Krishnamurti Bhadriraju. The Dravidian Languages

Подождите немного. Документ загружается.

2.2 V

owels

Centralization of front vowels before retroﬂexes is accompanied by rounding in some

cases and not in others, e.g.

∗

¯e

z >

∗

¯o

z > ¨o

lu as opposed to ¨e in k ¨e-(<

∗

k¯e

l-) ‘to hear’.

The Ku

rumba (Kapp 1978, 1987) dialects have two centralized vowels ¨ı and ¨e (short

and long) which in effect are i and e retracted to central position when followed by

retroﬂex and alveolar consonants originally, e.g. k¨ı

li ‘parrot’: Ta. ki

li ‘parrot’[584], ¨e

‘eight’: Ta. e

tu ‘eight’[784]; recent loanwords from English, in which /t d/ are pro-

nounced as retroﬂexes, do not show this change. Retroﬂex /

z/ are lost word-ﬁnally after

centralizing the

preceding vowels, e.g.

¨e (-p,-t) ‘hear’: Ta. k¯e

l (k¯e

tp-, k¯e

t-), k¨ıe ‘place

below’: Ta. k¯ı

z [1619]. In P¯alu Ku

rumba, retroﬂex

z and

l merge with y (<

∗

y)orare

assimilated to the following obstruents, promoting the allophones to a phonemic status.

Alu Ku

rumba also

the cause of centralization is

practically the same, e.g.

¨ı

de ‘place’:

Ta. i

tai ‘place, space’[434]. J¯enu Ku

rumba has ¨ı ¨e ¨u and ¨a, of which the conditioning

factors for ¨u and ¨a are not clear (Zvelebil 1988).

dagu also has four front centralized vowels ¨ı and ¨e (short and long) in root syllables

(when not preceded by palatal or labial consonants), conditioned by following retroﬂexes

and alveolar

∗

r (Emeneau 1970b, Balakrishnan 1976). Emeneau calls these ‘back un-

rounded vowels’, high and mid (1970b: 145). Word-ﬁnally it has ¨ı corresponding to the

Tamil enunciative vowel (1970b: 147ff.).

daga, another Nilgiri language with 125,000 speakers, is the largest of the Nilgiri

tribes, but it has only the inherited ten vowels (Hockings and Pilot-Raichoor 1992: xvi).

Badagas are said to have moved to their new habitat from Mysore around the sixteenth

centur

y. Hockings and Pilot-Raichoor

further maintain that it is

not a dialect of Kanna

as presumed by several scholars earlier. None of the dialects studied by these authors

shows any evidence of retracted (retroﬂexed) vowels that Emeneau had alluded to in

his 1939 paper. For instance, they have given several variants for the word meaning

‘seven’ (PD

∗

¯e

z), namely iu,¯ı u ,¯ıyu, iyyu, i

lu (in Merkun¯a

d dialect) besides ¯e

lu (Todan¯a

dialect). It was in the Merkun¯a

d dialect that Emeneau recorded, sixty years ago, [

¨ı¨u],

[

¨e¨u] for this word. In his foreword to the book cited above, Emeneau suggests that further

ﬁeldwork is needed to sort out this problem. The huge Ba

daga–English Dictionary by

Hockings and Pilot-Raichoor

has no evidence of retracted vowels.

Emeneau (1989: 135) thinks that the Nilgiri languages must have split from Pre-Tamil

before palatalization of velars (in the environment [+V, +Front]) started in Tamil–

Malay¯a

lam, i.e. ‘round about the beginning of the Tamil recorded texts’ (ﬁrst to third

century BCE). The absence of palatalization in Old Tamil (and Malay¯a

lam) when front

vowels were followed by retroﬂex (some alveolar) consonants suggests that the central-

ization of the front vowels was an inherited feature of all these languages, originally at

the subphonemic level. The only puzzle is that Kota does not show this feature.

Zvelebil calls it an independent language, ‘rather Kanna

da like’, but without supporting

arguments.

52 Phonolo

gy: descriptive

Old Kanna

da has the inherited ten vowels /ieaou¯ı¯e¯a¯o¯u/. Modern Kanna

da,

like Tamil–Malay¯a

lam, has added /

æ/ through borrowings from English. Tu

lu has the

same core system but it has added ε (front low unrounded vowel, historically from -ay

word-ﬁnally) and ¨ı (high central unrounded) (Bhat 1998), which mainly occur ﬁnally. ¨ı

and u result from a split of older /u/ and ¨ı corresponds to the enunciative vowel of the

other Southern languages (also see K. P. Kekunnaya 1994: 21–2). The long counterparts

of ¨ı and ε are extremely restricted.

