka can be found in Remijsen & Gilley (Reference Remijsen and Gilley2008).
Tone
Tone patterns distinguish both unrelated lexical items – cf. (12) – and inflections within inflectional paradigms – cf. (13).
(12)
(13)
There are four distinctive tone patterns (tonemes) – Low, High, Rise, and Fall. Each morpheme carries one, irrespective of its number of syllables. This toneme associates with the rightmost syllable, which is usually the only syllable. In polysyllabic words, the penultimate syllable may have high pitch. This high pitch is not distinctive. Rather, it is predictable given the toneme on the final syllable.
F0 tracks illustrating the realisation of the four tonemes are presented in figure 3. As seen from figure 3, Low and Fall differ in F0 alignment and also in F0 height. The same is true of the distinction between High and Rise. This can be seen in figure 3b. In utterance-final context, the Rise is realised as a level tone in the middle of the speaker's register (cf. figure 3a).
F0 tracks illustrating the four tonemes of Luanyjang Dinka on monosyllabic words with a mid or long vowel and a sonorant coda in two contexts, uttered by the same speaker. High: grey dashed; Rise: black dashed; Low: black dotted; Fall: black plain. The F0 tracks are aligned on the start of the vowel of the target word, which is marked by a vertical line. In the utterance context (2b), the end of the target syllable is marked by a dot. From Remijsen & Ladd (Reference Remijsen and Ladd2008).
The string of underlyingly specified tonemes is distorted by contextual effects and by tone sandhi processes. The most notable contextual process affects the realisation of the Rise. While this toneme is realised with rising F0 in most contexts (cf. figure 3b, black dashed line), it is realised as mid level F0 in phrase-final position (cf. figure 3a, black dashed line). The most important sandhi process is Dissimilatory Lowering, which turns a High tone into Low when it is preceded by another High tone. The behaviour of the tonemes under Dissimilatory Lowering motivates an analysis whereby the contours – Rise and Fall – are interpreted as combinations of level tones, i.e. as LH and HL, respectively. That is, Dissimilatory Lowering is triggered by LH and H alike, and both H and HL undergo its effect. Finally, the application of tone processes is to some extent conditioned by the morphosyntax. For example, Dissimilatory Lowering does not apply in existential yes/no questions. Further details on the Luanyjang Dinka tone system can be found in Remijsen & Ladd (Reference Remijsen and Ladd2008).
We have shown above that syllables vary considerably in their internal quantity, due to the three-level vowel length distinction. We could conceive of this vowel length distinction affecting the association of tones, with more complex tone patterns occurring in syllables where the vowel is mid or long. As illustrated in the XTone Cross-Linguistic Tonal Database (Allison, Hyman & Mortensen Reference Allison, Hyman and Mortensen2006), this situation is familiar from many tone languages, and it can be accounted for in an autosegmental analysis by postulating that the mora (i.e. a language-specific syllable-internal weight unit) is the tone bearing unit (TBU). In Luanyjang Dinka, though, syllable-internal weight does not affect the association of tone patterns at all – level and contour tonemes alike are found on syllables with short, mid, and long vowels. Because of this, we postulate that the TBU in Luanyjang Dinka is the syllable. This hypothesis is supported by the minimal pair in (13) above, where a short vowel has a level toneme (L) in the 2nd singular inflection, and a contour toneme (HL) in the passive. In this way, the HL appears regularly in the paradigms of short transitive verb stems. Among nouns, the contour tones are attested but rare – examples include b
ɲ ‘cup.s’ and mi-kw
c ‘bull.p-spotted.p’. These noun and verb examples show that complex tone patterns are found on monomoraic vowels, falsifying a TBU = mora account.
The alignment of tonemes provides an opportunity to evaluate the hypothesis that prevocalic glides are consonants rather than initial elements of diphthong vowels (cf. the ‘Consonants’ section above). As stated above, the Low tone on the noun argument is realised as a fall in F0. The starting point of this fall is aligned with the boundary between the onset and the vowel. This early alignment distinguishes it from the Fall (HL), which dips well into the vowel – see figure 3 and figure 4a. Importantly, when the vowel is preceded by /j/ or /w/, as in figures 4b and 4c, the F0 fall sets in at the boundary between the glide and the following vowel, as predicted under a consonantal interpretation. If these glides were part of the vowel, the alignment patterns observed in figures 4b and 4c would lead us to infer that their toneme is the HL. Crucially, this is not the case. The behaviour of rj
ec ‘age set.p’ and lw
ɔᵑ ‘iron.p’ in the tonal phonology is identical to that of nòoon ‘grass.s’
F0 tracks illustrating the realisation of the Low toneme on nouns without a prevocalic glide (a), and with a prevocalic glide (b, c).
Between-dialect variation
The segmental system reported here for Luanyjang Dinka corresponds closely to those found in descriptions of other dialects and Dinka in general (Tucker Reference Tucker and Perrot1981, Andersen Reference Andersen1987, Malou Reference Malou1988). By contrast, the suprasegmental system does show some between-dialect variation. That is, while all dialects appear to use voice quality, tone, and vowel length in the lexicon and in the morphology, the exact constellations vary.
