Competition between syllabic and metrical constraints in two Bunun dialects *.

Huang, Hui-Chuan J.

Abstract

This article shows how syllabic and metrical constraints interact differently in two dialects of Bunun, despite their similarity in that both dialects exhibit modifications of vowel clusters in response to the ONSET constraint. In Isbukun Bunun, foot forms are constrained both syllabically and moraically. Stress shifts from the unmarked penultimate syllable to final position under duress to satisfy both ONSET and the requirement that heavy syllables must not stand in a prosodically weak position. In contrast, in Takituduh Bunun, metrical wellformedness is considered more important, so the preferred disyllabic foot forms are maintained at the cost of creating onsetless syllables. The analysis is formalized within Optimality Theory by ranking syllabic and metrical constraints differently in the two dialects. OT is advantageous in analyzing the data for two additional reasons: 1) the correlation between stress assignment in nonsuffixed and suffixed words in Isbukun can be directly captured by an output constraint, and 2) the metrical influence on syllabification in Takituduh can be readily handled by allowing syllabic and metrical constraints to interact in the same hierarchy. An implication of the Isbukun data for the properties of surface glides is that pre-peak glides can be moraic or not, depending on whether they follow a tautosyllabic consonant.

1. Introduction

Bunun is an Austronesian language spoken by about 30,000 people in central and southern Taiwan. According to the classification in Li (1988), Bunun can be divided into five dialects: the northern dialects Takituduh and Takibakha, the central dialects Takbanuad and Takivatan, and the southern dialect Isbukun. Given that there have been relatively few studies on the phonology of this language, the objectives of this article are to describe and analyze the prosodies of Takituduh and Isbukun based on first-hand data. Specifically, the article will show how syllabic and metrical considerations interact differently in the two dialects: in Isbukun, syllabic wellformedness constraints may force a marked foot form, but in Takituduh, foot form constraints may force an otherwise illegitimate syllable.

Geographically, Isbukun is spoken in the widest area among the five dialects, including Nantou, Gaoxiong, Hualian and Taidong Counties; in contrast, Takituduh is spoken only in the Renai township of Nantou County (Li 1988). There have been only a few studies on Takituduh (Li 1988, 1992; Jeng 1999), but a number of works on Isbukun Bunun, including He et al. (1986), Li (1988, 1997), Huang (1997), Lin (1996), Nojima (1996), Lin et al. (2001), and Zeitoun (2000). (1) Many phonological descriptions about Isbukun are based on the variety in Nantou County (e.g., Li 1997; Lin 1996), except He et al. (1986) and Lin et al. (2001), which are based on speakers from Gaoxiong County. A noticeable subdialectal difference is that while the Isbukun dialect in Nantou exhibits final stress (Li 1997: 306), Gaoxiong Isbukun in general stresses the penultimate syllable (He et al. 1986: 7).

The Isbukun data in the article are based on speakers from Gaoxiong County. (2) The findings indicate that the stress patterns of Gaoxiong Isbukun (hereafter, Isbukun) are more intricate than straightforward unmarked penultimate stress in that the weight of the final syllable plays a role. All words that end with a heavy syllable have final stress, except for a few forms of reduplicated shapes and onomatopoetic words, which stress the penultimate despite the final heavy syllables. Words ending in a light syllable generally have penultimate stress, but some nonsuffixed forms ending in a light syllable exceptionally carry final stress, many of which are historically contracted forms. The fact that Isbukun words avoid unmarked penultimate stress when the final syllable is heavy suggests that the language constructs a quantity-sensitive trochaic (left-prominent) foot at the fight edge of words. The weight of the final syllables in suffixed forms is dependent upon a number of modifications at the morpheme boundaries, which take place due to the restriction that syllables must have onset consonants. The article will show how these syllable-related processes determine the metrical structure of Isbukun.

Takituduh Bunun exhibits very different interactions between syllabic and metrical requirements. The placement of stress is insensitive to syllable quantity, and furthermore, the syllabic constraint ONSET can be violated under some circumstances. The interactions between the syllabic and metrical considerations in the two Bunun dialects are analyzed within the framework of Optimality Theory (McCarthy and Prince 1993; Prince and Smolensky 1993), which is well suited to capturing dialectal variations via constraint rankings. It will be shown that the prosodic differences between Isbukun and Takituduh arise from ranking syllabic constraints and foot form constraints differently in the two dialects.

The article is organized as follows. Section 2 presents the data and an analysis of Isbukun Bunun, showing how the onset requirement in the dialect affects the placement of stress. Section 3 discusses the case of Takituduh Bunun. Section 4 summarizes and compares the prosodic similarities and differences of the two dialects. Section 5 discusses the theoretical implications of the analysis and concludes the article.

2. Isbukun Bunun

In order to show how syllabic considerations affect the metrical structure of Isbukun, this section will first present an overview of Isbukun phonemes and syllables (Section 2.1), show the ways that the constraint ONSET forces heteromorphemic vowel clusters to undergo changes (Section 2.2), and illustrate the stress patterns of nonsuffixed forms (Section 2.3) and those of suffixed forms (Section 2.4). Section 2.5 gives an optimality-theoretic analysis of the stress patterns.

2.1. Isbukun phonemes and syllables

The phonemic inventory of Isbukun Bunun includes fourteen consonants /p t k ? b d s h [>>] l m n q/and three vowels/i u a/(He et al. 1986; Li 1997). Surface glides [j w] are derived from underlying vowels/i u/, as will become evident in the following discussion. A palatalization rule affects/t s/that appear before a high front configuration [ij] and gives rise to the alveolo-palatal allophones [t[??][??]], respectively. (3)

A maximal Isbukun syllable is CGVC or CVGC, both of which come from underlying CVVC sequences; complex syllable margins are not allowed. High vowels in an underlying VV sequence undergo gliding to avoid onsetless syllables, such as /ua au ia ai/ strings being realized as surface [wa aw ja aj], respectively; the surface manifestation of two adjacent high vowels /iu ui/, however, varies as to which vowel constitutes the nucleus. If we recognize the crosslinguistic generalization that more sonorous vowels tend to occupy the nucleus position, variation as to either/i/or/u/remaining syllabic is expected due to their comparable sonority.

Notice that the term "glide" and the symbols j and w here simply refer to the vocalic elements that on the surface do not exclusively occupy the nucleus of a syllable and do not exhibit steady vowel formants phonetically; the terminology does not entail whether these segments are moraic or not. (4) Data from stress assignment, which will be discussed in later subsections, suggest that glided vowels in Isbukun are moraic except when they occur in syllable-initial position.

Patterns of static distribution of segments within syllables can all be handled on the same proposition: whether the glided vowels are moraic or not depends on their structural affiliation within a syllable. In Isbukun, there are no CGVG syllables, despite the fact that they have the same number of segments as legitimate CGVC/CVGC syllables. The lack of CGVG syllables can be readily understood if we adopt the following assumptions: coda consonants are nonmoraic, glided vowels are moraic, and syllables are maximally bimoraic. Under these assumptions, CGVG syllables are not allowed because they violate the bimoraic upper limit of a syllable by having three moras, while CGVC/CVGC syllables obey the constraint by having only two moras. The lack of CGVG syllables cannot be attributed to a ban on the linear sequence of a vowel flanked on both sides by glides, because GVGC syllables are allowed. The existence of GVGC syllables, in contrast with the absence of CGVG, further suggests that when there is no preceding consonant to fulfill the onset requirement, syllable-initial glides are forced into the onset position and become nonmoraic, thus allowing the syllable to obey the bimoraicity constraint. The variable structural status of Isbukun prepeak glides, which depend on the presence of a preceding tautosyllabic consonant, resembles the case of Spanish discussed in Harris and Kaisse (1999).

While a maximal syllable contains a bimoraic sequence with at most one consonant on either side, a minimal syllable in Isbukun is CV, because onsets are obligatory and coda consonants are optional. An examination of Isbukun content words shows that there are in general no words containing onsetless syllables.

2.2. The onset requirement in Isbukun Bunun morphophonemics

The need to avoid onsetless syllables in Isbukun not only affects intramorphemic vowel sequences but also vowel clusters that arise through morpheme concatenation. When a vowel-final stem takes a vowel-initial suffix, the vowel sequences undergo gliding or coalescence, depending on the qualities of the adjacent vowels. If the vowels are of different height, gliding takes place, similar to the scenario in an intramorphemic vowel sequence.

(1) Vowel sequences across morpheme boundaries: unlike vowels (The suffix -un is a patient voice marker, and -av marks nonagent voice in imperatives. The agent voice marker is ma-, m-, or null in the following data.)

 Agent voice forms Suffixed forms

a. /sadu/ [sadu] /sadu-av/ [sadwav]
b. /silili/ [[??]ilili] /[??]ilili-av/ [[??]ililjav]
c. /tupa/ [tupa] /tupa-un/ [tupawn]
d. /ta[??]a[??]a/ [ta[??]a[??]a] /ta[??]a[??]un/ [ta[??}a[??]awn]

 Agent voice forms Gloss

a. /sadu/ [sadu] /' see'
b. /silili/ [[??]ilili] /'imitate'
c. /tupa/ [tupa] /'tell'
d. /ta[??]a[??]a/ [ta[??]a[??]a] /'listen to'

If the vowels are identical, the article argues that the two vowels coalesce into a bimoraic vowel, for reasons to be explained in the following subsections:

(2) Vowel sequences across morpheme boundaries: like vowels (The suffix -a marks agent voice in imperatives.)

 Agent voice forms Suffixed forms

a. /matutu/ [matutu] /tutu-un/ [tutun]
b. /ta[??]a[??]a/ [ta[??]a[??]a] /ta[??]a[??]a-a/ [ta[??]a[??]a]

 Agent voice forms Gloss

a. /matutu/ [matutu] 'pour out'
b. /ta[??]a[??]a/ [ta[??]a[??]a] 'listen to'

The above data show that underlying vowel sequences undergo glide formation or vowel coalescence in order to satisfy the requirement that syllables must have onsets. When the two vowels are of different heights, the high vowel loses its syllabicity; when the two vowels are identical, they coalesce into one vowel. Either gliding or coalescence prevents the occurrence of surface onsetless syllables, which would have arisen from underlying vowel sequences upon morpheme concatenation. (5)

2.3. Stress assignment in nonsuffixed forms

Nonsuffixed forms in Isbukun exhibit both final and penultimate stress patterns. Although minimal pairs contrasting final and penultimate stress can be found, such as [tapa] 'shelf' versus [tapa] 'blanket', the unmarked position of stress is clearly the penultimate syllable, because words with penultimate stress greatly outnumber those that carry final stress. It is therefore proposed that penultimate stress is the norm and that words with final stress are marked. The unmarked penultimate stress suggests that the last two syllables are in general parsed into a trochaic foot at the right edge of words in Isbukun.

