5.0. Introduction

The present chapter is an attempt to step beyond the segmental approach of Bangla phonology. With respect to the current developments in the field of phonology, segments and their features are the micro units of phonology comparable to the molecules and atoms of the physical sciences respectively. The macro unit scheme of phonology starts with the syllable and goes up to the level of lexical category of word via the foot level.

In Bangla phonology, however, till date there is hardly any work done in terms of any of these higher level macro units. Hence the present chapter offers a preliminary characterization of Bangla syllables, the first and foremost macro unit of phonology, in terms of the concepts of metrical phonology.

As the initial work of its own kind, naturally the formulation given in the present chapter is full of gaps and ad hoc assumptions, which will provide ample open areas for future research.

The present chapter assumes that:

i) Phonetically, in terms of the prominence theory, as has been shown in chapter 2, each 'peak' of sonority corresponds to the centre of a syllable. Peaks are best illustrated by more sonorous sounds like vowels, whereas less sonorous sounds such as stops mark the valleys of prominence.

ii) Phonologically the syllable may be treated as a minimal phonotactic unit, say with a vowel as a 'nucleus', flanked by consonantal segments or legal clusterings, or the domain for stating rules for accent, tone, quantity and the like, as has been discussed in chapter 2.

On the basis of these two definitions the present chapter will attempt to

i) correlate between the phonetic 'peak' and the phonological 'nucleus' of the syllable in terms of sonority;

ii) list the phonotactic constraints on syllable structure in terms of a sonority hierarchy;

iii) suggest a template for Bangla syllable; and

iv) describe syllable boundary phenomena in terms of ambisyllabicity.

Section 1 will discuss the syllable as a relevant unit of phonology; section 2 will deal with the internal structure of and the phonotactic constraints on Bangla syllables in terms of the scale formalized as the sonority hierarchy; section 3 will provide a template for Bangla monosyllables; section 4 will deal with the problem of the syllable boundary in polysyllabic words, which will be followed by a conclusion.

5.1. Syllable: the minimal phonotactic unit of language

The recent developments in the field of generative phonology have established the reality of the syllable as the basic macro unit of a phonological string in terms of cross-linguistic evidence. In a number of languages the syllable has been proved to be the optimal domain for stating rules for stress, tone, length and the like. Mention may be made of a few typical case studies like that of English stress (Liberman and Prince, henceforth LP, 1977), Old English noun declensions (Lass, 1984), cases of vowel deletion and syllabification of consonants in many languages (Lass, 1984) and so on.

The SPE-based approach to English stress is a segmental approach. It treats stress as a property of the individual segment. Moreover, the assignment of stress here is determined by sequences of particular segments.

But stress is commonly regarded as a suprasegmental phenomenon. Metrical phonology prefers to consider phonological strings as not merely linear sequences, but as having hierarchical organization based on the syllable. And the major achievement of metrical phonology is to extend such a hierarchically based analysis to stress. The paper by LP (1977) is the first major work in this area.

LP (1977: 262) point out that the SPE-based theory of stress has seven special properties because of which it deviates from the rest of the generative phonological theory. This does not necessarily make the theory incorrect though, according to Hogg and McCully, henceforth HM, "It would clearly be advantageous if a theory of stress could be discovered which was not so radically out of line with the remainder of the phonological theory or which demonstrated that stress assignment had certain special properties which entailed an account of stress radically different from accounts of other phonological phenomena" (1987: 62).

However, LP (1977) present a radical solution to the problems of the SPE-based theory of English stress assignment in terms of a theory that looks (HM, 1987: 64) "at the relative prominence of syntactic and morphological constituents rather than the (quasi-) paradigmatic assignment of absolute stress values to vowels". Although several aspects of this theory have been questioned in the later developments along these lines, it contributed one of the essential insights of the theory of metrical phonology, the insight that (HM, 1987: 64) "the crucial factor in understanding stress patterns is that they reflect, to some extent at least, relations of prominence between constituents" in terms of hierarchical organization of the phonological strings based on the syllables.

