catalan obstruents and opacity: a case for representational

23
Catalan obstruents and opacity: A case for representational ambisyllabicity Juli Cebrian University of Toronto Catalan obstruents are involved in voicing and continuancy alternations that are dependent on both segmental and syllabic environments. The interaction of syllabically conditioned processes (spirantization, final devoicing and regressive voicing assimilation) and resyllabification results in cases of surface opacity. This paper puts forward an analysis that distinguishes representationally between word-internal and resyllabified onsets by considering the latter as ambisyllabic, thus eliminating opacity and allowing for an adequate analysis within a monostratal framework such as Optimality Theory. 1. Voicing and continuancy alternations in Catalan obstruents Catalan exhibits several processes that determine voicing and continuancy values of obstruents. These include the stop/spirant alternation (traditionally referred to as spirantization), final obstruent devoicing and regressive voicing assimilation. 1.1 Stop/spirant alternation in Catalan (SSA) The voiced stops [b,d,g] alternate with the voiced spirants [”,¶,'] in Catalan. This alternation is similar to that found in other Iberian languages including Spanish, Galician, Portuguese, and Basque. Stops are found word-initially after a pause, or preceded by a non-continuant sound, such as a nasal. The spirants surface whenever the obstruent is preceded by a continuant sound. 1 In addition, Catalan spirant alternants are found only in syllable onset; that is, a voiced alternant in coda position always surfaces as a stop, even if it is preceded by a continuant sound (Recasens 1993). This syllabic restriction is well documented in the literature and distinguishes Catalan from other Iberian languages with no such restriction (see, for instance, Mascaró 1984; Palmada 1994, 1997). The distribution of stops and spirants is summarized and illustrated by the examples in Table 1. 1 I follow standard analyses in the uncontroversial assumption that vowels, approximants, fricatives and affricates pattern as continuants with respect to the stop/spirant alternation. It should be noted, though, that the three obstruents differ in their distribution when following the laterals and the labiodental fricative. More specifically, following the laterals /l/ and /æ·/, the bilabial and the velar surface as spirants while the coronal surfaces as a stop. On the other hand, the coronal and the velar are spirants after /f/, but in the case of the bilabial the stop is favoured (Mascaró , 1991; Recasens, 1993). The stops in both cases have been explained in terms of a restriction on spirantization in homorganic sequences.

Upload: others

Post on 12-Sep-2021

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Catalan obstruents and opacity: A case for representational

Catalan obstruents and opacity: A casefor representational ambisyllabicityJuli CebrianUniversity of Toronto

Catalan obstruents are involved in voicing and continuancyalternations that are dependent on both segmental and syllabicenvironments. The interaction of syllabically conditionedprocesses (spirantization, final devoicing and regressive voicingassimilation) and resyllabification results in cases of surfaceopacity. This paper puts forward an analysis that distinguishesrepresentationally between word-internal and resyllabifiedonsets by considering the latter as ambisyllabic, thuseliminating opacity and allowing for an adequate analysiswithin a monostratal framework such as Optimality Theory.

1. Voicing and continuancy alternations in Catalan obstruents

Catalan exhibits several processes that determine voicing and continuancy values ofobstruents. These include the stop/spirant alternation (traditionally referred to asspirantization), final obstruent devoicing and regressive voicing assimilation.

1.1 Stop/spirant alternation in Catalan (SSA)

The voiced stops [b,d,g] alternate with the voiced spirants [”,¶,'] in Catalan. Thisalternation is similar to that found in other Iberian languages including Spanish,Galician, Portuguese, and Basque. Stops are found word-initially after a pause, orpreceded by a non-continuant sound, such as a nasal. The spirants surface wheneverthe obstruent is preceded by a continuant sound.1 In addition, Catalan spirantalternants are found only in syllable onset; that is, a voiced alternant in codaposition always surfaces as a stop, even if it is preceded by a continuant sound(Recasens 1993). This syllabic restriction is well documented in the literature anddistinguishes Catalan from other Iberian languages with no such restriction (see, forinstance, Mascaró 1984; Palmada 1994, 1997). The distribution of stops andspirants is summarized and illustrated by the examples in Table 1.

1I follow standard analyses in the uncontroversial assumption that vowels, approximants,fricatives and affricates pattern as continuants with respect to the stop/spirant alternation. It should benoted, though, that the three obstruents differ in their distribution when following the laterals and thelabiodental fricative. More specifically, following the laterals /l/ and /æ·/, the bilabial and the velarsurface as spirants while the coronal surfaces as a stop. On the other hand, the coronal and the velarare spirants after /f/, but in the case of the bilabial the stop is favoured (Mascaró , 1991; Recasens,1993). The stops in both cases have been explained in terms of a restriction on spirantization inhomorganic sequences.

Page 2: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 59

Table 1.Distribution of stops and spirants in Catalan

Obstruent Environment Example GlossStop - initially (after pause): [b]ar.ca boat

- after non-continuant: em.[b]ar.car to board (a ship)- in the coda: a[b].di.car abdicate

Spirant - in the onset after a continuant: la. [”]ar.ca the boatcar. [”]ó carbon, coalsu. [”]lim sublime

The facts concerning continuancy alternation in Iberian languages have beenexplained by assuming the stops to be underlying and deriving the spirants by meansof a spirantization rule (Wheeler 1979; Palmada 1997; Harris 1969). The oppositeview, that spirants are underlying and stops are the result of a process of fortition, isalso represented in the literature (e.g., Bakovic 1994 for Spanish). Alternatively, anumber of analyses have proposed that the feature [continuant] is unspecified forvoiced non-strident obstruents, arguing that there is no compelling evidence foreither alternant being underlying (e.g., Lozano 1979; Hualde 1991; and Mascaró1987, 1984). In this paper I will follow this third alternative, given that theoptimality theoretic framework I adopt allows the correct surface realizations tofollow from the crucial ranking of the constraints governing their distribution, sothat neither the stop nor the spirant need to be postulated as underlying. FollowingMascaró, I use capital letters to represent the archisegments (/B, D, G/).