If we regard the distribution of final or penultimate stress as completely unpredictable in nonsuffixed forms, however, there is a mysterious gap in the distribution of stress when the relationship between stress assignment and syllable quantity is taken into consideration. Assuming that CVV(C) syllables are heavy and CV(C) syllables are light, there is an asymmetry that may shed light on the nature of stress assignment in the language: among words that carry final stress, the final syllable may be heavy or light, but for words with penultimate stress, the final syllable is almost always light, except for words with reduplicated shapes, such as /muliahliah/ [muljahljah] 'tear apart' and/sinaunau/ [sinawnaw] 'make a face', and onomatopoeic words such as /tiptiah/ [tsiptsjah] 'thunder, explode' (based on the data collected so far). The data (3) below illustrate the tendency for nonsuffixed forms with final CVV(C) syllables not to exhibit unmarked penultimate stress patterns; instead, they overwhelmingly show final stress.

(3) Stress assignment in nonsuffixed forms with final CVV(C) syllables

a. [kavja[??]] 'friend'
b. [mahdjal] 'slippery'
c. [mapit[??]ja] 'cook'
d. [mahan[??]jap] 'understand'
e. [tanja] 'ear'
f. [mindja] 'pick up'
g. [tahaj] '(male name)'
h. [savaj] 'win'
i. [mat[??]i[??]baj] 'escape'
j. [??]alwa[??]] 'rat'
k. [salinhanwa[??]] 'dragonfly'
l. [mindudwa[??]] 'young'
m. [tanakaw] '(male name)'
n. [minbulaw] 'blister'
o. [takjus] 'slender'
p. [ha[??]umuj] 'testicles'

The sample here excludes nonsuffixed forms with final CVV(C) syllables such as [ma[??]jal] 'good', [makwan]] 'bad', [makwi[??]] 'narrow/slender', [madajn] 'big', [minpja] 'divide into how many', and [kasuj] 'make money', in which there is a clear prefix (ma-, min-, ka-) immediately preceding the final CVV(C) syllables. These words are excluded because the final stress can be attributed to the inability of the preceding affix to carry stress, rather than to any quantity effect.

The paucity of penultimately stressed words with final heavy syllables suggests that Isbukun obeys the Weight-to-Stress principle (WSP) (Prince 1983, 1990; Prince and Smolensky 1993), which dictates that heavy syllables cannot stand in prosodically weak positions. This quantity-sensitive stress system builds a monosyllabic foot on the final bimoraic syllable or a disyllabic left-headed foot when the final syllable is not heavy. The variable penultimate or final stress patterns in Isbukun nonsuffixed words are therefore phonologically predictable for the great majority of the vocabulary, as long as we recognize WSP and adopt the assumptions that CVV(C) is heavy and CV(C) is light.

2.4. Stress assignment in suffixed forms: consonant-final versus vowel-final stems

The effect of syllable quantity on stress placement is clearly manifested in suffixed forms. Before we proceed to the presentation of stress patterns of suffixed forms, a brief introduction to Bunun suffixes is needed.

Isbukun suffixes can be classified into two types based on whether they induce stress shift upon suffixation, a common scenario for languages across the world. A STRESS-SHIFTING SUFFIX will cause the stress to move rightward; a NON-STRESS-SHIFTING suffix, in contrast, has no effect on the stress position of the stem. Stress-shifting suffixes are illustrated by -a '(imperative agent voice marker)', -av '(imperative nonagent voice [NAV] marker)', -un '(patient voice marker)', and -an '(locative voice marker)'. Non-stress-shifting suffixes include -in '(perfect marker)', -an '(durative aspect marker)', and nominatives -a/-tia, and -an/-tan. The bound forms (clitics) of personal pronouns in the language behave like non-stressshifting suffixes; they include -ik '(1st person singular agent voice)', -im '(lst person plural exclusive agent voice)', -as '(2nd person singular agent voice)', and -am '(2nd person plural agent voice)'. The oblique case marker mas can optionally integrate with the preceding word through contraction, especially in connected speech, e.g., /saiv-an mas/ [sajvanmas] 'give-NAV Obl.' becoming [sajvas] (Li 1997: 338), and it has no effect on the stress location of the preceding word no matter whether contraction occurs or not.

The following data illustrate the different behaviors of the two types of suffixes:

(4) Stress-shifting suffixes:

 Agent voice forms Suffixed forms

a. /masabah/ [masabah] /masabah-a/ [masabaha]
b. /tisdadan/ [t[??]i[??]dadan] /tisdadan-a/ [t[??]i[??]dadana]
c. /haiap/ [hajap] /haiap-un/ [hajapun]
d. /ma[??]uman/ [ma[??]uman] /[??]uman-av/ [[??]umanav]
e. /masaiv/ [masajv] /saiv-an/ [sajvan]

 Agent voice forms Gloss

a. /masabah/ [masabah] 'sleep'
b. /tisdadan/ [t[??]i[??]dadan] 'run'
c. /haiap/ [hajap] 'know'
d. /ma[??]uman/ [ma[??]uman] 'remove'
e. /masaiv/ [masajv] 'give'

(5) Non-stress-shifting suffixes:

 Agent voice forms Suffixed forms

a. /makuan/ [makwan] /makuan-in/ [makwanin]
b. /minkailas/ [minkajlas] /minkailas-in/ [minkajlaSin]
c. /supah/ [supah] /supah-in/ [supahin]
d. /mapata[??]/ [mapata[??]] /mapata[??]-in/ [mapata[??]in]

 Agent voice forms Gloss

a. /makuan/ [makwan] 'bad'
b. /minkailas/ [minkajlas] 'getup'
c. /supah/ [supah] 'many'
d. /mapata[??]/ [mapata[??]] 'kill'

The data in (4) show that no matter whether stress falls on the penultimate or final syllable, stress moves to the unmarked penultimate position when taking a stress-shifting suffix. In contrast, a non-stress-shifting suffix in (5) does not change the stress location of its stem, no matter where the stress falls in the stem. Therefore, stress may fall on the antepenult of a suffix form, such as (5b), (5c), (5d), constituting an apparent counterexample to the generalization of a penultimate stress pattern. Since stress assignment is determined by its location in the stem in the case of a nonstress-shifting suffix (or a function word such as mas), rather than by phonological environments, the following discussion will focus on the data on stress-shifting suffixes.

All the examples given in data (4) end with a consonant, and they show uniformly that stress falls on the penultimate syllable of the suffixed form, the unmarked position for stress. Complexity arises when considering stems that end with vowels. All the stress-shifting suffixes here are vowel-initial, and they would potentially create illegitimate onsetless syllables when attaching to a vowel-ending stem. As mentioned in Section 2.2, Isbukun does not allow onsetless syllables on the surface; vowel hiatus is avoided by gliding or coalescence. As a result, when one of these stress-shifting suffixes follows a vowelending stem, gliding or coalescence takes place and stress falls on the final syllable of the suffixed form, in contrast to the penultimate stress pattern for consonant-ending stems as shown in (4). In other words, consonant-ending and vowel-ending stems exhibit different stress patterns in order to conform to the syllabification requirement of the language.

The following data (6)-(9) illustrate how consonant-ending versus vowel-ending stems exhibit different stress patterns upon suffixation due to pressure to satisfy the onset requirement. Examples of consonantending stems are given in (6) (as well as in [4] above) and (7) below: although there is a contrast between penultimate and final stress in nonsuffixed forms, stress neutralizes to the penultimate position of the suffixed form. As to vowel-ending stems, the suffixed form uniformly shows final stress, no matter whether the stem carries penultimate stress, as in (8), or final stress, as in (9).(6)

(6) Consonant-ending stems with penultimate stress

a. /maludah/ [maludah] /ludah-av/ [ludahav] 'hit'
b. /mindana[??]/ [mindana[??]] /[??]ind- [?]?inda-
 ana[??]-av/ na[??]av] 'help'
c. /kalat/ [kalat] /kalat-un/ [kalatun] 'bite'

(7) Consonant-ending stems with final stress

a. /minbas/ [minbas] /[??]inbas-av/ [??]inbasav] 'revenge'
b. /minsa- [minsa- /pinsan- [pinsan 'become
 nama[??]/ nama[??]] ama[??]-un/ ama[??]un] (nothing)'
c. /kusbai/ [kusbaj] /kusbai-a/ [kusbaja] 'fly'

(8) Vowel-ending stems with penultimate stress

a. /tupa/ [tupa] /tupa-un/ [tupawn] 'tell'
b. /tahu/ [tahu] /tahu-an/ [tahwan] 'talk'
c. /sadu/ [sadu] /sadu-an/ [sadwan] 'see'
d. /min[??]uni/ [min[??]uni] /min[??]uni-a/ [min[??]- 'become'
 unja]

(9) Vowel-ending stems with final stress

a. /piskaubu/ [pi[??]kawbu] /piskaubu- [pi[??]kawbwav] 'cheat'
 av/

The data above show that when consonant-ending stems are suffixed, the suffixed forms exhibit the unmarked penultimate stress pattern. The final stress pattern in the suffixed forms of vowel-ending stems is reminiscent of the fact that in nonsuffixed vocabulary, final heavy syllables attract stress due to WSP. Based on the evidence from stress, therefore, the article argues that a glided vowel nonetheless contributes to the weight of the syllable where it occurs. That is to say, words such as [tu.(pawn)](7) carry final stress because the final syllable is a bimoraic heavy syllable; stress does not fall on the penult *[(tu.pawn)] in that the heavy syllable cannot stand in the nonhead position of the trochaic foot. In this way, by upholding the principle of WSP in Isbukun, we can connect the phenomenon of suffixed forms of vowel-final stems exhibiting marked final stress with the lack of nonsuffixed forms with both penultimate stress and final heavy syllables.

