The Language of the Modhupur Mandi (Garo), Volume 1

Introduction (A)

pp. 17
Page  17

This chapter describes the pronunciation of the significant sounds of the Mandi and A'chik dialects of Garo. For the most part, these two dialects are so similar in pronunciation that a single description can serve for both. The few cases where they differ will be pointed out.

One source of difference comes from the varied influence that Bengali has had upon the Garo dialects. Bengali and closely related dialects of Assamese are spoken on three sides of the Garo Hills. These have probably been influencing Mandi and the other Garo dialects for a good many centuries. Borrowed words are so ubiquitous that to pretend to isolate "genuine" Garo vocabulary or "genuine" Garo phonology (pronunciation) is a bit artificial, because the large number of words that have been borrowed from Bengali have introduced new sounds and new sound patterns that all Garos now use easily. Nevertheless, by focusing first on the nonborrowed part of the vocabulary we begin from a reasonably stable and consistent starting point. Regularities emerge that might otherwise be missed. This description begins, therefore, with the nonborrowed core of Garo phonology. Later, the additions and modifications that have been introduced into the sound system as a result of contact with Bengali, and to a lesser but still important extent with English, will be described. Some people are far more fluent than others in Bengali and English, and individual speakers vary greatly in their success at carrying foreign sounds into their stream of Garo. This means that the stable phonological core of native Mandi phonology has to be seen within the context of extreme variation.Page  18

This and the next two chapters deal with pronunciation. The slanting brackets, / /, that enclose many of the examples, indicate that it is the pronunciation that is being considered. The symbols within the slanting lines represent the significant sounds ("phonemes") of the language. When the focus is less specifically on pronunciation, as it is in the later chapters, examples from Mandi and other dialects of Garo will be written in italics. Glosses (translations into English) will be enclosed in single quotes.

Syllables

(A)

pp. 18

The first characteristic of the Garo sound system that needs to be understood is the crucial importance of the syllable. While Garo is not a tone language, the structure of its syllables closely resembles that of the tone languages of East and Southeast Asia. As is typically the case with these tone languages every Garo syllable can be described as having: 1) an initial consonant or consonant cluster; 2) a central vowel; and 3) a final consonant. In addition, 4) the glottal stop plays such a special (and even "tone-like") role in the language that it deserves to be considered as the fourth major component of the syllable. Of these four components, only the vowel is obligatory. Every syllable must have a vowel, but some have no initial or final consonant and no glottal stop. Like the tone languages with which it is related, Garo allows more consonants and consonant clusters at the beginning of a syllable than at the end. Many aspects of the Garo sound system have to be described by reference to the position of sounds within the syllable.

Each Garo syllable is pronounced as a distinct unit, and even where the same letter is used to represent both an initial and a final consonant, the letter must usually be pronounced in rather different ways in the two positions. Consonants that come between vowels in the middle of a word can be consistently interpreted as belonging either to the preceding, or to the following syllable, and they are pronounced in different ways depending on the syllable to which they belong. This means that a careful transcription of the language must show which syllable a word medial consonant belongs to. To make this clear I have separated the syllables of polysyllabic words with hyphens.

In addition to its phonological importance, the syllable is often a unit of meaning. Put differently, syllable boundaries often correspond to morpheme boundaries. (A morpheme is a minimal unit that carries meaning—-the base of a word, a prefix or, more often in Garo, a suffix.) Hardly a single Garo morpheme is shorter than a syllable, but each syllable of a Garo word is often a morpheme that contributes its own particular meaning to thePage  19total meaning of the word. Many Garo words have three or four syllables, and words with as many as seven or eight syllables are by no means uncommon. Each of these syllables, however, is likely to have a separate meaning, and a separate morphological identity. Borrowed words form an important exception to this generalization. Many borrowed words are polysyllabic, and the syllables of these words are much less likely to have identifiably separate meanings than are the syllables of older Garo words. Even among words that show no signs of having been borrowed, bisyllabic morphemes do exist, but more often each syllable contributes its own meaning.

Syllable Initial Consonants

(A)

pp. 19

Table 2—1 lists all the consonants and consonant clusters that can be used to begin a syllable in the older vocabulary of Garo. In addition to these, Garos who are fluent in Bengali or English sometimes use Bengali and English consonants, or at least approximations to them, when using borrowed words, and the sounds that have been carried into Garo along with the borrowed words will be described later. For the sake of linguists who are unlikely to be able to turn for help to a native speaker of Garo, I will use phonetic terminology to describe the Garo speech sounds. Readers who want to learn the language will generally have the help of a Garo speaker who can be asked how the examples should be pronounced. You can imitate the pronunciation that you hear and ignore any phonetic terminology that is unfamiliar.

p-	t-	k-
b-	d-	g-
m-	n-
s-	ch-	j-
r-	h-	w-	(l-)	∅
sp-	st-	sk-
pr-	tr-	kr-
br-	dr-	gr-
mr-
sr-	chr-	jr-
spr-		skr-

/p-/, /t-/, and /k-/ are, respectively, bilabial, dental and quite far back velar stops. All are voiceless and aspirated. /p-/ and /t-/ are pronouncedPage  20quite like English p and t when these come at the beginning of a word before a vowel, and they are similar to the Bengali sounds that are written as XX and XX. /b-/ and /d-/ are the voiced equivalents of /p-/ and /t-/, and they are quite similar to English b and d or to the Bengali XX and XX. /k-/ and /g-/ are pronounced a bit further back than English k and g and they are occasionally somewhat affricated. They are similar to Bengali XX and XX.

/ch-/ and /j-/ are, respectively, voiceless and voiced affricates and /s-/ is a voiceless sibilant. The language lacks a voiced counterpart of /s-/ (i.e. it has no z sound). /s-/ is pronounced in a position intermediate between that of English s and sh , but it is closer to English s than is the sh -like sibilant of Bengali. Since Garo has only one sibilant (i.e. it has no distinct sh sound), even a rough approximation it is unlikely to be confused with anything else. You are unlikely to do any harm, except to your own vanity, if you do not pronounce it exactly right. /ch-/ and /j-/ are fairly similar to the English sounds written with the same letters, but they are pronounced with the tongue in the same position as Garo /s-/.

/m-/ and /n-/ are bilabial and apical nasals, very much like the English nasals and in no way remarkable. /r-/ is a flap in which the tongue brushes very quickly against the roof of the mouth until it is blown away. This means that it is like the Bengali r rather than the English r . /h-/ and /w-/ are very much like the equivalent English consonants. /h-/ is a rather unusual sound in A'chik and in most other dialects of Garo spoken in India, but it occurs more frequently in the Mandi dialects of Bangladesh. A number of words that never have initial /h-/ further north, are regularly heard with /h-/ in Mandi. Even in Mandi, however, initial /h-/ is extremely variable. Many words that sometimes have initial /h-/ can be heard on other occasions without it. This variation will be described in more detail in Chapter 4 where it is included among other examples of variation.

In the table, /l-/ is in parentheses because it occurs only in borrowed words and must, therefore, be assumed not to be a part of older Mandi phonology. In the 1950's, dialects could still be found in the Garo Hills that completely lacked initial /l-/. Speakers of these dialects pronounced borrowed words such as /lem/ 'lamp' as /rem/. I have never met a Mandi in Bangladesh who had any trouble with initial /l-/, however, and it is pronounced easily by many speakers in the Garo Hills as well. Initial /l-/ is one of the two best established sounds that have come into Garo with borrowed words (the other is syllable final /-s/), and it is a bit artificial to exclude /l-/ even from the core of Mandi sounds. /∅/, which can be read as 'zero' is included in Table 1 to indicate that syllables can be found that have no initial consonant at all.Page  21

The clusters given in the table as /sp-/, /st-/, and /sk-/ raise some fairly complex issues that will be considered in more detailed in the next chapter. Briefly, these clusters are pronounced with distinctly more aspiration (a stronger puff of breath) than are the s clusters of English. This aspiration keeps them distinct from the syllables which I write as /sip/, /sit/, and /sik/ which have such short vowels that they can sound like clusters to the untrained foreign ear. Since the /p/, /t/, and /k/ of these syllables are unaspirated, it is reasonable to consider them as syllable finals even though the vowels are very short. As will be described more fully later in this chapter, this shortening of /i/ is one part of a more general tendency for this vowel to be shortened under various circumstances.

