Syllable Structure in Tamil:
A case study of the Eastern Perspective.
Dissertation for MA Linguistics
University College London
Xavier Emmanuel
Student Number 69051654
12 September 2008
Contents
Section I
1.0 Introduction 3
1.1 Syllables and syllabification 4
1.1.1 Traditional view of syllabification 5
1.1.2 Syllable structure and sonority 6
1.1.3. Sonority Vs linguistically significant modulation 6
1.1.4 Sonority, perceptibility and syllable typology 7
1.1.4.1 Vowel length contrast 7
1.1.4.2 Consonant length contrast 8
1.1.5 Sonority, perceptibility and languages without
complex syllable edges. 8
1.2.0 Tamil Language: Introduction 9
1.2.1 Diglossia and dialects 10
1.2.2 Jaffna Tamil 10
Section II
2.0 The phonemes of Tamil and the writing system 11
2.1 The writing system 11
2.1.1 Class 1: Graphemes representing single vowels or
vowels-only sequences (i.e. V, V:, or V1V2) 12
2.1.2 Class 2: Graphemes representing Consonant-vowel
sequences (i.e. CV, CV: or CV1V2) 13
2.1.3 Class 3: Graphemes representing single consonant segments 14
2.1.4 The writing system and syllable typology:
the Eastern perspective 15
2.2 Phonotactic restrictions at specific positions 16
2.2.1 Word initial positions 16
2.2.1.1 Initial Vowels. 16
2.2.1.2 Word initial consonants 17
2.2.2 Word Final Positions. 18
2.2.2.1 Word final Vowels: 19
2.2.2.2 Word final Consonants: 20
2.2.3 Word internal consonants 22
2.2.3.1 Word internal Single consonant segments 22
2.2.4 Consonant clusters in Tamil 24
Section III
3.0 Syllable structure of Tamil and implications for theories
regarding the syllabic status of word-end consonants
and geminates. 25
3.1 Syllable typology of Tamil: an overview 25
3.2 The Tamil writing system and the word-final onset hypothesis 26
3.2.1 The ‘pulli’ grapheme as the Dull Syllable 31
3.3 Syllable Typology of Tamil and the word-end consonant. 32
3.4 The final onset hypothesis: evidence from Segmental
restrictions and lenition 35
3.5.0 Long consonants (‘geminates’) and vowel length in Tamil 38
3.5.1 Obstruent long consonants and vowel length 40
3.5.2 Non-obstruent long consonants and vowel length 42
3.5.3 Three-consonant clusters 43
3.6 Vowel length and consonant length in monosyllabic free roots 43
Concluding remarks 45
Acknowledgements 46
Bibliography and references 47 Appendix: The complete set of Graphemes in Tamil 48
Section I
1.0 Introduction
Traditionally, syllabification is taken as a process of mapping from the lexically specified segment strings of words or utterances onto a syllable sequence. While there is no doubt that the sequence specification of the segment strings are necessarily lexical, it is not clear in this view which, if any, aspects of syllable structure are underlyingly represented independently of segment sequence specification. This view has several well known inadequacies (reviewed in Harris and Gussmann, 2002). A radical alternative to this view is to postulate that syllable structure is specified independently of segmental sequence specification or location in the word. Implied in this alternative view is the possibility that there could be ‘lexically’ specified syllabic positions which are not occupied by any phoneme. This view seems to be embodied in the syllabic writing systems of many Eastern languages. An aspect of this view which has attracted particular interest recently concerns the syllabic status of word-final consonants. Proponents of the view that the word final consonant is in fact the syllable onset of a word final dull syllable have dubbed this view the ‘eastern prospect’, as distinct from the conventional ‘western prospect’ which views the word final consonant as occupying the coda position of the word final syllable (Harris and Gussmann, 2002). In this dissertation, I wish to use data from Tamil to explore the question of the syllabification of word final consonants and the status of the geminates. It is hoped that data from two particular source domains may be prove to be especially useful, namely data and insights on how the writing system of Tamil is used to represent its phonology, and data relating to dialect differences within Tamil.
Tamil is a major and in many ways an archetypal language of the Dravidian family of languages. There is a considerable body of literature on the grammar of the language ranging from classical texts to more recent monographs by western and Indian scholars. However, most of the recent literature is descriptive or concerned broadly with historical linguistics. The language has not been the subject of modern phonological study in the generative tradition to the extent that it deserves. This is particularly so in the area of syllable structure and syllabification. A necessary preliminary to this study has therefore been the assembling of phonological data of particular relevance to syllable structure and syllabification in the language.
The dissertation is presented in three sections. Section I presents overviews of two areas: firstly, a review of the concept of the syllable and syllabification, with a bias toward covering language types which avoid consonant clusters at word edges, and, secondly a brief background to the Tamil language. Section II is largely devoted to martialing the data relevant to syllable structure in Tamil. The data are assembled and presented, with appropriate comments and summary statements, as sub-sections according to points of particular interest for elucidating syllable structure. Section III is an attempt to use the data to define and account for the syllable structure of Tamil, and particularly to assess the implications for theories regarding the syllabic status of word-end consonants and geminate or long consonants.
1.1 Syllables and syllabification
The concept of the syllable is hard to define in phonetic terms. Never the less, it is required in both classical and modern phonological theory as an abstract but indispensable unit on which languages base their characteristic organisation of the sound patterns of words and word strings. Since languages vary in their sound patterns, this implies that the exact repertoire of shapes of these organisational units will vary from language to language, though we may expect to see some cross-linguistic (universal) preferences for certain shapes.
Aspects of phonological theory where reference to the concept of the syllable is necessary or convenient include statements of restrictions on the sequencing of sounds at particular positions of the word (i.e. phonotactic constraints), and phonological processes such as segmental ‘epenthesis’ and ‘deletion’ which are explainable on the basis of a requirement to conform to preferred syllable templates.
Though there is general agreement that the concept of the syllable and the cognitive process of syllabification are essential for modern phonological theory, much remains to be understood, and this is an active area of phonological research. The study of syllable structure and syllabification in different languages is a major plank of this research effort. The study of the differences between dialects would be particularly valuable since it may allow us to look at individual variables in isolation. The concept of the syllable and syllabification is reviewed in Kenstowicz (1994), pages 250-298.
1.1.1. Traditional view of syllabification
Traditionally, the syllable is conceived of as having an essential core of high sonority nucleus, preceded optionally by a low sonority onset and followed, again optionally, by a low sonority coda. Typically, the nucleus is a vowel (V), and the onset and coda are consonantal (C). If, for the sake of simplicity, a string of segments in a word or utterance is thought of as an assorted string of Vs and Cs, and if we assume a rule driven process of syllabification, the possible basic syllable types can be derived by the application of three primary rules. The first rule assigns each V to a syllable nucleus and the second rule assigns each prevocalic C to the syllable onset and a third rule assigns any unassigned C to the coda of the syllable formed by the preceding V. This yields the following set of basic syllable shapes: V, CV, CVC and VC. Since these rules apply sequentially, VC syllables would be expected to be found at word beginnings only.
Many languages allow more than one C to be assigned to the onset and coda positions, leading to C clusters at these positions. But there are a significant number of languages, including Tamil, which seem to exploit only the three basic rules for mapping segments to syllables. This implies that the repertoire of syllable shapes in these languages is limited to {V, CV, VC, and CVC}. A corollary to this is that there is maximally only one C at word beginning and at word-end, and word internally there are maximally only two Cs, at coda-onset junctions. When strings with C clusters which do not conform to this limit are encountered in a language of this type (as a result of morphological concatenation or borrowings from languages which do allow such clusters) phonological processes are set in train, firstly to alter the segment string by introduction of or deletion of segments as appropriate, and then to resyllabify the altered string to achieve the preferred syllable shapes for the language.
1.1.2 Syllable structure and sonority
In languages with clustered (complex) onsets and codas, the principle of sonority sequencing has been extensively invoked to account for the severe restrictions on the sequencing of segments within these clusters (reviewed in Kenstowicz, 1994 pp 250-298).
In particular, the Sonority Dispersion Principle (ascribed to George Clements, reviewed in Kenstowicz, 1994, pp 283-284) proposes that there is a general preference to maximise the sonority slope at the onset-nucleus half of the syllable at the expense, as it were, of the nucleus-coda half of the syllable. In languages without complex onsets or codas, descriptions and explanations of syllable structure and syllabification need not be so dependant on the relative sonority of adjacent segments. Indeed, in these languages, it is possible to discuss syllable structure and syllabification with little or no reference to sonority.
1.1.3. Sonority Vs linguistically significant modulation
In a general critique of the role of sonority as a determinant of phonotactic restrictions, Harris (2006) emphasises the distinction that needs to be made between the ‘audibility’ of the speech signal and its content in terms of specifically linguistic information. The phonological knowledge that speaker-listeners have which enables them to decode the information rich modulations of the carrier signal is based on the distinctive phonetic features or qualities of the adjacent segments and not on the differences in their sonority. Sonority itself has no direct correlate to phonetic features, and while it may well contribute to the audibility of the speech signal, it does not contribute to encoding the linguistic message. The most audible part of the signal is the vocalic nucleus of the syllable, but it offers the least scope for the phonetic modulations that are capable of encoding linguistic information in a way that is accessible to the core phonological knowledge of the speaker-listener. Conversely, the less sonorous parts of the syllable i.e. onsets, codas and particularly the intervocalic onset-coda clusters, consisting as they do of multiple interphases between adjacent feature bundles (in the form of consonantal segments), offer much greater scope for the phonetic modulations that are capable of encoding linguistically significant information.
The vocalic nuclei provide a baseline carrier signal from which the linguistically salient acoustic landmarks or cues associated with the enunciation of the coda-onset clusters can be evaluated. These cues include the abrupt drop in amplitude corresponding to the closure phase of the stop, formant discontinuities associated with laryngeal activity, bursts of aperiodic energy on the release of plosives, sustained aperiodic energy associated with fricatives, the frequency characteristics of the noise burst, changes in spectral shape that help cue the place category of a stop, formant characteristics distinctive of nasals and formant transitions during the approach and release phases.
The alternative to the sonority based view of the factors which determine the patterns of phonotactic restrictions contends that the relative preference or dispreference for particular sound sequences owes much to whether the particular sequences allow or hinder the production and perception of these cues. The native speaker phonological knowledge that enables us to attach specific linguistic meaning to these modulations or cues is based on the distinctive features of the sounds and not on their relative sonority. The evidence and arguments for this view are marshaled in Harris, 2006.