Notice that the term "consonant-ending" here refers to stems that end with a nonsyllabic element on the surface, including glides, which are actually underlying vowels in Bunun. A surface stem-final glide becomes the onset of the following syllable upon suffixation, and the syllable-initial glide in the suffixed form patterns with true consonants in inducing penultimate stress, as shown in (7). The data from stress assignment, therefore, provide additional evidence that syllable-initial onset glides are nonmoraic, as mentioned in Section 2.1.

From a derivational point of view, stress falls on the final syllable of a suffixed form in vowel-ending stems because gliding/coalescence takes place after stress is assigned to the second-to-last vowel. A rule-based analysis of the patterns is given below.

(10) A rule-based analysis

a. Rules:

Intervocalic gliding: [V, +high] [right arrow] [-syllabic] / V_V

Stress assignment: Stress the penultimate syllable (except in
 the case of lexical exceptions)

Gliding: [V, +high, a place] [right arrow]

 {_[V, a place
 [-syllabic] / [V, -a place]_}

Coalescence: Adjacent identical vowels coalesce.

b. Derivations:

 /tupa-un/ /tutu-un/ /kusbai/ /kusbai-a/ /haiap/
Syllabi- tu.pa.un tu.tu.un kus.ba.i kus.ba.i.a ha.i.ap
fication
Inter- -- -- -- kus.ba.ja (8) ha.jap
vocalic
gliding

Stress tu.pa.un tu.tu.un kus.ba.i kus.ba.ja ha.jap
Gliding/ tu.pawn tu.tun kus.baj -- --
coa- [tupawnl [tutun] [kusbaj] [kusbaja] [hajap]
lescence

The intermediate forms/tu.pa.un/,/tu.tu.un/, and/kus.ba.i/show that the penultimate stress generalization holds true at a level before Gliding/ Coalescence takes place. To put it another way, the gliding/coalescence rule that applies later obscures the penultimate stress pattern and gives rise to final stress. The derivation of/kusbai-a/[kusbaja] illustrates that the Intervocalic Gliding rule must apply before the stress rule; otherwise, stress would be incorrectly placed on the second-to-last input vowel */kusbaia/if no additional adjustments are made.

Although the rule-ordering analysis is straightforward, the article will adopt a constraint-based analysis of the data in the following section and argue that such an approach provides more insight into the interaction between stress assignment and syllabification in the language.

2.5. An Optimality-theoretic analysis of the Isbukun data

The stress patterns in Isbukun nonsuffixed forms are correlated with the ways stress assignment interacts with syllabification in suffixed forms, and such a correlation can be better captured by a constraint-based approach such as Optimality Theory rather than by rules because the framework of OT allows the correlation to be directly embodied by an output markedness constraint that prohibits heavy syllables in prosodically weak positions. Specifically, the constraint WSP is responsible for the paucity of nonsuffixed forms with penultimate stress and final heavy syllables as well as for the final stress pattern in suffixed forms of vowelending stems.

The predictable pattern of penultimate or final stress is captured by ranking FtBIN[sigma], which dictates that a foot must be disyllabic, below WSP and a foot type constraint FTTYPE(trochaic), (9) which prefers leftheaded feet. When no final heavy syllable is present, WSP is not relevant, so that FTTYPE(trochaic) and [FTBIN.sub.[??]] lead to penultimate stress, as seen in the unmarked stress patterns of nonsuffixed forms and in suffixed forms of consonant-ending stems. When there are final heavy syllables, it is impossible to satisfy WSP, FTTYPE(trochaic), and [FTBIN.sub.[??]] at the same time. The dilemma is resolved by resorting to violation of the lower ranked [FTBIN.sub.[??]] constraint, creating a final stress pattern on the surface.

Notice that [FTBIN.sub.[??]] is actually one version of the FooTBINARITY (FTBIN) constraint, which refers to either syllables or moras in the literature. The article recognizes two separate FTBIN constraints, one referring to syllables ([FTBIN.sub.[??]]) and the other to moras ([FTBIN.sub.u]), because they are better treated as separate constraints and ranked in different places in Isbukun. (10) [FTBIN.sub.u] is ranked below WSP and FTTYPE(trochaic), so a monosyllabic foot form is forced to appear in the presence of a final heavy syllable. The constraint [FTBIN.sub.u], however, is relatively highranked. Except for lexically marked final-stressed words such as [piskawbu], [FTBIN.sub.u] is not only obeyed in words with penultimate stress but also in words which exhibit final stress. Moreover, the bimoraic requirement of a foot is supported by the observation that monosyllabic words are long although they are short if affixed, e.g., /dan/ [da:n] 'road' versus/ka-dan/[kadan] 'to build roads' (Lin 1996:42-44). (11) The lengthening of vowels can be attributed to the bimoraic minimum of a content word, which is derived from the Prosodic Hierarchy (Selkirk 1980) when taken together with Foot Binarity (Prince 1980; McCarthy and Prince 1993). That is, a content word, being a prosodic word, must consist of at least one foot, which in turn consists of two moras. Vowels in Isbukun monosyllabic roots lengthen in order to form a bimoraic foot and fulfill the minimal word requirement. Since there is no positive evidence showing that [FTBIN.sub.u] is violated on the surface except in the case of lexically marked stress, (12) the article assumes that the constraint is not crucially ranked below WSP and FTTYPE(trochaic) as FTBIN[sigma] is, in order to highlight the bimoraic property of a foot in the language. Consequently, the constraints [FTBIN.sub.[??]] and [FTBIN.sub.u] are ranked in different places.

The constraints most relevant to the set of data here are summarized below, including the three constraints WSP, [FTBIN.sub.[??]], and FTTYPE(trochaic) already mentioned above:

(11) a. WSP: Heavy syllables are stressed.
 b. [FTBIN.sub.u]: A foot must be disyllabic.
 c. FTTYPE(trochaic): Feet have initial prominence. (13)
 d. ALL-FT-RIGHT: Every foot stands at the right edge of the
 Prosodic Word.
 e. PARSE-SYL: Syllables are parsed by feet.
 f. ONSET: Syllables must have onsets.
 g. V-NUCLEUS (V-Nuc): Every [-consonantal] segment must be
 linked to the nucleus (Rubach 2000: 274) without sharing it
 with other elements.
 h. UNIFORMITY-IO: No element of the output has multiple
 correspondents in the input (McCarthy and Prince 1995a).
 i. Max-IO-V: Input vowels must have a correspondent in the
 output (McCarthy and Prince 1995a).
 j. MAx-IO-u: The mora associated with an input segment must
 be preserved in its output correspondent.
 k. OCP-PLACE: glide-vowel or vowel-glide sequences of identical
 place features, such as [wu uw ji ij], are disallowed.

The constraint ALL-FT-RIGHT crucially ranks above a number of constraints and is presumably undominated in the language. The ranking of ALL-FT-RIGHT above PARSE-SYL accounts for the fact that stress is noniterative (Kager 1999) in Bunun. Only a single foot is formed at the right edge because any other foot structure would incur additional violations of the higher-ranked ALL-FT-RIGHT. The constraint ALL-FT-RIGHT must also rank above WSP in order to account for the possible occurrences of unstressed heavy syllables in pretonic positions, such as /piskaubu-av/ [pis.kaw.(bwav)].(14) If ALL-FT-RIGHT is dominated by WSP, [pis.kaw .(bwav)] would incorrectly lose to *[pi[??].(kaw).(bwav)]. Given the established rankings ALL-FT-RIGHT >> WSP and WSP >> [FTBIN.sub.u], by transitivity, ALL-FT-RIGHT also ranks above FTBIN[sigma], which accounts for the fact that forming a monosyllabic foot at the right edge is more optimal than forming a disyllabic nonfinal foot, e.g. [min.[??].(nja)] rather than *[(min.[??]u).nja]. The two constraints ALL-FT-RIGHT and PARSE-SYL will not be included in the following discussion in order to simplify the presentation.

The constraint ONSET motivates gliding or coalescence in vowel clusters. The suffixed forms of vowel-final stems exhibit a final stress pattern, because ONSET forces a final heavy syllable; the suffixed forms of consonant-final stems carry penultimate stress, because ONSET can be satisfied without creating a final heavy syllable. In Isbukun, the syllabic constraint ONSET must crucially rank above the foot form constraint [FTBIN.sub.[??]]; otherwise the correct form [sa.(dwav)] would lose to *[sa.(du.av)]. The rankings of both ONSET and WSP above [FTBIN.sub.[??]] lead to the different stress patterns in the suffixed forms of vowel-final versus consonant-final stems.

The article adopts the ranking ONSET >> V-Nuc to represent the changes of /i u/to [j w], which needs further discussion because of the complexities associated with the differences between syllable-initial and noninitial glides. The constraint V-Nuc as formulated in Rubach (2000) penalizes glides in the onset or in the coda since they are [-consonantal] elements that are not linked to the syllable nucleus. Strictly speaking, although syllable-initial glides in Isbukun violate V-NUC, nonsyllable-initial glides satisfy V-NUC because they may share, and thus link to, the Nucleus node with another vowel. Consequently, the ranking ONSET >> V-NUC is not sufficient to express the changes of/i u/ to [j w] in the data. One way to solve the problem is to assume that noninitial glided vowels in Isbukun are motivated by the ranking of ONSET above NoDIPH (No DIPHTHONGS, Rosenthall 1997a: 142) while syllable-initial glides are motivated by ONSET >> V-Nuc. Since it is argued in the article that postvocalic vowel-glide or glide-vowel sequences are in fact bimoraic, we may term such sequences diphthongs and treat them formally as diphthong formation rather than glide formation. Another solution is to modify the definition of V-Nuc so that the same constraint is violated by both syllable-initial and noninitial glides: every [-consonantal] segment must be linked to the nucleus without sharing it with other elements. In the tableaux given below, the constraint V-Nuc, with its revised definition, is adopted. (15)

The three constraints V-Nuc, UNIFORMITY-IO, and Max-IO-V are included to illustrate the various possible strategies used to repair an ONSET violation. V-Nuc proscribes the gliding of vowels, UNIFORMITYIO prevents coalescence, and Max-IO-V prohibits vowel deletion. While ONSET >> V-Nuc leads to gliding, the ranking ONSET >> UNIFORMITYIO accounts for vowel coalescence. The constraint V-Nuc should rank below UNIFORMITY-IO, because gliding affects vowel sequences in a wider variety of contexts. Max-IO-V is ranked higher than both V-Nuc and UNIFORMITY-IO, SO that vowel deletion is not observed here.