Clusters can be formed from most initial consonants with /r-/: /pr-/, /tr-/, /kr-/, /br-/, /dr-/, /gr-/, /chr-/, /jr-/, /sr-/, /mr-/, /spr-/, and /skr-/. Clusters of /r-/ with /w-/, or /st-/, however, are not found, and with /n-/ they are marginal. The /r-/ clusters, of course, are made with the usual Mandi flapped /r-/ so they are quite different from the /r-/ clusters of English.

Garos write three other sequences of consonants as clusters: /kn-/, /gn-/, and /sn-/. For a practical orthography, this is an entirely reasonable way to write these sound sequences, but for somewhat complex reasons that will be considered later, I prefer to consider them to be full syllalbes, so I write them as /kin-/, /gin-/, and /sin-/. For this reason, they are not included here as clusters.

p-	pal-a 'sell', pil'-a 'return', am-pok 'low stool'
t-	tang-a 'live', te'-rik 'banana', git-tam 'three'
k-	kin-a-a 'hear', kil 'cotton', ang-ko 'me'
b-	bol 'tree', bil-ak-a 'strong', ta'-bol-chu 'manioc'
d-	da'-o 'now', dam 'price', man-di 'person'
g-	gal-a 'throw away', bi-gil 'skin', dal'-gip-a 'big one'
m-	man'-a 'can', mal-a 'crawl', bi-ma 'female'
n-	nok 'house', nam-a 'good', nam-nik-a 'like'
ch-	ching-a 'we', chu 'rice beer', bit-chi 'seed'
j-	jak 'hand', jong 'younger brother', cha'-ja 'doesn't eat'
s-	sok-a 'arrive', song 'village', mi-si 'millet'
r-	ra-ma 'path, road', re'-a 'go, come', ba'-ra 'cloth'
h-	ha'-a 'ground', heng-gok 'snore' ha'-we 'yes'
w-	wak 'pig', wal 'night', ha'-we 'yes'
l-	lem 'lamp', le-ka 'paper, write'
∅	am 'mat', i'-ang-a 'go, move away'
sp-	spo-a 'blow'
st-	stip-at 'as much as', stik-a 'cover'
sk-	sko 'head', skang-o 'before'
pr-	prak 'each', prap 'kind of banyan tree'
tr-	tring-a 'weave', me'-tra 'woman'
kr-	krong 'post', nok-krom 'heir to a household'
chr-	chra 'kinsmen', chrik-a 'shout, scream'
br-	bring 'jungle', te'-brong 'jack fruit'
dr-	drak-a 'badly torn', pi'-sa-drang 'children (plural)'
gr-	grang 'feathers', grong 'horn', greng 'bone'
mr-	mrang-a 'redish', Mri 'name of a Mandi lineage'
jr-	jro-a 'chili hot', jring-jring 'always'
sr-	sruk 'quietly, secretly', sri 'slice'
spr	sprong-a 'blow off, as dust', spru 'snail' skr- skrik-a 'remove the skin, peel', skrok-a 'insert through a loop or hole'

Syllable Final Consonants

(A)

pp. 22

The syllable final consonants are shown in Table 2—2. As the table shows, fewer consonants and consonant clusters can occur at the end of a syllable than at the beginning, and many of those that do occur are quite different from any initial. The table includes the glottal stop (written as the Garos usually write it, with a raised dot) and clusters of /m/, /n/, /ng/, and /l/ with a glottal stop. As will be pointed out below, the glottal stop has such a different role from any other consonant that it can be considered to form class by itself, neither a consonant nor a vowel, but something of its own. For now, it can be included in the inventory of final consonants.

-p	-t	-k		∅
-m	-n	-ng	-l
-m'	-n'	-ng'	-l'	-'
(-s)

Syllable final /-p/, /-t/, and /-k/ are neither voiced nor aspirated. Generally they are not even "released". This means that the articulators are not opened after the consonant as they generally are in English. Thus they are quite different from any stops that occur as syllable initials, butPage  23their pronunciation is somewhat like that of the Bengali sounds that are written XX, XX, and XX.

Of all the syllable final sounds of native Mandi words, only /-m/ and /-n/ are enough like the initials to be in any likelihood of confusion with them. In these two cases only, syllable initial and syllable final consonants seem to be nearly the same, and when an /m/ or an /n/ occurs between two vowels it is less obvious than with other consonants whether they belong to the preceding or the following syllable. In addition to /-m/ and /-n/, a velar nasal, /-ng/, can be used as a syllable final, but this never occurs as word (or syllable) initial. In conformity with conventional Garo spelling, I write this velar nasal as a digraph (two letters) rather than with an n with a tail, which is the way linguists like to write it.

/-l/ is a lateral, but it is more retroflexed than English l . This means that the tongue tip is curled further back than in English and so, to an English speaker, it has a somewhat r -like quality. In native Mandi words, /r-/ occurs only as syllable initial, while /-l/ occurs only as syllable final. The two sounds are similar, and in borrowed words Garo speakers replace final /-r/ with /-l/ even more often than they replace initial /l-/ with /r-/. An old fashioned phonemicist would, of course, have considered syllable initial [r-] and syllable final [-l] to be allophones of the same phoneme. Apart from the fact that borrowed words with initial /l-/, have now become well established, however, ambiguity can be avoided for intervocalic r- 's and -l 's, even in native words, by following the conventional Garo spelling, and by using different letters in the initial and final positions. The following words illustrate the syllable final consonants when there is no glottal stop.

-p	bi-tip 'nest', nap-a 'enter
-t	mat 'squirrel', kat-a 'run'
-k	nok 'house', dak-a 'do'
-m	mil-am 'head hunting sword', dim-a 'buzz'
-n	gan 'song', man-di 'person'
-ng	song 'village', ching-a 'we'
-l	bil 'strength', mil-a 'fat'
∅	ci 'water', mi-rong 'husked rice'

The remaining Mandi finals include a glottal stop. This is a catch in the throat made by closing the vocal cords very briefly. English speakers usually put a glottal stop in the middle of "ohoh", meaning "oops". Some of the most complex problems of Mandi phonology involve the glottal stops, and these problems will be treated more fully in a later section. Now, as a first approximation, the glottal stop can be considered as one of the final consonants and as a constituent of several final clusters. In conventionalPage  24written Garo, the glottal stop is written with a raised dot, or sometimes with an apostrophe. I will follow Garo practice and use a raised dot instead of the dotless question mark that is conventional among linguists. Garos refer to this letter as the raka or 'hard' letter, a word that rolls more easily off the tongue than "glottal stop" so I will adopt the word raka here, along with the raised dot. Unfortunately, since the raised dot is not an ordinary letter of the English alphabet, many Mandis seem to feel that it is not a fully legitimate letter in Mandi either, and this leads them, quite often, to omit it from their writing even though it stands for an essential sound. The following minimal pairs illustrate the difference between words that are identical except that one is pronounced with a raka and the other without.

bi-a	'he, she'	bi'-a	'request, beg'
bu-a	'pierce'	bu'-a	'tell a lie'
ka-a	'tie'	ka'-a	'bitter'
ja-ko	'month' (accusative)	ja'-ko	'leg' (accusative)
so-a	'rot'	so'-a	'burn'

As several of these examples show, rakas often occur intervocalically, and for reasons to be given shortly, such intervocalic rakas must always be interpreted as belonging to the previous syllable rather than to the syllable that follows. Ordinarily, however, they do not occur as word final. More precisely, they do not occur as a part of a word's final syllable. Nor do they ever occur as syllable initial. The base, or "combining form", for the word meaning 'leg, foot', for example is /ja'-/, and various suffixes can be attached to this base: /ja'-pa/ 'sole of the foot', /ja'-si/ 'toe', /ja'-ko/ 'foot' (accusative case), etc. When used by itself however, some adjustment is made to the base, with the result that a word final raka is avoided. In northern dialects of Garo, the vowel of the syllable is echoed after the raka, so the subject form, which has no suffix, is pronounced /ja'-a/. Occasionally this pronunciation can be heard among Bangladeshi Mandis, usually under very strong emphasis, but the more usual Mandi way to avoid final rakas is simply to drop them entirely whenever they would otherwise occur in the final syllable of a word. The usual Mandi pronunciation of the subject form of the word for 'foot', then, is simply /ja/. In the farthest northeast corner of the Garo area, near the area whose dialect served as the model for orthographic Garo, I have heard case markers attached to forms with echo vowels. Thus the accusative of ja'-a 'foot' would be ja'-a-ko/ . Much more common in my experience, even among educated speakers, is ja'-ko where the accusative suffix is attached to the base form that lacks the echo vowel.

