Onset Clusters and the Sonority Sequencing Principle in Acehnese

Transcription

1 Onset Clusters and the Sonority Sequencing Principle in Acehnese Abstract Dr. Awwad Ahmad Al-Ahmadi Al-Harbi This paper provides a description of Acehnese cluster phonotactics. We argue that, contrary to appearances, Acehnese syllable structure is rule-governed. The examination of the role of sonority in the construction of Acehnese complex onsets reveals that the structure of onsets is controlled by universal principles, notably by the Sonority Sequencing Principle, which holds that within an onset sonority must rise towards the syllable nucleus. It is true that Acehnese has complex clusters that do not conform to the Sonority Sequencing Principle but these clusters are created by post-cyclic syllabification rules like Stray Segment Adjunction which links stray liquids to syllable onset. We also argue that a single sonority scale characterizes sonority in all languages and that this scale subdivides the obstruents into stops vs. fricatives subclass, with fricatives being more sonorous than stops. The Acehnese data provide the strongest evidence for the well-formedness of Stop+Fricative clusters and for the role of continuancy in determining sonority distinctions. Introduction [م [21/07/30 12:23:47 33) نم (1

2 Sonority figures prominently in recent phonological research. It is traditionally believed that the organization of segments within the syllable and across syllables is guided by principles of sonority that rank segments from least to most sonorous. Segment sequencing in onsets, for example, is explained with reference to the Sonority Sequencing Principle, which says that, within onset, sonority must rise towards the syllable peak. The Syllable-Contact-Law (Murray and Vennemann 1983) holds that the preferred contact between sequences of adjacent but heterosyllabic segments is achieved when the coda of the first syllable is higher in sonority than that of the segment that begins the second syllable. Although phonologists agree on the role sonority plays in the arrangement of segments within the syllable (the most sonorous segment occupies the peak position, while the less sonorous ones are relegated towards the syllable boundary), there is a lack of agreement on its nature and a hot debate on whether sonority scales are language-specific or there is a single scale common to all languages and whether phonological principles governing syllable structure, such as the Sonority Sequencing Principle, express absolute universals or tendencies. These are some of the issues that will be the focus of our discussion when we present data from Acehnese, a language with remarkable properties. The striking property of Acehnese is that its sonority violations are not restricted to one end of the syllable domain. Study of many languages (such as English, Russian etc) with syllables that offend the Sonority Sequencing Principle has led to the conclusion that a language that exhibits reversals word-initially, for example, does not usually tolerate them word-medially. When analyzing Acehnese data, it is obviously seen that complex clusters that do not conform to the Sonority Sequencing Principle occur in word-initial position as well as in word-internal position. Thus, Acehnese has a special rule allowing the offending segments to be incorporated into the syllable onset word-initially and word-internally. It is the purpose of this article to show that the anomalous onset clusters of Acehnese are of particular interest, however, because they are not defined in terms of position. That is, they are not restricted in their distribution to the periphery of the syllabification domain, where they do not give rise to problems of syllable division. This is what sets Acehnese apart from other languages that exhibit sonority violations. Another distinctive property of Acehnese is that it is the only language on record [م [21/07/30 12:23:47 33) نم (2

3 whose phonotactics reveals a preference for syllables that conform to the Sonority Sequencing Principle with respect to the assumption that fricatives are more sonorous than stops. As a matter of fact, the only obstruent clusters allowed in this language are stop plus fricative sequences. This issue is important for syllable phonology because most theories of the syllable and of sonority in particular do not recognize the need for comparing these two classes of obstruents. Since the sonority scale is worked out mainly on the basis of European languages, it is clear that these two classes are basically never adjacent in any word, which makes it unnecessary to compare directly their position on the sonority scale. Consequently, the sonority scale does not include a subdivision of obstruents into stops and fricatives. This is because the range of phenomena that involves sonority does not appear to require any further subdivision of the sonority scale. Acehnese syllable structure offers evidence that such a subdivision is clearly necessary and that these clusters are well-formed with respect to the Sonority Sequencing Principle. The implication of this phenomenon for syllabic theory is that onset clusters consisting of S+Stop are not regular onset clusters i.e. they are marked. These onset clusters constitute a major outstanding problem in current phonological theory. They violate the Sonority Sequencing Principle. But despite this fact, these types of clusters are relatively common across a significant number of languages that allow complex onsets. In this paper we propose an extension of Clements (1990) Dispersion Principle, an evaluation metric that determines the relative complexity of cluster types on the basis of their degree of distance from the optimal syllable and conclude that the reason why S+Stop onsets are relatively frequent across languages is because they are the least marked within the class of sonority reversals. The remainder of the paper is organized as follows. Section 2 gives pertinent background on the basic facts of Acehnese syllable structure. Section 3 provides theoretical background of sonority. In section 4 we discuss Acehnese onset clusters and show how they bear on current issues in sonority theory. Section 5 summarizes the most important conclusions. 2 Acehnese Acehnese is an Austronesian language spoken by about two and a half million people in the province of Aceh, the Northern part of Sumatra, Indonesia. There are four main dialect groups; (i) Pase: Aceh Timur (langsa) and Aceh Utara [م [21/07/30 12:23:47 33) نم (3