In the case of various Old English noun declensions also (Lass, 1984: 250) it has been shown that the problems of suffixes vs. endingless forms of nominative and the forms with vowel syncopation vs. those without vowel syncopation in the genitive singular are elegantly handled in terms of the internal structure of the syllable. In these cases the -VCC and -V"C types of syllable behave in a similar way - a fact that speaks in favour of these two types of syllable forming a natural class. This natural class, according to Lass (1984: 252), "occurs again and again elsewhere in Germanic and Indo-European, and indeed in other language families; and not only in phonology proper and morphology, but in metrics as well, in verse traditions as scattered as Old English, Greek, Latin, and Kannada". On the basis of the metrical patterns reported in the previous chapter one more language, viz. Bangla, may be added to this list.

However, again according to Lass (1984: 22) "It's worth trying to account for this (natural) class, because of its persistence".

Another argument for the primacy of syllables comes from the processes of vowel deletion in connected speech in many languages resulting in a transfer of its syllabicity to a contiguous consonant. For example (as is shown in Lass, 1984: 261), in Ger. [le:g"n] à [le:gn`] à [le:n`] (legen 'to lie') where the underlying form is divocalic and disyllabic, but the outputs are monovocalic but still disyllabic. This problem may be taken care of again in terms on the internal structure of syllables by saying that the syllabification in this case involves transfer of a consonant from coda to peak, where it replaces the original inhabitant of the peak.

Apart from the above mentioned ones there remain a good number of cross-linguistic studies that speak for the primitivity of the syllable as a descriptive unit.

Even in Bangla, as is shown in chapters 2, 3, and 4, syllables are apparently quite versatile macro-units of phonological analysis. Chapter 2 shows that the domain of the predictable stress in Bangla is the syllable. Chapter 3 shows that the classification of verb stems largely depends on the number and structure of the syllables therein. These may be considered as the indirect evidence for the reality of Bangla syllables. Direct evidence is available from the field of indigenous metrical studies of Bangla, as shown in chapter 4.

This body of evidence clearly supports a description of the nature of Bangla syllables in terms of the tools of metrical phonology.

5.2. The internal structure of Bangla syllables

In this section I shall introduce the version of the sonority scale proposed by HM (1987: 33) and then try to capture the internal phonotactic structure of Bangla monosyllables in terms of this scale.

5.2.1. Sonority scale

In the literature it has long been shown that the sounds of language can be classified in terms of their sonority relative to each other. For example, it is quite uncontroversial that the low / open vowels are the most sonorous ones while the least sonorous sounds are the voiceless stops.

On the basis of three studies on sonority, viz. Saussure (1966), Vennemann (1972), and Hooper (1976), HM (1987: 33) propose a scale of sonority in terms of which they formalize the sonority hierarchy of English monosyllables.

As per the requirements of the present chapter I wish to situate the sixteen canonical patterns of Bangla syllables in the context of this sonority scale.

The sonority scale of HM (1987: 33), where the most sonorous sounds are given the highest value and the least sonorous the lowest value, is as follows:

sounds sonority value examples
low vowels 10 /a, a/
mid vowels 9 /e, o/
high vowels 8 /i, u/
flaps 7 /r/
laterals 6 /l/
nasals 5 /m, n, N/
vd. fricatives 4 /v, ð, z/
vl. fricatives 3 /f, q, s/
vd. stops 2 /b, d, g/
vl. stops 1 /p, t, k/

It may apply to the sounds of Bangla as follows:

sounds sonority value examples
low vowels 10 /E, a, O/
mid vowels 9 /e, o/
high vowels 8 /i, u/
flaps and trills 7 /r, R, Rh/
laterals 6 /l/
nasals 5 /m, n, N/
vd. fricatives 4 /h/
vl. fricatives 3 /s, S/
vd. stops 2 /g, gh, j, jh, D, Dh, d, dh, b, bh/
vl. stops 1 /k, kh, c, ch, T, Th, t, th, p, ph/

HM (1987: 42) assign the sonority value 8 to /y, w/ too as these two are "the consonantal versions of the high vowels /i, u/".

Following the above principle of HM and extending it a little more I shall assign the semivowels and the nasalized vowels the same sonority values as their vocalic and oral counterparts respectively.