In addition to being used by itself at the end of a syllable, the raka can be used along with a syllable final /-m, -n, -ng/, or /-l/. In thesePage  25"clusters", the stop actually occurs nearly simultaneously with the other sound, so that the cluster that I write /-m'/, for example, starts with a nasal, has that nasal interrupted very briefly by a raka, and then continues as a nasal. The following minimal pairs illustrate the difference between finals with and without a raka.

jom-a	'sick'	jom'-a	'go stealthily, sneak'
rim-a	'take, bring'	rim'-a	'catch, hold, work'
chon-a	'small'	chon'-a	'finish'
sin-a	'like'	sin'-a	'cold'
song-a	'set up (as a post)'	song'-a	'cook'
ring-a	'drink'	ring'-a	'sing'
bil-a	'fly'	bil'-a	'roll up'

Native Mandi syllables (or words) never end in /-s/, /-ch/, /-j/, /-h/, /-w/, or /-r/, but syllable final /-s/ competes with syllable initial /l-/ for the role of the best established borrowed speech sound. All Mandis pronounce a number of borrowed words with final /-s/. Some speakers can also be heard to use a final /-r/ that contrasts with final /-l/ but this is much less well established than final /-s/. Final /-ch/ is even less common than final /-r/ but it can occasionally be heard. No distinction between p, t, k on the one hand, and b, d, g on the other is made in syllable final stops except among very fluent speakers of Bengali. (That is, no contrast is found between such pairs as -p and -b , -t , and -d , or -k and -g .) No final clusters are possible except for those with rakas. Word medially, however, any final consonant or cluster of one syllable can be followed directly by any initial consonant or cluster of the next syllable. This means that fairly complex consonant sequences are possible between two vowels, but these can always be interpreted as divided between the two syllables. Here are some examples of the various finals:

-p	ko-tip 'turban', chip-a 'close', gap-a 'full'
-m	kam 'work', nok-krom 'heir', nam-a 'good'
-m'	nom'-a 'soft', rim'-a 'hold, bring'
-t	chit-a 'rip, tear', dut 'milk', rat-a 'cut'
-n	chun 'lime', ken-a 'fear', min-a 'ripe'
-n'	ran'-a 'dry', ron'-a 'give', den'-a 'cut'
-k	mik 'eye', wak 'pig', nok 'house'
-ng	ring 'boat', rang 'gong', song 'village'
-ng'	bang'-a 'lots', rong'-te 'stone'
-l	wal 'night', bol 'tree', pul 'flower'
-l'	dal'-a 'big', chel'-a 'far' pil'-a 'return'
-'	pi'-a 'break', na'-a 'you', do'-o 'bird'
-s	dos 'ten', mas 'month'
∅	ci 'water', mi-rong 'husked rice'

Page  26

Vowels

(A)

pp. 26

Garo has only five distinctive simple vowels. These are conveniently and reasonably written as i, e, a, o, u . Only the first of these raises any real complications.

/i /. I follow conventional Garo orthography in writing this vowel as /i/. In open syllables (those without a final consonant) and in syllables that end with a simple raka (rakas that are not a part of cluster) /i/ is pronounced as a high front unrounded vowel, somewhat like the vowel of English see . In closed syllables (syllables that end with a consonant other than a simple raka), however, it is a high back unrounded vowel. That is, it is pronounced with the tongue high in the mouth but drawn distinctly to the back. The lips are not rounded. This closed syllable vowel is unlike any vowel in either English or Bengali. The closest that English gets to it is an unstressed vowel such as the vowel in the first syllable of 'contain', but the Garo vowel is higher, further back, and often more stressed than the vowel of 'contain'. The open and closed syllable pronunciations of /i/ are strikingly different, but in native Mandi words there is perfect complimentary distribution of the two phones.

The difference between the two pronunciations of /i/ means that Mandi words such as /bi-ma/ 'female' and /rim-a/ 'take, bring' have very different vowels. These words do not rhyme. The first word sounds rather like "be ma". The first syllable of the second word sounds more like the first syllable of "remark" when that word is said without stress on its first syllable, but in Garo there is good stress on the /rim-/. Similarly, /bi-ka/ 'liver' and /bik-a/ 'carve' have very different vowels. Mi 'rice' and pi'-a 'break' have front vowels, while mik-gil 'eyelid' and rim-it 'yellow' have back vowels. Bi-tip 'nest' has one of each. Differences like this make it crucial to attend to Mandi syllable boundaries.

Under certain conditions /i/ is very short. It is distinctly shorter in closed syllables than in open syllables and it is particularly short when following /s-/. When /i/ occurs between syllable initial /s-/ and a syllable final consonant it is, in effect, shortened from both sides, and it can almost be shortened out of existence. Garos conventionally write a number of words as if they have consonant clusters that I prefer to analyze as syllables with very short /i/'s. Written sni 'seven' kni 'head hair' and gni 'two' represent words that can sound, to the English ear, as if they begin with clusters. Writing sni, kni and gni poses no problem for a practical writing system, but it results in the odd limitation that Mandi would completelyPage  27lack such syllables as sin, kin , and gin before a syllable beginning with a vowel. This would be an eccentric limitation that would be quite out of harmony with the rest of the language. Greater regularity is achieved by interpreting these words as bisyllables : /sin-i/, /kin-i/ and /gin-i/, and that is how they will be written here.

In the spelling used throughout this book, then, i means a high front vowel when it is in an open syllable but a high back unrounded vowel when it is in a closed syllable. In a very few exceptional circumstances the high back unrounded pronunciation appears in what seems to be like an open syllable. In such cases I will use "ɨ" for that high back vowel. In other exceptional circumstances, and also in less exceptional borrowed words, high front vowels appear in closed syllables. For reasons that will be explained later, I will use "ii" for these high front unrounded vowels where they occur in what seem to be closed syllables, as in tiin , the Bengali word for 'three', which sounds quite like English 'teen'.

The four remaining Mandi vowels vary less from one phonological environment to another than does /i/, so they pose fewer problems.

/e /. Mid front and unrounded. This vowel is just a bit higher than the e of English bet, neck , a bit closer to paid or bait , but it is not diphthongized. It is slightly higher in open syllables than in closed. se 'husband', pek-a 'intoxicated', be'-en 'meat'.

/a /. Low central and unrounded. Similar to, or just a bit further back than the English a of father or calm : chang-sa 'once', cha'-a 'eat'

/o /. Mid back and rounded. Similar to the o of caught or paw in those dialects of American English that pronounce these words differently from cot and pa . In open syllables o may be pronounced a bit higher than in closed syllables, so that it sounds almost like the vowel of English go , or even more like the vowel of French mot 'word': no 'younger sister', nok 'house', cho'-a 'dig'.

/u /. High back and rounded. Similar to the vowel of English food, rude but not quite so far back. Since /i/ in closed syllables is pronounced quite far back, and since /u/ is sometimes further front than the "back" vowels of many languages, the contrast between /i/ and /u/ in closed syllables is that of unrounded versus rounded rather than front versus back. In open syllables, however, /i/ is also decisively further front than /u/: chu 'rice beer', pul 'flower', su'-a 'stab, pierce'.

All vowels are shorter in closed than in open syllables, though the differences for most are less dramatic than for /i/. Nevertheless, differences in the vowels of open and closed syllables, particularly differences in their length, are an important clue that allows Mandis to easily distinguish words that differ only in the position of the syllable break. In the following pairsPage  28of words, those in the first column have closed first syllables and relatively short vowels. The vowels in the second column occur in open syllables and they are pronounced slightly longer. Words with an intervocalic /p/, /t/, or /k/ are distinguished by the presence or absence of aspiration as well as by vowel length. Aspiration occurs only with syllable initial stops, not with syllable finals. The aspiration makes the location of the syllable boundary quite obvious. Locating the position of the syllable break in words with intervocalic nasals depends primarily upon the quality of the vowel, obvious in the case of /i/ but more subtle for the other vowels.

kat-a	'run'	ka-ta	'word'
ram-a	'dry' (transitive)	ra-ma	'road'
chon-a	'small' (neutral)	cho-na	'thresh' (infinitive)
bik-a	'carve'	bi-ka	'liver'

The Mandi dialects of Bangladesh differ from A'chik in the pronunciation of certain vowels. Most syllables that have /e/ or /e'/ but no other final consonant in A'chik are pronounced with /i/ or /i'/ by most Mandis. Thus A'chik /be'-a/ 'break' is usually pronounced /bi'-a/ by Mandis. A'chik /se/ 'husband' becomes /si/ for Mandis. A'chik /be-be/ 'true' becomes Mandi /bi-bi/. Where A'chik has /e/ in closed syllables (syllables ending in a consonant), however, Mandi also has /e/. Both A'chik and Mandi have /pek-a/ 'intoxicated', /ken-a/ 'be afraid' and /beng'-a/ 'forbid', and so on. /Sik-a/ 'sew' and /sek-a/ 'elope, seduce' are as different in Mandi as in A'chik. /se/ 'husband' and /si-/ 'die' are different in A'chik, but the same in Mandi.