4 (lhokseumawe), (ii) Pidie: Aceh Pidie (Sigli), (iii) Banda Aceh: Aceh Besar (Banda Aceh), and (iv) Meulabah: Aceh Barat (Meulabah). All dialects are mutually comprehensible. The main distinguishing features of the dialects are pronunciation (Pidie: canru? porridge, Pase: cagru?) and vocabulary (Pidie: ear, other dialects It goes without saying within each major dialect there are a number of subdialects. [See Asyik (1972), (1978) and Durie (1985) for more information on different aspects of Acehnese grammar]. The data on Acehnese presented here come from the Pidie dialect and have been elicited from four different informants; (i) Muhammad Yusran Hadi from Lemteh, Banda Aceh. Currently College of Shari a Islamic Law at Islamic University in Medina (his mother is Pidie and his father a Pase), (ii) Biktiyar a teacher of social sciences in an Indonesian school in Jidda, (iii) Mashhuudi a student of UKM in Malaysia, majoring in economics, and who visits Makkah to perform Umrah during Ramadhan, and (iv) Sulayman Ashi, currently an employee of Umm Al-Qura University in Makkah (All other informants are from Siglie, i.e. speak Pidie dialect). 2.1 Segments and syllables The consonant and vowel inventories of Acehnese are given in (1) and (2) respectively. (1) Acehnese consonant phonemes bilabial alveolar palatal velar glottal Stops p t c k? b d Š g Fricatives s S h Nasals m n J N Trill r Lateral l Glides w j [م [21/07/30 12:23:47 33) نم (4

5 Notes: All transcriptions are given in IPA. /s/ is a special IPA symbol used here for a laminal alveodental fricative. (2) Acehnese vowel phonemes Monophthongs Oral i 1 u o E V O a Nasal ĩ 1 u E V O a Diphthongs Oral i@ 1@ u@ E@ O@ Nasal [م [21/07/30 12:23:47 33) نم (5

6 The maximum syllable in Acehnese is C(C)V(C). On the surface, underived lexical items belong to the following syllable types: (3) Monosyllables a. CV bu boiled water b. CVC goh to rub c. CCV cru to start running slippery d. CCVC grah thirsty bambo In addition to the monosyllabic types in (3), the following polysyllables obtain: (4) Disyllables (Note that here a period is used to indicate syllable boundaries). a. CV.CV ra.ga basket ba.te@ stone b. CV.CVC la.wan enemy na.l1@n grass c. CV.CCV m1.rha to dissolve su.kla very black d. CV.CCVC si.krop scoop e. CVC.CV lin.to bridegroom ban.ci hate [م [21/07/30 12:23:47 33) نم (6

7 f. CVC.CVC?1 N.kot fish man.dum all g. CVC.CCV?1N.khuj to beat slippery h. CVC.CCVC?1 n.tr1? later ma n.dre t type of drink made of ginger or cardamom (5) Trisyllables a. CV.CV.CV si.ku.la school k1.bi.ri sheep b. CV.CV.CVC mu.pa.kat to confer k1.mu.jan incense c. CVC.CV.CV pan.cu.ri thief 2.2 Onsets Onset consonants in Acehnese are obligatory. All single consonants can be syllable-initial. In addition, syllable onsets may consist of a cluster of stop or liquid plus the glottal fricative h. Also permitted are clusters of stop plus liquid. (6) Onset clusters in Acehnese a. stop+h phon first bhep pocket thon year dho@ forehead khem to laugh [م [21/07/30 12:23:47 33) نم (7

8 gha m to beat chen to hop Šho to push headlong b. liquid+h lho? deep rhah to wash c. stop + liquid pran war ploh ten bri to give blo@ to buy trom to kick dro@ self, person croh to fry Šru@? to put in salty water for a day or two (of fruits and eggs) kru@n river kli? to cry gra? to lift up glu@ slippery 2.3 Codas Syllable-final consonant inventory is severely limited. The attested coda consonants are depicted in (7a) and exemplified in (7b). (7a) Acehnese syllable-final consonants Stops p t? Fricatives h [م [21/07/30 12:23:47 33) نم (8

9 Nasals m n N (7b) rap near h1òt desire ba? tree peh grind te m tin can?on leaf kvn strong Let us close the discussion of syllable codas with an important observation. Although the palatal glide /j/ can apparently close the syllable as in (8) buj koj (8) rhej a ruler pig relatives,it best be viewed as a diphthongal offglide rather than a coda consonant. Diphthongs abound in this language. Nothing in the following discussion hinges on this point, however. 3 Sonority and the Sonority Sequencing Principle 3.1 Sonority There has been little agreement on the question of what sonority is and how it should be defined. Phoneticians have proposed different phonetic parameters to characterize sonority. Based on acoustic intensity, Ladefoged (1993) defines sonority as the perceptual saliency or loudness of a particular sound. In Selkirk (1984), it is interpreted in terms of degree of opening; vowels are the most open, i.e. sonorous, sounds followed in decreasing order by liquids, nasals, fricatives, [م [21/07/30 12:23:47 33) نم (9