2.2.2 Other subgroups

South-Central Dravidian (SD II), namely Telugu, Gondi, Ko

da, Kui, Kuvi, Pengo and

Man

da, all have ﬁve short vowels and ﬁve long vowels, the core inherited system.

Modern Telugu, however, has added /

æ/ derived from internal changes

as well as from

loanwords from English, e.g. c

–

ru ‘tamarind soup’, t¯a

–

ku (t¯a

ti + ¯aku) ‘palm leaf’,

–

nku ‘bank’ (Krishnamurti and Gwynn 1985: 29–30). In some of the non-literary

languages (Gondi, Ko

da, Kui), short and long vowels contrast only in word-initial (root

syllable) position. In speech, Telugu, like the other literary languages, has subphonemic

glides [y

w] added to word-initial front

and back (rounded) vowels respecti

vely. Win

ﬁeld

(1928: 1–2) talks

of a hiatus in Kui when

‘vowels

come together,

laa young w

oman

’; this

hiatus is ‘prevented by the insertion of v or j or n, between the contiguous vowels, e.g.

l¯avenju (l¯a-v-enju), young man’. Winﬁeld perhaps meant a glottal stop while referring to

hiatus. He also suggests that vowel sequences au and ai are rendered av and ay, e.g. k¯au →

k¯av ‘a fruit’, m¯ai → m¯ay ‘our’ (Winﬁeld 1928: 2). Kuvi and Ko

da have a phonemic

glottal stop.

The ﬁve-vowel system with an additional phoneme of length also occurs in Central and

North Dravidian languages. The only exception is Brahui which does not have short e and

o owing to the inﬂuence of surrounding Indo-Aryan and Iranian languages, which lack

contrast between e/¯e and o/¯o (Emeneau 1962d: 7–20; Elfenbein 1998: 391–2). Emeneau

says that in non-initial (unaccented) syllables only, short e and o occur as allophones of

/e o/ which are always long in initial (accented) syllables. Emeneau draws attention to

limited contrasts between e and ¯e in inﬂected forms, non-initially (1962d: 7, fn.2).

2.3 Consonants

There is greater variation among the consonants of different subgroups. In Proto-

Dravidian voicing and aspiration were not distinctive features among the consonants.

Word-initially apical (alveolar and retroﬂex) consonants did not occur. There were no

initial consonant clusters. Medial clusters were usually geminate obstruents or sequences

of nasal + a homorganic stop (+ a homorganic stop); a semivowel or liquid followed by

an obstruent also occurred less frequently.

2.3 Consonants 53

2.3.1 South Dravidian (SD I)

The inventory of consonants of Old Tamil is very similar to that of Proto-Dravidian. It

has seventeen consonants, viz. /p t t

tk,mn˜n

n, r

z, l

l,yw

h/. The last segment is

my transliteration of Tamil ¯aytam ([k

]inTamil Lexicon), which is derived from a Proto-

Dravidian laryngeal

∗

H (see section 4.5.7.2.3 below; also Krishnamurti 1997b). The

rest are represented by the same symbols. Old and Modern Tamil write both dental [n]

and alveolar [

n] nasals but evidence for their contrast is insigniﬁcant. Lehmann (1998:

77) and Steever (1998: 14, 16) considered ¯aytam an allophone of Tamil /y/ before

obstruents, but there is no real evidence for this assumption. Old Tamil has eytu ‘to

approach’ [DEDR 809], and cey-tu ‘having done’ where /y/ is not phonetically [

h]. It is

important to notice that Old Tamil had three distinct coronal obstruents – t

t (dental,

alveolar and retroﬂex); the system is also preserved by Malay¯a

lam, old and modern.