A parallel change has taken place in Modhupur but does not seem to be established in the Mandi areas that border the Garo Hills. In Modhupur syllable final /o/ has generally become /u/. A'chik /go-a/ 'thrown' becomes /gu-a/ in Modhupur. A'chik /so'-a/ 'burn' becomes Mandi /su'-a/. Once again, in closed syllables there is no difference between A'chik and Mandi: /nok/ 'house', /grong/ 'horn', /ong'-a/ 'to be' are pronounced the same in both dialects. In both cases, Mandi mid vowels are raised to high vowels in open syllables and in syllables ending with a simple raka, but they are not raised in Mandi closed syllables.

Oddly there are exceptions to the raising of the vowels, notably in suffixes. Several very common suffixes, including /-o/ 'locative', /-ko/ 'accusative', and /-bo/ 'imperative' have syllable final /o/ in Mandi as well as in A'chik. The raising of /o/ is less complete than the raising of /e/, however and the subordinating /-e/ of A'chik is generally pronounced /-i/ in Bangladesh. Nevertheless, there are suggestions of some recognition of the distinctiveness of both pairs of vowels even if the distinction is often lost. It would seem that neither /e/ and /i/ nor /o/ and /u/ in open syllables havePage  29totally merged, even in Modhupur. Certainly words exist, even for Modhupur speakers, that continue to be spoken with syllable final /e/ as well as with syllable final /o/. Considerable individual and regional variability complicates the picture, and I do not understand all the variables.

In the face of this variability it seems safest in the book to write the vowels as they are pronounced in A'chik, that is with /-e/ or /-o/ depending on the A'chik pronunciation. I do this with some qualms, for it suggests a distinction that is not always made in the dialect to which this grammar most specifically refers. However, it will be relatively easy to learn to read syllable final /e/ and /o/ as if they are pronounced like /i/ and /u/, whereas if I were to obscure the difference between the vowels by always writing /i/ and /u/ there would be no way to recover the pronunciation used by speakers who make the difference. In this respect, then, I will write words in a more "standardized" form than they usually have when spoken in Bangladesh. This has one unfortunate consequence. It is all too likely that I have learned some words in Modhupur that are pronounced with either /i/ or /u/, but never had a chance to hear the word used in other dialects. In such cases, I have no way of knowing whether A'chik would have /e/ instead of /i/ or /o/ instead of /u/, and in my ignorance I have no choice but to write them as I have heard them, with /i/ or /u/.

Diphthongs

(B)

pp. 29

Since neither syllable initial nor syllable final consonants are required in Garo, it is inevitable that vowels will sometimes follow each other directly with no intervening consonants. In some cases, it can be difficult to decide whether vowel sequences should be considered as consisting of two simple vowels, each belonging to a different syllable, or as diphthongs that constitute just one syllable.

In some cases the best choice is a sequence of two simple vowels. This is particularly clear when the vowels belong to different morphemes. bi-a 'he' (nominative) clearly consists of two morphemes, /bi-/ 'he' and /-a/ 'nominative', as shown by such words as ang-a 'I' (nominative), bi-ni 'his' (genitive), bi-ko 'him' (accusative) etc. In a case like this, there is no temptation to consider the vowels to constitute a single complex diphthong. Nor does the phonetics of /bi-a/ encourage the idea that /i/ and /-a/ form a single unit. The two vowels remain distinct and do not merge. In other cases, however, complex vowels can quite naturally be interpreted as diphthongs. I describe these in approximately the descending order of their frequency and of their diphthong-like nature.Page  30

/ai/ starts as a low central vowel and moves to high front. It is unrounded throughout, and is very much like the vowel of northern American English my . Mandis use it in many words, though a high proportion of these are either interjections ( ai-au 'wow!'; hai 'let go!') or borrowed ( jai-ga 'place', hai-brens 'bench with a high back', sai-kel 'bicycle'). This suggests a somewhat marginal status or a recent development. However, even A'chik, with much less influence from Bengali, has well established words with /ai/, such as mai- 'what?' and hai-da 'don't know'. Like such Mandi words as kai-sim 'salt' or bai-sik 'how much?', these do not seem to be borrowed. All dialects, then, can be reasonably interpreted as having a diphthong /ai/.

/au/ , like /ai/, starts as a low central vowel but it then proceeds to a high back rounded position. It is very much like the vowel of northern American English now . Like /ai/, it occurs in a number of borrowed words such as lau 'gourd' but it is also found in words that show no sign of having been borrowed: jil'-au-a 'flair up'; mik-au-a 'wake up, open eyes', ai-au 'wow!'

The status of /ai/ and /au/ as diphthongs is confirmed by a few words where they are used with rakas. Simple vowels are often followed by a raka, and a sequence of two vowels in different syllables is often interrupted by a raka: pi'-a 'break', be'-en 'meet'. In chau'-a 'rob, steal', gau'-a 'break open' and hai'-a 'know', the raka occurs with a diphthong. These are unlikely to be borrowed words, since borrowings rarely, if ever, have rakas. When a speaker is asked to pronounce these words slowly and carefully, the raka seems almost to interrupt the vowel as [cha'-u-a], [ha'-i-a], but in ordinary fluent speech [chau'-], [gau'-], and [hai'-] are single syllables, quite different from a two syllable sequence such as [go'-a] 'throw' or [cha'-e] 'having eaten'. (The square brackets indicate that it is the phonetic details that are under consideration, while slant lines enclose the significant sounds or "phonemes".) Thus /ai/ and /au/ are like simple vowels in being able to occur either with or without a raka, but they are more common in syllables that have no rakas. They are rare in syllables that are closed by consonants other than a raka, although a few such syllables occur in borrowed words: dauk 'a kind of bird'; bait 'low wet area suitable for rice'.

/oi/ is very much like the English diphthong in boy . This is a common vowel in Bengali, while in Mandi and other Garo dialects it is probably found only in borrowed words. Some of these, such as boi 'book' and oi-kor 'letter' are used easily by Mandis, but are rarely or never used by A'chik speakers. I cannot imagine a Garo speaker of any dialect who does not use so-moi 'time', however.

/ai/, /au/, and /oi/ are the best established diphthongs in Garo. TwoPage  31others occur with high enough frequency, at least in the Mandi dialect, to be considered as candidates for diphthongs.

/eo/ starts as a mid front unrounded vowel and then moves back to a mid back rounded position. It occurs in a handful of Mandi words, none of them common: seo'-seo-i 'speaking with a hoarse voice', kreo-kreo 'with a cracking sound', bit-e-deo 'a kind of ground-living bird', mreo-mreo 'faintly, dimly', breo-a 'an unpleasant taste', gil'-eo-gil'-eo 'glowing, of coals'. Surely there are other such words but it is unclear whether there are enough to justify considering /eo/ to be a diphthong rather than a sequence of two vowels /e-o/. It would be interesting to compare these words with others whose bases end in /e/ but are followed by a locative /-o/ suffix, so as to determine whether any phonetic differences distinguish them.

/ui/ starts as a high back rounded vowel but moves to the front and becomes unrounded. It is not unlike the vowel of phooey or gooey . It is found in kui-cha 'eel', in borrowed words such as sui-tar 'sweater' and kui-nain 'quinine', and in A'chik ui-a 'know' and gui 'areca nut'. Its status as a diphthong is marginal and it could be regarded as simply a sequence of two simple vowels.

Beyond the diphthongs and marginal diphthongs already listed, any two vowels can succeed each other in adjacent syllables. It might be possible to find examples of all twenty-five possible vowel sequences, including repetitions of the same vowel, but some of these would surely be too rare to invite an interpretation as diphthongs. Borrowed examples include: i-u-ni-on 'union' the local administrative district in Bangladesh, ko-e-ri 'brown', me-i-e 'young girl'. Sequences of two morphemes can result in a wide variety of vowel sequences: u-a 'that', i-a 'this', chi-o 'in the water', etc. Two instances of the same vowel can occur side by side. Sometimes these are reduced to a single vowel, but the result is often a long vowel: sa-a 'sick, in pain'. The assimilation of such vowels is described in Chapter 4 in the section on "Morphophonemics".