The fact that glided and coalesced vowels retain their underlying moras is captured by the high ranking of MAx-IO-u. MAX-IO-u ensures that a final heavy syllable is created in the suffixed forms of vowel-ending stems no matter whether gliding or coalescence occurs. The advantages of adopting a high ranking MAX-IO-u, rather than IDENT-IO-u, will be discussed shortly along with the illustrating tableaux.

The output constraint OCP-PLACE disallows vowel-glide or glide-vowel sequences that are identical in terms of their place features. In this article, OCP-PLACE functions to block gliding in the case of identical vowel sequences, leading to coalescence on the surface.

The crucial rankings among these constraints are summarized below:

(12)

[ILLUSTRATION OMITTED]

The tableau in (13) illustrates how constraint interactions account for the unmarked penultimate stress pattern in the suffixed forms of consonantending stems. A consonant-ending stem allows the formation of a final light syllable in the suffixed forms, so that [FTBIN.sub.??] can be satisfied without violating other high-ranked constraints, resulting in penultimate stress.

Candidate (13a) shows that when the stem ends with a consonant, it is possible to satisfy the constraints ONSET, WSP, FTTYPE (trochaic), and FTBIN?? at the same time. The stem-final consonant is assigned to the onset of the following syllable, and the formation of a disyllabic trochaic foot at the fight edge does not lead to a violation of WSP. Placing stress in the final syllable violates [FTBIN.sub.??] if a monosyllabic foot is formed, as illustrated by candidate (13b), and the possibility of final stress is ruled out by FTTYPE (trochaic) if disyllabicity is satisfied at the cost of forming an iambic foot, as illustrated by (13c). Whether a stem carries penultimate or final stress does not make a difference in the stress placement of suffixed forms; therefore, only one constraint table is given to illustrate consonant-ending stems.

As to vowel-ending stems, the vowel hiatus created through suffixation would lead to violation of the high-ranked ONSET constraint, which must be avoided through the violation of some lower ranked constraint. The tableau in (14) illustrates how gliding occurs in response to the syllabic constraint ONSET, and how stress placement is affected accordingly.

Candidates (a) through (e) show that underlying vowels become glides to satisfy ONSET, at the cost of violating the lower ranked V-NUC. Candidates (a), (b), and (c) all place stress on the desired final syllable, but only (a) is the optimal output because (b) forms an iambic foot, violating FTTYPE (trochaic), and (c) is ruled out by MAx-IO-u due to the loss of an input mora. Candidates (d) and (e) place stress on the penultimate syllable. Candidate (d) violates WSP because the final bimoraic syllable is in the weak position of the foot. Candidate (e) avoids a WSP violation by turning the underlying vowel into a nonmoraic glide in the output, but this move again violates the equally high ranked constraint MAx-IO-u. Candidate (f) shows that given the need to obey ONSET, forming a disyllabic foot on the last two vowels of the suffixed form is impossible in the case of vowel-ending stems. A comparison of (a) and (f) suggests that the need to satisfy the onset requirement is more important than maintaining the unmarked disyllabic foot form. Candidates (g) and (h) illustrate the cases in which an ONSET violation is repaired through coalescence. The fact that candidate (g) loses to (a) shows that V-Nut is ranked lower than UNIFORMITY-IO. Candidate (h) shows that penultimate stress is impossible even if coalescence takes place, because MAX-IO-u demands that coalesced vowels preserve their input quantity. Candidate (i) satisfies the ONSET constraint by deleting one of the vowels in hiatus; although this form avoids violation of MAX-IO-u by saving the mora of the deleted vowel, it is ruled out by the constraint MAX-IO-V, which prevents deletion of underlying segments.

Tableau (15) shows that when the stem-final vowel and the suffix-initial vowel are identical, the ONSET constraint is satisfied through coalescence rather than gliding. The placement of stress in the final syllable is accounted for by assuming that coalescence, just as gliding, preserves the moras of the input vowels.

Candidates (a) through (i) in tableau (15) correspond to those in tableau (14) in terms of stress placement and the overall makeup of syllable shapes; the only difference is that the two vowels at the morpheme boundary in tableau (15) are identical whereas they are different in tableau (14). In tableau (15), candidate (g), which resorts to coalescence, wins over candidate (a), which employs glide formation, because given the sequence of two identical high vowels, gliding would create a configuration that is blocked by the high ranked OCP-PLACE.

If the two identical vowels are low vowels, gliding is not a possible strategy to repair hiatus, because low vowels lack corresponding glides. Therefore, coalescence is observed.

The fact that IDENT-IO-u cannot substitute MAX-IO-u to account for the moraic status of glided and coalesced vowels is illustrated in tableaux (15) and (16). For example, the mora-deleting candidate (16c) *[([tu.sub.u]*pa (1,2) uV)] satisfies IDENT-IO-u because either the preceding or the following vowel corresponds to a vowel associated with one mora in the output, and the optimal (16b) [tu.sub.u]*([pa.sup.(1,2).sub.uu]V)] would incorrectly lose to (16c) by violating the IDENT-IO-u constraint because each of the coalescing vowels is associated with two moras. The constraint MAx-IO-u. better captures the generalization in the data that input moras must survive in the output despite the type of segmental modification that occurs.

One might question the validity of the assumption that the correct output representation for the final syllable is a bimoraic vowel, since the evidence for assuming such a representation has so far come from stress placement, which is also what is being accounted for here. Another piece of evidence that independently supports the bimoraic analysis comes from dialectal comparison. We have mentioned in Section 2.3 that a few lexical items in Isbukun Bunun exceptionally place stress on final syllables. Some of these final stressed vowels in fact correspond to two vowels flanking an intervening glottal in Takituduh; for example [tapa] 'blanket' and [[??]uva[??]] 'child' correspond to [tapaha] and [[??]uva[??]a[??]], respectively, in Takituduh. It is possible that diachronically, the loss of glottals [h] and [??] in Isbukun (see Li 1988) is followed by the merger of adjacent vowels, and that final stress results because the adjacent two vowels both contribute a mora to the final syllable. The advantage of assuming a bimoraic analysis is that it not only renders transparent the final stress pattern in the suffixed forms of vowel-ending stems, but also accounts for the exceptional final stress in many nonsuffixed forms from a diachronic point of view. If we assume outright deletion of one of the identical vowels together with its mora, the placement of stress in [tutun] and [tupav] would pose a type of metrical opacity, which is an unnecessarily complicated analysis for the Bunun data. (16)

To summarize, the unmarked penultimate stress in Isbukun is captured through the constraint [FTBIN.sub.[??]]. In the suffixed forms of vowel-ending stems, the ranking of [FTBIN.sub.[??]] below ONSET and other relevant constraints forces the creation of final stress. The marked final stress pattern is not observed in the suffixed forms of consonant-ending stems, because in these cases [FTBIN.sub.[??]] can be satisfied without violating other higher-ranked constraints. In Isbukun Bunun, the syllabic constraint ONSET is crucially ranked above the constraint [FTBIN.sub.[??]], a metrical constraint demanding unmarked foot forms. The syllabic influence on metrical structure in Isbukun contrasts with the case of Takituduh Bunun, in which metrical influence on syllabification is observed.

3. Takituduh Bunun

In contrast to Isbukun Bunun, in which ONSET and WSP rank above [FTBIN.sub.[??]], this section will show that in the dialect of Takituduh, the rankings between the two sets of constraints are reversed. That is to say, the [FTBIN.sub.[??]] constraint that demands unmarked disyllabic foot forms is more important than the syllabic constraint ONSET, and [FTBIN.sub.[??]] also ranks above WSP. Section 3.1 presents a brief introduction of Takituduh phonemes, syllables and stress. Section 3.2 will show how the vowel clusters in Takituduh are modified in response to the ONSET constraint. Section 3.3 argues that ONSET ranks below [FTBIN.sub.[??]], based on evidence from minimal words. The OT analysis of the Takituduh data is given in Section 3.4.

3.1. Takituduh phonemes, syllables and stress

The phonemic inventory of Takituduh Bunun includes sixteen consonants /p t k q [??] b d ts s h v [??] l m n/ and three vowels/i u a/(Li 1988). Compared with Isbukun, Takituduh has two additional consonant phonemes: the voiceless uvular stop/q/ and the voiceless alveolar affricate/ts/. Surface glides are derived from the underlying vowels /i u/ as in Isbukun. The palatalization rule affects/ts s/that appear before a high front articulation [i j] and gives rise to the allophones [t[??] [??]], respectively.

Similar to Isbukun, a maximal syllable in Takituduh is CGVC or CVGC. Usually syllables are minimally composed of CV, but onsetless syllables may appear on the surface under some specific circumstances, as discussed in the following subsections.

In Takituduh Bunun, stress in general falls on the final syllable. Stress moves to the final syllables of affixed forms when stress-shifting suffixes are attached. Syllable quantity in Takituduh does not play a role in predicting stress patterns as it does in Isbukun.

3.2. The onset constraint in Takituduh Bunun

Recall that in the case of Isbukun, the onset constraint is strictly obeyed: vowel sequences are resolved by either gliding or coalescence, depending on whether the adjacent vowels are alike. In Takituduh, however, although gliding is also a strategy used to avoid onsetless syllables, vowel hiatus can be observed on the surface in some cases.

Similar to Isbukun, gliding takes place in Takituduh as a result of the relatively high-ranked ONSET constraint, as manifested in the formation of glides in tautomorphemic and heteromorphemic vowel sequences.