10 and stops. Multiple phonetic parameters have been proposed as well; Butt (1992) presence or absence of voicing and frictional noise; (Ohala and Kawasaki (1984) amplitude, periodicity, spectral structure, and fundamental frequency. Phonologists, on the other hand, disagree on whether the sonority of a given segment is determined by a multi-valued feature that either makes major class features seem superfluous or eliminates them entirely from the theory of phonotactics, or whether it should be derived from major class features. While Hankamer and Aissen (1974) assign the feature sonority a range of values from 1-9 with stops near 1 end of the scale and vowels at the high end of the continuum, Selkirk (1984) assigns a sonority index to individual segments that reflects position on the sonority hierarchy. In other words, there is a single n-ary feature [sonority], which captures natural classes of segments. Kiparsky (1981) arrives at a sonority scale which is defined with reference to the major manner class features [syllabic], [consonantal], [sonorant], [nasal], and [continuant]. Clements (1990) derives sonority from the number of plus specifications for the features [syllabic], [vocoid], [approximant], and [sonorant]. Some phonologists have suggested that the relative sonority of a particular segment is formally represented in terms of geometric tree structure. In Rice (1992) the more class nodes in the tree structure, the greater the sonority of the segment in question. The liquids have two nodes in addition to the Root node, nasals have one and stops have none. In Dogil (1992) the more class nodes in the tree structure, the lesser the sonority. Thus, the segments of lowest sonority are laryngeals that consist of a laryngeal gesture. [For excellent review of phonological models of sonority, see Cser (2000)]. 3.2 Sonority scales There are a number of competing sonority scales in the literature that rank segment types in order of their relative sonority. Views differ widely on how the sounds of speech can be arranged along a scale of sonority. The main question revolves on whether sonority ranking is language particular (9), or whether there is a single universal sonority scale that governs phonotactic patterning in all languages (10). (9) Representative sample of language-specific sonority scales (from least to most sonorous) [م [21/07/30 12:23:47 33) نم (10

11 a. Steriade (1982) for non-syllabic segments (i) for Classical Latin: p, b, k, g < t, d < f < s < m < n < l < r (ii) for Attic Greek: p, t, k < b, d, g < s < z < n, m < l < r b. Hankamer and Aissen (1974) for Pali stops < s < nasals < l < v < y < r < vowels c. Davis (1990) for Italian voiceless stops < voiced stops < noncoronal fricatives < coronal fricatives < n < m < liquids < vowels (10) Representative sample of universal sonority scales a. Blevins (1995) for non-syllabic segments voiceless plosives < voiced plosives < voiceless fricatives < voiced fricatives < nasals < liquids < glides < non-low vowels < low vowels b. Butt (1992) voiceless obstruents < voiced obstruents < nasals < liquids < glides < vowels c. Clements (1990) obstruents < nasals < liquids < glides < vowels In several languages, it has been reported that further subdivisions of sonority scales with universal values by means of additional features such as [voice], [coronal], and [place] are motivated. Place features, for example have been proposed by Steriade (1982) for Ancient Greek and by Levin (1985) for Klamath. Blevins (1995), however, argues quite convincingly that there are no languages that display sonority rankings for place features and asserts that in the case of Ancient Greek and Klamath alternative analyses are available. Further-more, both Clements (1990:296) and Butt (1992:50) argue that language-particular modifications of the sonority scale can be explained independently i.e. without reference to sonority. Finally, we concur with Clements (1990:296) that once we allow the [م [21/07/30 12:23:47 33) نم (11

12 relative sonority of segments to vary in its identity from one language to another, we seriously undermine its explanatory role by increasing the number of ways in which it will accommodate potential exceptions, thus reducing the number of cross-linguistic generalizations that it accounts for Therefore we assume the following universal ranking of segments: (11) Universal Sonority Scale stops < fricatives < nasals < liquids < glides < vowels The sonority scale given in (11), unlike that proposed by Clements, includes a partition of obstruents into stops and fricatives and assumes the addition of the feature [continuant] to his set of major class features in order to be able to subrank obstruents. The proposal of a difference in sonority between stops and fricatives will be substantiated by evidence from Acehnese phonotactics. 3.3 The Sonority Sequencing Principle The crucial role of sonority in defining possible and impossible onsets and codas is uncontroversial. There is general agreement on this in spite of theoretical debates on the concept of sonority and its substantive nature. It has been known for over a century that the position of segments within a syllable is determined by sonority; the most sonorous segment on the sonority scale constitutes the nucleus of the syllable or syllable peak while all the other sounds are grouped around the peak in such a way that segments of high sonority ranking stand closer to the peak and segments of low sonority ranking stand closer to the syllable boundary. This principle is referred to in the phonological literature as Sonority Sequenc-ing Principle and requires onset clusters to have an ascending sonority pattern toward the syllable peak and coda clusters to have a descending pattern of sonority away from the nucleus. We may state this principle as follows: [م [21/07/30 12:23:47 33) نم (12

13 (12) The Sonority Sequencing Principle (henceforth SSP) (Following Morelli 1999) Sonority increases towards the syllable peak and decreases towards the syllable margins Under this principle, a number of cross-linguistic generalizations on the distribution and sequencing of segments is explained with reference to the scale in (11). As a matter of fact, tight restrictions are imposed on the sequencing of segments on either side of the syllable peak. Thus, the principle allows clusters of the type pla, tra, for example in the onset and their inverse in the coda; plan vs. help, track vs. cart and prohibits such nonoccurring clusters as *lpa, *rta, *apl, and *atr. It follows from such an account of segment sequencing that two-member clusters fall into two types: (1) core clusters; clusters showing rising sonority profile towards the syllable peak and thus conforming to the SSP and (2) sonority reversals; clusters that constitute violations of the principle. Clements (1990) proposes the Dispersion Principle, a complexity metric for core clusters or unmarked clusters that measures their deviation from the optimal syllable i. e. the syllable in which sonority of the demisyllable (onset plus peak and peak plus coda) is maximized as we move from onset to nucleus and then minimized from nucleus to coda. As we show below, our assumption that stops and fricatives are assigned different sonority values calls for a slight emendation to his rankings of initial three-member demisyllables, but has far and reaching consequences for universal syllable phonotactics. According to his analysis fricatives and stops constitute a single class with respect to sonority. The observed patterns of syllable organization in Acehnese provide evidence for a distinction between stops and fricatives. Extending equation (13) that he uses to rank core clusters in terms of the dispersion in sonority within a demisyllable, to incorporate splitting the obstruents into two separate classes with respect to sonority gives us the rankings in (18). [ D = dispersion; degree of sonority distance from the optimal syllable. d = the distance in sonority rank between each ith pair of segments in the demisyllable. m = the number of pairs in the demisyllable, equals, where n is the number [م [21/07/30 12:23:47 33) نم (13