In this way, the tone system of Luanyjang Dinka – with four tonemes – may well be more complex than that of at least some of the other dialects: Andersen (Reference Andersen1987) reports only three tonemes for Agar Dinka (Low, High, Fall). Also, a more complex voice quality system has been postulated for the Bor dialect of Dinka (Denning Reference Denning1989, Edmondson & Esling Reference Edmondson and Esling2006). Edmondson & Esling (Reference Edmondson and Esling2006) distinguish breathy, modal, creaky, and hollow as phonetically distinct patterns. As noted by Denning (Reference Denning1989: 151f.), voice quality distinctions tend to affect several vocalic and suprasegmental parameters. It would be interesting to know whether and how, in Bor Dinka, the richer voice quality system interacts with the rest of the phonology.
Transcription of the recorded passage
Forward slashes in the transcription and in the translation mark major phrasing breaks in the recording, typically involving a pause.
‘The North Wind and the Sun’
Anyiköl Yom Tueᵑ ku Ak
l (Dinka orthography)
aɲik
ol 
jòom twéeᵑ k
akɔl (phonemic transcription)
‘The story of the North Wind and the Sun’ (translation)
Yom tueᵑ ku aköl aake tɛɛr e kamken / ku b
k ᵑic Ɣe ᵑa ril.
jòom twèeᵑ k ak
l àa-kè tέɛɛr kέɛm-kèen / k bi ᵑic eᵑàa rl
‘The North Wind and the Sun were contesting between themselves / so as to know which is more powerful.’
Et
n ke raan kɛny / bä ka cath / ci rot kuäth / alath yic / alanhdɛɛn / cen gup mɔr thin.
-t
ɛn kè ràaan kέɛɲ / b
kà-càa / c
rót kwa / aláa-c / aláa-d
ɛn / cén ɡw
ɔp m
ɔɔr n
‘At that time there was a traveller / he came walking by / he had wrapped himself up / in a cloth / his cloth / in which his body is warm.’
Gokë gam kamken na de raan / Ɣen bi kɔᵑ Ɣen bi nyur kɔu bei / alath / ken e ran ci riɛl.
ɡò-k ɡâam kέɛm-kèen nàa děe ràaan / ɰèn b k
ɔᵑ ɰèn b ɲw
ɛɛr kw bèej / aláa / kenέ ràaan c rjɛl
‘They agreed between themselves if there is a person / who will be the first to take off from the back / the cloth / this one would be the powerful one.’
Tnë aci yom tueᵑ jɔgɔ jɔk / bi jɔ puöt arëdt / je puöt-ë pu pu / kä puöt yen puöt / lök ke mɛt / alath / raan guöp / alath lök ke mɛt e raan gup.
-tɛn
à-c jòom twéeᵑ 
ɔ 
ɔɔk / b- ɔ pw
ot ar
et dit / è pwot -pj
w-pj
uw / k
pwot yèen pwot / l
ok kè m
ɛt / aláa / ràaan ɡwɔp / aláa lok kè mɛt è-ràaan ɡwɔp
‘Then the north wind suddenly started / to blow very hard / it suddenly blows whoow-whoow / but the more it blows / the more it attaches / the cloth / the man's body / the cloth increasingly attaches to the man's body.’
Na le dhal e thɛmde / ka luɛcpiny.
nàa lée àl è ɛm-dè / k-lwέɛc-pᵑ
‘When his attempt failed / he gave it up [lit. he put it down].’
Etn aci Aköl jɔ ba jɔɔk / bi ruɛlde jɔ ba looi / je tuc e ruelde je [luit] / je duitic ardt ba tuc apɛi.
-tɛn á-c akcl ɔ bà-ɔɔk / b rwɛl-dè ɔ bà-lóooj / èe tw
ɔc è-rwɛl-dè [luit] / èe dw
iit-ic aret dit b twɔc ap
ɛj
‘Then the Sun suddenly started / his shining to do / his shining was getting warm / getting to become extremely hot, to become very warm.’
Na la raan cath gup tuc / go alath [go] nyur bei yekɔu.
nàa làa ràaan c
a ɡwɔp twc / ɡòo aláa [ɡòo] ɲwɛɛr bèej jè-kw
‘When the body of the person walking became warm / he untied the cloth from his back.’
Tnë aci yom tuëᵑ jɔl gam / lɔn yenë aköl yenë ril tënë yeen.
-tɛn à-c [e] jòom twéeᵑ ɔl ɡàam / l
ɔn jèen
- akl à-rl tnè- jèen
‘Then the North Wind accepted / that the sun was powerful of the two.’
Yene akan.
jèen- akàan
‘That is it.’
Abbreviations
Acknowledgements
We thank Makol Jongkuc Athian, the speaker of the data presented in figure 3. All other speech data come from the second author. We are also grateful to Peter Ladefoged, who helped us in the analysis of /ɰ/. The paper has benefited from discussions we had with Torben Andersen, and from comments from Leoma Gilley and Bob Ladd. This research is supported by the British Academy (Small Grant no. 39265) and by the Arts and Humanities Research Council (Research Grant no. MRG-AN11781/APN19394). We gratefully acknowledge their support.