1.2 Final Obstruent Devoicing

Final Obstruent Devoicing (FOD) is responsible for the neutralization of voiced andvoiceless obstruents in coda position before a pause and before a voicelessconsonant. Before a vowel-initial word, word-final stop/spirant alternants alsoundergo devoicing, whereas final continuant obstruents are subject to regressivevoicing assimilation, as discussed in the next section. Examples of FOD are givenin Table 2, which also provides examples of word-medial intervocalic contrast.

Table 2.Final Obstruent Devoicing (FOD))

Finally Word internallyObstruent Phoneme FOD Gloss Contrast Gloss

Continuants/z/ va[s] glass va[z]os glasses/s/ go[s] dog go[s]os dogs

Stops/B/ re[p] s/he receives re[”]re to receive/p/ escu[p] s/he spits esco[p]ir to spit

In the example involving /B/ in Table 2, the intervocalic voiced alternantsurfaces as a spirant due to SSA, as discussed above.

Page 3: Catalan obstruents and opacity: A case for representational

60 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

1.3 Regressive Voicing Assimilation (RVA)

Another neutralization process affecting coda obstruents is Regressive VoicingAssimilation (Wheeler 1979; Mascaró 1983; Palmada 1994; Recasens 1993, amongothers). Regressive voicing assimilation (RVA) is responsible for the obligatoryvoicing agreement in heterosyllabic clusters of an obstruent followed by anotherconsonant. A coda obstruent is voiced before a voiced consonant, and it is voicelessbefore a voiceless consonant. Examples of RVA are given in Table 3.2

Table 3.Final obstruents in consonant clusters

Obstruent RVA Gloss FOD GlossContinuants

/z/ va[z] # gran large glass va[s] # petit small glass/s/ go[z] # negre big dog go[s] # petit small dog

Stops/B/ re[b] # res receive nothing re[p] # cartes receive letters/p/ escu[b] # molt spit a lot escu[p] # tot spit everything

In addition to RVA in consonant clusters, another regressive voicing processaffects word-final fricatives in the environment of a word-initial vowel, as illustratedin Table 4. This prevocalic voicing does not affect underlying voiceless stops (/p, t,k/) and voiced obstruents unspecified for continuancy (/B, D, G/), however, whichremain voiceless or surface as voiceless stops as a result of FOD, respectively (Table5). Thus, preceding a word-initial vowel, word-final /p, t, k, B, D, G/ are neutralizedto voiceless while word-final fricatives are neutralized to voiced.3

Table 4.Prevocalic word-final fricatives: RVA

Continuant obstruents Example Gloss/z/ va[z] # ample large glass/s/ go[z] # astut cunning dog

Table 5.Prevocalic word-final stops: FOD

Stops/spirants Example Gloss/B/ re[p] # això receive this/p/ escu[p] # això spit this

2Notice that, contrary to what happens in other languages with RVA such as Dutch, voicing inCatalan is triggered not only by voiced obstruents but also by sonorants ('go[z] negre', 're[b] res').Cebrian (1997), following Rice and Avery (1989), accounts for this pattern by assuming that bothsonorants and voiced obstruents have the feature Spontaneous Voicing (SV) in their representation andthat regressive assimilation involves this feature SV. See also Avery (1996), Rice (1993) and Piggott(1992) for further discussion on SV theory and voicing assimilation patterns.

3In principle, prevocalic regressive voicing also affects affricates. However, affricates aresubject to some variation. There is a tendency in the speech of some Eastern Catalan speakers toproduce voiceless affricates (e.g., [ ]) instead of voiced ones ([ ]) not only in the context for RVA butalso word-internally (Recasens, 1993). For this reason, and for the sake of simplicity, I will refer onlyto fricatives as the targets of prevocalic voicing.

Page 4: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 61

1.4 Surface contrast in VCV sequences

In general terms, syllable structure in Catalan is built according to universalprinciples that favour unmarked syllable shapes such as the Onset Principle (Itô1989). Although onsetless syllables are possible, a vowel is always provided withan onset if there is one available. Thus, at the phrasal level, syllabification acrossword boundary takes place whenever a word-initial vowel is preceded by a word-finalconsonant. An example is given in (1a). As illustrated in (1b), however, word-finalconsonants are not syllabified with a following word-initial consonant, even thoughthe resulting cluster may be a possible branching onset. In other words,syllabification across word boundary is limited to onset satisfaction (conversely,word-internal syllabification is characterized by onset maximization).

1. a. cap + amic --> ca. pa. micno friend

b. cap + limit --> cap. li. mitno limit

The segmental processes discussed in sections 1.1 through 1.3 demonstrate theclose relationship between syllabification and the distribution of Catalan obstruents;spirants are restricted to onset position, and final devoicing and regressive voicingaffect obstruents in the coda. In fact, it is the syllabic position rather than thesegmental environment that ultimately determines the nature of the stop/spirantalternant. This becomes evident when we compare two cases involving the sameobstruent in the same segmental environment but occupying a different syllabicposition, as illustrated in (2).

2. a. su. [”]limsublime

b. su[b]. lu. narsublunar

Both examples in (2) show a bilabial alternant (/B/) following a vowel andpreceding a lateral. The alternant in (2a), being word-internal, is found in an onsetand surfaces as a spirant. In the case of (2b), the prefix-final alternant is found in thecoda and consequently surfaces as a stop which is voiced because it precedes aheterosyllabic voiced consonant.

There is a case, however, of two different surface forms for the same underlyingobstruent in the same syllabic position. This case involves intervocalic obstruentsfrom different morphological sources. This is illustrated in (3), which shows twodifferent surface representations of underlying /B/ in onset position following acontinuant. Whereas the word-internal alternant surfaces as a spirant (3a), thealternant syllabified across morphological boundary surfaces as a voiceless stop (3b).In other words, it appears in the onset but shows the result of a process affectingcodas, i.e. final devoicing. The environment of devoicing is opaque as a result ofsyllabification.