Vowel sequences present one problem that I have not solved. A tiny number of words may, or may not, have a syllable with an intial w- . For example, Garo has a word meaning 'curved, as a crescent moon or a crooked smile'. I cannot find firm grounds for deciding whether to interpret this word as /ku'-eng-a/ or as /ku'-weng-a/. In the parallel case of ku'-wang-a 'with a wide open mouth' I interpret the word as having a w because there is a related word, na-wang 'a monster with a very large mouth who swallows the moon in an eclipse'. I know of no use of eng/weng that would force the choice of one or the other. It may be that there is simply no phonetic difference between the two.Page  32

The Glottal Stop or Raka

(C)

pp. 32

Unlike most Tibeto-Burman languages, Garo does not have contrastive tones. Nevertheless, many aspects of Garo phonology resemble those of the tone languages of east and southeastern Asia, and it is abundantly clear that the Garo contrast between a raka and its absence is cognate to the two way tone contrast of several closely related languages such as Tiwa, Boro, and Rabha (Joseph and Burling, 2001). Even without real tones, the raka, together with the structure of Garo syllables, gives its phonology a somewhat tone-like character. Among other things, the raka is subject to more complex morphophonemic rules than any other phonological unit. Vowels, and all the consonants except the raka, are very stable, but rakas come and go according to quite complex rules. Phonetically, of course, rakas are stops, but they effect, and are effected by, their surroundings in quite different ways than other Garo stops:

1. All other syllable final consonants cause a preceding /i/ to assume its backed allophone, but in syllables closed with a raka /i/ has the fronted pronunciation that is otherwise found only in open syllables. Thus, the raka does not close the syllable in quite the same way as (other) final consonants.

2. The raka is more transparent to vowel assimilation than are other consonants. An /i/ that follows immediately after another vowel sometimes assimilates to the preceding vowel but assimilation never occurs if some consonant other than a raka intervenes between the vowels. Assimilation can take place if the vowels are separated by nothing more than a raka.

3. Open syllable /e/'s and /o/'s that are found in A'chik have been raised in some Mandi dialects, and they have fallen together, or almost fallen together, with /i/ and /u/. In closed syllables, all Mandi dialects that I am aware of retain /e/ and /o/ as distinct from /i/ and /u/. In syllables closed by a raka, however, /e/ and /o/ have been raised in Mandi in the same way that they have been raised in open syllables. Once again, syllables seem less securely closed by a raka than by other consonants.

In these three ways, the raka acts quite differently from other syllable final consonants, and in all cases the syllables in which it occurs act more like open syllables than do syllables closed with other consonants. It should be noted that syllables ending with /-m'/, /-n'/, /-ng'/, or /-l'/ act like fully closed syllables.

The most striking difference between the raka and other consonants, however, is that rakas disappear under some circumstances. The following generalizations seem to account for whether or not a raka will be realized in a particular syllable.

1. Underlying Rakas. Each syllable can be characterized as either having or not having an underlying raka, just as each syllable is characterPage  33ized by particular consonants and a particular vowel. Syllables that do not have an underlying raka (almost) never acquire them, but syllables with underlying rakas often lose them. With a few minor exceptions, other vowels and consonants are very stable, but the raka is highly variable. The raka appears most reliably when it is part of the first syllable of a word, but where at least one other syllable follows. Numerous minimal pairs are unambiguously distinguished by the presence or absence of a raka in the first syllable. The raka (with some complications to be mentioned below) occurs toward the latter part of the syllable and it can occur both in syllables with no (other) final consonant, or in syllables where it co-occurs with final /-l/, /-m/, /-n/, or /-ng/. In the latter case it overlaps or interrupts the co-occurring final consonant, but in conformity with conventional Garo spelling, I write it after the other consonant. As will become apparent, however, it can be helpful to think of the raka as a feature of the syllable rather than as a member of a final cluster.

2. Final Syllable Deletion. As mentioned briefly earlier, rakas are not expressed in the last syllable of a word, but they are avoided in different ways in Mandi and in A'chik. In Mandi, a raka is rarely expressed in a single syllable word even when the syllable must be interpreted as having an underlying raka. For example, a word such as wal 'fire' must be considered to have an underlying raka, because a raka consistently appears in inflected forms such as wal'-ni 'of the fire', wal'-ko 'fire, accusative case', etc., as well as in compounds that have wal'- as their first constituent, such as wal'-ku-a 'smoke' and wal'-mi-si 'spark'. Many one syllable Mandi noun bases act like wal'- . They have a raka in their combining form but it disappears when the syllable is used alone.

The rule for deleting rakas from final syllables of words requires care about which sequences are to be considered as separate words and which are to be considered compounds. Indeed the rule can be used to help to define word boundaries. han'-chi 'blood' (lit. 'body-water', 'body liquid') must be regarded as a compound since a raka appears in han'- . If han were a separate word it would have no raka since it would be the final syllable of the word. In han jak 'whole body', (literally 'body hand',) no raka occurs with han , and this requires us to interpret the han of this phrase, as a separate word.

Many one syllable verb bases occur in Mandi, but since verb bases almost always have a suffix, those that have rakas hang on to them more consistently than do noun bases, which are regularly used with no suffix at all. Verb bases are usually protected from losing their rakas by at least one suffix. One form of the negative imperative, however, is formed by a verb prefix rather than a verb suffix. (This is the only fully productive prefix inPage  34the language.) In this case the verb base need not be followed by any suffix at all, and then any underlying raka is lost from a Mandi verb base, just as it is more often lost from a noun base:

cha'-a

'eat'

da'-cha

'don't eat'

This example is not conclusive, however, since the rule of second syllable deletion (see below) would remove the raka even if it were not in a final syllable. One other rather eccentric construction shows that rakas are dropped from word final verb bases, just as they are dropped from word final noun bases. Mandi has a reduplicated construction meaning "whether or not". This is by no means a common construction, but it requires the verb base to be used twice, both times without any affixes at all. da'- is prefixed to the second occurrence, but the first is found without an affix, the only situation I have encountered where a verb base is used with no affix whatsoever. In this case, any raka usually found in the verb base is lost:

When used with a suffix, both cha'- 'eat' and man'- 'get' have rakas, but since in this construction the raka is lost from the first use of the verb base as well as from the second, its first use must be regarded as a separate word. Verb bases which ordinarily contrast in their rakas, such as ring-a 'drink' and ring'-a 'sing', lose their contrast in this construction: ring da'-ring could mean either 'whether or not (he) drinks' or 'whether or not (he) sings'.

The fate of final syllables with underlying rakas is quite different in northern dialects of Garo. In A'chik, the raka is not lost from final syllables but, instead, it is separated from the end of the word by an echo vowel: ja'-a 'foot, leg', do'-o 'bird'. When a raka co-occurs with /-m/, /-n/, /-ng/ or /-l/, the raka is separated from the other consonant by an echo vowel, the raka coming before the echo and the other consonant after. In both cases the raka ends up between two identical vowels. Thus in A'chik, when underlying /wal'-/ 'fire' occurs without a suffix, it is pronounced /wa'-al/, rather than /wal/, which is its form in Mandi. Forms with echo vowels are heard occasionally from Mandis, usually in situations of heavy emphasis, but Mandis more often drop the raka completely when it would otherwisePage  35threaten to occur in the final syllable of a word or as part of a final cluster. Other examples of words where the two dialects have different free forms are shown in Table 2—3.