(17) Glide formation in tautomorphemic vowel sequences

a. /madain/ [aj] 'big'
b. /ma[??]aiv/ [aj] 'give'
c. /kuntsais/ [aj] 'change'
d. /mindia[??]/ [ja] 'pick up'
e. /madia[??]/ [ja] 'many'
f. /laksial/ [ja] 'slippery'
g. /tusauts/[aw] 'sing'
h. /mahau[??]/[aw] 'angry'
i. /maldadauk/ [aw] 'still'
j. /matsua[??]/ [wa] 'cultivate'
k. /silaluan/ [wa] 'cheat'

The data below suggest that in the case of heteromorphemic nonidentical two-vowel sequences, the second vowel turns into a glide if it is a high vowel.

(18) Glide formation in heteromorphemic vowel sequences

a. [tan[??]a] /tan[??]a-i/ (17) [aj] 'listen' (imp. patient
 voice)
b. [[??]asa] /[??]asa-i/ [aj] 'like' (imp. patient
 voice)
c. [tan[??]a] /tan[??]a-un/ [aw] 'listen' (patient voice)
d. [[??]asa] /[??]asa-un/ [aw] 'like' (patient voice)
e. [[??]ilulu] /silulu-i/ [uj] 'pull' (imp. patient
 voice)
f. [pislulu] /pislulu-i/ [uj] 'cause to know' (imp.
 patient voice)
g. [pin[??]uni] /pin[??]uni-un/ [iw] 'cause to become'
 (patient voice)

If the second vowel in a heteromorphemic nonidentical two-vowel sequence is nonhigh, glide formation is not observed, resulting in vowel hiatus on the surface. Huang (2002) attributes the nonoccurrence of gliding in the first member of the sequence to the ranking of ONSET below an output-to-output constraint, which demands that the stressed vowel in the base must remain syllabic in affixed forms (IDENT-BA-[??]).(18)

(19) Vowel hiatus on the surface: unlike vowels across morpheme
boundaries

a. [[??]i[??]ili] /sisili-a/ [ia] 'imitate' (imp. agent voice)
b. [kitlalavi] /kitlalavi-a/ [ia] 'follow' (imp. agent voice)
c. [[??]ilulu] /silulu-a/ [ua] 'pull' (imp. agent voice)
d. [matsalpu] /kajtsalpu-an/ [ua] 'sad' (locative voice)

If the two vowels across morpheme boundaries are identical, coalescence does not take place as in the case of Isbukun Bunun. Instead, vowel hiatus is observed:

(20) Vowel hiatus on the surface: like vowels across morpheme
boundaries

a. [kitlalavi] kitlalavi-i [ii] 'follow' (imp. patient voice)
b. [tan[??]a] tan[??]a-a [aa] 'listen' (imp. agent voice)

The lack of coalescence in (20) implies that the UNIFORMITY-IO constraint ranks higher than ONSET in Takituduh.

Regardless of the appearance of vowel hiatus in Takituduh, the two dialects are similar in that they both resort to gliding as a way of avoiding onsetless syllables, suggesting the ranking of ONSET above V-NUC in both dialects. The two dialects are different in the ways ONSET is ranked with respect to other constraints. In Takituduh, ONSET is dominated by the output-to-output constraint IDENT-BA-[??] and by UNIFORMITY-IO, leading to vowel hiatus on the surface. In Isbukun, however, onsetless syllables are disallowed because the syllabic constraint ONSET must be strictly obeyed.

3.3. Minimal word requirement and the violation of the ONSET constraint

The patterns of resolving vowel clusters in Takituduh suggest that ONSET is dominated by some constraints, and the fact that Takituduh stress is quantity-insensitive suggests that FTBIN[??] is ranked above WSP. The ranking relationship between the metrical constraint FTBIN[??] and the syllabic constraint ONSET remains to be determined. The data from the minimal word phenomenon may shed light on how the dialect deals with the conflict between syllabic and metrical considerations. In Takituduh, tautomorphemic nonidentical vowel sequences fail to undergo glide formation when there are just two vowels in the input, creating violations of ONSET on the surface.

(21) Nonoccurrence of glide formation in #CVV(C)# words (19)

a. /tai[??]/ [ai] 'taro'
b. /haip/ [ai] 'now'
c. /pia[??]/ [ia] 'how many'
d. /via[??]/ [ia] 'knife'
e. /vau[??]/ [au] 'eight'
f. /qau[??]/ [au] 'pillar'
g. /naun/ [au] 'cat'
h. /kaut/ [au] 'female name'
i. /buan/ [ua] 'moon'
j. /puaq/ [ua] 'flower'
k. /tsui[??]/ [ui] 'money'
l. /quin/ [ui] (20) 'body hair'
m. /niun/ [iu] 'female name'

In addition, a vowel in Takituduh must lengthen if it is the only vocalic element in the input:

(22) Vowel lengthening in Takituduh

a. /sak/ [saak] 'smell'
 /sak-a/ [saka] 'smell!' (imp. patient voice)
 /sak-un/ [sakun] 'smell' (patient voice)

b. /spat/ [paat] 'four'
 /spat-in/ [patin] 'four' (imp.)
 /RED-spat/ [saspat] 'four persons'

Given that Bunun does not have phonemic long vowels, the data (22) must involve a vowel lengthening rather than shortening rule, based on the morphophonemic alternations shown. These phonetically long vowels involve one Root node linked up to two timing slots, with the second timing slot heading an onsetless syllable (see tableau [29] below).

The disyllabic forms in (21) and (22) suggest that the ranking of [FTBIN.sub.[??]] above ONSET forces the creation of onsetless syllables, under the assumptions that a grammatical word is a prosodic word, that the prosodic word must consist of at least one foot, and that the foot must be disyllabic. When there are only two adjacent vowels in the input, glide formation fails to take place, because its occurrence would create a grammatical word that consists of a single monosyllabic foot. When there is only one vocalic element in the input, lengthening occurs in order to form a disyllabic foot, even though an onsetless syllable is thus created. The generalization that a minimal word in Takituduh must be disyllabic explains the lack of glide formation in #CVV(C)# words as well as the lengthening of vowels in #CV(C)# words. (21)

Recall that in the discussion of the FTBIN constraint (Section 2.5), lengthening in Isbukun is also reported in monosyllabic words such as /dan/[da:n] 'road,' which takes place to satisfy the bimoraicity requirement. Because the lengthening effect is much more prominent in Takituduh than in Isbukun based on the data collected, it is proposed in the article that a minimal word is monosyllabic and bimoraic in Isbukun but disyllabic and bimoraic in Takituduh. Isbukun ranks ONSET higher than [FTBIN.sub.[??]], so minimal words are monosyllabic and obey the onset requirement. Takituduh ranks [FTBIN.sub.[??]] higher than ONSET, SO onsetless syllables are created in minimal words.(22) The proposal here conforms to the hypothesis that the minimal word is bimoraic in quantity-sensitive languages and disyllabic in quantity-insensitive languages (McCarthy and Prince 1995b: 321-322).

3.4. An Optimality-theoretic analysis of the Takituduh data

The rankings between the following pairs of constraints have already been mentioned in the above discussion of the Takituduh data: (23)

(23) a. ONSET >> V-NUC
 b. UNIFORMITY-IO >> ONSET
 c. [FTBIN.sub.[??]] WSP
 d. [FTBtN.sub.[??]] >> ONSET

The ranking ONSET >> V-Nut indicates that gliding is the strategy used to avoid onsetless syllables, but gliding is prohibited if it would create a configuration that violates other higher ranked constraints such as [FTBIN.sub.[??]]. The ranking UNIFORMITY-IO >> ONSET means that the language would not resort to coalescence to avoid an onsetless syllable. The ranking of [FTBIN.sub.[??]] above WSP reflects the quantity-insensitive character of Takituduh; that is, foot forms are not constrained by the moraic composition of a syllable. The metrical structure of Takituduh is not constrained syllabically either, since [FTBIN.sub.[??]] ranks higher than ONSET. In contrast with Isbukun, in which syllabic considerations may affect foot forms, Takituduh exhibits a metrical influence on syllabification.

The ranking relationship among the four pairs of constraints is shown in the diagram below, together with other relevant constraints:

(24) Constraint rankings for Takituduh Bunun

[ILLUSTRATION OMITTED]

Candidates (c), (d), (e) show that the language employs neither vowel deletion nor consonant insertion nor coalescence, respectively, to avoid onsetless syllables, because MAx-IO-V, DEP-IO-C and UNIFORMITY-IO rank above ONSET and V-Nuc. Notice that in tableau (25), the metrical constraint [FTBIN.sub.[??]] does not play a role in winnowing out the correct output, because the input contains more than two vocalic elements. The gliding of one vowel still leaves enough vowels to form a disyllabic foot.

If adjacent vowels are identical, vowel hiatus is observed on the surface, because UNIFORMITY-IO prohibits the coalescence of the two vowels.

If the two adjacent identical vowels are high, the candidate with an onsetless syllable still wins, despite the possibility of gliding one of the high vowels and thus violating the lowest ranked V-NUC, because glide formation in this case would create a configuration violating the higher ranked OCP-PLACE.

Candidate (b) loses to the optimal (a) because glide formation creates a vowel-glide/glide-vowel sequence with identical place features, violating the constraint OCP-PLACE. In a word, the constraint tables (25)-(27) show that although the syllabic constraint ONSET forces glide formation to take place in some cases, Takituduh exhibits several cases of vowel hiatus due to the interaction of ONSET with other higher ranked constraints. The constraint [FTBIN.sub.[??]] does not have any effect in these cases, because there are more than two vowels in the input.

In the case of exactly two vowels in the input, each of the vowels remains syllabic in the output due to the disyllabic requirement, as shown below:

The constraint [FTBIN.sub.[??]] rules out candidates (b), (c), and (e) because gliding, vowel deletion, and coalescence all reduce the number of syllables from two to one. Although insertion of an epenthetic consonant would allow the formation of a disyllabic foot and avoid violating [FTBIN.sub.[??]], as illustrated by candidate (d), the ranking of DEP-IO-C above ONSET favors (a) with an onsetless syllable. (25) Notice that if [FTBIN.sub.[??]] is ranked below ONSET as in the case of Isbukun, the incorrect output (b) would be selected instead.

When there is just one vowel in the input, vowel lengthening takes place in order to meet the requirement of a disyllabic foot, as shown in tableau (29) below.