According to HM (1987: 33) "the usefulness of this concept of a sonority scale in the definition of a syllable lies in the fact that where sonority is greatest we have the centre of a syllable, whereas where sonority is lowest we are near the edge of a syllable. This, of course, can easily be deduced from the fact that whereas vowels are to be closely associated with syllable centres, voiceless stops are never so associated". For example, the number of syllables of a few examples along with their segmental sonority value is as follows:

1. form gloss sonority value no. of syllables
mondir 'temple' 5-9-5-2-8-7 2
anondo 'joy' 10-5-9-5-2-9 3
SOkal 'morning' 3-10-1-10-6 2
Sondha 'evening' 3-9-5-2-10 2 etc.

In these examples the correlation between columns 3 and 4 may be explained in terms of the concept of sonority peak - which is defined as follows (HM, 1987: 33):

"If some segment X has a sonority value which is higher than the sonority values of both of its immediately adjacent segments (or its only adjacent segment just in case it is at the end of a word), then the segment X constitutes a sonority peak".

Thus the number of sonority peaks in the examples of 1 is:
for mondir (5-9-5-2-8-7) 2; for anondo (10-5-9-5-2-9) 3; for SOkal (3-10-1-10-6) 2; and for Sondha (3-9-5-2-10) 2 and these numbers coincide with the number of syllables in each of them. In other words, the sonority peak represents the centre of a syllable, i.e. the syllable nucleus.

With the help of this concept of sonority scale it is possible to state quite a few phonotactic restrictions on the internal structure of Bangla syllables.

5.2.2. Phonotactic structure of Bangla syllables

The Sonority Sequencing Generalization, henceforth SSG, is stated by Selkirk (1984a: 116) as follows:

"In any syllable, there is a segment constituting a sonority peak that is preceded and/ or followed by a sequence of segments with progressively decreasing sonority value".

In other words,

"where we find a sequence of segments preceding or following a sonority peak, then sonority values increase as one nears that peak, and decrease as one moves away from that peak".

Now let us see, mainly in terms of monosyllables, how far the SSG applies to Bangla.

On the basis of the data and observations made in three research papers, viz. Mallik (1960), Sarkar (1986), and Mallik et al. (1998), I shall work out the applicability of SSG to Bangla.

Mallik et al. provide a fairly large body of Bangla words collected according to the principle of random sampling, from which at least 680 monosyllables are considered here. Sarkar (1986) provides 16 canonical patterns of Bangla syllables (as is shown in chapter 2), not all of which are attested in the monosyllables of Mallik et al. Mallik (1960) provides some cues about the clusters in Bangla. Moreover, some new data too will be provided here wherever necessary in order to fill up the gaps in these three studies.

Sarkar (1986) correctly observes that only a few borrowed elements show final CC sequences. On the whole SCB phonology tends to avoid more than one consonant word finally, even in the case of borrowed items, and as such borrowed items are often from English, a language that agrees to SSG, so here I exclude the canonical syllable pattern CVCC. I exclude the pattern V for obvious reasons. And I also exclude the pattern CCCVV as there is no exponent for this in Bangla monosyllables. The remaining 13 patterns of Sarkar (1986), monosyllables exemplifying them, and their sonority values are as follows:

2. pattern monosyllable gloss value
CV ghi 'ghee' 2-8
na 'not' 5-10
ki 'who' 1-9
CVC kaTh 'wood' 1-10-1
nil 'blue' 5-8-6
lal 'red' 6-10-6
Sukh 'happiness' 3-8-1
SeS 'end' 3-9-3
VC Ek 'one' 10-1
aT 'eight' 10-1
uT 'camel' 8-1
VV aY 'income' 10-9
uy 'white ant' 8-8
ey 'this' 9-8
oy 'that' 9-8
CVV koy 'where' 1-9-8
ney 'not' 5-9-8
bhOY 'fear' 2-10-9
CCV bhru 'eyebrow' 2-7-8
sri 'Mr.' 3-7-8
CCVC sthan 'place' *3-1-10-5
kleS 'trouble' 1-6-9-3
gram 'village' 2-7-10-5
hrOd 'lake' 4-7-10-2
srot 'current' 3-7-9-1
CVVC powS 'a Bengali month' 1-9-8-3
mayl 'mile' 5-10-8-6
gaYn 'singer' 2-10-9-5
CCVV krOY 'buying' 1-7-10-9
praY 'almost' 1-7-10-9
CCVVC prayj 'prize' 1-7-10-8-2
CCCV stri 'wife' *3-1-7-8
CCCVC strir 'wife's' *3-1-7-8-7
VVC ayn 'law' 10-8-5
awS 'a kind of paddy' 10-8-3