Modern Tamil has only dental and retroﬂex stops. Also Modern Tamil shows voice

contrast in the stop series, mainly in borrowed words. The stops were lenis medially in

old Tamil allophonically [w

r ] and voiced in postnasal position. Old Tamil had nine

consonants in word-initial position /p t ckmn˜n y v/, and nine in word-ﬁnal position /m n

nrl

z v y/. Words end in vowels in Modern Tamil. Malay¯a

lam has developed voiced and

aspirated stops (voiced and voiceless) through extensive borrowing from Sanskrit. It also

has six phonemic nasals corresponding to six stops /p t t

t c k/, because they contrast in

postvocalic gemination /mm nn

n˜n˜n˙n˙n/. Only /m n/ occur word-ﬁnally; an optional

[

ə]i

s added to words ending in /

nrrl

l y/,

e.g.

k¯al → k¯al

‘leg’. Proto-Dravidian retroﬂex

frictionless continuant /

z/, which occurred in all the literary languages until the medieval

period, remains only dialectally in Modern Tamil but is preserved intact in Malay¯a

lam.

In most Tamil dialects it has merged with /

l/ or /y/ (Lehmann 1998: 75–99, Annamalai

and Steever 1998: 100–28, Asher and Kumari 1997: 405–50).

Kota (Emeneau 1944: 15ff.) has six pairs of voiceless and voiced stops /p b t d t

ˇc j k g/, four nasals /m n

n˜n/, two laterals /l

l/, one ﬂap /r/ and two semivowels /v y/. /ˇc/

can be replaced freely by [s] and /j/ by [z]; before a retroﬂex, [s] is pronounced as [

s].

The Toda consonantal system is atypical and unique (Emeneau 1957, 1984). It has

seven pairs of stops, voiceless and

voiced /

pbtdcj t

dˇcˇ

d k g/, [c = ts, j = dz],

four nasals /m n

ŋ)/, seven fricatives, i.e. three voiceless /f ␪ x/ and four pairs of voice-

less and voiced sibilants /s z

s zˇsˇz

z/, three trills /r r

r/, two pairs of voiceless and

voiced laterals /



l/, and two ‘continuants’, palatal and velar /y w/. Emeneau classiﬁes

these into variable and invariable series with the feature +/− voice. The variable voiceless

consonants are /f ␪ x

r 



sˇs

s/ and the variable voiced consonants, /m n

n y/.

The variable consonants have variable voice feature allophonically (Emeneau 1984: 14,

Shalev et al. 1994: 32). In Toda voiced obstruents (stops and fricatives) do not occur

word-initially. Shalev et al. have discussed at length the acoustic and articulatory analysis

54 Phonolo

gy: descriptive

of the sibilants to answer the question how so many sibilants could be distinguished in

Toda and what cues are used for such distinction.

Many of the Nilgiri languages preserve the three coronal stops, tt

and

t. Kota (Emeneau

1944–6), Iru

la (Zvelebil 1973) and P¯alu Ku

rumba (Kapp 1978),

Alu Kurumba (Kapp

1987), Sh¯olega ( Zvelebil 1990b), J¯enu Ku

rumba (Zvelebil 1988) retain the distinction

between r and

r, of which the latter is derived historically from PD

∗

t in intervocalic

position. None of the Nilgiri languages preserves PD

∗

z. They all retain the other retroﬂex

consonants

l. The stops in ﬁve positions have phonemic voiced counterparts /p b t d

d)cj

d k g/ in all Nilgiri languages and Ko

dagu. In P¯alu Kurumba there is no reason

to represent

t, d and r as separate phonemes (see Kapp 1978). It is also not necessary to

have two n’s [n] and [n

]asi

s done in Tamil.

Kapp gi

ves a pair of forms to indicate

r contrast, nira(mu) ‘colour’: n¯ata(mu) ‘stench’(1978: 520), but it is highly suspicious

since, historically, [

r] is an allophone of /t/ in intervocalic position.

dagu has three sibilants

sˇs/ because of borrowings

from Sanskrit, and

ﬁve nasals

/m n

n˜n˙n/, corresponding to the ﬁve stops (Balakrishnan 1976: 1–12). All the nasals

contrast intervocalically in gemination, e.g. ca˙n˙n¯ati ‘friend’, a ˜n ˜n¯ana ‘´ıgnorance’. Word-

initially /˜n/ occurs in Ko

dagu as in Malay¯a

lam in native vocabulary (Balakrishnan 1976:

11, 26).