14 of segments]. In this equation the number of pairs m equals and di is the distance between each ith pair (segments on the sonority scale). It states that D, the dispersion in sonority within a demisyllable, varies according to the inverse of the squared values of the sonority distance between the members of each pair of segments within it. Consider table (14). (14) Sequence S L V Order Difference 4 1= = 2 6 1= 5 For the ith pair SLV, if we substitute in equation (13) the net dispersion will then be: (15) For the ith pair SNV, we have the following values: (16) Sequence S N V Order Difference 3 1= = 3 6 1= 5 So if we substitute in equation (13), the net dispersion for that ith pair will be: [م [21/07/30 12:23:47 33) نم (14

15 (17) Assuming the sonority scale in (11), this gives the following values of D for three-member demisyllables in initial position: (18) Complexity rankings for three-member demisyllables in initial position based on the sonority scale S < F < N < L < G < V and the order 1-6. For S, F, N. equal to 1, 2, 3, respectively. Dispersion Complexity Ranking SLV, SNV FLV SFV, SGV FNV, FGV NLV, NGV LGV The metric assigns the complexity ranking, 1, to the demisyllable with the lowest value, the rank 2 to the next highest, and so forth. The demisyllable SLV and the demisyllable SNV, for example, have the lowest value for D (0.40), and therefore the lowest complexity rank (1); FLV has the second highest value for D (0.56), and thus the second lowest complexity rank (2); and so forth. Thus the degree of complexity is not proportional to D itself, but rather to the ranking defined by D. The complexity rankings in (18) define a hierarchy over Acehnese core onsets and place its maximal complexity for initial two-member cluster at rank 3. Among unmarked demisyllables of length 3, then we find SFV, and SLV. Missing are SGV and SNV which are cross-linguistically un-common. We propose to rule out these two types of unmarked demisyllables on sonority-independent grounds; OCP effects barring tautodemisyllabic segments underlyingly: [-consonantal] and [-continuant] respectively. Two qualifications to this statement are in order here. First, while SGV [م [21/07/30 12:23:47 33) نم (15

16 demisyllables are present in English and French, for example, in both cases the status of these clusters as genuine onset clusters is called into question. Jensen (1993) observes that the only English onsets consisting of a consonant followed by glide j are ones in which j precedes the vowel /u/ [tune, cute, dune...etc]. Under his proposal j is inserted by a phonological rule since it never participates in onset clusters at all in core syllabification. This severely limits SG clusters to /tw, dw, kw, and gw/ [twin, dwell, quick, and guava]. In French, under Clements analysis CGV syllables [ dieu god, mieux best, rien nothing...etc] are surface exceptions that are derived from underlying CVV by the rule of Glide Formation. Second, sporadic instances of SN sequences are reported in Ancient Greek and French, for example. As a matter of fact, if SG and SN sequences exist in a particular language they will be the result of language-specific stipulations overriding OCP constraints. We have now reached the conclusion that SLV is the most optimal three-member initial demisyllable, followed by FLV, which is in turn followed by SFV. The rankings in (19) thus obtain: (19) CCV: SLV > FLV > SFV Following the intuition of Clements that the complexity of sonority reversals increases in proportion to the extent of the reversal, we arrive at the ranking of reversals in (20). (20) CCV: LSV > LFV > FSV In other words, LSV is more complex than LFV, LFV than FSV. It may reasonably be argued, then, that reversals in Acehnese and English correspond to preferred core clusters that are not present in their respective systems; namely FL and SF, which suggests that reversals are reflexes to missing optimal sequences. As a consequence, we are in a position to consider the phenomenon of onset clusters consisting of s plus stop sequences. There has been considerable debate in the literature on the status of these consonants (Steriade 1988, Clements 1990, Harris 1994, Morelli 1999; among others). Despite the fact that these clusters constitute sonority reversals and violate the SSP, they are relatively common. We [م [21/07/30 12:23:47 33) نم (16

17 propose that the reason why these onsets are attested in a significant number of languages is because they are the default reversals i. e. the least complex among the marked with respect to the SSP. Given the above fact, we are led to conclude that these offending clusters may best be characterized as substitutes for missing optimal sequences. 4 Acehnese cluster phonotactics A common phonotactic pattern of onset maximization is observed in Acehnese. Consider the following examples: (21) a. pha thigh tim.phan a kind of Acehnese cake that very rin.tha? to pull plvh tountie mam.plam mango prip a whistle sum.pru t to fall (of trousers) b. gu.bho hanging loosely c1.dra to have a defect bu.klam last night c. lhat to hang t1.lhat to hang (intr.) rh1t to cut off m1.rh1t to cut off (intr.) We note, firstly, that complex clusters are not restricted to word-initial position. [م [21/07/30 12:23:48 33) نم (17