3. a. re. [”]u su. [”]limI receive sublime

Page 5: Catalan obstruents and opacity: A case for representational

62 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

b. re. [p]ai. xò (re/B/ # això) su. [p]e. di. tor (sub+editor)receive this assistant editor

The contrast is between an onset in a morphologically non-derived environment (3a)and one in a derived environment (3b), a contrast that has been accounted for in rule-based approaches by means of cyclic derivation and multiple phonological levels(Mascaró 1983; Harris 1993, among others). In his analysis of constituent-building,Harris assigns the processes of FOD, spirantization and RVA to the final stratum inthe derivation, that is, following word-internal syllabification but preceding phrasalresyllabification. The fact that phrasal resyllabification applies only after thesyllable-sensitive processes have applied obtains the correct derivation.4 Thus, FODapplies to the coda obstruent in (3b), yielding /rep. a. § /, and, subsequently, phrasalresyllabification provides the onsetless vowel with an onset (/re. pa. § /). Theopacity of FOD in the case of the resyllabified obstruent is explained by thestratified derivation that permits the obstruent to be in different syllable positions atdifferent stages in the derivation. It is precisely the unavailability of suchstratification that poses a challenge to an OT approach. However, I will show thatstratification is not necessary if we distinguish the surface forms in (3a) and (3b) bytheir prosodic structure.

This interaction between syllable-sensitive processes and syllabification is notunique to Catalan. Other examples are found in the processes of /s/-aspiration andnasal velarization found in some Spanish dialects (also discussed in Harris 1993).The former is responsible for the change of /s/ into [h] in final position, and thelatter involves the velarization of final nasals. Both processes affect coda segmentsthat may eventually be resyllabified into onset position (e.g., las osas [la .ho. sah],'the bears fem.', sin eso [si. e. so], 'without that'). As with the Catalan processes,the surface opacity of the coda processes can be eliminated if we adopt anambisyllabic structure for edge consonants. This is discussed in the next section.

2. Ambisyllabicity

The proposal put forth in this paper is that word-internal onsets and 'resyllabified'onsets differ in their prosodic structure in that while word-internal onsets are onlyonsets, 'resyllabified' prevocalic final consonants are in fact ambisyllabic. Thesurface contrast between derived and non-derived onsets can thus be explainedstructurally by distinguishing between singly linked onset obstruents andambisyllabic obstruents. A final prevocalic consonant, being ambisyllabic, occupiessimultaneously a coda and an onset position, which renders the environment forsyllabically conditioned processes visible. Hence, I argue that by assumingambisyllabicity, the distinction between syllabification and resyllabification is nolonger relevant, and the surface opacity that results from that distinction disappears.

4Under this analysis, the stop/spirant alternation is analyzed as a rule of spirantization affectingunderlying voiced stops regardless of syllable position. Thus, in addition to applying segmentalprocesses before resyllabification, there is a crucial ordering and bleeding relation between FOD andspirantization. FOD also precedes RVA.

Page 6: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 63

2.1 Segmental properties of ambisyllabic consonants

The way the segmental properties of an obstruent derive from the prosodic positionit occupies is best illustrated with examples involving stop/spirant alternants, giventhe syllabic conditioning of the continuancy and voicing alternations. The proposedambisyllabic structure is shown in (4) with examples of VCV sequences involving abilabial alternant (/B/). In (4a), the alternant is found in an onset following acontinuant and surfaces as a spirant. In (4b), the alternant is ambisyllabic and itsurfaces as a voiceless stop.4. a. Onset /B/ b. Ambisyllabic /B/

σ σ σ σ σ

r u ∫ i r ´ p \ ß øruby receive this

Notice that there is only one possible phonetic realization in each case. On theone hand, only a spirant is possible when the alternant is found in the segmentalenvironment and syllabic position in (4a). On the other hand, given that theambisyllabic consonant occupies simultaneously an onset and a coda position, onlya voiceless stop is possible in (4b). I am assuming, then, that an ambisyllabicconsonant in a certain segmental environment has to be both a possible coda and apossible onset it that environment. This point is further illustrated by comparing thesegmental alternatives in (5).

5. a. Ambisyllabic [”] b. Ambisyllabic [b] c. Ambisyllabic [p]

σ σ σ

r ´ p \ ß ø

σ σ σ

r ´ ∫ \ ß ø

σ σ σ

r ´ b \ ß ø

* *

Of the three ambisyllabic structures in (5), (5c), which repeats (4b), is the onlypossible surface realization. The voiceless stop in (5c) is a well-formed onsetbecause there are no restrictions on voiceless stops in the onset, and it is a well-formed coda since coda stops are only voiced when preceding a voiced consonant.However, the spirant in (5a) is not possible in an ambisyllabic representation.Although a spirant alternant is expected in onset position following a continuant, itdoes not constitute a permissible coda because stop/spirant alternants are alwaysstops in the coda. Finally, the ambisyllabic voiced stop in (5b) constitutes neither apossible onset, being in the environment for a spirant, nor a possible coda, since astop would be devoiced in this context. Therefore, the segmental properties of thecorrect surface realization, the voiceless stop (5c), follow from its ambisyllabicnature. Hence, the introduction of the distinction between onset consonants andambisyllabic consonants correctly accounts for the surface contrast found in VCVsequences (illustrated in (4) above), eliminating opacity and the need for rule orderingand stratification.