Old Kanna

da had ﬁve stops with a four-way distinction – voiceless unaspirated, voice-

less aspirated, voiced unaspirated and voiced aspirated – four fricatives /s ´s

s h/, three

nasals /m n

n/, two laterals /l

l/, two trills /r

and one retro

ﬂex

approximant /

which

was becoming archaic (Ramachandra Rao 1972: 1–29). K¯e´sir¯aja also listed [˙n] and [˜n]

following the Sanskrit model, but they were not phonemic. Modern Kanna

da has added

æ/ among the vowels and /z f/ among the consonants. Older /r/ and /r/ were replaced by

/r/ (Steever 1998: 130). Ramachandra Rao (1972: 4) treats [˙n] as an allophone of /m/,

and [˜n] as an allophone of /n/. Word-initially /

n y r/ did not occur in Old Kanna

da. Initial

d/ occurred only in loanwords; only one case of word-initial

r and a few of l are found

in Pampa Bh¯arata (tenth century CE). All consonants except /r

z s/ could occur double

in Old Kanna

da (1972: 10–11).

lu (D. N. S. Bhat

1998: 160

–1, Kekunnaya 1994:

11) has

ﬁve pairs of stops,

voiceless

and voiced /p b t d

dcjkg/,ﬁvenasals /m n

n˜n˙n/, one sibilant /s/ and two semivowels and

two liquids /v y l r/. The Brahmin dialects add three more fricatives /´s

s h/ and /

l/ besides

the aspirated stops in both the voiceless and voiced series. Word-initial front vowels

have a y-onglide and the rounded vowels a w-onglide in speech. Bhat says the aspirated

stops are in free variation with the corresponding unaspirated ones even in the Brahmin

dialect (1998: 161). The South Common (non-Brahmin) dialect is said to preserve the

contrast of l and

l;/˙n

l/ do not begin a word. All unaspirated stops and sonorants have

The author said, in a private communication to me in August 1998, that it was a mistake.

2.3 Consonants 55

a contrast between single and double, including the four nasals (Kekunnaya 1994: 29).

No consonant occurs in ﬁnal position.

2.3.1.1 Phonotactics

The ﬁrst syllable of a word can be either V

or C

; any short or long vowel can

occur as V

is not usually an apical [−Distributed] consonant /t

z/; in Modern

Tamil, Malay¯a

lam, Kanna

da, Tu

lu, Iru

la and Ko

dagu, in recent times, a limited number

of words with word-initial apicals have started emerging through aphaeresis or loss of

a short word-initial vowel, e.g. Ta. re

du (OTa. ira

tu), Ma. ra

du ‘two’, Iru

la ra

du,

du,Ko

dagu da

d¨ı,Tu

lu ra

d¨ı. This tendency i

s much more widespread in Iru

la and

lu. The diphthongs are treated as long vowels metrically in the literary languages

because of their equivalence to long vowels in the duration of articulation (two morae).

Phonologically they can be considered sequences of a vowel + consonant, where the C =

semivowel. There are no word-initial consonant clusters in the native element.

2.3.2 South Dravidian II (SD II)

These languages have word-initial apicals /

d/ and consonant clusters unlike South

Dravidian I. All languages of this subgroup developed word-initial consonant clusters

with obstruents as ﬁrst members and sonorants as second members. Gondi and Ko

do not show such initial clusters now, but there is evidence that they once existed and

were simpliﬁed. Old Telugu had such clusters with Cr-(C= an obstruent other than an

affricate, s and w), but they are simpliﬁed by loss of -r-(<

∗

z) in Modern Telugu. The

remaining languages, Kui, Kuvi, Pengo and Man

da, still have initial consonant clusters

and the sound change is still in progress (Krishnamurti 1978a).

In Old Telugu, the inherited consonantal system was fairly well preserved except for

incorporating the feature of voicing from the earliest known period: /p b t d

dcjkgsm

z y w/; Sanskrit borrowings from a very early period added two more sibilants

/´s

s/ and the voiced and voiceless aspirated

series /ph bh th dh ch jh

dh kh gh/.

In Modern Telugu there is only an alveolar ﬂap /r/; even by the thirteenth century /

was totally eliminated as a distinctive sound after it had merged with /

d r/ in different

environments. A new phoneme /f/ is added to Modern Telugu in such loanwords as

/¯af¯ısu/ ‘ofﬁce’, and /k¯af¯ı/ ‘coffee’. /

l/ do not begin

a word. All words ended in vowels

in Old Telugu. In Modern Telugu only /mnwy/canoccur word-ﬁnally.