CCVVC is represented by common English loans. Except for the canonical patterns VV, CCVC, CCCV, and CCCVC all the others agree with SSG. VV, CCVC, CCCV, and CCCVC patterns apparently do not agree with SSG.

Among these four VV poses a minor problem for SSG. In a VV sequence, i.e. in a diphthong, sometimes two consecutive segments show the same sonority value, e.g. /uy/ (8-8). In such a case, if we adhere to SSG, it is not easy to identify the peak.

This, I think, has been caused by the limitations of this particular sonority scale. According to HM (1987: 32), "Since sonority is in part to be correlated with degrees of obstruction of the air-stream, it follows that high vowels are less sonorous than low vowels" etc.

Hence it is possible that a more delicate scale than this one may show that the semivowels are less sonorous than their pure vowel counterparts. However, I leave the question of a more delicate scale open, and handle the so-called limitations of this scale in terms of (1) a phonotactic claim made by Sarkar (1985-86: 16) that Bangla has only falling diphthongs, and (2) a template condition. The phonotactic claim made by Sarkar (1985-86: 16) suggests that in a falling diphthong the first element is the syllabic one. Hence between two equal-valued segments the one on the left is the peak in Bangla. On the basis of this suggestion a template condition, or to be more specific a coda template condition, which will be discussed later, will stipulate the permissible arrangement of the terminal nodes of a syllable structure tree.

Though the other problem apparently arises from the CCVC, CCCV, and CCCVC clusters, it is not restricted only to these three canonical patterns. In fact, this type of disagreement with SSG appears in the case of all the clusters having an /s/ and a voiceless stop as their first and second members respectively, e.g. sthan (3-1-10-5), stri (3-1-7-8) etc.

In English also forms like spin, stile, strip, sprint etc. pose a similar problem. HM (1987: 49) handles this problem by treating /sp, st/ as single constituents in syllable onsets on the basis of the fact that (HM, 1987: 48) "/s/ appears to have extremely close links with the following stop consonant" which "must always be voiceless".

This logic holds good even for Bangla as in Bangla too sequences like /*sb, *sd, *sgh, *sg/ etc. are nonexistent.

Moreover, in Bangla, as has been discussed in chapter 2, phonotactically /s/ is the only sibilant allowed as the first member of clusters, e.g. snan 'bath' but proS-no 'question'; skOn-dho 'shoulder' but pu-roS-kar 'prize'; smi-ta 'a name' but SuS-mi-ta 'a name' etc.

In these examples /s/ is the member of a cluster; whereas /S/ is the member of a mere sequence. So in Bangla the existence of /s/ as the first member of the initial sequence of a word ensures close links between /s/ and its following segment.

Hence the /sth, st, sp, sph/, in sthan 'place', stri 'wife', spOndon 'beat', and sphoTik 'crystal' respectively may be treated as single constituents in syllable onsets. Thus sthan and stri will have the following structures:

3. s s

O R O R

P Co st r i

sth a n

and the SSG will not be violated.

Moreover, with such a treatment of the sibilant+stop consonant clusters it could be claimed that phonotactically Bangla may have at most two members in the onset of a syllable.

The claim, on the one hand, is supported by Mallik's (1960) observation that /str/ is the only CCC cluster in Bangla, (though /spr/ is also available in Bangla polysyllabic forms), as is mentioned in chapter 2; and on the other hand, it can explain why other cluster types (e.g. /*tmr, *mlr, *plr/ etc.) are not tolerated in Bangla.