1.1.4 Sonority, perceptibility and syllable typology
How do these considerations relate to the syllable typology of languages? Given that the vocalic nucleus of the syllable has relatively low information carrying capacity compared to the syllable edges and coda-onset junctions with their feature-rich consonantal segments, do languages like Tamil which eschew complex onsets and codas incur a relative limitation in their capacity to meaningfully modulate the speech signal? If so, are there any compensatory devices which are reflected in the syllable inventories of these languages?
1.1.4.1 Vowel length contrast
Let us take second look at the potential of the vocalic nucleus for linguistically significant contrast. Lexical contrasts in vowel sounds are based on gross categorical differences in just one qualitative property, namely their characteristic formant pattern. This severely limits the number of phonemic vowel sounds, often to five or less. However, languages can and do harness the quantitative property of vowel length to increase their repertoire of vowel contrasts. In speech production and perception, this property also divides categorically, there being at most only binary contrast. This may appear at first sight to be of limited potential, but when fully exploited, it can double the number of vocalic phonemes. The repertoire of vocalic phonemes may be further augmented by exploiting diphthongs. Thus the syllable inventory available for traditional syllabification expands to V, VV, V1V2, CV, CVV, CV1V2, CVC, CVVC, CV1V2C, and VC, most of which are attested and fully exploited lexically in Tamil.
1.1.4.2 Consonant length contrast
Length contrast in consonants may also be exploited lexically in languages that have non-complex syllable edges, such as Tamil, to augment their phonemic inventory. Geminates do not differ from single consonants in the scope they offer for linguistically significant acoustic modulations of the carrier signal. Binary contrast in the quantitative property of length however does offer a basis for lexical contrast. Phonemic long consonants or ‘geminates’ of obstruent stops, nasals, laterals and glides are frequently found word medially. In Tamil, length contrasts in vowels and consonants appear to be independent of each other for obstruent stops, but not for sonorants. A further point of interest is the possibility, hitherto unremarked in Tamil, of phonemic length contrast at the word final position in monosyllabic free roots. The data relating to these and the implications for syllabification are discussed in detail in later sections.
1.1.5. Sonority, perceptibility and languages without complex syllable edges.
Even though the Sonority Dispersion Principle was formulated mainly in relation to languages with complex onsets and codas, languages lacking complex onsets or codas, such as Tamil, also seem to conform to it in that they tend to show a sharp rise in sonority from onset to nucleus and a less marked fall in sonority from nucleus to coda. However, this says no more than that there is in these languages a cyclical variation in sound amplitude which maps with the internal structure of the syllables. Sonority itself does not fully account for the patterns of qualitative segmental restrictions at word initial, word final and intervocalic (coda-onset) positions in these languages. These qualitative restrictions and their alignments/misalignments are well exemplified in the case of Tamil.
1.2.0 Tamil Language
Introduction
Tamil (t̪ ɑmiɭ ) is the archetypal member of the Dravidian family of languages, which is native to South India. There are about 60 million native speakers in the state of Tamil Nadu. The Tamil speaking community of about four million in the north and east of
Sri Lanka dates back a thousand years or more. Sizable communities dating from the 19nth century exist in Malaysia, Singapore, Fiji, Burma, South Africa, Mauritius, Trinidad and Tobago and British Guyana. More recent immigrant communities, mainly of Sri Lankan Tamils, are found in the major conurbations of the UK, Canada, the USA, Australia and France.
Tamil has an extensive corpus of classical, medieval and modern written literature and a codified grammar. The oldest grammatical text, the Tolka:piyam, is said to be from the second or third century BC. Of all the Dravidian languages, Tamil is the least influenced by contact with Indo-European, the other major language family of India. This is in part attributed to the campaigning of the highly influential ‘pure’ Tamil movement from the early twentieth century and to the dominance of the classical literary tradition. Even in the non-literate individual, there is a high degree of awareness of non-native phonological and morphosyntactic forms.
There are several recent descriptive grammars of the Tamil language written in or translated into English. Among these, the following were extensively consulted as background reading and sources for language data for this dissertation: (Asher (1985), Schiffman (1999), Krishnamurti (2003), Andronov (2003), Steever (1998), Burrow and Emeneau (1984).
1.2.1 Diglossia and dialects
As with many other long established literary languages, there is marked diglossia among native Tamil speakers. A ‘high’ or literary variety (referred to in this dissertation as Literary Tamil or LT) is spoken by educated speakers in formal contexts, and is the norm in the media, public speaking and the stage. There is relatively little dialectical variation in spoken LT. In colloquial speech by contrast, there is very marked dialectical variation across a number of dimensions: geographical, social class, caste and ethnicity. It is claimed that a standard spoken dialect of South India has emerged, being the informal speech of educated speakers used in the major urban areas and increasingly used in the cinema and in television drama and chat shows (Asher 1985). It is this ‘standard’ dialect that is described in the recent grammars listed above and is used as one of the dialects for this dissertation. The dialect is referred to simply as Spoken Tamil or ST.
Though it is not used in everyday informal contexts, exposure to LT is so pervasive that competence in it (as listeners at any rate) is more or less universal. Tamil therefore shows classic diglossia (Ferguson, 1959). Colloquial dialects of Tamil are regarded popularly and by historical linguists (see for example Andronov, 2003, Krishnamurti, 2003) as having evolved or derived from classical Tamil. However, as seems to be the case with diglossia in Arabic (Owens, 2001), it is quite likely that the colloquial dialects of Tamil co-existed from pre-literate times with the dialect which came to be the ‘high’ or literary dialect.
1.2.2 Jaffna Tamil
The community of native Tamil speakers in the north and east of Sri Lanka have had a degree of geographical separation from the mainland communities for several centuries, and their informal spoken dialect is notably different from the mainland dialects. There are no published works which fully document this dialect, and the descriptions and illustrative examples used in this dissertation are my own or from other native speakers of the dialect now living in London. This dialect is referred to in this dissertation as Jaffna Tamil or JT (from the most populous town and district in northern Sri Lanka). In some respects, JT appears to conserve forms from the classical language to a greater degree than the mainland dialects, but it also shows many language internal changes. It is suggested that Tamil became established in the North-eastern parts of Sri Lanka as a result of penetration during the first millennium CE by ‘elite’ groups of traders, administrators and mercenaries from South India who may have extensively used the ‘high’ dialect (Indrapala, 2005).
For the purposes of this dissertation, only those aspects of dialect variation which may have a bearing on syllable structure are examined.
Section II
2.0 The phonemes of Tamil and the writing system
As a necessary preliminary to a discussion of syllable structure in Tamil, this section provides a brief account of the phoneme and grapheme inventory of Tamil, and a description of relevant phonotactic restrictions seen in the language. The illustrative examples given here are supplemented by more extensive lists in the appendix.
It is convenient to organise the description of the sounds used in Tamil in the order that is used in traditional descriptions of the writing system of the language. This is the approach that is usually employed in the recent descriptive grammars dealing with Tamil, and indeed with Indian languages in general. The writing system is, for the most part, ‘syllabic’. A basic premise in this dissertation is that attention to the design and the workings of the writing system may provide insights into the ancient grammarians’ view of the syllable structure of the language. In the datasets presented here, I have included, where appropriate, representations in the original script in addition to the IPA representations.
2.1 The writing system
The writing system of Tamil is designed to represent all of the phoneme inventory that is required to cover the full range of underlying representations in the native lexicon without redundancy. In general the graphemes for consonants stand for place of articulation and phonemic manner characteristics. Context sensitive allophonic variants are not separately represented in the writing system, and the allophone that finds representation in the writing system has traditionally been regarded as the ‘underlying representation’.
If one takes the traditional view of syllable structure, the script has elements of the alphabetic and syllabic systems. An individual grapheme may represent one or more segments. The script is written left to right, and top to bottom.
I have attempted to base the classification of the graphemes directly on the classical grammatical text of Tamil (The tolka: piyam), rather than on current pedagogical charts. The classificatory principles in the older text seem to more closely reflect a rigorous phonological analysis of the language. The other reference source used was Bright, 1998.
The graphemes can be grouped into three classes:
(1) single vowel segments, long vowels or diphthongs (i.e. V, V: or V 1V2)
(2) consonant-vowel compounds (i.e. CV, CV: or CV1V2)
(3) single consonant segments.
(There are no graphemes or ligatures in Tamil to represent consonant clusters.)
2.1.1.
Class 1: Graphemes representing single vowels or vowels-only sequences (i.e. V, V:, or V1V2)
Qualitatively, there is a five-way vowel contrast. All vowels have a two way quantitative contrast. There are two diphthongs. This allows for twelve phonemic vowel formations. When the vowels occur word initially (i.e. in zero-onset syllables) each of these twelve vowel formations is represented in the writing system by it own grapheme. The script never uses these vowel-only graphemes word-medially or word-finally.
அ [ɑ, æ] ஆ [ɑː] இ [i] ஈ [iː] உ [u] ஊ [uː] எ [e] ஏ [eː] ஜ [ai] ஒ [o] ஓ [oː] ஔ [au]
2.1.2. Class 2: Graphemes representing Consonant-vowel sequences (i.e. CV, CV: or CV1V2)
These account for the vast bulk of the graphemes in the script. They consist of a basic set and a larger derived set. The basic set consists of 18 consonant symbols, each incorporating the default or ‘inherent’ vowel |ɑ|
க ங ச ஞ ட ண த ந ப ம ய ர
kɑ ɲɑ cɑ ŋɑ ʈɑ ɳɑ t̪ ɑ n̪ ɑ pɑ mɑ yɑ ʀɑ
ல வ ழ ள ற ன
lɑ vɑ ʎɑ ɭɑ rɑ nɑ
The derived set is made up by combining the individual graphemes of the above set with linear diacritics specific for each of the eleven vowels (leaving out the | ɑ| or அ, which is inherent in the basic set). To illustrate:
கி ஙி சி ஞி டி ணி தி நி பி மி யி ரி
ki ɲi ci ŋi ʈi ɳi t̪ i n̪ i pi mi yi ʀi
லி வி ழி ளி றி னி
li vi ʎi ɭi ri ni
கா ஙா சா ஞா டா ணா தா நா பா மா யா ரா
kɑ: ɲɑ: cɑ: ŋɑ: ʈɑ: ɳɑ: t̪ ɑ: n̪ ɑː pɑː mɑː yɑː ʀɑː
லா வா ழா ளா றா னா
lɑː vɑː ʎɑː ɭɑː rɑː nɑː
and so on ……… The complete set of these graphemes is given in the Appendix.