Page 7: Catalan obstruents and opacity: A case for representational

64 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

2.2 Ambisyllabicity vs. resyllabification

Although it could be argued that the ambisyllabicity analysis results in a morecomplex and abstract surface structure, I argue that it is preferable to a solutioninvolving intermediate representations and resyllabification. The need forresyllabification has been questioned in earlier works on syllabification. Rice (1990)argues against resyllabification on the grounds that whereas syllabification rulesgenerally operate in a structure-building way (Steriade, 1988), resyllabificationinvolves destroying syllable structure that has already been built. Thus, instead of amodel in which a prevocalic final consonant is initially syllabified in the rime, andlater delinked from rime position of the first syllable and resyllabified into onsetposition of the second syllable, Rice argues for a model which allows suchsequences to be initially syllabified as V.CV, never going through a stage of VC.V.This is achieved in her model by means of final extraprosodicity. Final consonantsare extraprosodic and may become visible at levels later than the word, at whichpoint they are syllabified in the coda if no vowel follows, or they may go in theonset with a following vowel-initial syllable. Thus, instead of requiring twoprocesses for filling onsets, syllabification and resyllabification, her analysis treatsbetween-word syllabification and within-word syllabification in the same way.Extraprosodicity alone, however, cannot account for the Catalan facts because finalobstruents need to be in the coda, at least at some stage, to be subject to devoicing.5

The proposal of ambisyllabicity presented in this paper explains the behaviour ofCatalan final obstruents and captures Rice's arguments against resyllabification sinceambisyllabicity does not involve destroying structure and it integrates word-level andphrase-level syllabification.

Interestingly enough, further support for ambisyllabicity can be found inSelkirk's (1982) work arguing against such a structure. Selkirk rejects Kahn's (1976)ambisyllabicity account of English aspiration, glottalization, tapping and release byarguing that ambisyllabic segments never need be syllable-initial and syllable-finalat the same time. She argues that all phonetic properties characteristic ofambisyllabic segments are derived from their coda status. Her analysis involvesresyllabification of an onset to the preceding syllable, forming a coda. Selkirk'sresyllabified structure, however, runs counter to the universally observed preferencefor onsets, expressed above as the Onset Principle. In fact, Selkirk's claim thatambisyllabic segments tend to have the properties characteristic of coda segmentslends support to an ambisyllabicity approach and makes it preferable to aresyllabification analysis that delinks final consonants from their original syllablesand links them to a following vowel-initial syllable. The ambisyllabic structure, butnot the resyllabified one, retains the structural coda position and thus explains thecoda properties of the ambisyllabic segment. On the other hand, an ambisyllabicstructure, but not Selkirk's resyllabified coda structure, is able to satisfy the Onset

5In her discussion of extraprosodicity, Rice (1990) examines languages with word-finaldevoicing in which the word-final consonant fails to devoice when followed by a vowel-initial word.Extraprosodicity renders the final consonant invisible to devoicing. When extraprosodicity is turned off,the prevocalic final consonant must get syllabified with the following vowel and thus remains voiced.Notice, however, that in Catalan we find the opposite case, that is, extraprosodicity would have to beturned off earlier so that the final consonant was visible to final devoicing while still in the coda andprior to being syllabified with the following onset. Therefore, a non-resyllabification account ofCatalan requires ambisyllabicity.

Page 8: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 65

Principle. Therefore, only ambisyllabicity can explain why such segments haveproperties characteristic of coda segments and at the same time satisfy the universaltendency to fill an empty onset.

2.3 Morphological properties of ambisyllabic consonants

An ambisyllabic obstruent in Catalan can be found word-finally, morpheme-finallyin the case of some prefixes (e.g., su[p]editor, from sub and editor, 'assistant editor'),and in final position of words in compounds (e.g., su[t]est, from sud and est, 'southeast'). I assume ambisyllabicity in Catalan only affects coda consonants at amorphological edge and that a morpheme-internal consonant cannot be ambisyllabicin Catalan.6 On the other hand, not all edge consonants are ambisyllabic, only thoseoriginally occupying a coda position and required by onset satisfaction to syllabifywith a following vowel.

Ambisyllabicity in Catalan is therefore justified by syllabification principlesand the coda properties of the ambisyllabic segment. In addition, an ambisyllabicityaccount is able to encode the close relationship between phonological structure andmorphological structure. Morphologically, it distinguishes a segment at amorphological edge from a morpheme-internal segment. Phonologically,ambisyllabic consonants occupy both an onset and a coda, and, consequently, theyare subject to the constraints affecting those two syllabic constituents. The role ofambisyllabicity in capturing the relationship between prosody and morphology hasalso been defended in other analyses of word-final phenomena such as Hong (1997).Hong, presenting evidence from Korean phonological phenomena and Englishflapping, argues against unique-onset syllabification of prosodic word-finalconsonants preceding a vowel-initial word, and proposes ambisyllabicity as the onlytenable solution.

Another analysis that assumes ambisyllabicity affecting a morphological edgeis Merchant (1996). Merchant accounts for the distribution of the palatal fricativeand the velar fricative in German by allowing for ambisyllabic consonants stem-finally but not root-finally. A stem-final consonant followed by a vowel-initialstem-attaching suffix is ambisyllabic, while a root-final consonant followed byvowel-initial root-attaching suffix patterns as word internal. The condition on thedistribution of the dorsal fricative is that it follow a tautosyllabic back vowel. Thiscondition is matched when the fricative is stem-final and ambisyllabic but not whenit is root-final or word-internal. Notice that the distinction between root-final andstem-final could also apply to Catalan given that ambisyllabicity is postulated forsequences involving word-edges and sequences of derivational prefixes and stems, butsequences involving root-final consonants followed by vowel-initial root-attachingsuffixes pattern as word-internal. The issue of whether morphologically markedambisyllabic consonants are universally restricted to domains larger than the rootrequires further research, but the examples from Korean, Catalan and German seem

6Ambisyllabicity in Catalan is not motivated by phenomena along the lines of syllable shaperequirements for stress assignment or preceding vowel quality. These are typically put forward asarguments for ambisyllabic structures in word or morpheme-internal position, as has been done forlanguages such as Dutch (van der Hulst 1984) or English (Kreidler 1995). It could be speculated thatthat kind of evidence is needed to support the ambisyllabicity of word or morpheme-internalconsonants.