	Combining	A'chik	Mandi
	free form	free form	free form
tuber	ta'-	ta'-a	ta
stick	gol'-	go'-ol	gol
bug	jong'-	jo'-ong	jong
body	han'-	ha'-an	han

3. Second syllable deletion. In both Mandi and A'chik rakas are uniformly deleted from the second syllable of a word. They are retained in third syllables. In two syllable words this rule simply duplicates the effect of Rule 2 in Mandi, though not in A'chik where an echo vowel is introduced by rule 2. In longer words, the effects of the rule become clear in both dialects. The most salient situation where these raka deletions can be noticed is in verbs with a derivational syllable that follows directly after a single syllable verb base. Many of these derivational syllables have rakas in other circumstances, generally when they are used as verb bases, but the rakas disappear when the syllable is in second position. Pil'- 'go back, return', for example, has a raka when it is used as either a verb base (and thus as the first syllable of the word), or when occurring as the third syllable of a verb. It loses its raka when it occurs in second position. Similarly, dal'- 'big' has a raka when used as a verb base but loses it in a number of compounds where it comes second. The operation of this rule can be seen in the following examples. -ba'- 'in this direction' does not occur as an independent word but since it has a raka when in third syllable position it must be regarded as having an underlying raka that is lost in second position

pil'-a	'return'
kat-pil-a	'run back'
kat-ba-a	'run in this direction'
kat-ba-pil'-a	'run back in this direction'
kat-pil-ba'-jok	'has run back in this direction'
tam'-bing-a	'round
mik-tam-bing-a	'round faced'
dal'-a	'big'
rong-dal-a	'big, of round things'
ning'-a-o	'inside'
ha'-ning-a-o	'in the ground'
ro'-a	'long'
mang-ro-a	'long of an animal'
tong'-sa	'one', of cylindrical objects
rat-tong-a	'cut', of cylindrical objects
kal'-a	'play'
mang-kal-a	'fool around, spook'
tom'-a	'gather'
ka-tom-a	'tie in a bundle, pack' (ka-a'tie')

Page  36

As soon as it is recognized that rakas occur easily in first and third syllables, but not in second syllables, it is natural to wonder about fourth syllables. It is not easy to maneuver syllables with underlying rakas into the fourth syllable of a word, but the following examples, formed either with the two syllable verb base gu-ri 'wander around' or with the two syllable affix, -ru-ra'- 'back and forth', suggest that the raka is deleted on the fourth syllable just as it is on the second.

gu-ri-ba'-pil-jok	'wander back in this direction'
gu-ri-pil'-ba-jok	'wander back in this direction'
i'-ru-ra'-pil-a-ring-a	'returning in a back and forth way'

It is more natural for -ba'- to precede -pil'- than to follow, so gu-ri-pil'-ba-jok is not a fully convincing example. Gu-ri-ba'-pil-jok is an entirely natural construction. The raka does seem to be lost from -pil'- in this word, and also in i'-ru-ra'-pil-a-ring-a . However, I do not find it easy to be confidant of my hearing in such long words, and I have been unable to devise longer words that would test later syllables. Verbs of six or seven syllables are easy enough to construct, but none of the syllables that easily come late in these words have underlying rakas. Note that the final syllable deletion rule and the second syllable deletion rule work together to make it (almost) impossible for rakas to occur in adjacent syllables (but see Rule 4, below, for intrusive rakas).

The second syllable raka deletion rule accounts for the way rakas are used in words formed by reduplication. Reduplication is a modestly productive process in Mandi and many words are formed by reduplication. Such words may have a raka on the first syllable, but never on the second:

jem'-jem-a	'constantly'
jing'-jing-a	'sift'
mol'-mol-a	'request, beseech'
dil'-dil-a 'shake with cold or fear'

Page  37When two syllable words are reduplicated, as happens very frequently with adverbs, any raka on the first syllable is faithfully copied into the third syllable as well.

je'-et-je'-et	'very heavy'
so'-om-so'-om	'very soft, of flour, mattresses, blankets'
sol'-i-sol'-i	'by small chops, as when carving'

Reduplication apparently leads to one partial exception to the preservation of the raka on the third syllable of a word. Reduplicated syllables are occasionally suffixed to a verb base. If the underlying single syllable has a raka it would not appear in the second syllable of the reduplicated form and once suffixed to a single syllable verb base the raka also disappears from the first syllable of the reduplication. In such cases the raka does not reappear on the second syllable of the reduplication even after it has moved into the third syllable of the word. This can easily be accounted for if we assume that suffixation takes place after reduplication. By the time suffixation occurs it seems to be too late to recover the underlying raka. I have not, however, uncovered many examples where a raka of this sort might be looked for in the third syllable, and too much faith should not be placed in this generalization.

jem'-jem-a	'repeatedly'
den'-jem-jem-a	'cut repeatedly'

A particularly interesting example is provided by han dal'-a 'big of body'. I believe I have heard this pronounced, in Modhupur, both as han'-dal-a and as han dal'-a . If this is correct, han'-dal-a can be regarded as a compound, where the first syllable retains its raka, but the second syllable undergoes the second syllable raka deletion rule. Han dal'-a would have to be regarded as two separate words, with the raka lost from the first word because it occurs in the final (only) syllable. (A'chik, of course, would probably have a'-an dal'-a .) Of the two Mandi pronunciations, han dal'-a is easier to elicit clearly. It is the pronunciation used whenever the linguist asks a speaker to repeat carefully. han'-dal-a seems to occur with rapid articulation, but its very rapidity makes it is more difficult to hear or to interpret with confidence. If my hearing is accurate, it suggests, reasonably enough, that with rapid speech, the two words merge into a compound. In no case is this ever heard (in Modhupur at least) with two rakas. I find is surprisingly difficult to locate the raka in cases like this. The raka is very clearly audible. It is entirely clear that there is one and only one raka, but the raka seems to float in a space that is separated from the other segments, almost as if it is a suprasegmental. Thus I find it remarkably difficult even to be confidant that han dal'-a and han'-dal-a are different. The point isPage  38that the phrase (or word) as a whole has exactly one raka, but where in the sequence of segments the raka falls is of secondary importance. Some instrumental phonetics ought to be able to sort this out but it makes no difference to native speakers, and not much difference even to this speaker of Mandi as a second language.

4. In only one situation can a raka be introduced where none exists in the underlying syllable. When a two syllable verb stem (generally consisting of a one syllable verb base and a one syllable derivational suffix) ends with /a/, or less often when it ends with another vowel, a raka is often introduced before the present-neutral suffix -a . In effect, the raka separates the two /a/'s. This rule can override Rule 3, with the result that, in this one circumstance only, a raka can appear in the second syllable of some words, even when that syllable directly follows a raka in the first syllable: i'-ba'-a 'come', i'-ba-pa'-a 'come with', but i'-ba-bo 'come' (imperative), i'-ba-pa'-bo 'come with!' (imperative). There is, however, one important and clear exception even to this exception. The negative -ja- does not acquire a raka when occurring before -a . In fact the two /a/'s of -ja- and -a do not keep their distance at all, but merge. Verb suffixes that must be considered as formed from underlying -ja- plus -a are pronounced simply as /-ja/: i'-ang-a 'go', i'-ang-ja 'not go'. (See the section on "Vowel Loss" in Chapter 4.)

It is tempting to view the collapse of -ja- and -a into -ja as if there is some difficulty about pronouncing two a 's in a row. Two a 's seem to be in danger of merging into a single vowel unless they are kept apart by a raka. The vowel of -ja- merges, while a raka is inserted in other cases. This cannot be a satisfactory "explanation" however, because there is no problem at all about letting the -a suffix follow directly after a one syllable verb base that ends in -a : sa-a 'serve food', ga-a 'climb'. In such cases, no raka is introduced, but under even modestly careful articulation the vowel is pronounced long enough to count as two syllables. We are left with a peculiar and rather eccentric rule that allows a raka to be introduced in second syllables in the single situation where there is following -a . In no other situation are rakas added.

V-Raka-V Sequences

(C)

pp. 38

Mandi and other dialects of Garo have more words than would be expected by chance in which the raka occurs between two identical vowels in successive syllables. Relatively few words have the form CV'-VC where the two vowels are different, but many are found with identical vowels. Consider, first, those in which the second syllable is closed with an /l/ or a nasal:Page  39

be'-en	'meat'
ke'-em-a	'excrement, dung'
be'-el-a	'spit out food'
so'-om-so'-om	'comfortable, soft'
sa'-al-a	'swell up, as when rice cooks'

These look very much like the result of expansion from an underlying single syllable by the addition an echo vowel. Such expansions are a regular feature of A'chik, but not of Mandi. For example, A'chik jo'-ong 'insect, worm' is expanded from underlying jong'- , which is the combining form used with case markers and in a large number of compounds. The impression that the forms in the list have been expanded from something shorter is strengthened by the existence of a few examples in which expanded and unexpanded forms alternate, even in Mandi:

bol'-a, bo'-ol-a	'rotten, of eggs'
jol'-a, jo'-ol-a	'rise, of water levels'
mo'-ong-a, mong'-a	'ask someone to go along'
se'-em-a, sem'-sem-a	'tiny'

So we have a few examples of alternations in Mandi between CVN' and CV'-VN, where N stands for any nasal or /l/. In A'chik, this is a fully productive alternation. Not all words with a V'-V form have unexpanded partners, however, so in the present language they cannot all be derived from unexpanded forms. Be'-en 'meat' is a very common word in both Mandi and A'chik but there is no related /*ben'-/, even when followed by a suffix. (The asterisk means that the form that follows does not exist.) Nor, so far as I am aware does /*som'-/ ever appear as an alternate pronunciation of so'-om- 'soft.' Words like be'-en and so'-om-so'-om have the look of expansions that have become lexicalized, frozen in their expanded form, though we can only guess at their actual historical derivation. The only way to describe the synchronic language is to list the various forms that each word assumes, and admit a lack of consistency.