Maintaining the length of the input vowel would violate the constraint [FTBIN.sub.[??]], as illustrated by candidate (c). Candidate (d) shows that insertion of both a consonant and a vowel to satisfy [FTBIN.sub.[??]] and ONSET is considered less optimal than violating just the ONSET constraint. Notice that candidate (a) with vowel lengthening wins over (b) with vowel insertion, no matter whether DEP-IO-V, the constraint against vowel insertion, ranks above or below DEP-IO-[u], the constraint against vowel lengthening, because violation of DEP-IO-V implies violation of DEP-IO-[u]; in cases where the violation of DEP-IO-[u] is necessary in order to satisfy [FTBIN.sub.[??]], DEP-IO-V will favor a representation with a linked structure rather than the insertion of a whole segment. (27)

The proposed OT analysis of the Takituduh data shows that although the syllabic constraint ONSET plays an important role in the modification of vowel clusters, ONSET is dominated by other constraints, leading to the occurrences of onsetless syllables under some circumstances.

4. Summary: a comparison of the two dialects

The prosodic similarities and differences between Isbukun and Takituduh can be summarized as follows. The two Bunun dialects are similar in that they both employ glide formation as a strategy to deal with underlying vowel sequences, which is reflected in the ranking ONSET >> V-NUC.

Although Isbukun and Takituduh are similar in adopting gliding as a strategy to avoid vowel clusters, they are different in how sequences of rising sonority and sequences of identical vowels are dealt with in suffixed forms. In Isbukun, vowel clusters are systematically avoided both within and across morphemes through the application of gliding between adjacent nonidentical vowels and coalescence between identical vowels. In Takituduh, however, modifications of vowel clusters fail to take place when suffixation creates a sequence of rising sonority or identical vowels even though in other combinations gliding takes place as in the case of Isbukun. The differences are attributed to the undominated status of ONSET in Isbukun and the ranking of ONSET below IDENT-BA-[??] and UNIFORMITY-IO in Takituduh.

Takituduh and Isbukun are also different in their stress patterns. Takituduh stress falls on the final syllable, and its quantity-insensitive character is reflected in the ranking [FtBin.sub.[??]] >> WSP. Isbukun stress is basically penultimate, but exhibits final stress when gliding or coalescence creates a final heavy syllable. In Isbukun, the quantity-sensitive character is captured by the ranking WSP >> [FtBin.sub.[??]]; ONSET also needs to rank higher than [FtBin.sub.[??]], in order to account for the final stress pattern.

Another difference between the two dialects is how the minimal word requirement is fulfilled. While a content word in Takituduh must be minimally disyllabic, an Isbukun content word can be monosyllabic and bimoraic. Based on the assumption that a content word is a prosodic word, which consists of at least one foot, the analysis captures the difference by proposing that [FTBIN.sub.[??]] is ranked differently with respect to other constraints in the two dialects. Specifically, Takituduh ranks [FTBIN.sub.[??]] above ONSET, which leads to the appearance of onsetless syllables when the input contains no more than two vowels, and Isbukun ranks [FTBIN.sub.[??]] below ONSET, which is responsible for the regular occurrence of gliding even if the input contains just two vowels as well as for the marked final stress pattern.

5. Discussion and conclusion

The article has illustrated the prosodic characteristics of the two dialects of Bunun and formalized their similarities and differences by means of the constraint ranking mechanism in OT. The two dialects under comparison are distinct from each other in how syllabic and metrical constraints rank with respect to each other, although they are similar in that violation of ONSET is primarily resolved through glide formation. In Isbukun, the syllabic constraint ONSET has priority over the metrical constraint [FTBIN.sub.[??]]; when the concatenation of a vowel-ending stem and a vowel-initial suffix renders the satisfaction of both ONSET and [FTBIN.sub.[??]] impossible, the dialect chooses to maintain syllable wellformedness at the cost of creating marked monosyllabic feet. In contrast, Takituduh opts for the preferred disyllabic foot form when the demands of syllabic and metrical constraints are in conflict, ranking ONSET below [FTBIN.sub.[??]]. The different results of the competition between syllabic and metrical constraints within the same language family are captured naturally within the framework of OT by ranking the constraints ONSET and [FTBIN.sub.[??]] differently in the two dialects.

An output-oriented theory such as OT is advantageous in analyzing the facts in the two Bunun dialects. First of all, the framework of OT allows syllabic and metrical constraints to interact in the same hierarchy, and such parallelism makes it possible to account for both the top-down influence of foot forms on syllabification in Takituduh and the fact that foot forms are constrained syllabically and moraically in Isbukun. As already pointed out in previous studies (Prince and Smolensky 1993; Rosenthall 1997b), it is difficult to capture the effect of metrical structure on syllabification in an approach that builds prosodic structures from the bottom up. Moreover, in Isbukun, the relatively high-ranked constraint WSP directly expresses the correlation between the static distribution of stress in nonsuffixed words and the dynamic assignment of stress in morphophonemic alternations. That is, the need to avoid heavy syllables in the nonhead position of a foot is responsible for the lack of words with penultimate stress and final heavy syllables in nonsuffixed words, and also for the different stress patterns induced by consonant-ending versus vowel-ending stems in suffixed words.

One important implication of the analysis of Isbukun is that glides that derive from an underlying vowel may pattern together with consonants or vowels on the surface, a counterexample to the prediction in Levi (2003) that if a language has only derived glides but not underlying glides, the derived glides in the language should pattern with vowels. In Isbukun, glide-ending stems pattern with consonant-ending stems in inducing penultimate stress in the suffixed forms, despite the fact that surface glides are exclusively derived from underlying vowels. The Isbukun data suggest that the behavior of a derived glide in stress assignment is not tied to its underlying status but is dependent on its syllabic affiliation on the surface.

Furthermore, the coalescence analysis of Isbukun vowel clusters shows that IDENT-IO-[u] is problematic in winnowing out the correct output; the analysis can be successful only if we adopt the constraint MAx-IO-[u], which makes direct reference to whether the mora of an input segment is preserved in the output. The high ranking of MAx-IO-[u] and the low ranking of V-NUC in Isbukun reflect the fact that underlying vowels remain moraic on the surface even though they do not exclusively occupy the nucleus of a syllable on the surface due to the ONSET constraint. The data from Isbukun Bunun shows that the loss of syllabicity does not necessarily entail the loss of moraicity of a segment.

Received 27 August 2003 Revised version received 5 October 2004

References

Bernhardt, Barbara H. and Stemberger, Joseph P. (1998). Handbook of Phonological Development." From the Perspective of Constraint-Based Nonlinear Phonology. San Diego, CA: Academic Press.

Garrett, Edward (1999). Minimal words aren't minimal feet. In Papers in Phonology 2, UCLA Working Papers in Linguistics 1, Matthew K. Gorden (ed.), 68 105. Los Angeles: UCLA Linguistics Department.

Harris, James W. and Kaisse, Ellen M. (1999). Palatal vowels, glides and obstruents in Argentinian Spanish. Phonology 16, 117 190.

He, Ru-fen; Zeng, Si-qi; Li, Wen-shu; and Lin, Qing-chun (1986). Gaoshanzu Yuyan Jianzhi (Bunenyu) [An outline of the Formosan native languages (Bunun)]. Beijing: Min-Zu.

Huang, Hui-chuan J. (2002). Zhuoshe Bunongyu de huayin xingcheng guilu [Glide formation in Takituduh Bunun]. Tsing Hua Journal of Chinese Studies 32(2), 441-468. Hsinchu, Taiwan: National Tsing Hua University.

Huang, Lillian (1997). Gaoxiongxian de bunongyu [Isbukun Bunun in Gaoxiong County]. In Gaoxiongxian Nandao Yuyan [Formosan Austronesian languages in Gaoxiong County], Paul J.-K. Li (ed.), 351-410. Gaoxiong: Gaoxiong County Government.

Jeng, Heng-hsiung (1977). Topic and Focus in Bunun. Institute of History and Philology, Special Publication No. 72. Taipei: Academia Sinica.

Jeng, Heng-syung (1999). Bunun tense and aspect. In Selected Papers from the Eighth International Conference on Austronesian Linguistics, Symposium Series of the Institute of Linguistics (Preparatory Office) No. 1, Elizabeth Zeitoun and Paul Jen-kuei Li (eds.), 455-487. Taipei: Academia Sinica.

Kager, Rene (1999). Optimality Theory. Cambridge: Cambridge University Press.

Levi, Susannah (2003). Representing underlying distinctions between vowels and glides. Paper presented at the 77th annual meeting of the Linguistic Society of America, Atlanta.

Li, Paul J.-K. (1987). The preglottalised stops in Bunun. In A Worm of Language: Papers Presented to Prof S. A. Wurm on His 65th Birthday, Pacific Linguistics 100, Donald C. Laycock and Werner Winter (eds.), 381 387. Canberra: Pacific Linguistics.

--(1988). A comparative study of Bunun dialects. Bulletin of the Institute of History and Philology 59(2), 479 508.

--(1992). Taiwan Nandao Yuyan de Yuvin Fuhao Xitong [Orthographic systems for Formo- san languages]. Taipei: Educational Research Committee, Ministry of Education, Taiwan.

--(1997). Nantouxian Xinyixiang de bunong nanbu fangyan [Isbukun Bunun in Xinyi Township, Nantou County]. In Gaoxiongxian Nandao Yuyan [Formosan Austronesian languages in Gaoxiong County], Paul J.-K. Li (ed.), 300-350. Gaoxiong: Gaoxiong County Government.

Lin, Hsiu-hsu (1996). Isbukun phonology: a study of its segments, syllable structure and phonological processes. Unpublished Master's thesis, National Tsing Hua University.

Lin, Tai; Zeng, Si-qi; Li, Wen-shu; and Ismahasan, Bukun (2001). Isbukun Bunongyu Goucifit Yanjiu [A study on the morphology of Isbukun Bunun]. Taipei: Ducewenhua.

McCarthy, John and Prince, Alan (1993). Prosodic morphology I: constraint interaction and satisfaction. Unpublished manuscript, University of Massachusetts, Amherst and Rutgers University.