Gondi has several dialects, but there are only a few systematic descriptions. Natarajan

(1985: 32–3) set up the following consonants for Abhuj Maria, a Southern dialect of

Bastar: /p b t d

dcjkgmn˙nslr

r w y/. According to G. U. Rao (1987b:

§2), the

South Bastar Gondi has an overall pattern of twenty-one consonants; he adds /

nh/to

Natarajan’s list. Hill Maria has, in addition to the above, three new segments: uvular /R/,

a glottal stop /ʔ/ and a retroﬂex /

n/. Western Gondi has twenty consonantal segments

56 Phonolo

gy: descriptive

including a phonemic /h/ and no retroﬂex /

n/. Adilabad Gondi has twenty consonants

including a phonemic /h/. Adilabad Gondi described by Subrahmanyam (1968a) has also

aspirated consonants. Even some native words have the feature of aspiration, e.g. ph¯o

‘sun’ (cf. Ta. po

zutu). The Southeastern Gondi has twenty phonemes including /

n/, but

with a marginal status for /h

ŋ/. G. U. Rao (1987b: §3) set up twenty-three consonants

for Proto-Gondi: /pbtd

dcjkgmn

ŋshr

r rl

l w y/.

da has twenty-three consonantal segments /p b t d

dkgʔ szhR

rmn

n ŋ r

r l v y/. /ʔ R

ŋ/ do not occur word-initially. Voiceless and voiced sibilants and trills

/s z R

r/ pattern with stops. In consonant clusters /ʔ/ is never the ﬁrst member and /R/ is

never the second member of a consonant cluster (Krishnamurti and Benham 1998).

Kui has eighteen consonants /p b t d s j

dkg(s)hmn

r l v/. Winﬁeld lists [s] twice,

a palatal [s] with a corresponding

voiced [j], the second a sibilant

[s], but there is no

phonological difference between the two. There is no semivowel /y/ in Winﬁeld’s Kui.

Where it is expected on the basis of PD

∗

y, it is replaced by i or j in Kui; v ∼ b occurs

in dialects mostly inﬂuenced by Oriya. Winﬁeld calls the stops ‘strong consonants’, and

the sonorants, ‘weak consonants’, because the sonorants ‘may be lost in declension or

conjugation’. In the case of obstr

uents, only voice feature changes

(Win

ﬁeld 1928:

§1).

The Kuvi consonantal system is essentially similar to that of Kui: /p b t d

dkgʔ c

j,sh,mn

ŋ ,l,r

r, v y/; /c j/ are phonetically /ts ds/ freely varying with /s z/ (Israel

1979:

§1); /ʔ ŋ /donot

begin a word; only /

ŋ y m/ occur ﬁnally.

Pengo has twenty-two consonants: /pbtd

dcjkgszhmn

ŋr

v y/. Nasalized

vowels occur in Indo-Aryan loanwords like g

¯o ‘wheat’. Aspiration occurs in Oriya

loanwords, e.g. dhan ‘wealth’. Native h can combine with stops, e.g. k¯uk-hi- ‘having

heard’, where -hi is perfective participial sign. We do not know if the sequence is pro-

nounced as a single aspirated stop or as a sequence of k + h (Burrow and Bhattacharya

1970:

§1).

Man

da is closer to Pengo than to any other member (Burrow 1976), but we do not

know much about the phonological system. It has /r

r l/ besides ﬁve pairs of voiceless

and voiced stops.

2.3.3 Central Dravidian (CD)

None of the

Central Dravidian languages shows either apical consonants or consonant

clusters word-initially in native vocabulary. This feature distinguishes them from the

South Dravidian II subgroup. Except for the Kolami–Naiki subgroup, aspirate conso-

nants are not phonemic. Particularly Naiki shows voiceless and voiced aspirated stops

initially in native words (see section 2.3.3.1).

B. Ramakrishna Reddy collected data on Man

da over a decade ago but he has not published his

analysis of the language yet. He also reported that there were two other dialects or languages

closely allied to Pengo and Man

da, which he called Indi and

Awe.

2.3 Consonants 57

Kolami has four pairs of stops, voiceless and voiced /p b t d

d k g/, two affricates

/c j/, two sibilants /s z/, three nasals /m n

ŋ/, two liquids /r l/ and two fricatives /v y/.