This class of graphemes may appear at any position in the written word, though there are restrictions at the word initial and word final positions (see later). They obviously cannot contribute any consonant segments to word final or ‘syllable closing’ positions.
Though Graphemes in this class represent combined consonant and vowel segments, the qualitative properties of the individual segments represented within each one remain distinct and unchanged ( Tolka:piyam chap 1, verses 10. 15).
2.1.3. Class 3: Graphemes representing single consonant segments
These are derived from the basic set of 18 graphemes in class 2, by adding a special diacritic which cancels the ‘inherent’ vowel |ɑ| , leaving graphemes with this diacritic to represent a single consonant segment. This ‘cancelling diacritic’ is a dot or ‘puɭɭi’placed above the grapheme, and the class of graphemes is known as ‘puɭɭi’ graphemes’, (The two graphemes at the end of the set are occasionally used for writing Sanskrit loan words.)
க் ங் ச் ஞ் ட் ண் த் ந் ப் ம் ய் ர்
k ɲ c ŋ ʈ ɳ t̪ n̪ p m y ʀ
ல் வ் ழ் ள் ற் ன் Sanskrit ( ஸ் ஷ் )
l v ʎ ɭ r n loans : ( s ʃ )
The ‘puɭɭi’graphemes for sonorants and glides may appear at word-end in written representations, and are phonetically realised in ‘high speech’ and in some colloquial dialects. All of the graphemes in this class may appear word-medially, to contribute to consonant clusters. Since the only consonant clusters allowed in Tamil are word medial and, in root or root-derived morphemes, consist of geminates or half geminates, word internal puɭɭi graphemes are usually followed immediately by a homorganic consonant-vowel grapheme from class 2. Puɭɭi graphemes never appear word initially.
It may be useful at this stage to summarise the positional correlations of the different classes of graphemes described above.
• Onset-less left edge positions can only be filled by the vowel-only (class 1) graphemes, and these graphemes are not used in any other word position.
• Consonant-vowel compound graphemes (class 2) may be used in any word position, but there are quality and quantity based restrictions at word initial positions and word final positions (see later).
• Single consonant segment graphemes (class 3) may occupy the right edge if the segment is a sonorant or glide. They may all be used word medially to form homorganic consonant clusters.
2.1.4. The writing system and syllable typology: the Eastern perspective
Taking the traditional approach to syllabification, and also taking into account the fact that vowel length is always phonemic in Tamil, the syllable repertoire in the language was listed in section 1.1.4.1 as V, VV, V1V2, CV, CVV, CV1V2, CVC, CVVC, CV1V2C and VC. In this traditional view, the Tamil writing system would be interpreted as an alphabetic- syllabic hybrid. But this interpretation is the result of imposing the ‘Western’ view of syllable structure on the facts of the Tamil writing system. An alternative is to treat the Puɭɭi grapheme as a complete representation of a ‘dull syllable’ i.e. an onset with unexpressed nucleus (C0). This alternative view would allow the writing system to be seen as fully syllabic, if we take every instance where a Puɭɭi grapheme appears as representing the presence of a dull syllable (C0). The complete syllable repertoire of the language then simplifies to V, VV, V1V2, CV, CVV, CV1V2, and C0. In terms of the three grapheme classes, class 1, class 2 and class 3 graphemes would simply represent the three core syllable types in the language i.e. onset-less syllables (V), CV syllables and the dull syllable (C0) respectively. We could make the assumption that this was indeed how the ancient Eastern grammarians understood the syllable structure of languages to be underlyingly represented. In this dissertation, this assumption is used to explore the question of word-end consonants and geminate consonants in Section III.
2.2 Phonotactic restrictions at specific positions
In the traditional view of syllable structure, word edges are taken to reflect syllable edges. This view is seen as self evident in monosyllabic words. The view is extended to polysyllabic words using the sonority sequencing principle to implement it in practice, particularly to deal with the syllabification of intervocalic consonant clusters. However, detailed and critical examination of the relationship between word edges and syllable edged shows that syllable edges, as defined by independent phonological phenomena, do not always align with word edges. These discrepancies have fundamental implications for theories of syllable structure. The following data-sets from Tamil are presented as the basis for a critical evaluation of the relationship between word edges and syllable edges.
The illustrative examples given here include cognate terms in the dialects included in this study. The ‘high’ variety of spoken Tamil is referred to here as Literary Tamil (LT). The colloquial variety of spoken Tamil in mainland South India and described in the reference sources used here (Schiffman, 1999; Asher 1985) is referred to as Spoken Tamil (ST). The spoken dialect in the Jaffna peninsula of Sri Lanka is referred to as Jaffna Tamil (JT). ‘High’ speech in South India and Sri Lanka may be taken as virtually indistinguishable.
2.2.1 Word initial positions
2.2.1.1 Initial Vowels. All vowels in the inventory are attested in the word initial position. Note however that in ST, vowel harmony consistently results in lowering of word initial high short vowels. Note also in ST the insertion of onset glides before mid vowels ( /y/ before |e| and /w/ before |o|). Onset-less left edge positions can only be represented in writing by the vowel-only (class 1) graphemes, and these graphemes are not used in any other word position. This suggests that Tamil bars V•V hiatus (see also section 3.1 for further discussion of this point).
—————-LT ST T
1. அம்மா ɑmmɑ ɑmmɑ ɑmmɑ mother
2. அடிமை ɑʈimai ɑʈime ɑʈimai slave
3. ஆடு ɑːʈu ɑːʈɨ ɑːʈɨ goat
4. ஆலயம் ɑːlɑyɑm ɑːlɑyõ ɑːlɑyɑm temple
5. இலை ilɑi ele ilai leaf
6. இனிப்பு inippu inippɨ inippɨ sweet
7. ஈனம் iːnam iːnõ iːnam disrespect
8. ஈ iː iː iː fly (insect)
9. உடம்பு uʈɑmbu oʈɑmbɨ uʈɑmbɨ body
10. ஊளன் uːɭɑn uːʈæ̃ uːɭɑn lame man
11. ஜந்து aint̪u aint̪ɨ aint̪ɨ five
12. எச்சம் eccɑm yeccõ eccɑm remainder
13. ஏலம் eːlɑm yeːlõ eːlɑm cardamom
2.2.1.2 Word initial consonants
For words which begin with an onset, the initial grapheme is always class 2 i.e. CV, CV: or CV1V2. It follows therefore that there are no word-initial consonant clusters in the native lexicon.
All the consonantal segments represented in the 18 graphemes in the basic set of class 2 graphemes may appear in the word initial position, with the notable exception of the apical segments ʈ, l, ɭ and r. These exceptions are still observed in LT, but ST, and to a lesser extent JT, are somewhat more tolerant in allowing these in loan words and also in native words when the word initial vowel is elided.
There are no distinctions in the writing system between homorganic voiced and unvoiced obstruents nor between stops and homorganic fricatives. These are always allophones as single segments. In Tamil, voicing is not phonemic. In speech, word initial obstruents in native words are unvoiced as are word-internal obstruent geminates. Intervocalically, single obstruent stops usually show lenition. In word internal half-geminates, the post-nasal obstruents are voiced, but this is not reflected in the orthographic representation. In loan words, voiced obstruents may be used word initially, particularly in ST. Speakers of LT or JT tend to use loans either as forms which are fully adapted to the native system or use the loans as forms which remain strictly true to their original foreign pronunciation (e.g. code switching between English and JT).
LT ST JT
14. காது ka:ðɨ ka:ðɨ ka:ðɨ ear
15. நெத்தி net̪ti net̪ti net̪ti forehead
16. நடித்தல் nɑʈit̪tɑl nɑʈit̪tɑl nɑʈit̪tɑl actinɡ
17. தோளன் t̪ oːɭɑn t̪ oːɭæ̃ t̪ oːɭɑn friend
18. பிரிவு pirivɨ pirivɨ pirivɨ separation
19. பொறுப்பு poruppɨ poruppɨ poruppɨ responsibility
20. முதுமை muðumai muðumɛ muðumai maturity
21. மிருகம் miruxɑm miruxõ miruxɑm animal
22. சேவல் ceːvɑl ceːvɑl ceːvɑl rooster
23. கிளவன் kiɭɑvɑn keɭavæ̃ kiɭavɑn old man
24. இரண்டு irɑnʈɨ rɑnʈɨ rɑnʈɨ two
The retroflex sounds | ɭ | and |ɖ| do not occur word initially.
2.2.2 Word Final Positions.
The word final segment in Tamil can be a vowel, a sonorant or, in a few instances, a glide. Obstruents never occur at word-end in the native vocabulary. In written representations, the word final grapheme is either class 2 ( for vowel endings) or the sonorant and glide subset of class 3.
2.2.2.1 Word final Vowels:
All the orthographically represented long vowels except the diphthong |au| are attested at the word final position. There was a range of restrictions on the short vowels that may appear word finally in old Tamil (Tolka:piyam), but the only remaining restriction in modern Tamil seems to be the absence of the short vowels |e| and |o|. The orthographic vowel |u| is nearly always realised as /ɨ/ in the word-final position in ST, JT and even in LT except in more formal speech. Of the long vowels, |u:| is attested only as a vocative; |a:|, |e:| and |o:| are commonly used suffixes serving respectively as the yes/no question morpheme, emphatic particle and ‘dubitative’ question morpheme. The diphthong |ai| is the suffix marking accusative case, but also appears in some underived forms and in forms derived by root level morhology.
Long vowels are uncommon in the word-final position in underived forms and in forms derived by root-level morphology.
LT ST JT
25. காது ka:ðɨ ka:ðɨ ka:ðɨ ear
26. உடுப்பு uʈuppu uʈuppɨ uʈuppɨ clothing
27. பிறவி pirɑvi perɑvi pirɑvi incarnation
28. பல்லி palli palli palli lizard
29. நட nɑʈɑ nɑʈɑ nɑʈɑ walk (imp.)
30. அவன் ɑvɑn ɑvæ̃ ɑvɑn (2nd per.M.sing Nom)
31. அவனா? ɑvɑnɑː ɑvɑnɑː ɑvɑnɑː is it him?
32. அவள் ɑvaɭ ɑvæ ɑvaɭ (2nd per.F.sing Nom)
33. அவளா? ɑvɑɭɑː ɑvɑɭɑː ɑvɑɭɑː is it her?
34. அவனோ? ɑvɑnoː ɑvɑnoː ɑvɑnoː could it be him?