--and Prince, Alan (1995a). Faithfulness and reduplicative identity. In Papers in Optimality Theory, University of Massachusetts Occasional Papers 18, Jill N. Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), 249-384. Amherst, MA: Graduate Linguistic Student Association.

--and Prince, Alan (1995b). Prosodic morphology. In The Handbook of Phonological Theory, John Goldsmith (ed.), 318 366. Cambridge, MA: Blackwell.

Nojima, Motoyasu (1996). Lexical prefixes of Bunun verbs. Gengo Kenkyu 110, 1-27. Prince, Alan (1980). A metrical theory for Estonian quantity. Linguistic Inquiry 11, 511-562.

--(1983). Relating to the grid. Linguistic Inquiry 14, 19-100.

--(1990). Quantitative consequences of rhythmic organization. Chicago Linguistic Society (26)2, 355-398.

--and Smolensky, Paul (1993). Optimality Theory: constraint interaction in generative grammar. Unpublished manuscript, Rutgers University and University of Colorado.

Rosenthall, Sam (1997a). Vowel/Glide Alternation in a Theory of Constraint Interaction. New York: Garland.

-- (1997b). Metrical influence on syllable structure in Lenakel. Linguistic Inquiry 28(1), 187193.

Rowicka, Grazyna (1999). Prosodic optimality and prefixation in Polish. In The Prosody Morphology Interface, Rene Kager, Harry van der Hulst, and Wim Zonneveld (eds.), 367-389. Cambridge: Cambridge University Press.

Rubach, Jerzy (2000). Glide and glottal stop insertion in Slavic languages: a DOT analysis. Linguistic Inquiry 31(2), 271-317.

Selkirk, Elisabeth (1980). The roles of prosodic categories in English word stress. Linguistic Inquiry 11,563-605.

Zeitoun, Elizabeth (2000). Bunongyu Cankao Yufa [A reference grammar of Bunun]. Taipei: Yuanliu.

Notes

* The article was presented at the 15th North American Conference on Chinese Linguistics. I thank the audience at the conference for helpful comments. Sincere gratitude also goes to the major Isbukun consultants Bukun Ismahasan and Lamata Istanda, and to the Takituduh consultants Maduvaian Savung and Talum Qalavangan. The research leading to this article was supported by NSC90-2411-H-007-025 in Taiwan. Correspondence address: Graduate Institute of Linguistics, National Tsing Hua University, 101, Section 2, Kuangfu Road, Hsinchu, Taiwan, 300. E-mail: hcjhuang@mx.nthu.edu.tw.

(1.) Also see Li (1987, 1988) for a cross-dialect perspective, and Jeng (1977) on Takbanua[??] Bunun.

(2.) Unless noted otherwise, all the data presented in the article are based on the author's own continuing fieldwork, which is comprised of sentence elicitations and short narratives so far. The generalizations regarding stress placement as presented in the article are consistent with either type of data, but the article does not address all the observed processes that affect underlying vowel sequences; see Note 5.

(3.) The palatalization rule may optionally change/s/to [[??]] after [ij].

(4.) This is different from the usage of "glide" as a mora-less vowel in standard moraic theory, as pointed out by an anonymous reviewer. As will become evident in the following discussion, the low-sonority high vowels in a vowel sequence are forced by the ONSET constraint to occupy the onset position and become nonmoraic, and when such vowels occupy nonsyllable-initial position, they remain moraic on the surface. Because both types of segments are derived by the onset requirement, the article uses one term "glide" to refer to them, despite the fact that they are different in moraicity.

(5.) A reviewer points out that there are data involving the additional changes of/ai, au/ to [i] and [[??]]: the change of /au/to [[??]] is observed especially in connected speech, and the change of/ai/ to [i] occurs within a word (e.g., maisna/misna 'from') as well as across words (e.g., minsuma-in/minsumin 'to have appeared'). Although the monophthongization of /au/to [[??]] and the change of /ai/to [i] are also processes that affect vowel sequences, they are not discussed here for the following reasons. The reduction nature of monophthongization is not directly related to the onset requirement, and its phonetic realization is gradient, manifesting much intra- and interspeaker variations. As far as the change of/ai/to [i] is concerned, it is unclear how many roots are affected by this optional rule (Li 1997: 306), and it seems that there are also variations in how speakers use the forms with the reduced [i]. The/-in/ suffix as in /minsumain/ is nonstress-shifting whereas the article includes only stress-shifting suffixes (see Section 2.4). Notice that no matter whether/au/monophthongizes to [[??]] and whether /ai/ changes to [i], they do not affect the generalization in the article that Isbukun avoids the unmarked penultimate stress when the final syllable is heavy. The apparent violations in cases such as /minsuma-in/ [minsumajn] do not constitute a counterexample because they involve different types of suffixes.

(6.) As the data given in (1), (2), (4), and (5), the data (6)-(9) contain an agent voice form in the left and a suffixed form to its right. The agent voice forms in the left column carry the agent voice marker ma-, m-, or null, which might not show up in the suffixed forms if there is a conflict in voice. The apparent alternations involving the beginning consonants in examples (6b), (7a), and (7b) are due to a separate process deleting the initial consonants in forms with the agent voice marker m-.

(7.) The dots indicate syllable boundaries, and the parentheses represent foot structures.

(8.) In a narrow transcription,/kusbai-a/and/haiap/are pronounced as [kus.baj.ja] and [haj.jap], respectively, in which the syllable-initial glides are also perceived as part of the preceding syllables.

(9.) Kager (1999: 172) uses the constraints RhType=T and RhType=l to state whether a rhythmic type should be trochaic or iambic.

(10.) Rowicka (1999: 373) also proposes that there should be separate [FTBIN.sub.[??]] and [FTBIN.sub.[u]] constraints.

(11.) The data in Lin (1996) are based on the Isbukun dialect spoken in Nantou County. The lengthening phenomenon is observed in Gaoxiong Isbukun too, as illustrated by /spat/ [pa:t] 'four'.

(12.) Presumably [FTBIN.sub.[??]] is ranked only below the faithfulness constraint that demands underlying stress to be preserved.

(13). The article assumes that a foot consisting of a single syllable does not violate FTTYPE (trochaic); a monosyllabic foot violates [FTBIN.sub.[??]] but not FTTYPE (trochaic).

(14.) In addition, to account for the fact that [pi[??]kaw(bwav)] is more optimal than *[pi[??](kawbwav)], we seem to need a separate higher-ranked WSP constraint, which prohibits only heavy syllables in the nonhead position of a foot but not those unstressed heavy syllables that are linked directly to the Prosodic Word node. Alternatively, a constraint that prefers the stressed syllable to be right-aligned would rule out the incorrect *[pi[??](kawbwav)].

(15.) Notice that the ranking ONSET >> IDENT-IO-[u] (e.g., Rosenthall 1997b) alone cannot capture the changes of /i u/to [j w] in Isbukun either, because the constraint IDENTIO-[u] is violated by syllable-initial glides but satisfied by noninitial glides in Isbukun.

(16.) Whether bimoraic vowels such as those in the final syllables of [tutun] and [tupav] are consistently longer than their monomoraic counterparts is an issue requiring a phonetic study. If these syllables are not phonetically longer, stress placement still provides phonological evidence for the abstract bimoraic representation. If these bimoraic syllables are significantly longer, we cannot infer the bimoraic representation from the duration alone because the length effect could simply be the result of being the head of a foot. Therefore, although the final heavy syllables in words such as [tutfin] and [tupav] sometimes sound longer, they are not marked for length throughout the article.

(17.) While the nonagent voice marker in imperatives is /-av/ in Isbukun, it is /-i/ in Takituduh.

(18.) Huang (2002: 453) also notes that these onsetless syllables might not be as obvious in fast speech.

(19.) The corresponding forms in Isbukun are monosyllabic, such as/buan/[bwan] 'moon' and/pia/[pja] 'how many'.

(20.) The vowel/u/is pronounced more like [o] when it is adjacent to the uvular/q/.

(21.) If the minimal word phenomenon is not connected to foot structure, as argued in Garrett (1999), the disyllabicity of minimal words does not necessarily argue for the high ranking of [FTBIN.sub.[??]], but for the high ranking of some other constraints which lead to the minimality requirement, such as the BE-LONG constraint proposed in Garrett. Although many languages surveyed in Garrett (1999) exhibit a lack of connection between foot structure and minimality, the data from Takituduh Bunun are compatible with either a foot structural explanation or an approach based on length. In either account, the comparison between Takituduh and Isbukun shows the different roles of the syllabic constraint ONSET in the two dialects.

(22.) There seems to be no positive evidence suggesting how [FtBin.sub.[u]], is ranked with respect to other constraints in Takituduh. Although ranking [FtBin.sub.[u]] above or below [FTBIN.sub.[??]] will give rise to different metrifications for words that contain more than two vowels in the input (e.g., /kavia[??]/ 'friend' [ka(vja[??])] or [(kavja[??])]), either of the rankings is consistent with the final stress pattern in Takituduh.

(23.) The ranking IDENT-BA-[??] >> ONSET in Takitituduh is omitted from the presentation of the OT analysis. Notice that ONSET must rank above IDENT-BA-[??] in Isbukun because the onset requirement is strictly obeyed in the dialect.

(24.) The symbol 'T' represents an epenthetic consonant.

(25.) The ranking of DEP-IO-C above ONSET predicts that insertion of a consonant to avoid onsetless syllables is not possible in either word-initial or word-medial position. In the Takituduh data collected so far, I have not observed vowel-initial words.

(26.) The ranking of DEP-IO-V above ONSET is justified by the realization of /spat/ 'four' as [paat] rather than *[sVpat] ('V' represents an epenthetic vowel). If ONSET ranks above DEP-IO-V, [paat] would incorrectly be considered less optimal than *[sVpat]. The preference of [paat] over *[sVpat] also suggests that DEP-IO-V must rank above Max-IO-C.