However, the available canonical forms of Bangla syllables along with the SSG can account for the non-occurrence of a few forms in Bangla as follows:

1. According to Sarkar (1985-86: 16) Bangla has only falling diphthongs, in which the first element is syllabic and the second one is non-syllabic. But a few diphthongs, viz. /*iY, *iW, *uY, *uW/ are not tolerated in Bangla as they violate SSG, e.g. the sonority value of all these four unacceptable diphthongs is *8-9.

2. Though CCVC is an existing canonical pattern, forms like *bnOk, *rtin, *rlop etc. are not acceptable in Bangla as they violate SSG, and so on.

Thus, SSG is a necessary condition for Bangla syllables as it can explain quite a few phonotactic restrictions of the language, as above, which the canonical pattern alone cannot deal with.

5.3. Template for Bangla monosyllables

A template may be defined as an abstract structure which all syllables have to fit in in order to be recognized as acceptable syllables in a particular language. In this section I shall attempt to formulate the template for Bangla monosyllables.

With the unitary treatment of the sibilant+stop consonant clusters, as is shown in the preceding section, the number of members of the onset clusters is restricted to two. In Bangla words, as is shown in chapter 2, the coda consists of a single member. Thus the normal maximal structure for Bangla monosyllable is CCVVC, which is arranged as:

4. s

O R

C C P Co

V V C

In terms of the sonority scale and the SSG a number of interesting phonotactic statements could be made about this maximal structure of syllable in Bangla.

The tree structure suggests that the syllable is organized hierarchically into Onset+Peak+Coda, where the latter two form a Rime constituent. Such an organization is supported by the fact that in Bangla the weight of a syllable (cf. chapter 2) and the mora-count of a syllable (cf. chapter 4), although crucial to constitution of the rime components, appear to play no part in the construction or the number of segments of the onset.

The rime so far has been divided into two parts, viz. peak and coda. But current research, as is reported in HM (1987: 45), suggests that this is a false division because the subdivisions under the rime node apparently do not play any crucial role, rather the rime node itself plays the crucial role, in phonology.

In Bangla at least two phonological phenomena, viz. syllable weight and mora-counting, directly depend on the branching or non-branching of the rime node, rather than that of the peak or the coda separately. For example, Bangla shows three types of heavy syllables, viz. (1) coda preceded by a monophthong - in which the rime branches, e.g. kan 'ear'; (2) coda preceded by a diphthong - in which the rime and the peak branch, e.g. ayn 'law'; and (3) a diphthong-final syllable - in which the peak branches, e.g. bhay 'brother', as follows:

5. s s s

O R R O R

P Co P Co P

k a n a y n bh a y

If the subdivisions of rime are done away with and 5 is represented as 6, then a single, simple and general statement like a syllable is heavy if and only if the rime branches (as was done in chapter 2) can take care of the heavy-light aspect of the syllable.

6. s s s

O R R O R

k a n a y n bh a y

In order to achieve generalization and simplicity, on the basis of the argument given in HM (1987: 45) and the one shown above, I reject the subdivisions under rime and accept the direct branching of rime. Thus 4, along with convenient indexing, may be flattened:

7. s

O R

O2 O1 R1 R2 R3
| | | | |
k r a y m
kraym 'crime'

(It is interesting that no native monosyllable appears to have branching of this type, using all the options).

In terms of the sonority scale and the SSG now I shall consider the template conditions of 7, i.e. the maximal structure of Bangla monosyllables.

I shall divide the template conditions into two groups, viz. a) general conditions and b) specific conditions.

a) General template conditions

These conditions apply to the maximal tree structure as a whole and comment on the presence-absence status of the terminal nodes.

i) R1 is obligatorily filled.
ii) O1 is optionally filled.
iii) O2 is filled if and only if O1 is filled.
iv) R2 is optionally filled.
v) R3 is filled if and only if R2 is filled.

b) Specific template conditions

These conditions apply to the specific branch of the tree and determine the sonority value, henceforth SV, of the terminal nodes.

