Moras or Skeletal Units? A Question of Parametric Variation

Catalan Working Papers in Linguistics (CWPL) 1991: 149-165 Universitat Authnoma de Barcelona - MORAS OR SKELETAL UNITS? A QUESTION OF PARAMETRIC VARIATION Maria-Rosa Lloret Universitat Aut6noma de Barcelona One major issue that is being addressed in recent phonological research is whether the phonological representation has to recognize the existence of moras or skeletai units as primitives. The present paper argues that both approaches are not necessarily mutually exclusive, since there exists the possibility that some phonological processes have to refer to both phonologicai entities in order to account for all the facts of a language. In fact, tbis dichotomy wrongly appeared by facing Hayes' (1989) moraic theory with other theories of syllable structure. However, if one takes into amunt that neither the skeleton can be disregarded nor syllable weight is always a phonological variable, a new ordering of phonological primitives can be suggested, which does not cmespond to altemative prosodic theories but reflects the parametric variation that languages show between moraic and syllablic structure. 1. Introduction. One major contribution of non-linear phonology to the general phonological theory has been the introduction of the skeleton (McCarthy 1979) as an intermediate level between syllabic constituents and melodic segments. (1) O O R ( = syllable A I O =onset N C R = ~hp N = nucbw I i i "vX 7 7 X = skelersl unit ) g a 1 a [gal The independence of this level with respect to the others plays an important role in the description of some segments (long vowels, geminate consonants, complex segments) as well as in the justification of certain phonological processes (total assimilations, compensatory lengthenings, reduplications). For example, the representation of long segments as two skeletal units associated with the same melody captures and satisfactorily expresses the ambiguous behavior of long vowels and consonants, which sometimes act as if they were two segments while others as if they were a single entity. In the Cushitic Oromo language, for instance, phonological processes that affect quantity, such as epenthesis, treat long consonants as a sequence of two segments: the epenthetic vowel [i], which is inserted to avoid triconsonantic sequences, occurs after two heterogeneous consonants (2a) as well as after a geminate consonant (2b). (2) a. arg+na -> argina ' we see' gonf+ta -> gonfita 'you (sing.) adorn' b.kenn+na -> kennina 'we give' mudd+ta -> muddita 'you press' XX X X XX XX X X XXX II \/+ II-II~~III k e n na ke n i n a [kennina] However, processes that involve melodic changes, such as labialization, treat long segments as a single entity: [w] spreads its labial characteristic to a preceding vowel, whether this is short (3a) or long (3b). (3) a. ilbiisaw+na -> ilbiisowna -> ilbiisofna 'we are simple' b. lakkaaw+na -> lakkoowna -> lakkoofna ("lakkaofna) 'we count' II \/VI+II-II I II1 a k a wna la k O yna XX X X X XXXX .I LAB The presence of a skeletal tier is also proved through the analysis of phonological processes such as compensatory lengthenings. In Latin, for instance, deletion of /s/ before anterior sonorants entails lengthening of a preceding vowel (4). If one assumes the existence of a skeletal tier, compensatory lengthening can be universally justified as a spreading process that repairs ill-formed structures resulting from the appearance of empty X positions, which othenvise could not be incorporated into the prosodic structure. (4) *kasnus -> kaanus 'gray' * kosmis -> koomis 'courteous' XXXXXX XXXXXX XXX XXX ka snu s ka nus ka nus [kaanw] Although the previous data seem to adequately justify the presence of a skeleton, during the last decade an alternative prosodic representation has been proposed (Hayes 1989, among others), according to which the skeletal units (X) and the syllabic constituents (O, R, N, C) are replaceú by moras (5). Using this approach, moras do not merely represent a segment (as X did) but play two different roles: they count as phonological positions (i.e. they are phonological primitives) and also represent the contrast between light (or monomoraic) and heavy syllables. The defenders of such an approach regard this theory as superior to the one that claims the existence of a skeleton and syllabic structure, because moraic representation does not only capture the duality of some segments (e.g. long segments) but justifies the phonological processes that involve syllable weight (e.g. compensatory lengthening) better than previous approaches. For exarnple, moraic theory explicitely captures the cross-linguistic generalization that only segments deleteú in coda position trigger lengthening of an adjacent segment. Thus, in Latin deletion of /s/ entailed compensatory lengthening in coda position (4 and 6a) but not word-initially (6b), because in this position /s/ does not have moraic value. Note that in order that the theory that assigns skeletal positions to all the segments accounts for the data in (6b), one has to idiosyncratically stipulate that melodies only spread to syllable-final empty positions, so that the empty word-initial X position is not filled by spreading but deleted through convention because of its stranded character (6b'). In moraic theory, however, this is independently motivated because only syllable-final consonants make their syllable heavy, i.e. they have moraic value. (6) a. *kasnus -> kaanus 'gray' * kosmis -> koomis 'courteous' b. *snurus -> nurus 'daughter-in-law' smereo -> mere0 'I deserve' Cf. b' O n + A P " N N A c X XXXX IIIIY XXXX X I s n u ru s 0 n u ru s 9 1111 I 1 As a consequence of this reanalysis of the prosodic structure, two alternative prosodic theories are presented nowadays in the literature: the theory of syllable structure , whose primitives are the skeletal units (X), the syllabic constituents (O, R, N, C), and the syllables (O), and the moraic theory , which replaces the skeletal units and the syllabic constituents by moras (p). (7) Prosodic theories: a. Theory of syllable structure (X; O, R, N, C; O) b. Moraic theory (p; O) The aim of this paper is to prove that both theories are not necessarily mutually exclusive but their differences correspond to some kind of parametric variation that languages show. In general, the distinction established by moraic theory between light and heavy syllables turns out to be crucial (i.e. phonologically relevant) for languages that show a phonemic contrast between short and long vowels, as it was already pointed out by Troubetzkoy (1939).1 However, syllable weight is not a phonological variable in languages without phonemic vowel length. In these languages, instead, the interna1 structure of the syllabic constituents becomes relevant. The article is organized as follows. We first provide data to support the claim that the dichotomy presented in (7) does not match the real facts of the languages. Section 2 justifies the level of skeletal representation even in languages with well-motivated moraic structures while section 3 shows that the role of the skeleton can not be taken over by melodic structure either. 154 With this done, we suggest a new ordering of phonological primitives from which the parametric variation that languages show can be conectly established (section 4). 2. Skeleton and Moras. Hayes (1989) claims that skeletal units are empirically unnecessary because compensatory lengthening does not take place systematically whenever a segment deletes leaving an empty prosodic position behind, but only comes along in certain positions, which ,not accidentally, always have moraic value.2 In Oromo, however, there exists a case of compensatory lengthening that can only be justified with the presence of the skelet~n.~ Oromo shows a phonemic contrast between short and long vowels and, thus, syllable weight functions as a phonological variable. The language treats CV and CVC syllables as monomoraic or light, while CVV and CVVC syllables are treated as heavy or bimoraic.4 (8) Orom: Light or monomoraic syllables: [CVIp, [CVCIp Heavy or bimoraic syllables: [CVIp [VIp, [CVIp [VCIp This division between the two different types of syllables turns out to be crucial, for instance, to justify the behavior of such syllables in relation to the general constraint that the language shows on the syllable structure of the words, namely Oromo does not allow long vowels in two successive syllables unless the second is also word-final. If concatenation of morphs puts together syllables with long vowels (9a), the vowel of the second syllable obligatorily shortens. This constraint, however, does not concern CVC syllables (9b). (9) a. gaal+oota -> gaa 1 o ta 'camels' (cf. nam+oota -> na moo ta 'men') afcaal+oom- -> afcaa 10 m- 'be kind' (cf. ko?+oom- -> ko ?oo m- 'be busy') b. fard+oota -> far doo ta 'horses' lugn+oom- -> lug noo m- 'be coward' With these precisions established on the weight of the Oromo syllables, let us now look at the peculiar compensatory lengthening process that the language shows. Oromo has a deletion process that erases the melodic content of a glottal consonant (i.e. glottal stop /?