18 Secondly, intervocalic clusters of two consonants in (21b) and (21c) are syllabified as V.CCV even though the two-consonant sequences in (21c) constitute ill-formed onsets (as can be seen from the fact that [b, d, k, l, and r] can not close the syllable on the left. These are illicit codas. Next, while the data in (21a) and (21b) exhibit a pattern in which a segment in the onset is followed by a segment of greater sonority as predicted by the SSP, the data in (21c) show a falling sonority profile. Consequently, all the clusters in (21a) and (21b) are classified as core onsets and all the clusters in (21c) are classified as reversals. 4.1 Core onsets The role of the SSP in the construction of Acehnese onsets is indisputable. If we ignore /lh/ and /rh/ for the moment, it is clear that all occurring onset clusters conform to the SSP. All consonants in the onset obligatorily rise in sonority Stop plus liquid Two-consonant onsets may be one of the following stop plus liquid clusters. (22a) lists all possible two-consonant onsets that consist of a stop followed by a liquid; examples follow in (22b). (22) a. pl / pr tr cr kl / kr bl / br dr Šr gl / gr b. plu@n to run plah to slice pru@ to spray prah to press ble t br1@h blan field rice container husked rice brat heavy [م [21/07/30 12:23:48 33) نم (18

19 tro? tran arrive to pull dro m a barrel droh to bark cron to scoop crah cracked ŠrOh ŠrO m beautiful to storm kle p to blink klo? to scratch with claw kreh to scrape kru@n river gleh clean glah a glass gro h to snore grah thirsty As shown, all of the labials p and b and the velars k and g combine freely with both l and r. The alveolars t and d and the palatals c and Š cluster only with r. Although /tl/, /dl/ /cl/, and /Šl/ clusters show rising sonority, they are systematically barred. Because many languages including English and French prohibit /tl/ and /dl/ onset clusters, this seems to indicate that the SSP alone falls short of deriving consonant sequences within an onset. In response to this [م [21/07/30 12:23:48 33) نم (19

20 problem, a number of common strategies have been proposed in the earlier literature. Butt (1992), for example suggests that the exclusion of /tl/ and /dl/ is due to articulatory economy ; the articulation of the stop involves the release of an apical constriction while the articulation of the lateral requires the apex to be in contact with the alveolar ridge. Since the features [anterior] and [coronal] are maximally utilized for the stops in Acehnese, this view is untenable. Steriade (1982) notes that it can be explained by reference to sonority; coronal stops are higher ranked on the sonority scale than noncoronal stops. The exclusion of /t/ and /d/ in initial clusters before /l/, then follows from what is called the minimal sonority distance which requires adjacent consonants in the onset to be separated by a minimal number of intervals on the scale, or else they are excluded. In other words, /t/ and /d/ are closer in sonority to /l/ than are the noncoronal stops. [See Ohala and Kawasaki (1984: ) and Clements (1990: ) for criticism of her approach]. Finally, Clements and Keyser (1983) motivate the use of the filter in (23) to rule out, on a language-particular basis, such clusters that are ruled well-formed by the SSP. (23) * +cor -stri [+lat] σ This filter should be formulated in Acehnese as: (24) * +cor -del rel [+lat] [م [21/07/30 12:23:48 33) نم (20

21 σ Perhaps the most promising strategy, following Borowsky s (1986) suggestion, for which he credits McCarthy, is to treat the liquid /r/ as noncoronal. Consequently, these co-occurrence restrictions may be accounted for by the OCP-motivated filter in (25) rather than by sonority constraints. (25) *[+cor] [+cor] Stop plus fricative The phonotactics of Acehnese reveals very clearly a preference for syllables that obey the SSP with respect to the assumption that fricatives are more sonorous than stops. A list of two-segment onset clusters in which a stop is followed by the glottal fricative /h/ is given in (26a) and exemplified in (26b). (26) a. ph th ch kh bh dh Šh gh b. pha thigh phet bitter ph1 pale bhan football, tire bhem an instrument used to make earthenware bhoj a kind of sweet made of flour thon year thu@p to plug that very dho@ forehead [م [21/07/30 12:23:48 33) نم (21

22 dhot to snap dhu m a large drum (used at the mosque) khan half khe N sound of the dog khop to cover gha m to beat molar tooth che? small chet paint chi? big Šh1@t bad, evil Šho to push headlong Šhi@ dry As (26) illustrates, stops occur freely before /h/. But this is only possible if the aspirate (pž, tž etc) and murmured (bž, dž etc) consonant series are analyzed as two-segment onsets. In (27), we quote from Durie (1985) the acoustic description of Acehnese stops and his arguments for treating the aspirated and murmured stops as clusters of stop plus /h/. (27) Durie (1985: 10-12; 19) Voiceless aspirated stops have a voice onset lag; voiceless unaspirated stops have coincident or near coincident voice onset timing; modal voiced stops have a modal voice onset lead; murmured stops have a whispery voice onset lead and a modal voice onset lag [م [21/07/30 12:23:48 33) نم (22