i) SV of R1 ³8, i.e. the SV of R1 cannot be less than 8.
ii) SV of O1 £9, i.e. the SV of O1 is not more than 9. For example, in the words like yar 'friend' and WaR 'cover (said of a pillow)' etc. the O1 position is occupied by y and W, the SVs of which are 8 and 9 respectively.
iii) SV of O1 ³6 iff O2 is filled, i.e. the SV of O1 is not less than 6 if and only if O2 is filled, e.g. /ml, gr, br/ etc.; but /*nm, *mt/ etc.
iv) SV of O1 £7 iff O2 is filled, i.e. the SV of O1 is not more than 7 if and only if O2 is filled, e.g. /gr, pl, sl/ etc.
v) SV of O2 £5, i.e. the SV of O2 is not more than 5, e.g. /ml, hr, sr, gr, tr/ etc.; but /*rl, *lr/ etc.
vi) SV of R2 £9, i.e. the SV of R2 is not more than 9, e.g. R2 can have at the most the mid semivowels, viz. Y and W.
vii) SV of R2 ³8, iff R3 is filled, i.e. the SV of R2 is not less than 8, if and only if R3 is filled, e.g. the semivowel of a diphthong followed by another consonant.
viii) SV of R2 £9 iff R3 is filled, i.e. the SV of R2 is not more than 9 if and only if R3 is filled.
ix) SV of R3 £7, i.e. the SV of R3 is not more than 7, e.g. gowr 'a name' etc.

Thus the terminal nodes of a syllable have to fulfil the above template conditions in order to be acceptable as a potential syllable in Bangla.

Regarding the upper limit of the SV of O1, i.e. condition bii, it should be mentioned that monosyllabic yar and WaR, the examples given here, often freely vary with disyllabic i-ar and o-aR respectively. A semivowel in the onset position, however, is more common in polysyllabic forms, e.g. khay-ye-che 'he has fed' etc.

5.4. Syllable boundary

As is hinted in chapter 2, a sonority-based approach to the syllable gives a useful general guideline, but it does not always indicate clearly where the boundary between adjacent syllables falls. Hence a syllable-based discussion of phonology, which in turn is sonority-based too, will face the problem of boundary determination.

According to HM (1987: 51), "a syllable-based description does not face any major problems in determining how many syllables a word has. But there certainly do seem to be problems in determining where the boundary between each syllable (in a polysyllabic word) might lie."

Lass (1984: 262) states the problem as follows: "How do we decide, given a string of syllables, what is the coda of one and the onset of the next? This is not entirely tractable, but some progress has been made. The question is: can we establish any principled method for bounding syllables, so that words are not just strings of prominences, with indeterminate stretches of material in between?"

Accepting the fact that the syllable boundary problem "is not entirely tractable", in this section I shall consider the principles available in the literature and attempt to justify their effectiveness in bounding Bangla syllables adequately in polysyllabic words.

Two basic approaches to the problem of bounding syllables are the universalist approach and the phonotactic approach.

According to the universalist approach, as exemplified by Hooper (1972, and 1976: chap. 12), a general procedure, along with necessary language specific deviations, for inserting syllable boundaries has been developed on the basis of assumptions about universally optimal syllable structures. This approach consists of principles ranging from the most obvious to the completely arbitrary. The most obvious one is that there is a rule that assigns a syllable boundary at the beginning and the end of a word. The other principles are like this - the syllable boundary must be placed between a [+syll] and a [-syll] followed by a [+syll] segment because CV is the minimal universally attested, and hence optimal syllable, and so on.

Lass (1984), however, shows that on the whole the principles of the universalist approach are too powerful as well as too restrictive because in many languages they often fail to "take account of the possibility of phonetic and phonotactic evidence for bounding".

For example, Lass (1984: 265) discussed the phonetic evidence of glottalization and aspiration, the two characteristic phonetic markers of word final and word initial positions in English respectively, and points out the limitations of the universalist approach in identifying them.

In accordance with the principle that CV is the optimal syllable etc., the rule of syllabification, as is given in Lass (1984: 263), is as follows:

Æ à $ / [+syll] ___ [-syll] [+syll]

$ indicates syllable boundary. Thus the English word 'butter' would always unambiguously be syllabified as /$bL$tr$/. But for a dialect with [bL?tr] the universalist principle of syllabification fails because for a dialect with such glottalization the correct syllabification is /$bLt$r$/, rather than /$bL$tr$/, as glottalization is a clear phonetic marker of word final, hence syllable final, position.