35. அவனே! ɑvɑneː ɑvɑneː ɑvɑneː it is him!
36. அவனை ɑvɑnɑi ɑvɑnɛ ɑvɑnɑi him (acc)
37. முதுமை muðumai muðumɛ muðumai maturity
38. இலை ilɑi ele ilai leaf
39. நாடா nɑːʈɑː nɑːʈɑː nɑːʈɑː tape (noun)
2.2.2.2 Word final Consonants:
The design of the writing system implies that word-final open syllables are represented by a grapheme of class 2, and that all words which have a final consonant segment end with a ‘puɭɭi’grapheme. Only ‘puɭɭi’graphemes for sonorants and glides appear at word-end in written representations, and are always phonetically realised in ‘high speech’. In ST however, word final nasals are ‘deleted’ and word final liquids have an ‘epenthetic’ /ɨ/ added. The sounds that can appear as word final consonants are:
ன் ண ம் ல் ள் ர் ழ் ய்
n ɳ m l ɭ r ʎ y
Obstruents never appear as word-final segments in Tamil.
There is a strong tendency in ST for all surface forms to appear without the word-final consonantal sound. Where the ‘underlying form’ does not end in a vowel, most colloquial dialects (particularly ST) achieve a CV ending by /ɨ / epenthesis for word final liquids, or, for word final nasals, by final consonant deletion and ‘vowel nasalisation’. (This is the conventional analysis. An alternative analysis is presented in section III).
This tendency is less strong in JT: the word final nasals |m|, |ɳ| and |n| are usually retained as word final segments in surface forms, as is |r|. The laterals | l| and |ɭ| have an epenthetic vowel added in monosyllabic words with short vowels, but are left unchanged when they follow long vowels and in polysyllabic underived or root-derived forms.
LT ST JT
40. கல் kɑl kɑllɨ kɑllɨ stone
41. கால் kɑːl kɑːlɨ kɑːl leg
42. சொல் col collɨ collɨ word
43. தோல் t̪ oːl t̪ oːlɨ t̪ oːl skin
44. பல் pɑl pɑllɨ pɑlllɨ tooth
45. பால் paːl paːlɨ paːl milk
46. கோவில் koːvil koːvil koːvil temple/church
47. பிரிதல் piriðɑl piriðɑl piriðɑl separation
48. நடித்தல் nɑʈit̪tɑl nɑʈit̪tɑl nɑʈit̪tɑl actinɡ
49. கள் kɑɭ kɑɭɭɨ kɑɭɭɨ toddy
50. தேள் t̪ eːɭ t̪ eːɭɨ t̪ eːɭ centipede
51. அவள் ɑvaɭ ɑvæ ɑvaɭ 2ndperFsingNom
52. தின் t̪ in t̪ innɨ t̪ in eat (imperative)
53. மீன் mi:n mi:nɨ mi:n fish (noun)
54. அவன் ɑvɑn ɑvæ̃ ɑvɑn 2nd perMsing Nom
55. ஊளன் uːɭɑn uːɭæ̃ uːɭɑn lame man
56. கிளவன் kiɭɑvɑn keɭavæ̃ kiɭavɑn old man
57. மண் Maɳ Maɳɳɨ Maɳ soil/earth
58. வீண் viːɳ viːɳɨ viːɳ waste
59. மரணம் mɑrɑnɑm mɑrɑnõ mɑrɑnɑm death
60. ஏலம் eːlɑm yeːlõ eːlɑm cardamom
61. மரம் mɑrɑm mɑrõ mɑrɑm tree
62. ஆம் ɑːm ɑːmɘ oːm yes
63. அவர் ɑvɑr ɑvɑrɨ ɑvɑr He (polite)
64. ஈர் iːr iːrɨ iːr louse
65. தேர் t̪ eːr t̪ eːrɨ t̪ eːr chariot
66. துளிர் t̪ ulir t̪ ulir t̪ ulir new shoot
67. தமிழ் t̪ɑmɨɭ t̪ɑmɨɭ t̪ɑmɨɭ Tamil
68. கைய் kɑy kɑyi kɑy hand
69. நாய் nɑːy nɑːyi nɑːy dog
2.2.3 Word internal consonants
2.2.3.1 Word internal Single consonant segments
Any of the consonant sounds described above may appear intervocalically as single segments. These single C segments are represented in the writing system as the initial part of the compound (class 2) graphemes each of which represent a unique CV, CVV or CV1V2 sequence. These single intervocaliv consonants are parsed unproblematically as onsets. Words which lack any word-internal consonant clusters ( ie a sequence of CV syllables) will therefore consist merely of a sequence of class 2 graphemes. Words which begin with onset-less syllables start with a class 1 grapheme (V, VV or V1V2), and words which end with sonorant consonant segment or a glide will terminate in a class 3 grapheme).
Intervocalic single obstruent stops always undergo lenition (e.g. P β or b; t̪ → θ or ð;
k → x or h). A more detailed discussion of lenition is given in Section III.
LT ST JT
70. பகுதி pɑxuði pɑxiði pɑxuði portion
71. விதவை viðɑvɑi viðɑvɛ viðɑvɑi widow
72. பரிதாபம் pɑriθaːβɑm pɑriθaːbõ pɑriθaːβɑm pathos
73. கிளிசகேடு kilicɑxeːʈɨ kilicɑxeːʈɨ kilicɑxeːʈɨ disgrace
74. தொளிலாளி t̪ oɭilɑːɭi t̪ oɭilɑːɭi t̪ oɭilɑːɭi workman
75. தவறணை t̪ ɑvɑrɑnɑi t̪ ɑvɑrɑnɛ t̪ ɑvɑrɑnɑi taverna (loan)
76. மகிமை mɑhimɑi mɑhimɛ mɑhimɑi glory
77. பிரிவு pirivɨ pirivɨ pirivɨ separation
78. முதுமை muðumai muðumɛ muðumai maturity
79. அடிமை ɑʈimai ɑʈime ɑʈimai slave
80. அளு ɑɭu ɑɭɨ ɑɭɨ weep
81. ஆடு ɑːʈu ɑːʈɨ ɑːʈɨ goat
82. உணவு uɳɑvu oɳɘvɨ uɳɑvɨ food
83. ஒளுகு oɭuxu woɭuxɨ oɭuxɨ leak
84. அனுபவம் ɑnuβɑvɑm ɑnubɑvõ ɑnuβɑvɑm experience
85. மரணம் mɑrɑnɑm mɑrɑnõ mɑrɑnɑm death
86. கிளவன் kiɭɑvɑn keɭavæ̃ kiɭavɑn old man
87. துளிர் t̪ ulir t̪ ulir t̪ ulir new shoot
2.2.4 Consonant clusters in Tamil
Consonant clusters do not occur word initially. This is also assumed to be case word finally and the orthography supports this view, but there is some reason to believe that phonemic consonant length contrast may occur in the word final position (further discussed in 3.6).
In the writing system, graphemes of class 3 (‘ pulli’ graphemes or consonant segments without the ‘inherent’ vowel), are never found at the left edge, and can only occur singly at the right edge. Word-medial C clusters require the interposing of a class 3 grapheme between two class 2 graphemes or between a leading class 1 grapheme and a following class 2 grapheme.
Word medial consonant clusters are common in Tamil, and close analysis of these clusters is crucial to the understanding of syllabification in Tamil. There is no dialect variation affecting word medial consonant clusters, and the examples choosen to illustrate this section are therefore not listed by dialects, and are mostly taken from JT.
Consonant clusters in Tamil may be:
1. Long consonants (‘geminate’) of obstuents
2. Long consonants (‘geminate’) of sonorants
3. non-homorganic two-segment clusters
4. three-segment clusters
Consonant clusters in underived or root derived forms in the native lexicon and in loan words which have been completely adapted to tamil phonology are geminates or are homorganic. It will be argued in this dissertation that obstruent geminates are tautomorphemic and may syllabify as C0•CV but non-obstruent geminates are heterosyllabic in the conventional sense. However, morpheme boundaries are often unclear. Non-homorganic clusters are usually demonstrably heteromorphemic and heterosyllabic.
The relationship between long consonants and preceding vowel length is of particular interest when considering syllable structure, and is explored in greater detail in section 3.5.
Section III
3.0 Syllable structure of Tamil and implications for theories regarding the syllabic status of word-end consonants and geminates.
3.1 Syllable typology of Tamil: an overview
As may be discerned from the segmental restrictions demonstrated in the datasets relating to word initial, word final and word medial positions, Tamil does not allow complex onsets or codas. It has also been demonstrated in the datasets above and discussed in section (1.1.4) that Tamil freely exploits length contrasts in vowels, allowing branching syllable nuclei consisting of long vowels or diphthongs. The syllable types that we may expect to encounter in the language are therefore {V, VV, V1V2, CV, CVV, CV1V2, CVC, CVVC, CV1V2C, VC}. I have presumed that long vowels are tautosyllabic in Tamil on the basis of how the writing system uses the different classes of graphemes in strings. As explained in section 2.2.1.1, class 1 graphemes are never used word internally, suggesting that Tamil bars V•V hiatus. Closed syllable shortening effects are seen preceding sonorant geminates, and this issue is discussed in 3.5.2
In the traditional view, the three primary rules of syllabification described in section 1.1.1 are sufficient to achieve syllabification of the native words and word strings. It has been noted in previous accounts of the grammar of Tamil (Krishnamurti, 2003) that in free roots or words derived by root level morphology, the preference is for forms with three morae which is achieved by the following restricted set of syllable strings: (C)V:•CV / (C)VC•CV / (C)V•CV•CV. However, our data sets show that there are also free forms, apparently without any morpheme boundaries visible to phonology, which are of the form (C)V:C:V where Tamil orthography and modern grammatical transliterations represent the long consonants as geminates. As mentioned before (1.1.1.4), consonant length contrast is phonemic in Tamil and is used extensively and independently of vowel length contrast (except when preceding sonorant geminates, see 3.5.2), so these forms are not infrequent in the word stock of the language. The problem of syllabifying strings of this type (i.e. (C)V:C•CV Vs (C)V:•CCV or (C)V:•C0•CV) is discussed in 3.6.
Words of the form (C)VC:V: and (C)V:C:V: are also frequently in use, but the word final long vowel in these words are suffixal vocative or emphatic markers or other grammatical markers and so the words may escape the three morae rule for free roots. However, syllabification of these forms is still debatable. The form (C)VC:V: may syllabify as (C)VC•CVV with four morae or as (C)V•CCVV or (C)V•C0•CVV with three morae. The form (C)VVCCVV may syllabify as (C)VVC•CVV with five morae and a superheavy syllable or as (C)VV•CCVV or (C)VV•C0•CVV with four morae and no super heavy syllables.