In loanwords from Marathi [ts

l] are used (Emeneau 1954:

§1). Subrahmanyam (1998)

represents Emeneau’s /z/ by /j /. Naik

ri has a four-way distinction of stops or affricates

with features of voicing and aspiration like /p ph b bh/ in all ﬁve positions . It also has

an alveolar /c/, three fricatives /v s h/, three nasals /m n

ŋ/ two laterals /l

l/, a trill /r/ and

a semivowel /y/. /c/ is said to be an alveolar affricate /ts/. Palatal affricates /ˇc j/ occur

mainly in Marathi loans. Contrast between long and short vowels seems to be conﬁned

to root syllables (Thomasiah 1986:

§1). In this work I have treated Naik

ri and Naiki

(Chanda) as dialects of the same language, distinct from Kolami (see section 2.3.3.1).

Parji has nineteen consonantal segments of normal status /p b t d

dcjkgmn˜n

ŋ r

r l v y/; /s h/ are marginal phonemes heard in rendering Halbi words (Burrow and

Bhattacharya 1953: 1–8). It appears that /˜n

ŋ/ could be phonemically represented as /nj

ng/, although Burrow and Bhattacharya say that the stop element is lost word-ﬁnally but

is restored (in the case of g of

g when followed by a vowel) (1953: 7).

Ollari has four pairs of stops /pbtd

d k g/, two pairs of affricates /ts dz c j/, ﬁve

nasals /m n

n˜n

ŋ /, three liquids /r

r l/, two sibilants /s z/ and two fricatives /v y/. The

segments /ts dz c j/ are said to have a marginal status, occurring mainly in loanwords

(Bhattacharya 1957: 10). As in Parji, it seems possible to represent /

ŋ ˜n/ as sequences of

/ng nj/ with phonetic realization as simple nasals in word-ﬁnal position (Bhattacharya

1957: 1–14).

Kon

dekor Gadaba has nineteen consonantal segments: /p b t d

dcjkgsmn

ŋ rlw

y/; /

ŋ/ do not begin a word. There are no word-initial consonant clusters (Bhaskararao

1980).

2.3.3.1 Voiced aspirate stops

A puzzling feature of

Central Dravidian is the presence of many words with voiced

(some

voiceless) aspirated stops in Naik

ri (which Burrow and Emeneau have listed as a dialect

of Kolami in DEDR, see index, pp. 688–91), corresponding to unaspirated voiced stops

of several other languages in native words. There are eighteen items of this kind, most of

which have etymologies, e.g. ghurrum ‘horse’: OTe. gu

rramu, Mdn Te. gurram [1711],

gh¯a

li ‘wind’: Te. g¯ali id.[1499], dh¯av ‘distance’: Te. davvu, Kol. [SR] davva [446] (from

∗

t-a- does not seem to be an appropriate etymology.) Even the items with voiceless aspi-

rated stops in Naik

ri have cognates in voiced unaspirated stops in many other languages,

e.g. kh¯ır ‘line’ [1623] has cognates with g- in Ka., Ko

d., Tu., Te., Kol. [SR], Ko

da, Kui

and Kuvi. This situation leads one to suspect that we have evidence of a voicing laryngeal

of Proto-Dravidian retained here. Naik

ri is deﬁnitely not a dialect of Kolami and it pre-

serves some archaic features which Kolami has lost, e.g. n¯em ‘we (inclusive)’ from PD

∗

˜n¯am [3647]. Note that there are three lexical items in Naik

ri which begin with h- before

58 Phonolo

gy: descriptive

word-initial vowels corresponding to its absence in the other languages, namely he

‘bull, ox’: Ka. e

ztu,Ta.erutu,Te.eddu,ple

dlu (aberrant phonology [815]), hegar

‘watchman’s raised platform’: Kol. ¯egar id. (no other language has cognates; [877]) and

rug ‘iguana’ (Te. u

dumu id. [592]). There are forms beginning with vowels, which

contrast with these. No explanation has been given for this phenomenon.

2.3.4 Nor

th Dravidian (ND)

rux has the following consonants: kkhxgghcchjjh

rhtthddhpphbbhsh

mnrlyw. Grignard (1924a:

§1) considers [˜n

n ŋ] as positional variants (i.e. allophones)

of /n/ before palatal, retroﬂex and velar consonants. He calls them ‘purely accidental

modiﬁcations of the “n” sound’ (1924a: 3).