Even for a dialect with phonetic [dIskh¢st] 'discussed' and [dIsk¢st] 'disgust' the correct syllabification would be /$dIs$kst$/ and /$dI$skst$/ respectively depending on the existence of aspiration, a clear marker of word initial, hence syllable initial, position. This fact too, the universalist approach fails to capture.

Such phonetic evidence exists in Bangla also. At this point I wish to refer to the phonotactic restrictions of Bangla syllabification - as discussed in chapter 2, sec 1. Though the rules of syllabification conform to the most obvious principles of the universalist approach, viz. the assignment of syllable boundaries at the beginnings and ends of words, they respect certain phonotactic cues also. For example, for the sequence NSk the syllable boundary is placed between NS and k on the basis of the phonotactic restriction that dental sibilant, rather than palatal sibilant, is the only sibilant in Bangla consonant clusters. In this case the application of R-8, i.e. the general rule for inserting syllable boundary between a CCC sequence would produce the counterintuitive *N-Sk.

Since the universalist approach fails to predict language specific syllabification rules (as in the case of NSk in Bangla) I reject it and try for the other approach, viz. the phonotactic approach, for Bangla syllabification.

The phonotactic approach starts with the universalist's obvious assumption of assigning syllable boundaries at the beginnings and ends of words. But unlike the universalist approach it recognizes the ambiguity and language specific phonetic informations.

Under this approach, according to HM (1987: 51), the three basic principles of syllabification are (a) the Principle of Maximal Codas, (b) the Principle of Maximal Onsets, and (c) the Principle of Maximal Codas and Maximal Onsets - all of which are loyal to the basic imperative that

8. No syllabification should yield syllables that do not conform with the syllable template of the language in question.

Now I shall consider the applicability of these three principles in Bangla.

According to the Principle of Maximal Codas as many intervocalic consonants as are permitted by the basic syllable template are assigned to the coda of the first syllable, e.g. the form notun 'new' should be syllabified as [not][un].

According to the Principle of Maximal Onsets maximum number of consonants are assigned wherever possible to the onset of the second syllable, e.g. [no][tun] for notun.

According to the Principle of Maximal Codas and Maximal Onsets the intervocalic consonants are assigned both to the coda of the first syllable and the onset of the second syllable provided that the syllable templates are not violated, e.g. [no[t]un] for notun.

According to Lass (1984: 267) the first two principles produce non-overlapping or properly bracketed syllabification; whereas the third one produces overlapping or improperly bracketed syllabification.

However, the Principle of Maximal Codas does not seem to apply to Bangla because of the following reasons:

i) In case of intervocalic single consonants it produces quite counterintuitive divisions, e.g. [not][un], [Sob][uj] for notun and Sobuj 'green' etc.

ii) Even the writing system does not support the Principle of Maximal Codas as in Bangla we write òîÅÂò , not * ÂòîÄ ëÂ×ò.

iii) In case of intervocalic consonant sequences the problem cases for this principle are the forms like prosthan 'exit', baSosthan 'residence' etc., where [s] cannot be unambiguously a member of the first syllable.

On the contrary, the Principle of Maximal Onsets seems to work for Bangla syllabification because:

i) The assumed syllable divisions of the polysyllabic words in Bangla support this principle, e.g. [no][tun], [So][buj], [a][kaS], [ba][nan] etc. for notun, Sobuj, akaS 'sky', and banan 'spelling' respectively.

ii) The rules of syllabification of the consonant sequences, viz. R-5, 7, 8, and 9, as discussed in chapter 2, mostly conform with the Principle of Maximal Onsets directly.

Though there remain a handful of controversial cases, which will be accounted for shortly, the Principle of Maximal Onsets agrees with the intrinsic spirit of Bangla syllabification quite satisfactorily.