A further observation from our datasets is that in LT, and in a few cases in JT, apparently monosyllabic free forms of the type CVC and CVVC are encountered. The word final C in these forms is a nasal or lateral (but not |m|). Taking a rule based derivational view, a preferred syllable string is achieved unproblematically for the forms |CVVC| by / ɨ/ epenthesis and resyllabification: CVVC CVV•Cɨ. This however, does not seem to hold for free forms of the CVC type. Here, again on the conventional derivational view, epenthesis involves an apparent gemination (C insertion) of the word final C in addition to /ɨ/ epenthesis: CVC1 CVC1C1ɨ. Alternative views on how to account for the apparent consonant doubling and the implications for syllabification are discussed in section 3.6.
3.2 The Tamil writing system and the word-final onset hypothesis
Proponents of the view that the word-final consonant is in fact the syllable onset of a word final dull syllable have dubbed this view the ‘eastern prospect’, as distinct from the conventional ‘western prospect’ which views the word final consonant as occupying the coda position of the word final syllable. There are of course several independent lines of evidence in support of the word-final onset view (Harris and Gussmann, 2002) but an arresting and thematic claim is that this view is embodied in the writing systems of the Eastern languages, including the Brahmi derived scripts used by the Dravidian languages of south India. Harris and Gussmann support this claim with a brief illustration from the Ethiopic Fidal writing system, but do not explore it in much detail. It is of course not claimed that exploring the details of a writing system would in itself provide us with robust and objective linguistic evidence. However, we could make the assumption that the writing system does indeed mirror the ancient Eastern grammarians’ understanding of how the syllable structure of the language is underlyingly represented. In this dissertation this assumption has been used as a touchstone in exploring the question of word-end consonants and geminate consonants.
Implicit in the assertion that the Eastern writing systems embody the concept of the word final onset, there appears to be two assumptions: that each and every grapheme stands for an entire syllable and that there are a set of discrete and individual graphemes which are used to denote the word final dull syllables.
The Graphemes for pure vowels (class 1 ) clearly stand for Vowel-only syllables (i.e. syllables of the type V, V: and V1V2) in their entirety. This syllable type is only attested word initially, and these graphemes are used only in this position (examples 88-92)
Compound graphemes of class 2 are used to represent syllables of the CV, CVV and C1V2 type in their entirety, as illustrated in the examples 93-97 and in all the non-initial syllables in 88-92.
The graphemes of class 3 represent single consonant segments, and are derived from the corresponding class 2 grapheme by adding the ‘pulli’ diacritic which indicates the cancellation of the default |ɑ| vowel e.g க (kɑ) க் (k) ; ம(mɑ) ம்(m);
ள (ɭɑ) ள்(ɭ). These graphemes are used to represent the word final single consonant segment in examples 98-107 (from JT).
88. அ•டி•மை ɑ•ʈi•mai slave
89. ஆ•டு ɑː•ʈɨ goat
90. இ•லை i•lai leaf
91. ஈ iː fly (insect)
92. உ•ண•வு u•ɳɑ•vɨ food
93. க•ள•வு kɑ•ɭɑ•vɨ theft
94. ம•னை•வி mɑ•nɑi•vi wife
95. கா•து ka:•ðɨ ear
96. தொ•ளி•லா•ளி t̪ o•ɭi•lɑː•ɭi workman
97. பி•ரி•வு pi•ri•vɨ separation
98. ஆ•ல•யம் ɑː•lɑ•yɑm temple
99. ஈ•னம் iː•nam disrespect
100. கி•ள•வன் ki•ɭa•vɑn old man
101. ஊ•ளன் u:•ɭɑn lame man
102. ஏ•லம் e:•lɑm cardamom
103. ம•ர•ணம் mɑ•rɑ•nɑm death
104. தோ•ளன் t̪ o:•ɭɑn friend
105. தின் t̪ in eat (imperative)
106. மீன் mi:n fish (noun)
107. கால் kɑ∶l leg
So far, the two assumptions that each and every grapheme corresponds to an entire syllable and that there are a set of discrete and individual graphemes which denote the word final dull syllables seems to be borne out. But this is not the whole story, since the above examples do not include any word internal syllable-closing consonants or long consonants (geminates). Consider examples 108-116, where the non-final (C)VC syllables are formed by two graphemes: a class 1 (vowel only) grapheme or a class 2 (compound) grapheme which stands for [(onset)+nucleus], and a class 3 ‘pulli’ grapheme which stands for what appears to be the syllable closing single consonant segment. These examples suggest firstly that there is not necessarily a one-to-one correspondence between graphemes and syllables and secondly that the ‘pulli’ grapheme is not exclusively used to stand for the word final dull syllable.
108. உ•டம்•பு u•ʈɑm•bɨ body
109. இ•ரண்•டு i•rɑn•ʈɨ two
110. அ•ம்பு am•bɨ arrow
111. தம்•பி t̪ɑm•bi younɡer brother
112. அன்•பு æn•bɨ affection
113. மண்•டி man•di sediment/dreɡs
114. சங்•கு cɑŋ•ɡɨ conch shell
115. தந்•தம் t̪ɑn•̪t̪ɑm tooth/tusk
116. வேண்•டு•தல் ve∶ɳ• du•ðɑl request (noun)
The above examples show syllabification on the traditional assumtions of seeing the word final consonant as a coda. Alternatively, we could claim that the ‘pulli’ graphemes stand for a dull syllable (C0 ) both word finally and word internally. Example 110, in this view, would syllabify as: அ•ம்•பு a• m•bɨ (V•C0• CV). And example 116 would syllabify as: வே•ண்•டு•தல் ve∶•ɳ•du•ðɑl (V:• C0• CV•CV • C0 ). This implies that the internal syllables are in fact not closed and so explains why there is no ‘closed syllable shortening’. Consider the further examples 117-134 which contain long consonants (geminates). Some examples also show lenition (fricativisation or flapping) of non-initial obstruent-stop onsets (128-134). The ‘weaker’ forms are allophonic with the word initial stops, and are not differentiated in the orthography.
117. தொக்கை t̪ okkɑi thick
118. பாக்கம் pɑːkkɑm district of town
119. அப்பா ɑppɑː father
120. தாத்தா t̪ɑːt̪taː grand father
121. இராத்திரி irɑːt̪t̪iri niɡht
122. கிளாக்கு kɨɭɑːkkɨ clerk
123. தும்மல் t̪ummɑl sneezinɡ (gerund)
124. அன்னம் ænnɑm swan
125. திண்ணை t̪iɳɳai verandah
126. கள்ளன் kɑɭɭɑn thief
127. முல்லை mullai jasmine
128. பகைத்தல் pɑxɑit̪t̪ɑl hatred
129. மிதித்தல் miðit̪tɑl trampling (gerund)
130. அபிப்பிராயம் ɑβippirayam opinion
131. தொகுப்பு t̪oxuppɨ classification
132. கட்டிடம் kɑʈʈiʈɑm buildinɡ (gerund)
133. நடித்தல் nɑʈit̪tɑl actinɡ (gerund)
134. உடுப்பு uʈuppɨ clothes
Some questions arise from the above examples. Can the geminate of [pulli grapheme+the consonant part of the following compound grapheme] be represented as a single segment in the melodic tier with double linking to the skeletal tier? How are the geminates to be syllabified – is the pulli grapheme the coda of the preceding vowel nucleus or part of the (complex) onset for the following, or is it indeed a syllable of its own (C0 •)? If the pulli grapheme is taken as the first segment of the geminate, and in view of the notion of inseparability of geminates (Kenstowicz 1994, page 410), can an onset ‘pulli’ grapheme in this context contain a silent vowel which is interposed between the geminates? And finally, in view of the obstruent lenition data, is the pulli grapheme used here to indicate the fact that the word internal lenition of obstruent-stops is ‘blocked’, allowing a ‘strong’consonant realisation with lexical contrast?
3.2.1 The ‘puɭɭi’ grapheme as the Dull Syllable
A simple statement that would seem to capture the facts regarding the distribution of ‘pulli’ graphemes would be that they are only used to represent syllable closing consonants, word internally or at word end. But this generalisation would be countered by the fact that obstruent geminates fail to impose length restrictions on the preceding vowel (see 3.5.1) and so the ‘pulli’ graphemes do not always seem to close the preceding syllable. This raises the possibility that pulli graphemes of obstruent stops may in fact be part of a complex (geminate) onset of the following syllable, but this loses the strong generalisation that Tamil does not have complex onsets.
This leads us to adopt the Eastern perspective: that the ‘pulli’ graphemes represent dull syllables i.e. an onset consonant segment and a silent vowel. This view provides a more comprehensive and more elegant and principled explanation for most of facts relating to the distribution of the graphemes. The one observation that seems to fall outside this generalisation is related to the sonorant geminate, the leading ‘pulli’ segment of which does seem to close the preceding syllable, as suggested by the fact that long vowels are not attested in the relevant nucleus (See 3.5.2.).
Another salient point, which strongly suggests that ‘puɭɭi’ graphemes are not confined to representing word-end consonants, is that the full set of ‘Pulli’ graphemes contains obstruent stops, a category which never appears as word end consonants.
The putative syllabic representational role of the ‘puɭɭi’ grapheme is further discussed in sections 3.5.0 to 3.6.
The above survey then seems to show that the design of the writing system reflects in many respects the ‘Eastern perspective’ of the underlying syllable structure of the language. The classical grammatical text of Tamil, the Tol ka:piyam, does not make any obvious references to the notion of syllables in its extensive section on the writing system. Rather, it seems to emphasise the discrete segmental nature of what the graphemes embody. For example:
மெய்யோ டியையினும் உயிரியல் திரியா.
“ Even though it may stand together with the consonant (in a compound letter), the nature of the vowel does not change”
(Chapter1, verse10)
மெய்யின் இயற்கை புள்ளியௌடு நிலையல்.
“ The nature of the consonant is as (seen) when it stands with the ‘pulli’ ”
(Chapter1 verse15).
However, the text is written in Old Tamil verse, and the commentaries and translations that are available interpret it as largely descriptive or prescriptive. Its insights from the ‘generative grammar’ standpoint could easily be missed in these commentaries and translations. The sequential arrangement and presentation of the material in this classical text certainly suggest a strikingly modern approach to the phonology and morphosyntax of the language.