/ and implosive lcf I) in s yllable-final position. This deletion entails lengthening of the preceding vowel ( 10). According to Hayes' moraic theory, this kind of deletion should not yield compensatory lengthening because the deleted glottal consonant does not leave an empty mora behlnd (10a). (Remember that in Oromo final consonants do not add weight to the syllable, i.e. they have no moraic value.) However, the data prove that such lengthening occurs, which can only be justified by the presence of a skeleton (10b). (10) fecf+na -> feena 'we wish' a?+ta -> aata 'you push' b. XXXXX XXXXX XX XXX f ecft a fe ta fe 1 a [feeta] Since Oromo seems to require the presence of a skeleton, one could think that the language does not need to refer to its moraic structure but, instead, its syllable structure would function as a phonological variable. We will next show, however, that syllable structures are unable to account for the Oromo facts. This will prove that in Oromo, as in any language with phonemic vowel length, syllable weight is a phonological variable, which is better expressed through moraic representation. If we assume that Oromo has the syllable structure shown in (1 l), we correctly capture the Oromo syllable division between light CV(C) and heavy CVV(C) syllables: the rhyme of light syllables consists only of nucleus while the rhyme of heavy syllables consists of nucleus and coda. However, the representation of (1 1) cannot justify the compensatory lengthening process (CL) that the language shows, because the deleted final consonant occupies a position different from that occupied by the second part of a long vowel, narnely the former is included within the nucleus while the latter constitutes the coda, and there is no reason to claim that the language has two different types of long vowels. light syiiables kasyllables The syllable structures represented in (12) and (13) are also untenable for Oromo. In this case the compensatory lengthening process can be adequately represented because the vowel lengthened as a consequence of the deletion of a final consonant results in a long vowel with the same structure of underlying long vowels (i.e. a complex nucleus in (12) or a branching rhyme in (13)). Note, however, that in these cases the division between light and heavy syllables is incorrectly established. A first version of this paper was presented at the Pnmer Coloquio de GramAtica Generativa held at Miraflores de la Sierra, Madrid, Spain, 20-22 March 1991. The work benefited from a Postdoctoral Research Grant and the PB 89-0344 C.1.C.Y .T. Grant. Notes 1 See Newman (1972) for a general statement on the importance of the syllable weight concept. 2 See the previous Latin example in (6). 3 The following Oromo data are taken from Lloret (1988). 4 Note that Oromo contrasts with Latin in treating CVC syllables as light. 5 There are only two exceptions, which have to be lexically marked: (i) k1irit'+ii -> k'irit' ii (*kfirit't'ii) 'chisel' (t' = ejective or glottalized t) (ii) hiik+ii -> hiik ii (*hiikkii) 'extension' 6 This phenomenon is traditionally referred to as liaison . 7 The other prosodic theories are segmental in the sense that the number of prosodic segments intuitively wrresponds to the number of melodic segments. References Clements, George N. and Jay Keyser (1983) CV Phonology , M.I.T. Press, Carnbridge, Massachusetts. Hayes, Bruce (1989) "Compensatory Lengthening in Moraic Phonology", Linguistic Znquiry , 20, 253-306. Hyman, Larry (1985) A Theory of Phonological Weight , Foris Publications, Dordrecht. It6, Junko (1989) "A Prosodic Theory of Epenthesis", NLLT ,7, 217-59. Kaye, Jonathan D. and Jean Lowenstamm (1984) "De la syllabicité" in F. Dell, D. Hirst, and J .-R. Vergnaud (eds.) Fome sonore du langage: structure des représentations en phonologie , Hermann, Paris. Lloret, Maria-Rosa (1988) Gemination and Vowel Length in Oromo Morphophonology , Ph.D. dissertation, Indiana University, Bloomington, Indiana. McCarthy, John (1979) Formal Problems in Semitic Phonology and Morphology , IULC, Bloomington, Indiana. McCarthy, John and Alan Prince (forthcoming) Prosodic Morphology , M.I.T. Press, Carnbridge, Massachusetts. Newman, Pau1 (1972) "Syllable Weight as a Phonological Variable", Studies in Apican Linguistics ,3,301-23. Troubetzkoy, N. S. (193911967) Principes de phonologie , Klincksieck, Paris. UAB Departamenl de Filologia Catalana 081 93 Bellatema (Barcelona)