Page 9: Catalan obstruents and opacity: A case for representational

66 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

to point to stem and word boundary VCV sequences as possible contexts forambisyllabicity.

2.4 Ambisyllabicity vs. gemination

It should be noticed that the ambisyllabic structure postulated in this paper is not tobe interpreted as a phonetic geminate consonant. An ambisyllabic consonant inCatalan is a singleton that belongs to two structural positions, the coda of onesyllable and the onset of the following syllable, but it occupies only one timing orskeletal position. I assume that geminates differ from ambisyllabic consonants inthat geminates involve two C slots or are moraic. In this respect I disagree withother approaches such as Borowsky et al. (1984) or van der Hulst (1984) thatpostulate the same prosodic representation for both structures. Borowsky et al. arguethat the formal representation of ambisyllabicity is identical to that of geminationand that phonetic rules interpret the representation with reference to the phonologyof the particular language. This view predicts that languages will not have bothambisyllabic consonants and phonetic geminates. Such a claim is not withoutcontroversy. The need for different representations for ambisyllabicity andgemination is argued for in Kang's (1991) discussion of Korean. In addition, anidentical representation is problematic for Catalan. Catalan has phonetic geminates,and, as argued in this paper, ambisyllabic singleton consonants.

The existence of both types is not problematic if a geminate consonant isrepresented as a sequence of two identical consonants, each occupying its owntiming position and belonging to a different syllable. The different structures aregiven in (6) with examples from Catalan. The example in (6a) shows a morpheme-internal geminate, whereas (6c) shows a geminate arising in a word sequence. Inboth cases the consonant is phonetically long, involving two identical articulations,each pertaining to a different syllable (Recasens, 1993). The examples in (6b) and(6d) show an ambisyllabic final consonant obeying onset satisfaction, but beingphonetically rendered as a single consonant. The phonetic contrast underlining thedifference in structure is evident in the pair formed by (6c) and (6d).

6. a. batlle b. bell home

σ σ σ σ σ

b a Ò Ò \ b e Ò ø m \mayor old man

Page 10: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 67

c. cap # port d. cap # hort

σ σ σ

k a p ø r tk a p p ø r t

σ

harbour no orchard

As demonstrated in this section, the ambisyllabicity account renders surfacerepresentations non-opaque, eliminates language particular stipulations concerningcyclicity and rule ordering, and captures the relationship between prosodic andmorphological structure. The Catalan voicing and continuancy alternations formerlyhandled by rule ordering now follow directly from representations and independentprinciples. This point becomes even more evident under an OT analysis, as will beshown below. I will show, too, how the voicing alternations found with underlyingfricatives follow from the same structure and principles as the voicing andcontinuancy alternations found with stops.

3. An OT analysis of voicing and continuancy alternations

I will present next an analysis of the Catalan voicing and continuancy alternationsunder an OT framework. For the sake of brevity, I will only briefly discuss themotivation of each of the different constraints governing the segmental processes(see Cebrian 1998 for more discussion). The aim of this section is to illustrate howthe assumption of ambisyllabic representations resolves the problem of apparentsurface opacity in a theory that evaluates output structures without resort tointermediate levels.

3.1 Ambisyllabicity and OT

In an optimality theoretic framework, syllabification of an output string isdetermined by the relative ranking of constraints governing prosodic structure suchas ONSET and NOCODA, and alignment constraints (McCarthy & Prince 1993).These constraints and their ranking for Catalan are given in (7).

7. ONSET: Syllables begin with an onset.NOCODA: Syllables end in a vowel.ALIGN(STEM,σ): Every final stem-edge matches to a final syllable edge.CRISP(σ): All segments are uniquely syllabified.

Ranking: ONSET >> ALIGN(STEM,σ) >> NOCODA, Crisp(σ)

The effect of this ranking on a series of outputs involving word-finalobstruents is illustrated in the tableau in (8) (syllable boundaries are represented bydots; a square bracket (']') marks a stem edge, a non-crisp edge is represented bymarking the ambisyllabic consonant as belonging to two syllables (\/)). Following

Page 11: Catalan obstruents and opacity: A case for representational

68 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

Merchant (1996), I assume that alignment constraints are not violated by multiplelinking. A specific constraint requiring prosodic units to have crisp edges, alsodefined in (7), is ranked low in Catalan. Therefore, the candidate with anambisyllabic final obstruent (8c) satisfies both ONSET and ALIGN and thus winsover the other candidates. Given the specific ranking of constraints for Catalan, theambisyllabic structure constitutes the best output for word-final prevocalicconsonants.7

8. Ambisyllabic vs. singly-linked intervocalic consonants

/kap # m /no man

ONSET ALIGN NOCODA CRISP(σ)

a. ka. p] . m ] *!b. kap]. . m ] *! *☞ \/c. ka p] . m ]

* *

\/ \/d. ka p] m ]

**! * *

The relevant constraints governing voicing and continuancy alternations arediscussed in the following sections. Having established the ambisyllabicity ofprevocalic final obstruents here, non-ambisyllabic candidates for final prevocalicobstruents and constraints on prosodic structure will be omitted from the tableaux.

3.2 Constraints governing SSA

As discussed in section 1.1, the stop/spirant alternants surface as stops after a pauseor in the coda. In the onset, they surface as stops after non-continuants and asspirants after continuants. I propose to account for the stop/spirant alternation bymeans of the two constraints listed in (9). ONSET[CONT] is a specific constraint onsequences of onset voiced stops preceded by a continuant and it disallows voicedalternants in onset position if they do not agree in continuancy with the precedingsegment. *[CONT] is a markedness constraint on continuancy. (See Cebrian 1998 forfurther discussion on the nature of the constraints and their ranking).