3.3 Syllable Typology of Tamil dialects and the word-end consonant.
One of the most obvious dialectical differences between literary Tamil (LT) and south Indian spoken Tamil (ST) is with regard to the word-end consonant. LT clearly allows word end nasals, liquids and glides though not obstruent stops. Many recent descriptive grammars blandly assert that there are no word-end consonants in modern ST, but this is an oversimplification in many respects, as we will see. Jaffna Tamil (JT) does allow many word-end consonants, but patterns with ST in some respects. We may well expect dialectical differences in this position to be relevant to an assessment of the final onset view of word end consonants. It is worthwhile therefore to delineate these differences in some detail. Consider first words ending in the nasals |m|, |n| and |ɳ| in examples 135-143
LT ST JT
135. தின் t̪ in t̪ innɨ t̪ in eat (imperative)
136. மீன் mi:n mi:nɨ mi:n fish (noun)
137. அவன் ɑvɑn ɑvæ̃ ɑvɑn (2nd per.M.sing Nom
138. கிளவன் kiɭɑvɑn keɭavæ̃ kiɭavɑn old man
139. பெண் peɳ peɳɳɨ peɳ woman
140. வீண் viːɳ viːɳɨ viːɳ waste
141. ஆம் ɑːm ɑːmɘ oːm yes
142. மரணம் mɑrɑnɑm mɑrɑnõ mɑrɑnɑm death
143. ஏலம் eːlɑm eːlõ eːlɑm cardamom
In LT and in colloquial spoken JT the word-end nasals are given full phonetic realisation as the terminal segments. In ST however, we see what the descriptive grammars call ‘ u’ epenthesis (realised actually as /ɨ/) in the case of monosyllabic roots, and, in the case of the words of more than one syllable, we see what the descriptive grammars refer to as “consonant deletion and nasalisation of the vowel”. Alternatively, the ‘epenthesis’ may of course be viewed as giving voice to the nucleus of the dull syllable (C0). And an alternative view of the ‘deletion and nasalisation’ may be to see it as a form of lenition of the nasals in a weak position where the stop element only is deleted but the lip rounding and the velar opening are retained, leaving behind nasality as the main segmental feature i.e. there is no complete deletion of the word final segment but the consonantal feature is lost.
Consider next words ending in the liquids |l |, |ɭ |, |r | and the glide |y| ). In all of the examples 144-156, there is vowel ‘epenthesis’ or ‘voicing of the dull syllable nucleus’ in ST. In the bisyllabic word (156) with the retroflex lateral ending, ST shows lenition, somewhat similar to the situation with the nasals. In JT, only the short closed monosyllabic word are different from the LT forms, and these show ‘vowel epenthesis’ (see section 3.6).
[ Interestingly, examples 154 and 155 suggest that ‘ay’ is not a true diphthong in the sense of a branching nucleus, but VC. This view is in fact specifically endorsed in the Tolka:piyam, Chap 2, verses 21 and 22 with respect to ‘ai’ and ‘au’]
LT ST JT
144. கல் kɑl kɑllɨ kɑllɨ stone
145. கால் kɑːl kɑːlɨ kɑːl leg
146. பல் pɑl pɑllɨ pɑlllɨ tooth
147. பால் paːl paːlɨ paːl milk
148. பகல் pɑxal pɑxal pɑxal day
149. கள் kɑɭ kɑɭɭɨ kɑɭɭɨ toddy
150. தேள் t̪ eːɭ t̪ eːɭɨ t̪ eːɭ centipede
151. அவர் ɑvɑr ɑvɑrɨ ɑvɑr He (polite)
152. ஈர் iːr iːrɨ iːr louse
153. துளிர் t̪ ulir t̪ ulirɨ t̪ ulir new shoot
154. கைய் kɑy kɑyi kɑy hand
155. நாய் nɑːy nɑːyi nɑːy dog
156. அவள் ɑvaɭ ɑvæ̃ ɑvaɭ 2nd per.F.sing Nom
In ST therefore, there is, at the very least, a strong tendency for words not to end in a segment containing the (-continuant) feature value, i.e. for the word final syllable to be realised as CV. How is this tendency to be explained? In support of the final onset hypothesis, it is suggested that the preference as to whether or not to allow word final consonants is parametric and is independent of whether or not the internal grammar allows codas word internally (See Harris, 2002). Languages like Italian, which allow CVC syllables word internally but only CV syllables word finally are sited as examples. Dialect differences are well recognised for affecting single parameters, leaving other related parametric choices unaffected. This seems to be the case for the differences noted in these examples, since the syllable templates in the dialects are otherwise the same. The dialect differences at the word final position in Tamil therefore provides strong support for the view that the word final C need not parallel the word internal codas i,e, could be a dull onset (C0).
3.4 The final onset hypothesis: evidence from Segmental restrictions and lenition
Comparing the data sets presented for illustrating word initial consonants and word final consonants (2.2.1.2 and 2.2.2.2 respectively), it is easy to see the asymmetry in their segmental quality. One may naively assume an alignment between word initial consonants and word medial syllable onsets, and a parralel relation between word final consonants and word medial codas. The previous grammatical descriptions (Schiffman, 2003, and Asher 1985) and Tamil orthography relating to intervocalic single consonant position show that this position, which would be expected to syllabify as onsets, is unrestricted with regard to segmental quality.
The restriction on word-end consonants mimics restrictions on presumed word-internal codas in terms of segmental quality, unless geminate obstruents are taken as heterosyllabic between a preceding CVC syllable and a following CV(C) syllable (see 3.5.1. for arguments agains this). Internal codas, if taken to be word internal consonant segments which close CVC syllables, are always sonorants or the glide |y| in Tamil, as are word final consonants. The figure below (relating to LT) shows that in terms of segmental quality, there is (1) identity between word final consonants and internal codas, (2) partial overlap between word final consonants and word internal onsets and (3) no overlap except for the nasals |m| and |n| between word final consonants and word initial onsets. This suggests an identity between word-final consonants and word internal codas, but, as the lack of complete identitity between word-initial consonants and word internal onsets shows, matching in terms of segmental identity does not in itself constitute evidence for identitity in terms of syllable constituency.
Figure 1
Word initial consonants
k c t̪ p y v m n
word internal codas
m n ɳ l ɭ r ʎ y
word internal onsets
k c t̪ ʈ p y v m n ɳ l ɭ r ʎ y
Word final consonants
m n ɳ l ɭ r ʎ y
The only systematic lenition that is observed affects the ‘weak positions’ of word- internal non-geminate obstruent onsets and word-end nasals. Examples 156-168 below show lenition of intervocalic singleton obstruent segments which the orthography invariably represents as unaspirated stops. Examples 169-175 illustrate lenition of word-end nasals, seen only in ST. The lenition takes the form of loss of the consonantal feature, leaving nasality behind as the residual segmental feature. In articulating the residual |m| segment, there is also significant lip rounding
LT ST JT
156. பகுதி pɑxuði pɑxɨði pɑxuði portion
157. விதவை viðɑvɑi viðɑvɛ viðɑvɑi widow
158. பரிதாபம் pɑriθaːβɑm pɑriθaːbõ pɑriθaːβɑm pathos
159. கிளிசகேடு kilisɑxeːʈɨ kilisɑxeːʈɨ kilisɑxeːʈɨ disgrace
160. மகிமை mɑhimɑi mɑhimɛ mɑhimɑi glory
161. முதுமை muðumai muðumɛ muðumai maturity
162. பகைத்தல் pɑxɑit̪t̪ɑl hatred
163. மிதித்தல் miðit̪t̪ɑl trampling
164. அபிப்பிராயம் ɑβippirayam opinion
165. தொகுப்பு t̪oxuppɨ classification
166. கட்டிடம் kɑʈʈiʈɑm buildinɡ
167. நடித்தல் nɑʈit̪tɑl actinɡ
168. உடுப்பு uʈuppɨ clothes
169. அவன் ɑvɑn ɑvæ̃ ɑvɑn 2nd perMsing Nom
170. ஊளன் uːɭɑn uːɭæ̃ uːɭɑn lame man
171. கிளவன் kiɭɑvɑn keɭavæ̃ kiɭavɑn old man
172. மரணம் mɑrɑnɑm mɑrɑnõ mɑrɑnɑm death
173. ஏலம் eːlɑm yeːlõ eːlɑm cardamom
174. மரம் mɑrɑm mɑrõ mɑrɑm tree
175. ஆம் ɑːm ɑːmɘ oːm yes
If one takes the stand that lenition targets weak syllable positions independently of segmental quality, this observation may be construed as evidence for the final onset view in the sense that it unifies word internal onsets with word-end consonants. As possible objection to this argument, one may posit a weak domain encompassing word internal codas and word end consonants, but that lenition of word internal codas does not occur in Tamil since these are sonorants and lenition of sonorants in this position would have lexical and phonological consequences.
(For liquids, loss of the stop or constrictive element leaves only sonorancy behind, and sonorancy in itself is not easily perceptible as containing lexical information and neither can it serve as a syllable boundary. For nasals at word-end, loss of the stop element leaves behind nasality as the only remaining segmental feature, but this can be manifest allophonically as ‘nasalisation’ of the preceding vowel. Word internally, is it possible for a segment with nasality as its only feature to be perceptually significant or to serve as a syllable boundary?). This limitation on the possibility of lenition also applies to sonorants in onset positions word internally, and may explain why lenition in word internal onset positions only affect obstruents. It needs to be pointed out however that Sonorant lenition is well attested in other languages (e.g. Sonorant → glide in Cibaeno Spanish) and that lexical consequences are frequently overlooked in sound change.
The evidence in this section suggests that in terms of qualitative segmental distribution, word-final consonants in Tamil could pattern with internal codas or internal onsets or with both. The available evidence from lenition would seem to unify word internal obstruent onsets and the word final nasals but in view of the limitations on the possibility of lenition of sonorants word internally, the strength of this evidence may be limited.
For word internal onsets, there are no restrictions on segmental quality. Obstruents never occur as word final consonants in Tamil. Word final consonants can only be sonorants or the glide |y|. Thus, if word final consonants are onsets, there is a severe segmental restriction on onsets in this position. But this restriction in itself is not an argument against word final consonants being onsets, since severe segmental restrictions also affect word-initial consonants which are clearly onsets (liquids and retroflex consonants are not found as word initial consonants). The word initial and word end consonants have a complementary distribution, and this may be rooted in the need to demarcate word boundaries.