23 Aspirated stops have several characteristics in common with the murmured stops which distinguish them from unaspirated voiceless and modal voiced stops: a delayed modal voice onset time, a noise source (at the glottis or place of articulation of the following vowel) during the period before the onset of modal voicing, a lowered Fundamental Frequency at the onset of the following vowel, and a delayed peak of intensity for the following vowel The aspirate and murmured consonant series are analyzed as phoneme sequences /Ch/. For example [bž] is analyzed as /b/+/h/. The unitary treatment of murmured and aspirated consonants can be motivated acoustically. Moreover, at least two of the acoustic features shared by these stops- a delayed modal voice onset time, and a noise source during the period before the onset of modal voicing- are also shared by /h/. The /h/ is psychologically real to native speakers, who intuitively use h to transcribe these consonants, and it simplifies the description of the phonotactics. For example /phet/ bitter forms a stem p1het which takes the infix /1n/ to give /p1n1het/ bitterness. Therefore, we can conclude that stop plus h sequences are tautosyllabic clusters that share the same distributional properties of stop plus liquid, and can hence be considered regular phonotactics. Now that we have reviewed the major evidence for the treatment of Acehnese aspirated and murmured stops as phoneme clusters, we turn to the arguments in favor of the ranking in (28) (28) Stops < Fricatives The major lines of evidence that have been cited for establishing a rank within the class of obstruents in such a way that fricatives are more sonorous than stops are: (i) drawing on the work of Sievers (1881) and Jespersen (1904), fricatives are clearly more open than stops and show a higher degree of auditory prominence (cf. Butt 1992), (ii) fricatives in Imdlawn Tashlhiyt Berber are better syllable peaks than stops (cf. Prince and Smolensky 1993), and (iii) Italian with [م [21/07/30 12:23:48 33) نم (23

24 its heterosyllabic [sc] clusters, provides indirect evidence for the subranking of obstruents. (29) vi:.ta life vin.ta defeated (fem.) vis.ta sight As the Italian examples in (29), from Harris (1994), illustrate [st] in medial position are heterosyllabic clusters because in Italian stressed vowels in open syllables are lengthened. Since the initial syllable of vista patterns with vinta in having a short vowel, these clusters are heterosyllabic, with the /s/ in the coda of the preceding syllable and /t/ in the onset of the following syllable. Hence, the special status of s plus consonant clusters in medial position is evident in their ill-formedness. For our purposes, the most important point is that no cases are on record where cluster phonotactics makes it necessary to compare directly the position of stops and fricatives in the syllable structure. Acehnese, a language with only SF clusters, provides the strongest evidence for a phonotactic pattern in the onset where these two classes of obstruents are basically adjacent and well-formed with respect to the SSP. To summarize the discussion, we see that SF clusters constitute well-formed onsets. Our data show that the preponderance of evidence at present is for splitting the obstruents into stops and fricatives: the resultant stop plus fricative sequence shows rising sonority and thus qualifies as a core onset. Further support for this comes from the syllable-contact-law of Murray and Vennemann (1983), which states that a syllable coda must exceed the following onset in sonority. The effects of the syllable-contact-law are attested in a number of languages including Acehnese. Examples are cited in (30). (30) sum.pru t to fall (of trousers) g1n.t1t can.koj to break wind hoe mas.?u t personal name [م [21/07/30 12:23:48 33) نم (24

25 sah.bat friend In each of these syllable contacts, there is a preference for the sonority of the final segment of the first syllable to be higher than that of the first segment of the second syllable. One can see readily enough that sonority falls across a syllable boundary in a descending pattern from either a nasal or a fricative into a stop. Before closing this section it is worth mentioning that the behavior of h in Acehnese constitutes a direct challenge to Clements (1990:322) claim that laryngeals fall outside the sonority scale. It systematically combines with a preceding stop as well as a preceding liquid to yield the most striking sets of Acehnese onsets; the combination [ph, bh, th, dh, ch, Šh, kh, gh] and [lh, rh] respectively. The fact that the only obstruent clusters allowed in Acehnese are stop plus h and the fact that h participates in onset clusters of falling sonority, provide clear evidence that the fricatives including h must be assigned a special place of their own on the sonority scale, contrary to Clements claim. 4.2 Sonority reversals In addition to the core clusters listed in (22) and (26), Acehnese possesses clusters involving a liquid and the glottal fricative /h/ that do not conform to the SSP. (31) lh1@h already lhv to churn up with feet rh@t to fall (from above) rhvm to throw As shown, the [h] of [L+h] clusters systematically violates the principles that govern the structure of Acehnese onsets. It violates the requirement of rising sonority by freely combining with a preceding liquid (in contrast to clustering with a preceding stop, for example). We can account for the exceptional behavior of [h] if liquids are incorporated by a language-particular adjunction rule. Thus lh1@h and rh@t have the analysis l h1@h and r h@t respectively, with [م [21/07/30 12:23:48 33) نم (25

26 angled brackets denoting the extrasyllabic consonants. A special ad-junction rule adds l to the core syllable h1@h and r to the core syllable h@t; given this rule, the initial liquids are now fully syllabified. On the present account, then, syllabification is governed by the SSP, which does not tolerate sonority reversals. The offending consonants [l] and [r] violate the SSP and hence cannot be syllabified. If these segments remain unsyllabified then they will be subject to Stray Erasure. Since these consonants surface phonetically despite the fact that they are extrasyllabic, Acehnese has a rule of adjunction which overrides the SSP and includes offending liquids into prosodic structure because only then can they escape Stray Erasure. Note now that while the consonants that are subject to adjunction in Acehnese are restricted to a particular class of segments; namely liquids, their distribution is not limited to word-initial position. (32) a. m1.rh1t to cut off (intr.) p1.rh@t s1.m1.rhvm to drop to throw (intr.) b. r1.lh@h broken n1.lh1@h the last p1.lhe@ to make three A similar analysis to that of Rubach (1997) on Polish can be extended to word-internal clusters in (32). But first, compare those to the clusters in (33) which have rising sonority. (33) su.kla very black si.krop scoop In the first step, the SSP constructs core syllables. We assume that Acehnese maximizes onsets; the entire [kl] in [sukla] thus enters the onset while only the second element of the [rh] cluster in [m1rh1t] (32a) does. [م [21/07/30 12:23:48 33) نم (26