More interesting are examples in which the second syllable ends in a stop.

ga'-ak-a	'fall down'
go'-ok-a	'come apart, come loose, fall out, of a tool handle or hair'
so'-ot-a	'kill'
de'-ep-a	'held by pressure, as with tongs'
se'-et-a	'pinch between two fingers'
ru'-ut-a	'stretch, as something elastic'
je'-et-je'-et	'very heavy'

Page  40

On the analogy with the nasals, it is tempting to look for an alternation between CVS' and CV'-VS, where S stands for /p/, /t/, or /k/. However, S' (a stop with a raka) is never found in Mandi or even in A'chik. Or at least there is no contrast between stopped syllable with and without a raka. This means that in the present language, words like ga'-ak-a 'fall down' cannot be considered as expansions from underlying single syllables, simply because clusters consisting of a raka plus a stop (which would have to be the underlying and unexpanded source for these words) do not exist, either in Mandi or in any other Garo dialect that I am aware of. However, the raka can appear in syllables along with any of the other possible syllable finals of native Mandi words. That is, a raka can co-occur with final /m/, /n/, /ng/, or /l/, or as part of an otherwise open syllable, but a raka cannot co-occur with /p/, /t/, or /k/. Nor can it co-occur with final /s/ or with any other final consonants that are found only in recently borrowed words. Still, we do find a fair number of words that have the form CV'-VS, where S stands for /-p/, /-t/ or /-k/. In these words, two identical vowels are separated by a raka. If such words could be derived from underlying syllables of the form CVS', this would imply a more symmetrical phonological pattern. It would mean that a raka could co-occur with any other syllable feature of the language (ignoring borrowings). In effect, any sort of syllable at all could either have a raka or not have a raka. This would suggest an even more tone-like quality for the raka.

The temptation to treat these words in this way is strengthened by an alternation that puzzled me for a long time.

so'-ot-a	'kill'
ra-sot-a	'kill with a knife, as a chicken'( ra- 'cut, slice')
rim'-sot-a	'kill by choking' ( rim'- 'grab, hold')

The syllable -sot- is too much like so'-ot- to be a simple coincidence, but alternations of this sort are not common in Mandi. Nevertheless, if both so'-ot- and -sot- are regarded as deriving from an underlying (but never realized) *-sot'- , an entirely regular pattern presents itself. In ra-sot-a and rim'-sot-a the raka disappears, as it would be expected to disappear in a second syllable. In so'-ot-a , on the other hand, an echo vowel is introduced which allows the raka to be preserved in a language that does not (any longer?) allow /-t'/ to be realized in a way that would contrast with simple /-t/.

je'-et-je'-et	'heavily'
on-jet-a	'place something heavy on something else'
sik-jet-a	'press down hard, squash'
go'-ok-a	'come apart, come loose'
gil-gok-a	'peel off, as bark' ( bi-gil 'skin, bark')
rong-gok-a	'break off, of a round thing'
dok-gok-a	'hit and separate' ( dok-a 'hit')
de'-ep-a	'squeeze, pinch'
rim'-dep-a	'depress with a finger'
ku'-dep-a	'having a flat mouth' ( ku'-sik 'mouth')
gek-gek-a, ge'-ek-ge'-ek-a	'groaning with a high fever'

Page  41

Words, such as ga'-ak-a 'fall' and je'-et-je'-et 'tight, forcefully' that have the form CV'-VS can have a distinctive glottalization that carries through both vowels, giving them a creaky quality that is otherwise unusual in the language. Creakiness is sometimes associated with glottal stops in Southeast Asian languages and it sometimes occurs with particular tones. It is not a widespread or salient feature of Mandi phonology, but its appearance in CV'-VS sequences both sets them apart as distinctive, and gives a unity to the syllable pair. This phonetic feature makes these seem more unified, more monosyllabic, than they otherwise would.

It is therefore very tempting to reinterpret the sequences that I have written as CV'-VS as deriving from underlying syllables of the form CVS'. It is true, however, that such syllables are very much rarer than CVN' syllables. Mandi had scores of words of the CVN' type in which the raka co-occurs with a nasal or /-l/. Far fewer syllables could be interpreted as having an underlying CVS'. Nevertheless, even these few call out for recognition.

There is, admittedly, one other possible source for apparent echo vowels. As will be described later, Garo has two very common suffixes that lack any initial consonant and that begin with an underlying /i/. The vowels of these two suffixes assimilate to the vowel of the previous syllable, but this happens only when the previous syllable is open or ends in a raka. This results in words in which a raka is surrounded by phonetically similar or identical vowels. It is possible to speculate, therefore, that a word such as be'-en 'meat' could have been derived from an earlier *be'-in , by assimilation of the /i/ to the previous vowel. There is no real support for this in the contemporary language. Nor is there any historical or comparative evidence that would either confirm or rule out such a speculation.

The raka has now been described in sufficient detail to make its special role in the language clear. It does not act like any other consonant. It can co-occur with the majority of other final consonants, and a case can be made for its underlying co-occurrence with all of the final consonants except those that have been recently borrowed. The raka is not limited inPage  42co-occurrence with any vowel or any syllable initial consonant. Syllables terminating in a raka act in several ways like open syllables. In all these respects the raka acts more like a feature of the syllable and less like a linearly ordered segment than do the vowels or the other consonants of the language. The raka also undergoes far more deletion than any other consonant or vowel. Summarizing these features, it seems fair to say that the raka acts in ways rather like the tones of other Southeast Asian languages. Glottalization works closely with tones in many of these languages. Glottal stops, and sometimes creaky voice, are characteristic of certain tones in some languages. The phonetics of the Garo raka do not permit it to be considered a real tone, but the language does present us with something that looks like a rudimentary tone system. Rabha, Boro, and Tiwa, languages that are closely related to Garo, all have two contrasting tones. The high tone of these languages is certainly cognate to the Garo raka (Joseph and Burling, 2001). I incline to the view that the Garo raka represents the last remnant of what was once an unambiguous tone system. Duanmu (1994) has presented an interesting argument that Garo represents, instead, a stage of incipient tonality, a stage that its sister languages have passed beyond. I leave this historical question open.

Conventional Garo Spelling

(B)

pp. 42

The Garo romanization that has been in use for about a century, reflects the sounds of the language very closely, and in this book I have chosen to follow the now well established spelling wherever possible, even where this deviates from established linguistic usage. Thus I write ch, ng and ' where linguists might prefer wedge-c, eng and a glottal stop symbol. I also write m', n', ng' and l' , although ' m, ' n, ' ng and ' l would be more satisfactory for describing the expansions with echo vowels. In spite of my acceptance of these conventions, my transcription still differs in two minor and one not so minor ways from the conventional orthography.

1. Syllable Boundaries . Conventional spelling does not mark syllable boundaries within words. This is defensible from a practical point of view, but it does result in a number of homographs for words that are in clear phonological contrast such as rama for both /ra-ma/ 'road' and /ram-a/ 'dry', and bika for both /bik-a/ 'carve' and /bi-ka/ 'liver'. I would not suggest that anyone use clumsy hyphens to mark all the syllable boundaries in a practical orthography, but a technical description needs to show them. Showing syllable boundaries has the additional virtue that, more often than not, it marks the boundaries of morphemes.

2. /i/ . In some situations I write i where conventional spelling uses no vowel letter at all. These are cases where the i is very short, so thePage  43temptation to omit it is understandable. Nevertheless, some aspects of pronunciation and grammar can be expressed more consistently if i 's are used to break up what otherwise look like clusters. Writing the i 's allows some morphemes to retain a more consistent spelling under varying conditions, and syllables such as /sin/, /sil/, and /gin/ become possible that would otherwise never occur directly before a vowel. For example, in conventional orthographic Garo, n is written as if it is the second member of a number of clusters: gn-, kn-, sn- . Even sl- is occasionally written as a cluster as in slai 'gun', although this places l prevocalically, something otherwise impossible in older Garo words. For present purposes, I prefer to write an i between the consonants, writing sil-ai 'gun' instead of slai , gin-i 'two' instead of gni , kin-i 'hair' instead of kni , and so on. These i 's are admittedly very short.