3.5.0 Long consonants (‘geminates’) and vowel length in Tamil
Long consonants in Tamil may be usefully categorised as
i. Obstruents: [pː], [kː], [t̪ː], [ɖː], [cː]
ii. Nasals, laterals and glide: [nː], [mː], [lː], [ɭː ], [yː].
The distinction is made because these two categories show some interesting differences, but before we come to considering the two categories in detail, it is useful to briefly discuss some general issues relating to long consonants or geminates.
The term ‘geminate’ implies that these consonants are pairs of segments with identical feature bundles which occupy sequential positions. This view is also implied in the Tamil orthography as well as in the phonetic transliterations used by many modern linguists working with the language. While it is clear that consonant length contrast is phonemic and highly productive in Tamil (see 3.5.1.an 3.5.2. below for examples from JT), the question of how best to represent this length contrast is debatable, and this debate has important implications for syllabification.
It is well recognised cross linguistically that, compared to non-geminate consonant clusters, geminates display certain properties which suggest that their two segments are tightly bound ( Kenstowicz, 1994, pp 410-416). In the linear model of representation, some aspects of this tight bonding may be captured by a radical approach which posits a segmental feature (+/- long) so that long and short consonants are in effect regarded as segmental minimal pairs differing only in this feature. In the non-linear autosegmental model, this bonding may be captured by the linking of the relevant consonant sound represented as a single segment in the segmental tier to two successive slots in the skeletal tier and positing the Uniformity Condition which states that in order for a rule to change the feature content of a segment, every skeletal slot linked to it must also satisfy the rule (Schein and Steriade, 1986).
Though vowel length and consonant length are both contrastive in Tamil, the writing system of the language takes differing approaches to representing length in vowels and consonants. While long vowels have individual graphemic representations distinct from their corresponding short vowels, long consonants are represented by the presence of a homorganic class 3 stop grapheme (i.e. the consonant stripped of the inherent vowel) in front of a class 2 grapheme. The graphemic representation of long consonants may therefore appear to reflect a view that they are composed of two sequential segments. The syllabification of these geminate segments is another matter: in the ‘Eastern’ view, the leading segment represented by the ‘pulli’ grapheme could be construed as a ‘dull syllable’ i.e the geminate context could be —(C)V•C0•CV• —
3.5.1 Obstruent long consonants and vowel length
Phonetically, these obsturents are all realised as voiceless, unaspirated stops where the stop phase is held for perceptibly longer than in the corresponding ‘short’ consonant. The obstruent single consonants segments as stops are only found word initially. Intervocalic single consonants are found always as the spirantised or flapped allophone though this phonetic distinction is not reflected in the orthography. Consonant length is always contrastive in Tamil and some minimal pairs (from JT) are shown in 176-181 below, and further examples in 182-194 show that consonant length and preceding vowel length are independantly contrastive:
176. தொகை t̪ oxɑi amount
177. தொக்கை t̪ okkɑi thick
178. பாகம் pɑːxɑm portion
179. பாக்கம் pɑːkkɑm district of town
180. பசை pɑcɑi paste (noun)
181. பச்சை pɑccɑi green
182. கட்டு kɑʈʈɨ tie up
183. காட்டு kɑ:ʈʈɨ show
184. சட்டி cɑʈʈi pot
185. சாட்டி cɑ:ʈʈi coastal hamlet
186. மடக்கு mɑʈɑkkɨ fold (imp.)
187. மொட்டாக்கு moʈʈɑ:kkɨ veil
188. தொகுப்பு t̪oxuppɨ classification
189. தோப்பு t̪oːppɨ orchard
190. அப்பா ɑppɑː father
191. தாத்தா t̪ɑːt̪taː grand father
192. இராத்திரி irɑːt̪t̪iri niɡht
193. கிளாக்கு kɨɭɑːkkɨ clerk
194. கட்டிடம் kɑʈʈiʈɑm buildinɡ
If one accepts the view that syllable rimes are maximally binary branching, the obstruent long consonants in the above examples with preceding long vowels would be parsed either as as tautosyllabic onsets or the leading member of the geminates (represented by the ‘puɭɭi’ grapheme) could be taken to be in fact a dull syllable (C0 ). Argumrnts in favour of the latter view have been stated earlier (see 2.1.4 and 3.2.1) and in the section immediately above .
In the case of obstruent geminates with preceding short vowels, the leading consonant represented by the ‘puɭɭi’ grapheme could be parsed either as a coda consonant of a heavy closed syllable or as a dull syllable preceded by a light syllable. Take example 194:
கட்•டி•டம் kɑʈ•ʈi•ʈɑm CVC•CV•CVC
OR க•ட்•டி•டம் kɑ•ʈ•ʈi•ʈɑm CV•C0•CV•CVC
This of course implies that the ‘puɭɭi’grapheme could also be used to represent the coda consonant in word internal closed (heavy) syllables.
Another observation to note from the above data is that the long consonant ‘resisits’ the lenition of word internal obstruents, and in terms of phonetic quality are the same as word initial obstruents. Morphologically, it seems likely that obstruent geminates are tautomorphemic, and represent or at least historically originate as various affixes to indicate, for example, causatives or as formatives). These affixes are classically represented with long obstruent at their right edge e.g. -kkɨ, t̪ t̪a, ʈʈi (Tholka:piyam, Chapter 4)
3.5.2 Non-obstruent long consonants and vowel length
In terms of segmental quality, these are aligned with word-final consonants, but they are also included in the set of sounds which can be word internal onsets (see figure 1 in section 3.4). Phonetically, the ‘long’ and ‘short’ sounds differ in only in the length for which the characteristic periodic sound is held. The length contrast is phonemic as the examples 195-202 show, but unlike the case with obstruents, this contrast is not independent of preceding vowel length. Examples 203-208 show short vowels preceding long non-obstruent consonants, but long vowels are not attested in this context.
195. முலை mulai breast
196. முல்லை mullai jasmine
197. புளி puɭi sour (tamarind)
198. புள்ளி puɭɭi dot (mark)
199. பலி pɑli sacrifice
200. பல்லி pɑlli lizard (gecko)
201. சனி sani saturn
202. சன்னி sanni pneumonia
203. அம்மா ɑmmɑ mother
204. தும்மல் t̪ummɑl sneezinɡ
205. மின்னல் minnɑl lightning
206. திண்ணை t̪iɳɳai verandah
207. கள்ளன் kɑɭɭɑn thief
208. பிள்ளை piɭɭɑi child
An obvious way to account for this restriction would be presume that these geminates are heterosyllabic, with the leading segment occupying the coda position of the preceding syllable and contributing to its weight. An alternative explanation would be to propose that the first part of a sonorant geminate is actually syllabic, i.e. occupies a nuclear position, but this explanation has no support elsewhere in the corpus of evidence available in Tamil. What is clear however, is that an explanation based on equating the ‘puɭɭi’grapheme with a dull onset will not serve in this case.
There is thus a notable difference in the syllabification of word internal-geminates which is determined by segmental quality, and it seems that word-internal ‘puɭɭi’graphemes can in fact be used to represent syllable closing sonorant consonant segments in the context of sonorant geminates.
3.5.3. Three-consonant clusters such as the ones below (209-212) are uncommon. They usually start with the rhotic ர் or the palatal retroflex approximant ழ் (usually realised as ɭ). The extra consonantal segment here may originate as part of suffixal word level morphology as explained in section 3.5.1. For example, 212 may then be syllabified as ம•கிழ்•ச்•சி mɑ•xiɭ•c•ci i.e. of the form CV•CVC•C0•CV
209. தீர்ப்பு t̪iːrppɨ verdict
210. தீர்த்ம் t̪iːrt̪t̪am ablution
211. நேர்த்தி neːrt̪t̪i sacred vow
212. மகிழ்ச்சி mɑxiɭcci happiness
3.6 Vowel length and consonant length in monosyllabic free roots
As in other languages, Tamil has a limited number of free standing monosyllabic root words. These words are well known to be minimally bimoraic, ie are heavy syllables (see review in Kenstowicz, 1994, page 297-298). There are notable dialect differences in Tamil which affect word end consonants, summarised in section 2.2.2.2 The differences as seen in monosyllabic free forms are illustrated in the examples below:
LT ST JT
213. தின் t̪ in t̪ innɨ t̪ in/ t̪ innɨ eat (imperative)
214. மீன் mi:n mi:nɨ mi:n fish (noun)
215. கல் kɑl kɑllɨ kɑllɨ stone
216. கால் kɑːl kɑːlɨ kɑːl leg
217. பல் pɑl pɑllɨ pɑllɨ tooth
218. பால் paːl paːlɨ paːl milk
219. நில் n i l n i l lɨ n i l/ n i l lɨ stand (imp.)
220. மண் mɑɳ mɑɳɳɨ mɑɳ/mɑɳɳɨ earth/soil
221. பெண் peɳ peɳɳɨ peɳ/peɳɳɨ woman
222. வீண் viːɳ viːɳɨ viːɳ waste
223. கள் kɑɭ kɑɭɭɨ kɑɭɭɨ Palm sap/toddy
224. தேள் t̪ eːɭ t̪ eːɭɨ t̪ eːɭ centipede
225. ஈர் iːr iːrɨ iːr louse
226. கைய் kɑy kɑyi kɑy hand
227. நாய் nɑːy nɑːyi nɑːy dog
228. ஆம் ɑːm ɑːmɘ oːm yes
The parametric difference between LT and ST is clear cut: LT allows word-end consonants and ST does not. The vowel endings in ST is achieved by ‘epenthesis’, or, taking the final dull syllable view, it does not allow syllable nuclei to be silent. In ST, these roots are all tri-moraic (either CVV•CV or CVC•CV), and when one surveys the Dravidian languages as a whole, it would seem that the archetypal free root form in Dravidian is tri-moraic.
There are two points to observe: (1) In JT, word final consonant ending is the rule after long vowels. After short vowels however, most speakers of JT prefer a vowel ending in informal conversation. (2) in both ST and JT, vowel epenthesis after short vowels seems to be accompanied by ‘doubling’ of the final consonant. This ‘doubling’ is clearly phonetically perceptible, and is consistently noted in IPA transliteration in the published descriptive grammars. The spoken dialect is not traditionally written down in the script, but recent Tamil transcripts of colloquial Tamil, such as popular cinema and drama, represent these forms in the same way as other geminates, with a pulli letter followed by a class 2 compound letter, as in these examples:
229. kɑllɨ கல்லு stone ;
230. mɑɳɳɨ மண்ணு soil/earth
An explanation may be put forward on the basis of the common observation in LT that the word-end consonant in short-vowel monosyllabic free roots appears to be held for a perceptibly longer period than in intervocalic single consonants or in word-end consonants in long-vowel equivalents. It seems therefore that although the written representation for the word-end consonant in the LT forms of short-vowel monosyllabic roots consists of a pulli grapheme only, these CVC forms in LT are in fact underlyingly CVCC.