27 (34) PW σ σ s u k l a As a result, extrasyllabic consonants in (32) remain unsyllabified because they violate the SSP. At this stage m1rh1t and r1lh@h, for example, are represented as follows: (35) PW σ σ m 1 r h 1 t PW σ σ r 1 l h Since the SSP offenders cannot close the preceding syllable (liquids are not permitted syllable-finally; see 2.3), it is, therefore reasonable to link them to the phonological word by a separate adjunction rule. This rule derives the surface representations in (36). (36) [م [21/07/30 12:23:48 33) نم (27

28 PW σ σ m 1 r h 1 t PW σ σ r 1 l h As a result, word-initial extrasyllabic consonants are adjoined to the syllable onset while word-medial SSP offenders are linked to the phonological word. However, it is difficult to maintain that this is desirable. Rather than assuming two adjunction rules, one for word-initial environments and another word-medially, let us, without arguments, introduce a new prosodic constituent into the prosodic hierarchy; the stem, and assume that extrasyllabic segments are adjoined directly to the stem node. Unifying the position to which these segments belong in the prosodic structure is thus achieved by prosodifying them by adjunction to the stem. Unfortunately, this plan won t work. First, the prosodic constituent status attributed to the stem is dubious. Second, there exist extrasyllabic liquids that surface phonetically and are not limited to the stem-initial position. (37) (a) hilha to pull (with force) t1lhon nude t1rho? to have a fit of choking m1lhv a fight [م [21/07/30 12:23:48 33) نم (28

29 t1lh1p disjointed (b) boh-rh1m a very sour fruit boh-rhomrhom a kind of sweet made of flour served in Ramadan breakfast (c) m1lha? very thin m1sirh1@? to spread (of news) to set (sun) In such forms either no morpheme boundary exists, as in (37a) or if it does, (37b) it has been fossilized. In cases such as (37c), it is impossible to relate the meaning of their dummy stems to the meaning of the real stem; lha? means to flay (an animal) while rh1@? means plait (hair). Clearly, then, word-medial extrasyllabic consonants cannot be prosodified by putting them into a stem. Therefore, the central question remains the same; to which node are extrasyllabic consonants adjoined? Note that they must be incorporated into prosodic structure by a rule of adjunction because they are pronounced. But is it an adjunction to the phonological word, the syllable node, or the onset. Acehnese favors adjunction to an onset of a core syllable that is otherwise well-formed with respect to the SSP. As evident from the preceding discussion, in Acehnese extrasyllabic segments are restricted to a certain class of segments: they are all liquids, but their distribution is not limited to the word-initial position. Word-initial and word-medial SSP offenders are included into prosodic structure. The node to which extrasyllabic consonants are adjoined is the same for both: it is ONSET. These consonants violate the SSP, which prohibits liquid-fricative onsets and hence cannot be syllabified. Yet they are pronounced. Postulating an adjunction rule solves the problem. It affects only and exclusively liquid segments. Syllabification is governed by SSP and consequently, not all segments can be syllabified. Given that they cannot by syllabified by regular syllabifica-tion rules, we are looking at a process of adjunction. It prosodifies extrasyllabic consonants by putting them into a syllable onset creating clusters in which the most sonorous segment (the liquid) occurs closer to the syllable margin than to the syllable peak. [م [21/07/30 12:23:48 33) نم (29

30 5 Conclusion In this paper, we have examined Acehnese onset clusters. Specifically, we have discussed the role universal phonological principles such as the SSP and the Maximal Onset Principle (Selkirk 1982, Harris 1994) play in defining well-formed sequences of phonological segments that occupy the onset. We have demonstrated that in order for a consonant sequence to be well-formed the first consonant must be less sonorous than the second as predicted by the SSP. Furthermore, we have shown that the syllabification of word-internal consonants respects the Maximal Onset Principle that states that a medial consonant or consonant cluster should be syllabified in the onset of the second syllable rather than the coda of the first. Interestingly, edge constraints (following Itô 1989; restrictions on the occurrence of segments at syllable margins) conspire to ensure that no Acehnese syllable can end with more than one consonant and that out of the large inventory of Acehnese consonants, given in (1), only those listed in (7a) may occur at the right margin of the syllable. Onset maximization is, thus, guaranteed by severely constraining the consonants that may appear in syllable-final position. This explains the limited number of consonants permitted syllable-finally. The restrictions exclude all voiced stops and all voiceless obstruents that are high; next only nasals are permitted in the coda, and among those the palatal nasal is excluded. In short, the distribution of syllable-final consonants in Acehnese results from edge constraints whose combined effect is onset maximization. In addition, we have attempted to account for the phonological behavior of liquid plus fricative clusters; these clusters do not fit the sonority profile of an onset since the second consonant of the sequence is less sonorous than the first. To this effect, we have suggested that their special status can be explained once they are analyzed as extrasyllabic segments. On this account, then, the SSP controls syllabification, these consonants will fail to enter the onset. But the offending consonants cannot be left unprosodified in surface representation. They surface phonetically. The desired effect is achieved by an adjunction rule that adds stray liquids into the onset. As a result, they form ill-formed tautosyllabic clusters. Finally, the data discussed in this paper support two kinds of conclusions. First, the role of continuancy with respect to sonority is determined. It is argued that obstruents should be divided into two separate classes: stops and fricatives with stops being less sonorous than fricatives. Acehnese data provide substantial evidence that SF clusters constitute well-formed onsets. Given the fact that the [م [21/07/30 12:23:48 33) نم (30