3. Mandi Specialties . Finally, most of my examples are drawn from the Mandi dialects of Bangladesh rather than from A'chik. In these examples, I generally spell words as they are pronounced in Bangladesh, even when this differs from the usual conventions of orthographic Garo. As explained earlier, I make one exception to this practice by writing e in many words that are generally pronounced as [i] in Bangladesh, and by writing o even when these become [u] in Modhupur. Spelling the words this way keeps them closer to conventional orthography and should not be too disturbing to readers most familiar with Bangladeshi Mandi. One must simply get in the habit of pronouncing most words written with syllable final o 's and e 's as if they are /u/'s and /i/'s.

I hope that these deviations from conventional spelling are sufficiently trivial that they will cause few problems even when readers shift back and forth between the examples in this book and materials that are written in conventional A'chik spelling.

Difficult Aspects of Pronunciation

(A)

pp. 43

The most difficult sounds for a learner to pronounce in A'chik or Mandi depend, of course, upon the language or languages already spoken by the learner. Most learners, however, should find many of the sounds quite easy. The following are those that often give English and Bengali speakers the most difficulty.

The raka . The speech sound that usually gives the adult learners the most trouble is the glottal stop or "raka". A raka is made by closing off the glottis (the space between the vocal cords) and then very quickly reopening it. Speakers of all languages close and then reopen the glottis with great frequency, but learning to control the glottis voluntarily, as required forPage  44speaking a language like Garo, can be a considerable challenge for people whose native language does not distinguish between words that have a glottal stop from those that do not. Listen to speakers of Mandi or A'chik when they make words with and without a raka, and keep imitating them until they accept your pronunciation. Practice oh-oh 'oops' or ah-ah-ah-ah as you might say it when mildly reprimanding a small child, until you get a feeling for just what you do when you make a raka. If you pronounce ah-ah-ah-ah very slowly, there will be an alternation between a humming sound and periodic abrupt interruptions with an interval of silence. (Don't whisper! You must say it out loud.) The humming is caused by vibration of the vocal cords and the interuptions occur when the vocal cords come together and "stop" the air from passing. This is the raka. Practice until you can easily distinguish such pairs as bi-a 'he, she' and bi'-a 'request, pray', ga-a 'climb' ga'-a 'step', wal-ko 'night (accusative) and wal'-ko 'fire (accusative)'.

Combining the raka with /m/, /n/, /ng/, or /l/ is likely to be particularly difficult. You might practice making an "m" sound, with the lips closed, but interrupt the humming sound now and then with rakas. The humming sound should stop each time you close your glottis and start again as soon as you separate your vocal cords. Practice with pairs such as ring'-a 'sing' and ring-a 'drink', gam-a 'cultivate' and gam'-a 'make a sound'. Listen carefully to native speakers and practice until they accept your pronunciation. The most difficult part of all will probably be remembering which words require a raka and which do not. You are likely to leave them out from some words where they belong and you may even put in extras where they do not belong. Generally, you will be understood even when you make mistakes,. You cannot expect people to stop and correct you, but you will sound funny if you don't get them in the right places. All you can do is to try, gradually, to get them in where they belong.

Backed /i/ . The second Garo speech sound that is often unfamiliar to learners is the /i/ of closed syllables (syllables that end in /p, t, k, m, n, ng/, or /l/). This is called a "backed /i/" because the tongue is drawn far to the back of the mouth. You should be able to do a passable version of the closed syllable /i/ if you put your tongue in the position that English speakers use for the "u" of food but instead of rounding your lips as you do for food , you must spread them wide as in feet . As always, listening and imitating is essential. Eventually, you will get a sense that a closed syllable simply cannot have a front /i/ (one that sounds similar to the vowel of English feet ) while an open syllable or a syllable ending in a simple raka cannot possibly have a backed /i/.Page  45

Syllable transitions . A somewhat trickier problem is posed by the transitions between syllables and it will probably take a bit longer to get a feel for these. As soon as you can, however, you should try to get a sense that there is a bit of a break between syllables. Each syllable often carries its own meaning in Garo, so the syllables are crucial building blocks of meaning as well as sound, and they are set off from each other in pronunciation somewhat more clearly than in many other Indian or European languages. (Syllable transitions are considered in considerable detail in the next chapter, but that discussion may be more detailed and technical than most learners will need or want. I offer only the most essential points here.)

Many more consonant sequences can occur in the middle of a Garo word than at the beginning or at the end, but even the most complex medial sequences can be split into two parts, the part that belongs with the preceding syllable and the part that that belongs with the one that follows. Any combination of an allowed syllable final consonant with an allowed syllable initial consonant or consonant cluster is possible in the middle of a word, but no others are. For example, medial l-r is possible because one syllable can end with an -l and the next can begin with an r- . On the other hand a medial * r-l is totally impossible. (The '*' means that what follows never occurs). Since vowels are slightly shorter (literally shorter because they take less time) in closed syllables than in open syllables, the length of a vowel is a clue to whether a following consonant is part of that syllable or part of the next. Once you get a feeling for the breaks between syllables, your knowledge of whether or not a syllable is closed will lead you to pronounce the length of the vowel shorter in closed syllables.

Here are some pairs and groups of words that have contrasting sequences of medial consonants. Listen to a native speaker's pronunciation of these words until you can hear the difference between them, and practice them until native speakers accept your own imitations.

cha'-che 'rice husks, chaff' ( cha'-chi in Mandi) chat-chi 'kinsman'

ram-a 'dry' ra-ma 'road'

ka-ta 'word' kat-a 'run'

bi-ka 'liver' bik-a 'carve'

cho'-a 'dig' (present-neutral')

cho'-na 'to dig' (infinitive)

chon'-a 'finish' (present-neutral)

chon-a 'little' (present-neutral)

cho-na 'thresh' (infinitive)

at-chu 'grandfather'

a-chak 'dog'

no 'younger sister' (nominative)

no-o 'near younger sister' (locative)

no-ko 'younger sister' (accusative)

nok 'house'(nominative)

nok-o 'in the house' (locative)

nok-ko 'house' (accusative)

Page  46

/r/ . Remember that Garo /r/ is made by flapping the tongue very quickly against the roof of the mouth. This is the way /r/ is made in most South Asian languages, so South Asians rarely have trouble with it, but /r/'s in many European languages are different, and an English /r/ sounds quite awful when it floats into a Garo sentence. Garos are tolerant, and since /r/ of any variety is unlikely to be confused with anything else, they will probably understand even a strong American /r/, but if you want your speech to sound right, you will have to learn to make /r/ the way the Garos do.

Stress and Intonation . The patterns of stress and pitch that are always a part of spoken language are often among the most difficult aspects of pronunciation for foreign learners to acquire, and there is little that you can do except try your best to imitate what you hear. I offer a rather technical account of some features of stress and intonation in the next chapter, and at some point in your studies this may provide a bit of guidance, but you are unlikely to want to get too deeply into this topic at first. Here are a few pointers to help you get started.

To start with, notice that many Garo words have a slight stress on their final syllable. That syllable may be said just a bit louder than earlier syllables and the pitch may rise, somewhat in the same way that an English word, such as behave has a stress and pitch rise on the second syllable. When pronounced slowly and clearly, a Garo sentence may seem to be a series of words, each set off by a slight rise at its end. An important exception to this is that many suffixes are not stressed, so that the stress stays on the syllable that would have been final had the word not had any suffixes. Some suffixes are stressed, however, and they can be thought of as attracting the stress from the stem of the word and pulling it to themselves. Listen to a speaker pronounce the following words and try to hear the point of stress. You must realize, however, that stress is generally less forceful than in English, and you may have difficulty hearing it at first. The words are written with an accent over the syllable that is most likely to be stressed.Page  47

Words with stress on the final syllable.

Words with stress on a nonfinal syllable.

The stress is fairly subtle but if you can hear the difference between words with final and nonfinal stress you will be on your way to a reasonable pronunciation.

In addition to word stress, there are larger patterns of intonation that involve more than one word. Questions, for example, are often set off by a rising intonation quite similar to the rising intonation of many English questions, but there are other intonational patterns that are quite different from English. You have to listen and imitate.

Page  48