The Principle of Maximal Codas and Maximal Onsets, which according to HM (1987: 50) is the principle of ambisyllabicity, too applies to Bangla. Ambisyllabicity is a principle that assigns a segment to more than one syllable at a time. In fact, in Bangla there is evidence that in some cases a segment cannot be unambiguously assigned to either the preceding or the following syllable. The Principle of Maximal Onsets alone cannot handle such cases. Two such cases are as follows:

i) Intervocalic semivowels

In forms like OYon 'orbit', khaWa 'treat' etc. the placement of the syllable boundary is not so simple. On the one hand, after the Principle of Maximal Onsets they should be syllabified as [O][Yon] and [kha][Wa].

On the other hand, as in Bangla hardly any word starts with a semivowel, a second type of syllabification, viz. [OY][on] and [khaW][a], too seems to be plausible.

Moreover, the native speaker's intuition does not help to resolve the problem as it accepts [OY][on] and [kha][Wa].

At this point a non-controversial syllabification of these forms would be [O[Y][on] and [kha[W]a], which reflects the principle of ambisyllabicity.

ii) s+th sequences

Two different judgements about syllabification are available in the case of forms with intervocalic s+th sequences, e.g. [ba][Sos][than] vs. [ba][So][sthan] for baSosthan 'residence'; [pros][than] vs. [pro][sthan] for prosthan 'exit' etc. Neither of them violets the syllable template.

This fact too would be better handled in terms of the principle of ambisyllabicity as follows:

[ba][So[s]than] and [pro[s]than]

Hence the principle of ambisyllabicity is supported by the intuitive facts of the language.

Moreover, ambisyllabicity is supported by one more fact that there are phonological rules in the language sensitive to the process of ambisyllabicity. Chapter 3 shows that in the segmental approach of Bangla verb morphology there are rules that are sensitive to the subcategories of disyllabic stems, viz. CVCV and CVCCV.

In terms of the concept of ambisyllabicity these two subgroups, viz. CVCV and CVCCV, may be described as a) potential to be ambisyllabic, and b) clearly non-ambisyllabic respectively as CVCV may be syllabified as [CV[C]V], but CVCCV is syllabified only as [CVC][CV].

Thus we see that ambisyllabicity is a valid principle in Bangla syllabification.

At this point I assume that the Principle of Maximal Onsets is the basic principle of the underlying syllabification of Bangla and then a rule, viz. the rule of ambisyllabicity, applies to that underlying syllabification and results in ambisyllabic segments, where necessary. The rule of ambisyllabicity, after HM (1987: 53), roughly looks like:

9. s s s s

R O R O
à
… Rj C …. Rj C …

Where Rj stands for the last rime segment, which because of the Principle of Maximal Onsets and the basic imperative 8, will not be greater than R2 in Bangla. Thus the ambisyllabification of OYon will be as follows:

10. s s s s

R1 C R R C R

R1 R2 à R1 R2
| | | |
O Y o n O Y o n etc.

Apparently such an assumption is quite acceptable in the present framework as it does not produce any clashing effect with any of the phonological assumptions made either in this or in any of the preceding chapters. Eventually, a reasonably thorough network of metrical phonology of Bangla in terms of the higher level units like mot, foot etc. will work out the details of the rule of ambisyllabicity and judge the actual merit of the same in the language.

5.5. Conclusion

Above is an attempt to sketch the basic tools of a syllable-based approach to the phonology of Bangla. The principal claims of this approach are in brief:

a) In Bangla phonology the syllable is a well-defined unit.

b) Syllables have hierarchically organized internal structures.

c) The basic structural possibilities of Bangla syllables are represented in terms of a language-specific template.

d) The concept of syllable boundary is a language-dependent one.

e) Ambisyllabicity is a valid principle in Bangla.

Among the unresolved problems left open here, one is that of the syllable template itself, for which no native monosyllable has been found, though the branching under this template is not at all counterintuitive. This problem may be taken care of in terms of some sort of global restrictions on the template stating that locally the onset and the peak may branch but globally not more than four branches are allowed under a syllable template in Bangla.

Another uninvestigated aspect is what Fudge (1969) (as is discussed in HM, 1987: 50) calls the collocational restrictions "which contract or expand upon the basic template".

May be an investigation combining these two questions will lead to some definite answers - a task that awaits rigorous future research along the same line.