Assuming that LT short-vowel free roots actually syllabify as CVC•C0• has considerable explanatory power. It explains the observed phonetic facts of consonant length, it allows short-vowel free roots to be counted as heavy in LT, and it obviates the need to postulate a derivational step of ‘consonant insertion’. It may also explain why JT allows word-final consonants after long vowels but not after short vowels in monosyllabic free roots, if we assume that the former have the form CVV•C0• and the latter the form CVC• C0•, and that the latter syllable sequences is highly marked and regularly appears only in LT.
Concluding remarks
Taking the stand that the writing system of Tamil is fully syllabic and that it accurately reflects the underlying syllabic structure of the language has been shown to provide a clear perspective on a number of areas of current interest in phonology. In particular, regarding the ‘puɭɭi’ graphemes as representative of ‘dull onsets’ allows elegant accounting for diverse phenomena involving areas such as dialect differences at word end, gemination and the interdependence between vowel length and consonant length. However, it may be that the writing system does not use these graphemes exclusively to represent dull syllables. Tamil does seem to include in its syllable repertoire word internal syllables which are closed by sonorant consonant segments which are represented in writing by ‘puɭɭi’ graphemes. Fresh studies of the ancient grammatical texts from a generative stand-point may provide further insights into the original ‘Eastern perspective’.
Further in this study, data relating to dialects and dialect differences in Tamil have been shown to provide support for the final-onset hypothesis. Particularly strong evidence comes from dialect differences in syllable typology. Lenition phenomena in South Indian spoken Tamil (ST) seem to support a unity between word internal onsets and word-end nasals. The ‘dull onset’ hypothesis also provides explanations for the phenomena relating to vowel length and geminate consonants, but the full exploitation of these data is hampered by the lack of reliable data on syllable prominence or stress in Tamil.
Acknowledgements
I wish to record my gratitude to John Harris and Richard Breheny for encouragement and helpful discussions in choosing the subject area for this dissertation. John was very patient in guiding me towards an explicit focus for the work, and made many valuable comments, suggestions and hints on early and late drafts.
Many friends and acquaintances in London who are native speakers of Jaffna Tamil provided illustrative examples and checked my examples for authenticity. Finally, I must thank my wife Jacintha and the rest of the family for forbearance and support during the two years of the MA Linguistics course and particularly over this summer as I toiled over the dissertation.
Bibliography and references
Asher R.E. Croom Helm Descriptive Grammars: Tamil. (1985) Routledge, London, New York
Andronov M.S. A comparative Grammar of the Dravidian Languages. (2003) Harrassowitz Verlag, Weisbaden
Bright W. The Dravidian Scripts. In: Steever, S (Ed): The Dravidian languages(1998). Routledge, London, New York. Pages 40-71
Burrow T. and Emene ` au M.B. A Dravidian Etymological Dictionary (1984) Oxford University Press.
Ferguson, C. Diglossia. Word (1959) 15: 325-40.
Harris,J. and Gussmann E. (2002). Word-final onsets. UCL Working Papers in Linguistics 14. ISSN: 0956-7194
Harris J. The phonology of being understood: Further arguments against sonority. Lingua (2006) 116: 1483–1494.
Indrapala, K. The evolution of an ethnic identity : the Tamils in Sri Lanka c.300 BCE to c. 1200 CE (2005). M.V. Publications for the South Asian Studies Centre, Sydney.
Krishnamurti B. Cambridge Language Surveys: The Dravidian Languages. (2003) Cambridge University Press.
Kenstowicz, M. Phonology in Generative Grammar (1994), Blackwell, Cambridge MA and Oxford UK
Owens J. Arabic Sociolinguistics: Linguistique Arabe: Sociolinguistique et Histoire de la Langue. Arabica, (2001), 48: 419-469
Schein, B and Steriade,D. On geminates. Linguistic Inquiry (1986) 17: 691-744
Schiffman, H.F. A reference Grammar of Spoken Tamil. (1999) Cambridge University Press
Steever, S (Ed): The Dravidian languages. (1998) Routledge, London, New York
Tolka:piyam. Author Unknown. With comentary by Pulyoor Kesikan (1961) Pari Offset Printers, Chennai, India.
Appendix: The complete set of Graphemes in Tamil
Class 1: Graphemes representing single vowels or vowels-only sequences (i.e. V, V:, or V1V2)
அ [ɑ, æ] ஆ [ɑː] இ [i] ஈ [iː] உ [u] ஊ [uː] எ [e] ஏ [eː] ஜ [ai] ஒ [o] ஓ [oː] ஔ [au]
Class 2: Graphemes representing Consonant-vowel sequences (i.e. CV, CV: or CV1V2)
These account for the vast bulk of the graphemes in the script. They consist of a basic set and a larger derived set.
The basic set consists of 18 consonant symbols, each incorporating the default or ‘inherent’ vowel |ɑ|
க ங ச ஞ ட ண த ந ப ம ய ர
kɑ ɲɑ cɑ ŋɑ ʈɑ ɳɑ t̪ ɑ n̪ ɑ pɑ mɑ yɑ ʀɑ
ல வ ழ ள ற ன
lɑ vɑ ʎɑ ɭɑ rɑ nɑ
The derived set is made up by combining the individual graphemes of the above set with linear diacritics specific for each of the eleven vowels (leaving out the | ɑ| or அ, which is inherent in the basic set).
கா ஙா சா ஞா டா ணா தா நா பா மா யா ரா
kɑ: ɲɑ: cɑ: ŋɑ: ʈɑ: ɳɑ: t̪ ɑ: n̪ ɑː pɑː mɑː yɑː ʀɑː
லா வா ழா ளா றா னா
lɑː vɑː ʎɑː ɭɑː rɑː nɑː
கி ஙி சி ஞி டி ணி தி நி பி மி யி ரி
ki ɲi ci ŋi ʈi ɳi t̪ i n̪ i pi mi yi ʀi
லி வி ழி ளி றி னி
li vi ʎi ɭi ri ni
கீ ஙீ சீ ஞீ டீ ணீ தீ நீ பீ மீ யீ ரீ
ki: ɲi: ci: ŋi: ʈi: ɳi: t̪ i: n̪ i: pi: mi: yi: ʀi:
லீ வீ ழீ ளீ றீ னீ
li: vi: ʎi: ɭi: ri: ni:
________________________________________________________________________
கு ஙு சு ஞு டு ணு து நு பு மு யு ரு
ku ɲu cu ŋu ʈu ɳu t̪ u n̪ u pu mu yu ʀu
லு வு ழு ளு று னு
lu vu ʎu ɭu ru nu
The u is usually realised as ɨ in the word final position
________________________________________________________________________
கூ ஙூ சூ ஞூ டூ ணூ தூ நூ பூ மூ யூ ரூ
Ku: ɲu: cu: ŋu: ʈu: ɳu: t̪u: n̪ u: pu: mu: yu: ʀu:
லூ வூ ழூ ளூ றூ னூ
lu: vu: ʎu: ɭu: ru: nu:
கெ ஙெ செ ஞெ டெ ணெ தெ நெ பெ மெ யெ ரெ
ke ɲe ce ŋe ʈe ɳe t̪ e n̪ e pe me ye ʀe
லெ வெ ழெ ளெ றெ னெ
le ve ʎe ɭe re ne
கே ஙே சே ஞே டே ணே தே நே பே மே யே ரே
ke: ɲe: ce: ŋe: ʈe: ɳe: t̪ e: n̪ e: pe: me: ye: ʀe:
லே வே ழே ளே றே னே
le: ve: ʎe: ɭe: re: ne:
கை ஙை சை ஞை டை ணை தை நை பை மை யை ரை
kɑi ɲɑi cɑi ŋɑi ʈɑi ɳɑi t̪ ɑi n̪ ɑi pɑi mɑi yɑi ʀɑi
லை வை ழை ளை றை னை
lɑi vɑi ʎɑi ɭɑi rɑi nɑi
கொ ஙொ சொ ஞொ டொ ணொ தொ நொ பொ மொ யொ ரொ
ko ɲo co ŋo ʈo ɳo t̪ o n̪ o po mo yo ʀo
லொ வொ ழொ ளொ றொ னொ
lo vo ʎo ɭo ro no
கோ ஙோ சோ ஞோ டோ ணோ தோ நோ போ மோ யோ ரோ
ko: ɲo: co: ŋo: ʈo: ɳo: t̪ o: n̪ o: po: mo: yo: ʀo:
லோ வோ ழோ ளோ றோ னோ
lo: vo: ʎo: ɭo: ro: no:
கௌ ஙௌ சௌ ஞௌ டௌ ணௌ தௌ நௌ பௌ
kɑu ɲɑu cɑu ŋɑu ʈɑu ɳɑu t̪ ɑu n̪ ɑu pɑu
மௌ யௌ ரௌ லௌ வௌ ழௌ ளௌ றௌ னௌ
mɑu yɑu ʀɑu lɑu vau ʎɑu ɭɑu rɑu nɑu
Class 3: Graphemes representing single consonant segments
These are derived from the basic set of 18 graphemes in class 2, by adding a special diacritic which cancels the ‘inherent’ vowel |ɑ| , leaving graphemes with this diacritic to represent a single consonant segment. This ‘cancelling diacritic’ is a dot or ‘puɭɭi’placed above the grapheme, and the class of graphemes is known as ‘puɭɭi’ graphemes’. The two graphemes at the end are used occassionally to represents sounds in loan words.
க் ங் ச் ஞ் ட் ண் த் ந் ப் ம் ய் ர்
k ɲ c ŋ ʈ ɳ t̪ n̪ p m y ʀ
ல் வ் ழ் ள் ற் ன் Sanskrit ( ஸ் ஷ் )
l v ʎ ɭ r n loans : ( s ʃ )
Xavieremmanuel.org 
Pingback: Sounds of Tamil, the writing system and dialect variations. Part I : The Vowels. | Xavieremmanuel.org
Pingback: A Linguistic Study of the Tamil Language (Co-Authored with Meghna C.N) – Historie Totale