9. ONSET[CONT]: No onset voiced stop preceded by a continuant.*[CONT]: A segment must not be continuant.

As illustrated in the tableaux in (10) below, the distribution of voiced stops andspirants is the result of ranking ONS[CONT] higher than *[CONT]. Although notincluded in the tableaux, I assume that the more marked status of a continuant withrespect to a stop follows from the ranking of *[CONT] over some constraint on non-continuant sounds (*[STOP]) (for the sake of accuracy, the outputs in the tableaux

7Notice that the constraints and the ranking proposed predict that ambisyllabicity will only befound with final consonants in Catalan, and not with word-internal middle consonants. As shown in(8d), a candidate with an ambisyllabic middle consonant (e.g., the [m] in the second word) incurs aviolation of NOCODA and thus loses to a candidate with a unique-onset middle consonant (8c).

Page 12: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 69

show vowel reduction and r-deletion, although these processes are not discussedfurther; as above, dots mark syllable boundaries). The candidate with the spirant isthe winner in (10a) given that the candidate with the voiced stop violates the highlyranked ONS[CONT]. In (10b) and (10c), the candidates with voiced stops do notviolate ONS[CONT] because they do not follow a continuant, like the two /B/'s in(10b) and the /D/ in (10c), or because it is in the coda, like the /B/ in (10c).

10. Voiced stop/spirant alternations in Catalan: ONS[CONT] >> *[CONT]

a. Alternant in onset position following a continuant

/aBi/grandfather

ONS[CONT] *[CONT]

☞ a.”i *a.bi *!

b. Alternants in onset position not following a continuant

/BomBa/bomb

ONS[CONT] *[CONT]

☞ bom.bbom.” *!”om.” *!*

c. Alternant in coda position

/aBDikar/abdicate

ONS[CONT] *[CONT]

☞ b.di.ka”.di.ka *! *”.¶i.ka *!*

The constraint *[CONT] is only meant to disallow voiced spirants that areinvolved in the stop/spirant alternation. Other voiced continuants are indeed foundword-initially, after nasals and in the coda. The alternating spirants differ from othercontinuants precisely in the fact that other continuants are underlyingly specified ascontinuant. Thus, a faithfulness constraint involving continuancy is ranked higherthan *[CONT], ensuring that underlying continuants are not penalized.

3.3 Constraints governing voicing and devoicing

Regressive voicing and final devoicing are responsible for the voicing neutralizationof coda obstruents. Coda obstruents are voiceless unless followed by a voicedconsonant. In addition, word-final fricatives are also voiced when followed by avowel-initial word. As seen in section 1.4, derivational accounts rely on the orderingof FOD before RVA and their interaction with cyclic syllabification.

Page 13: Catalan obstruents and opacity: A case for representational

70 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

A fair amount of attention has been devoted to voicing and devoicingphenomena in recent research in OT (e.g., Lombardi 1995, 1996; Beckman 1997;Butska 1997). Beckman (1997) assumes Lombardi's constraints in a more generaldiscussion of positional faithfulness across languages, including a discussion ofCatalan. Beckman argues that the voicing patterns found in different languagesillustrate the privileged position of the onset, given that when a language showsvoicing agreement in clusters, it is the voicing of the onset that determines thevoicing of the cluster. This privileged status of the onset position is implementedby relativizing identity to position (IDENTONSET(VOICE)). A high ranking of thisconstraint in languages like Catalan ensures the regressive nature of voicingassimilation.

In this paper I follow the general lines of Beckman's positional faithfulnessaccount. However, following Butska (1997), I replace Beckman's identity constraintswith featural correspondence constraints. Butska's analysis of voicing processes inUkrainian dialects demonstrates that Identity constraints cannot account for theassimilation patterns of Ukrainian. She adopts a featural correspondence approach tofaithfulness as advocated in a variety of works, including Causley (1996),Lamontagne and Rice (1995), Cull (1998) and others. The proposed constraints andconstraint ranking are presented in (11) and (12).

11. FAITH-ONS(VOICE): Output voiced segments in the onset must have acorresponding voiced segment in the input.

MAX(VOICE): Every feature [voice] of the input has a correspondent inthe output.

*VDOBSTR]σ: An obstruent in the coda must not be voiced.AGREE(VOICE): Clusters of heterosyllabic consonant sequences agree in

voicing.8

12. Constraint ranking: AGREE(V), FAITH-O(V) >> *VDOBSTR]σ >> MAX(V)

The way this ranking accounts for the Catalan voicing and devoicing patterns isillustrated in the tableaux in (13), (14) and (15). (13) shows that the high ranking ofthe positional faithfulness constraint guarantees voicing contrast in word-internal(i.e., singly linked) intervocalic obstruents. Final devoicing is the direct result of theeffect of the constraint on coda voiced obstruents, as shown in (14).

13. No neutralization word-internally

/Griza/grey (fem)

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

☞ gri. z ]gri. s ] *! *

8This constraint is formulated with reference to consonant sequences in order to capture RVAtriggered by both obstruents and sonorants, as is the case in Catalan, and explained in terms of thefeature SV (see footnote 2), as well as to capture RVA triggered by obstruents only and involving thefeature [voice] (see Rice and Avery 1989, Cebrian 1997 for further discussion).

Page 14: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 71

14. Final devoicing

/Griz/grey (masc)

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

griz *!☞ gris *

Finally, regressive assimilation is implemented by the combined effect of thehighly ranked positional faithfulness and agreement constraints.