Malto has ﬁve pairs of voiceless and voiced stops /p b t d

d c j k g/, a voiceless uvular

stop /q/, four fricatives /ð s h/, four nasals /m n ˜n

ŋ /, three liquids (trill, lateral, ﬂap)

/r l

r/ and two semivowels /w y/ (Mahapatra 1979:

§2). In consecutive syllables (except

when they are separated by a ‘formative boundary’) an initial consonant as onset of a

syllable is followed by an identical segment as the onset of the next syllable, qeqeldu

‘earth’, kake ‘comb’,

tuwa ‘leper’, etc. (1979: 38–40). /ðγ/ do not occur word-initially.

There are no word-initial clusters in Malto.

Brahui has four pairs of stops + a glottal stop and two affricates: /pbtdcj

dkgʔ/,

eight fricatives /f s z ˇsˇzx h/, four nasals /m n

ŋ/, two laterals / l/, two ﬂaps /r

r/,

and two semivowels /w y/ (Elfenbein 1997, 1998). The large number of fricatives is

owing to the inﬂuence of Balochi. It is reported that two-thirds of Brahuis are bilingual

in Balochi. Words with /f x / have good Dravidian cognates. Word-initially /f/ does

not occur. Voiceless lateral //, transcribed by Bray and Elfenbein as [l

h], is internal to

Brahui as are the retroﬂexes, which are not found in Balochi; /

ŋ / occurs only before

velar stops and can be replaced by /nk, ng/. Voiceless stops can optionally be aspirated in

speech. Voiced aspirates occur onl

y in loanwords from Indo-Aryan. Northern Brahui has

/h/ which is replaced in the Southern dialects by a glottal stop [ʔ] initially and between

vowels; it is lost ﬁnally and before consonants. ‘It is often preﬁxed to word-initial words

in the south’ (Elfenbein 1997: 800–4, 1998: 392–4).

2.4 Suprasegmental features

Vowel length is the only stable suprasegmental feature in all Dravidian languages. In

single morphemes (roots), which are monosyllabic, short and long vowels contrasted

in the proto-language. This feature is inherited by all Dravidian languages with the

exception of Brahui, which has lost the contrast in mid vowels e/¯e and o/¯o under the in-

ﬂuence of the neighbouring Indo-Aryan and Iranian languages. In non-radical syllables

also, in the literary languages, long and short vow

els contrast in in

ﬂection and deriva-

tion. In many non-literary languages, short and long vowels contrast only in the root

syllable.

2.4 Supr

asegmentals

2.4.1 Stress and intonation

Very little has been written about stress and intonation. The following statement is

typical: ‘Stress in Tamil is non-contrastive, and falls on the ﬁrst syllable of a word.

Intonation patterns have yet to be studied satisfactorily’ (Annamalai and Steever 1998:

104). Andronov (1969: 33) says that ‘the place of a stressed syllable in a word is not ﬁxed’

in Tamil and marks words differently, p´eyar ‘name’ vs. vir´al ‘ﬁnger’, ´ava

n vs. tam´ı

z.In

that case, stress becomes phonemic, but there is little supporting evidence for such an

analysis. Asher (1982: 230–4) says the role of stress is ‘to express emphasis’. It does

not distinguish words. ‘The main correlate of stress is greater intensity.’ Syllables with

longer vowels when stressed have extra duration. Statements, commands and question-

word questions have a falling intonation []; if the question is repeated with surprise,

it has a rising intonation []. Yes–no question sentences ending in an interrogative

particle have a rise–fall on the particle []. Exclamatory sentences ha

ve a mid-level

intonation [→].

In Malay¯a

lam stress does not distinguish one lexical item from another. Citing K. P.

Mohanan, Asher and Kumari (1997: 436–7) say that position and vowel-length play a

role in determining the placement of primary stress. The stress falls on the ﬁrst syllable

if it is short; it falls on the second syllable which follows a short syllable and if it has

a long vowel. ‘All syllables after the second which have a long vowel carry secondary

stress’, k

udira ‘horse’, pa

t¯

lakk

¯aran ‘soldier’. Declarative sentences end with a rising

pitch. Questions which end in interrogative particles -¯e or -¯o end with a falling pitch.

Question-word questions carry a rising tone. Non-ﬁnal clauses end on ‘level or rising

tone’. Commands end with falling pitch. Asher and Kumari say that intonation ‘´ıs very

much a primary area where research is needed’ (436). According to Mohanan ‘stress is

associated with word melody’.