(27.) The optimal candidate in (29) contains a doubly linked representation, which presumably violates a low-ranked constraint DOUBLYLINKED (Bernhardt and Stemberger 1998). The low ranking of DOUBLYLINKED raises the question of whether the correct phonological representation of words such as /buan/ should be [buwan], with the Root node of /u/ linked to both the nucleus of the first syllable and the onset of the following syllable, rather than [buan] as shown in Tableau (28). Technically, whether [buan] or the [buwan] with double linking would win depends on the ranking between ONSET and DOUBLYLINKED: DOUBLYLINKED >> ONSET gives rise to [buan] and the opposite ranking results in [buwan]. However, the output form [saak] does not provide positive evidence for the ranking between the two constraints; it only justifies the pairs of ranking [FTBIN.sub.[??]] >> ONSET and [FTBIN.sub.[??]] >> DOUBLYLINKED. Since onsetless syllables must be tolerated in minimal words such as [saak], the article assumes in (28) that the onsetless candidate [buan] wins.

HUI-CHUAN J. HUANG

National Tsing Hua University, Taiwan

(13) Tableau: consonant-ending stems

/ludah-av/ FTTYPE WSP ONS MAX- OCP- MAX-
 (trochaic) IO-u PLACE IO-V

* a. lu.(da.hav)
 b. lu. da.(hav)
 c. lu.(da.hav) *!

/ludah-av/ [FTBIN. UNIFORMITY- V-
 sub.[??]] IO NUC

* a. lu.(da.hav)
 b. lu. da.(hav)
 c. lu.(da.hav) *!

(14) Tableau: vowel hiatus resolved through gliding

/[sa.sub.u] FTTYPE WSP ONS MAX- OCP- MAX-
[du.sub.u]- (trochaic) IO-u PLACE IO-V
[a.sub.u]v/

* a. [sa.sub.u] *
 ([dw.sub.u]
 [a.sub.u]v)

 b. ([sa.sub.u] * *!
 [dw.sub.u]
 [a.sub.u]v)

 c. [sa.sub.u] * *!
 ([dwa.sub.u]v)

 d. ([sa.sub.u] * *!
 [dw.sub.u.]
 [a.sub.u]v)

 e. ([sa.sub.u] * *!
 [dwa.sub.u]v)

 f. [sa.sub.u] * *!
 ([du.sub.u] *
 [a.sub.u]v)
 g. [sa.sub.u] *

 h. ([sa.sub.u] * *!
 [do.sub.u]v)

 i. [sa.sub.u] * *!
 ([da.sub.uu]v)

/[sa.sub.u] [FTBIN. UNIFORMITY V-
[du.sub.u]- sub.[??]] IO NUC
[a.sub.u]v/

* a. [sa.sub.u] * * *
 ([dw.sub.u]
 [a.sub.u]v)

 b. ([sa.sub.u] * *
 [dw.sub.u]
 [a.sub.u]v)

 c. [sa.sub.u] * * *
 ([dwa.sub.u]v)

 d. ([sa.sub.u] * *
 [dw.sub.u.]
 [a.sub.u]v)

 e. ([sa.sub.u] * *
 [dwa.sub.u]v)

 f. [sa.sub.u] *
 ([du.sub.u] *
 [a.sub.u]v)
 g. [sa.sub.u] * * *!

 h. ([sa.sub.u] * *
 [do.sub.u]v)

 i. [sa.sub.u] * *
 ([da.sub.uu]v)

(15) Tableau: vowel hiatus resolved through coalescence

/[tu.sub.u][tu.sup.1.sub.u]- FTTYPE WSP ONS MAX- OCP-
[u.sup.2.sub.u]n/ (trochaic) IO-u PLACE

 a. [tu.sub.u]* *!
 ([tw.sup.1.sub.u]
 [u.sup.2.sub.u]n)
 b. ([tu.sub.u]* *! *
 [tw.sup.1.sub.u]
 [u.sup.2.sub.u]n)
 c. [tu.sub.u]* *! *
 ([tu.sup.1.sub.u]
 [w.sup.2]n)
 d. ([tu.sub.u]* *! *
 [tu.sup.1.sub.u]
 [w.sup.2.sub.u]n)
 e. ([tu.sub.u]* *! *
 [tw.sup.1]
 [u.sup.2.sub.u]n)
 f. [tu.sub.u]* *!
 ([tu.sup.1.sub.u]*
 [u.sup.2.sub.u]n)
* g. [tu.sub.u]*
 ([tu.sup.1,2.sub.uu]n)
 h. ([tu.sub.u]* *!
 [tu.sup.1,2.sub.u]n)
 i. [tu.sub.u]*
 ([tu.sup.2.sub.uu]n)

/[tu.sub.u][tu.sup.1.sub.u]- MAX- [FTBIN. UNIFOR- V-
[u.sup.2.sub.u]n/ IO-V sub.[??]] MITY NUC

 a. [tu.sub.u]* * *
 ([tw.sup.1.sub.u]
 [u.sup.2.sub.u]n)
 b. ([tu.sub.u]* *
 [tw.sup.1.sub.u]
 [u.sup.2.sub.u]n)
 c. [tu.sub.u]* * *
 ([tu.sup.1.sub.u]
 [w.sup.2]n)
 d. ([tu.sub.u] *
 [tu.sup.1.sub.u]
 [w.sup.2.sub.u]n)
 e. ([tu.sub.u]* *
 [tw.sup.1]
 [u.sup.2.sub.u]n)
 f. [tu.sub.u]*
 ([tu.sup.1.sub.u]*
 [u.sup.2.sub.u]n)
* g. [tu.sub.u]* * *
 ([tu.sup.1,2.sub.uu]n)
 h. ([tu.sub.u]* *
 [tu.sup.1,2.sub.u]n)
 i. [tu.sub.u]* *! *
 ([tu.sup.2.sub.uu]n)

(16) Tableau: low vowels in hiatus

/[tu.sub.u][pa.sup.1.sub.u]- FTTYPE WSP ONS MAX- OCP-
[a.sup.2.sub.u]v/ (trochaic) IO-u PLACE

 a. [tu.sub.u]* *!
 ([pa.sup.1.sub.u]*
 [a.sup.2.sub.u]v)
* b. [tu.sub.u]*
 ([pa.sup.1,2.sub.uu]v)
 c. [tu.sub.u]* *!
 ([pa.sup.1,2.sub.u]v)
 d. [tu.sub.u]*
 ([pa.sup.2.sub.uu]v)

/[tu.sub.u][pa.sup.1.sub.u]- MAX- [FTBIN. UNIFOR- V-
[a.sup.2.sub.u]v/ IO-V sub.[??]] MITY NUC

 a. [tu.sub.u]*
 ([pa.sup.1.sub.u]*
 [a.sup.2.sub.u]v)
* b. [tu.sub.u]* * *
 ([pa.sup.1,2.sub.uu]v)
 c. [tu.sub.u]* *
 ([pa.sup.1,2.sub.u]v)
 d. [tu.sub.u]* *! *
 ([pa.sup.2.sub.uu]v)

Tableau (25) shows that due to the ranking of ONSET above V-NUC,
gliding takes place to deal with vowel clusters that arise through
affixation.

(25)

/tan[??]a-un/ [FTBIN.sub.[??]] UNIFORMITY- MAX- DEP-
 IO IO-V IO-C

 a. tan[??]aun
* b. tan[??]awn
 c. tan[??]un *!
 d. tan[??]aTun (24) *!
 e. tan[??]on *!

/tan[??]a-un/ DEP- ONSET DEP- V-
 IO-V IO-u NUC

 a. tan[??]aun *!
* b. tan[??]awn
 c. tan[??]un *
 d. tan[??]aTun (24)
 e. tan[??]on

(26)

/tan[??][a.sup.1]-[a.sup.2]/ [FTBIN.sub.[??]] UNIFORMITY- MAX-
 IO IO-V

* a. tan[??][a.sup.1][a.sup.2]
 b. tan[??][a.sup.2] *!
 c. tan[??][a.sup.1][Ta.sup.2]
 d. tan[??][a.sup.1,2] *!

/tan[??][a.sup.1]-[a.sup.2]/ DEP- DEP- ONSET DEP- V-
 IO-C IO-V IO-u NUC

* a. tan[??][a.sup.1][a.sup.2] *!
 b. tan[??][a.sup.2]
 c. tan[??][a.sup.1][Ta.sup.2] *!
 d. tan[??][a.sup.1,2]

(27)

/[kitlalavi.sup.1]-[i.sup.2]/ [FTBIN.sub.[??]] UNIFORMITY-
 IO

* a. [kitlalavi.sup.1]-[i.sup.2]
 b. [kitlalavi.sup.1]-[j.sup.2]
 c. [kitlalavi.sup.2]
 d. [kitlalavi.sup.1][Ti.sup.2]
 e. [kitlalavi.sup.1,2] *!

/[kitlalavi.sup.1]-[i.sup.2]/ MAX- DEP- DEP- OCP- ONSET
 IO-V IO-C IO-V PLACE

* a. [kitlalavi.sup.1]-[i.sup.2] *
 b. [kitlalavi.sup.1]-[j.sup.2] *!
 c. [kitlalavi.sup.2] *!
 d. [kitlalavi.sup.1][Ti.sup.2] *!
 e. [kitlalavi.sup.1,2]

/[kitlalavi.sup.1]-[i.sup.2]/ DEP- V-
 IO-u NUC

* a. [kitlalavi.sup.1]-[i.sup.2]
 b. [kitlalavi.sup.1]-[j.sup.2] *
 c. [kitlalavi.sup.2]
 d. [kitlalavi.sup.1][Ti.sup.2]
 e. [kitlalavi.sup.1,2]

(28)

/buan/ [FTBIN.sub.[??]] UNIFORMITY- MAX- DEP- DEP-
 IO IO-V IO-C IO-V

* a. buan
 b. bwan *!
 c. ban *! *
 d. buTan *!
 e. bon *! *

/buan/ ONSET DEP- V-
 IO-u NUC

* a. buan *
 b. bwan *
 c. ban
 d. buTan
 e. bon

(29)

/sak/ [FTBIN.sub.[??]] UNIFORMITY- MAX- DEP- DEP-
 IO IO-V IO-C IO-V (26)
 u u
 RT
* a. saak

 u u *!
 RT RT
 b. saak

 c. sak *!

 d. saTak *! *

/sak/ ONSET DEP- V-
 IO-u NUC

 u u * *
 RT
* a. saak

 u u * *
 RT RT
 b. saak

 c. sak

 d. saTak *