15. Regressive voicing assimilation

a. Underlying Voiceless + Voiced cluster

/Gos#Blank/white dog

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

gos]. ”la ] *!☞ goz]. ”la ] *

gos]. pla ] *! *

b. Underlying Voiced + Voiceless cluster

/Griz#klar/light grey

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

griz]. kla] *! *☞ gris]. kla] *

griz]. 'la] *! *

The examples in (13), (14) and (15), as well as those in (10) above, involveobstruents which occupy either an onset or a coda position. Recall, however, thatfinal obstruents preceding an onsetless syllable and are syllabified with the followingvowel are analyzed as ambisyllabic, that is, as occupying both an onset and a codaposition. Recall also that continuant obstruents and non-continuant obstruents showa different pattern in this situation. As discussed in section 1.3, prevocalic word-finalfricatives are voiced, whereas underlying voiceless stops and stop/spirant alternantssurface as voiceless stops. This neutralization contrasts with the voicing distinctionmaintained word-internally. These facts have not been accounted for in previous OTanalysis of Catalan such as Beckman (1997) and Colina (1995). Under Beckman'sanalysis 'resyllabified' final obstruents are wrongly predicted to pattern just likeword-internal obstruents and maintain their underlying voicing specifications due tothe high ranking of FAITH-O(V). I show next how the segmental difference betweenword-internal and word-boundary intervocalic obstruents falls out from their differentprosodic structure and the proposed ranking of the relevant constraints.

Page 15: Catalan obstruents and opacity: A case for representational

72 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

3.4 Word-internal and ambisyllabic stop/spirant alternants

The segmental properties of ambisyllabic stop/spirant alternants is determinedby the set of constraints governing SSA and voicing that make reference to eitherthe coda or the onset (i.e., ONS[CONT], FAITH-O(V) and *VDOBSTR]σ). Theproposed global ranking of these constraints is given in (16).

16. ONS[CONT], AGREE >> FAITH-O(V), *[CONT] >> *VDOBSTR]σ >> MAX(V)

The tableaux in (17) and (18) illustrate how this ranking obtains the rightresults for both ambisyllabic and word-internal stop/spirant alternants, respectively(AGREE is omitted because it is not relevant here). The high ranking of ONS[CONT]rules out candidates with an intervocalic voiced stop (17a & 18a). The remainingtwo alternatives, (b) and (c), violate *[CONT] and FAITH-O(V), respectively. Anequal ranking of FAITH-O(V) and *[CONT] makes the choice between a voicelessstop and a spirant be determined by the lower ranked *VDOBSTR]σ. This constraintaffects ambisyllabic obstruents (17b) but not single onsets (18b), yielding the rightresults in all cases.

17. Ambisyllabic stop/spirant alternant

/r´B#aßø/receive this

ONS[CONT] FAITH-O(V) *[CONT] *VDOBS]σ MAX(V)

a. \/ r´ b]\. ßø]

*! *

b. \/ r´ ”]\. ßø]

* *!

c. \/☞r´p]\. ßø]

* *

18. Word-internal stop/spirant alternant

/aBi/grandfather

ONS[CONT] FAITH-O(V) *[CONT] *VDOBS]σ MAX(V)

a. a.bi] *!b. ☞ a.”i] *c. a.pi] * *!

The winning candidate in both cases results from the higher ranking of themarkedness constraint on coda voiced obstruents over the faithfulness constraint.Notice that in (18b), the spirant is not ruled out by the markedness constraint oncoda voiced obstruents precisely because the constraint affects coda voiced obstruentsand the spirant is in the onset. The candidate with a voiceless stop violates thegeneral faithfulness constraint (MAX(V)) and thus loses to the winning candidate(i.e., [a.”i]). The ambisyllabic spirant in (17b), however, violates *VDOBSTR]σ andloses to the candidate with a voiceless stop. This different pattern illustrates that therelevant markedness constraint on voiced obstruents has to refer specifically to the

Page 16: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 73

coda. Therefore, the correct outputs in word-internal and word-final stop/spirantalternants follow from their different prosodic structure and the syllable-basedformulation of the constraints.

3.5 Prevocalic word-final fricatives.

The last case to be accounted for is that of prevocalic final fricatives. Recall thatfinal fricatives pattern differently final stops (including /B, D, G/) in prevocalicenvironment. While stop/spirant alternants are affected by coda devoicing, fricativesare voiced as a result of prevocalic RVA. As with the stop/spirant alternants,ambisyllabicity allows us to differentiate between word-internal (non derived) onsetfricatives and resyllabified final fricatives. However, the effect of *VDOBSTR]σforces the same neutralization pattern on all obstruents, wrongly predictingambisyllabic fricatives to surface as voiceless instead of voiced.

Regressive voicing assimilation of prevocalic continuant obstruents can becaptured by a constraint along the lines of the agreement constraint that governsRVA in consonant clusters. Recall from (11) above that AGREE makes reference tothe heterosyllabic nature of the cluster involved. Given that prevocalic RVA affectsambisyllabic fricatives, and that ambisyllabic obstruents are both tautosyllabic andheterosyllabic with the following vowel, I propose that prevocalic regressive voicingof final fricatives is the result of an agreement constraint that requires heterosyllabicclusters of continuant obstruents followed by vowels to agree in voicing(AGREE(FRIC-VOWEL)).9 The effect of this constraint is illustrated in (19). In theirquality of being heterosyllabic with the following vowel, the ambisyllabicconsonants in (19) are subject to the agreement constraint. Thus, (19a) violatesAGREEFV, but (19b) does not. Notice that the heterosyllabic nature of the clustersaffected by AGREEFV stresses once more the relevance of prosodic structure. Word-internal intervocalic fricatives, which are simple onsets and thus only tautosyllabicwith the following vowel, do contrast in voicing.

9The different pattern found between stops and fricatives seems to indicate that, with respect tovoicing assimilation, fricatives are easier targets than stops, being voiced by both vowels and voicedconsonants, and consonants are better triggers than vowels, since voiced consonants trigger voicingassimilation in all cases but vowels only do when the target is a continuant obstruent. This pattern maybe related to the different articulatory properties of stops and continuants, and to the closer affinitybetween vowels and fricatives. Thus, fricatives are closer to vowels in their manner of articulation,both being produced with unimpeded airflow, whereas stops differ from both vowels and continuantobstruents in that stops involve total obstruction. An explanation along these lines may explicate whyvowels trigger regressive voicing of fricatives and affricates only. Nevertheless, the goal of this paperis to provide an account of voicing and continuancy alternations in Catalan, and a more thoroughexamination of the phonetic and articulatory details behind the phonological processes lies beyond thescope of this paper.