Emeneau (1984: 18) makes some signiﬁcant remarks on intonation in Toda: 1. A

declarative sentence has ‘a slight drop of pitch on the syllable or syllables after the last

strong stress of the sentence’. 2. ‘Imperative or prohibitive sentences have level pitch

continued to the end.’ 3. ‘Questions ending in an interrogative particle, imperatives in -¯o

or -m¯o, vocatives i

¯a or -¯o, ...some interjections ...have on the ﬁnal syllables a pitch

that is somewhat higher (level or slightly rising) than the pitch of the preceding syllables.’

Lisker and Krishnamurti (1991) make the following observations about Telugu stress

based on an experimental study in which two groups of speakers participated, ten native

Telugu speakers who also knew English and ﬁfteen native English speakers who had no

knowledge of Telugu. The judgements on placement of stress are broadly similar in both

the groups. The ﬁrst syllable carries a phonetic stress in words of two short syllables

(SS), e.g. g´adi ‘room’ (SS – [mean value of responses of Telugu speakers by the syllable

1st: 2nd: non-deﬁnite] 57% : 35 % : 8%]), or when the ﬁrst syllable is long and the second

short, e.g. t

¯ata ‘grandfather’ (LS – 77% : 20% : 3%). If the second syllable is long or if

both syllables are long, stress falls on the second syllable, e.g. dag

¯a ‘deceipt’ (SL – 40% :

60 Phonolo

gy: descriptive

59% : 1%), b¯ak

¯ı ‘debt’, r¯up

¯ay ‘rupee’(LL – 21% : 76% : 4%). In the case of trisyllabic

words, the ﬁrst syllable is stressed if it is long; otherwise, it is the penultimate syllable that

carries stress, e.g. k

¯uturu ‘daughter’ (LSS – 59% : 26% : 9% : 5%; the last value for unde-

cided or non-deﬁnite): pal´aka ‘slate’ (SSS – 25% : 59% : 10% : 5%), tap´assu ‘penance’

(SLS – 26% : 63% : 0% : 11%), tup

¯ak¯ı ‘gun’ (SLL – 15% : 57% : 21% : 7%), s¯at

¯ani ‘a

weaver caste’ (LLS: 25% : 64% : 0% : 11%). The difference in scores among the English

speakers is sharper for the same classes of items. There is a greater degree of variation

in repeated tests among T

elugu speakers than among the English

speakers.

‘If Telugu

were a stress language like English, we might expect distributions showing either con-

stant values of 0 or 100% or having the shape of step functions with values shifting

from 0 to 100%. Instead we ﬁnd distributions that, particularly for Telugu speakers,

are closer to being

continuously variable between

those extremes

’(4). ‘. . . the Telugu

group’s behaviour is a consequence of the nondistinctive function of word-stress in the

language’(5).

In Ko

da, short and long vowels contrast only in the initial syllable (Krishnamurti

1969a: 188–92) of a ‘phonological word’. A phonological word occurs between two

marked junctures /# + ,/ (# marks silence before or after an utterance; + boundary of

phonological word; /, / = marks a phonological phrase which is part of a higher structure).

A vowel in the ﬁrst syllable is stressed if it is long; if it is short the second syllable is

phonetically longer and louder and carries primary stress. All alternate syllables from

the stressed one carry ‘a second degree loudness’ or prominence, [

¯us.pa.z`ı.nad] ‘she is

applying

oil

’, [a

r.b´a.zi.n`ad] ‘she is cr

ying

’. A stop

preceding a stressed vowel

is markedly

fortis. ‘Once the word boundaries are indicated and the contrast of short and long vowels

in initial syllables marked, features of stress and length in the remaining syllables within

a phonological word become automatic and predictable’ (190). A morphological word

can be two phonological words, e.g. #n¯a + r¯ayna# ‘Oh God!’ (name of a Hindu God

n¯ar¯aya

na) is one morphological word, but two phonological words, because there are

two long vowels. A phonological word has only one

phonemic long vowel. In Ko

phonological and morphological words have been deﬁned differently (for details, see

Krishnamurti 1969a: 186–92).

It is clear from the existing descriptions that the ﬁrst syllable is stressed irrespective

of vowel length. If it is short and followed by a heavy syllable it is the latter that

receives primary stress. Sentences ending

in an interrogative particle /

¯a¯o¯e/ have a rising

intonation (section 8.4). Imperatives have a falling pitch when they end or constitute

whole utterances.

2.5 Sandhi or morphophonemics

Both internal (intra-word) and external (inter-word) sandhi is widely prevalent in

Dravidian. Since some of the processes are idiosyncratic to individual languages, only