31 only obstruent clusters allowed in Acehnese are SF sequences, it appears that, in addition to the major class features, the feature [continuant] needs to be invoked to express distinctions in sonority within obstruents. Second, the Acehnese data seem to present a putative counterexample to Clements (1990) claim that laryngeals fall outside the sonority scale. The stops combine with the glottal fricative to yield a sequence that conforms to the SSP while the liquids cluster with it to yield the only sonority violations in this language. This is clear evidence that laryngeals have their own position on the universal sonority scale. References Asyik, Abdul Gani (1972). Atjehnese Morphology. M.A. thesis, Institut Keguruan dan Ilmu Pendidikan, Malang, Indonesia. Asyik, Abdul Gani (1978). Bunyi Bahasa dalam Bahasa Aceh. Banda Aceh: Fakultas Keguruan, Universitas syiah Kuala. Blevins, Juliette (1995). The syllable in phonological theory, in J. Goldsmith (ed.), A Handbook of Phonological Theory. Oxford: Blackwell, Borowsky, Toni (1986). Topics in the Lexical Phonology of English. PhD dissert-ation, University of Massachusetts, Amherst. Butt, Matthias (1992). Sonority and the Explanation of Syllable Structure. Linguistische Berichte 137: Westdeutcher Verlag. Clements, George N. (1990). The role of the sonority cycle in core syllabific-ation, in J. Kingston and M. Beckman (eds.), Papers in Laboratory Phonology I: Between the Grammar and the Physics of Speech. Cambridge: Cambridge University Press, Clements, Goerge N., and Samuel Jay Keyser (1983). CV Phonology: A Generative Theory of the Syllable. Cambridge, MA: MIT Press. Cser, András (2000). Phonological models of sonority, The Even Yearbook 4: Davis, Stuart (1990). Italian onset structure and the distribution of il and lo, [م [21/07/30 12:23:48 33) نم (31

32 Linguistics 28: Dogil, Grzegorz. (1992). Underspecification, natural classes and the sonority hierarchy, in J. Fisiak and Stanisław Puppel (eds.), Phonologi-cal Investigations. Amsterdam: John Benjamins, Durie, Mark (1985). A Grammar of Acehnese on the basis of a Dialect of North Aceh. Dordrecht: Foris. Hankamer Jorge and Judith Aissen. (1974). The sonority hierarchy, in A. Bruck, R. Fox, and M. La Galy (eds.), Papers from the Parasession on Natural Phonology. Chicago Linguistic Society, Harris, James (1994). English Sound Structure. Oxford: Blackwell. Itô, Junko (1989). A prosodic theory of epenthesis, Natural Language and Linguistic Theory 7: Jensen, John T. (1993). English Phonology. Amsterdam: John Benjamins. Jespersn, Otto (1904). Lehrbuch der Phonetik. Leipzig: B. G. Teubner. Kiparsky, Paul (1981). Remarks on the metrical structure of the syllable, in Wolfgang U. Dressler, Oskar E. Pfeiffer and John R. Rennison (eds.), Phonologica Proceedings of the 4th International Phonology Meeting. Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft, Ladefoged, Peter (1993). A Course in Phonetics. New York: Harcourt Brace Jovanovich. Levin, Juliette (1985a). A Metrical Theory of Syllabicity.Ph. D. dissertation, MIT. Levin, Juliette (1985b). Reduplication in Umpila, MIT Working Papers in Linguistics 6: Morelli, Frida (1999). The Phonotactics and Phonology of Obstruent Clusters in Optimality Theory. Ph.D. dissertation, University of Maryland at College Park. Murray, Robert and Theo Vennemann (1983). Sound change and syllable structure in Germanic, Language 59: Ohala, John and Haruko Kawasaki (1984). Prosodic phonology and phonetics, Phonology Yearbook 1: Prince, Alan and Paul Smolensky (1993). Optimality Theory: Constraint interaction in generative grammar. Technical Report 2 of the Rutgers Center for [م [21/07/30 12:23:48 33) نم (32

33 Cognitive Science. Piscataway, NJ: Rutgers University. Rice, Keren (1992). On deriving sonority: A structural account of sonority relationships, Phonology 9: Rubach, Jerzy (1997). Extrasyllabic consonants in Polish: Derivational Optimality Theory, in Iggy Roca (ed.), Derivations and Constraints in Phonology. Oxford: Clarendon Press, Selkirk, Elizabeth (1982). The syllable, in H. van der Hulst and N. Smith (eds.), The Structure of Phonological Representations (Part II). Dordrecht: Foris, Selkirk, Elizabeth (1984). On the major class features and syllable theory, in M. Aronoff and R. T. Oehrle (eds.), Language Sound Structure. Cambridge, MA: MIT Press, Sievers, Eduard (1881). Grundzüge der Phonetik. Leipzig: Breitkopf and Hartel. Steriade, Donca (1982). Greek Prosodies and the Nature of Syllabification. Ph.D. dissertation, MIT. Steriade, Donca (1988). Reduplication and syllable transfer in Sanskrit and elsewhere, Phonology 5: [م [21/07/30 12:23:48 33) نم (33