Page 17: Catalan obstruents and opacity: A case for representational

74 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

19. Ambisyllabic /z/: cu[z] això (cf. cu[s] ~ cu[z]ir)sew this sew to sew

a. Ambisyllabic [s] b. Ambisyllabic [z]

σ σ σ

k u s \ ß ø

σ σ σ

k u z \ ß ø

*

cluster clusterviolates AGREEFV satisfies AGREEFV

For the sake of simplicity, I will define AGREE(V) in my analysis of Catalanas a unified constraint grouping both the constraint on consonant clusters andAGREEFV, as shown in (20).10

20. AGREE(VOICE): Clusters of heterosyllabic consonants or heterosyllabicfricative-vowel sequences agree in voicing.

The following tableaux illustrate how this constraint and the proposedconstraint ranking gives the right results for both underlying voiceless (21) andvoiced (22) continuant obstruents. Constraints governing SSA are omitted as theyare not relevant here.

21. Underlying voiceless fricative

/Gos # astut/cunning dog

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

\/ gos]\s.tut

*!

\/☞ goz]\s.tut

* *

22. Underlying voiced fricative

/Baz # alt/tall vase

AGREE(V) FAITH-O(V) *VDOBS]σ MAX(V)

\/ bas]al]

*! * *

\/☞ baz]al]

*

10The Agreement constraint governing fricative-vowel clusters can be assumed to belong to afamily of agreement constraints that would include the agreement constraint on consonant clusters(AGREE). Agreement constraints could be related in an implicational manner so that languages that

Page 18: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 75

Although this constraint requires more refinement and further support, it isnevertheless the best and only attempt at an adequate account of prevocalic fricativevoicing in OT and allows a unified analysis of Catalan RVA processes. This initself already constitutes an improvement over earlier analyses such as Beckman(1997) , Colina (1995), and Merchant (1997).

Notice that the tableaux in (21) and (22) illustrate that the agreement constraintis more highly ranked than the positional faithfulness constraint, as already indicatedin the global ranking given in (16) and repeated in (23). This final ranking ofconstraints governing continuancy and voicing alternations accounts for all the factsabout the distribution of Catalan obstruents discussed in this paper. These arerepeated and summarized below in the tableaux in (24) through (28). Theseconstraints, together with the constraints governing prosodic structure, and therepresentation of edge consonants as ambisyllabic do away with the problem ofopacity and allow an integrated analysis of Catalan continuancy and voicingalternation in a monostratal framework that evaluates output surface structures.

23. ONS[CONT], AGREE >> FAITH-O(V), *[CONT] >> *VDOBSTR]σ >> MAX(V)

24. Word-internal onset obstruentsOutput: - faithful to underlying voicing

- voiced spirant in the case of SSA

/aBi/ ON[CT] AGREE F-ON(V) *[CONT] *VDO]σ MAX(V)a.bi] *!

☞ a.”i] *a.pi] * *!

25. Final devoicing (all obstruents)Output: - voiceless obstruent

- voiceless stop in the case of SSA

/rEB/ ON[CT] AGREE F-ON(V) *[CONT] *VDO]σ MAX(V)rEb] *!rEB] *! *

☞ rEp] *

have AGREE(FRIC-V), a seemingly less common and more marked constraint, would also have AGREE,affecting consonant clusters.

Page 19: Catalan obstruents and opacity: A case for representational

76 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

26. Voicing agreement in consonant clusters (all obstruents)Output: - voicing agreement determined by the voicing of the onset C

- stop in the case of SSA

/aBDikar/ ON[CT] AGREE F-ON(V) *[CONT] *VDO]σ MAX(V)☞ «b. di. ka] *«B. Di. ka] *!* *«p. ti. ka] *! **«p. di. ka] *! *«B. di. ka] *! * *

27. Prevocalic final (ambisyllabic) stop-spirant alternantOutput: - voiceless stop.

/rEB#aSO/ ON[CT] AGREE F-ON(V) *[CONT] *VDO]σ MAX(V)\/

rE b]«.SO]*

\/ rE B]«.SO]

* *!

\/☞ rE p]«.SO]

* *

28. Prevocalic final (ambisyllabic) fricativeOutput: - voiced fricative

/Gos#astut/ ON[CT] AGREE F-ON(V) *[CONT] *VDO]σ MAX(V)\/

gos]«s.tut]*!

\/ ☞ goz]«s.tut

* *

4. Surface representations and OT analyses

The goal of this paper has been to account for the allophonic variation ofCatalan obstruents. The analysis presented examines all the facts concerning voicingand continuancy alternations and makes the correct output form in all cases followfrom a single crucial constraint ranking. In this respect, it obeys the basic premisesof OT, making the constraints and constraint ranking provide us with the actualphonetic representation and integrating in a monostratal framework the processestraditionally assigned to different levels of the phonology. Although the prosodicstructures assumed for the surface representations demand a certain level ofabstraction, they allow a uniform single-level account of the data and succeed inencoding the close relationship between prosodic structure and morphologicalstructure. Therefore, an adequate analysis has been obtained taking into account bothprosodic structure and surface realizations.

Page 20: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 77

While there are several other OT accounts involving Catalan obstruents(Beckman, 1997; Colina, 1995), the only account that analyzes the full range of datais Merchant’s (1997) account of continuancy and voicing alternations framed withinSympathy theory. Nevertheless, I argue against such an analysis on the grounds thatthe current proposal is simpler, better motivated, more thorough, and, ultimately,that a Sympathy account is not necessary once the apparent opacity has beeneliminated.11 On the one hand, an account based on the notion of ambisyllabicityrelies on a device that is able to relate the facts with the proposal, that is, the choiceof the device of ambisyllabicity is not arbitrary. On the contrary, the solution ofambisyllabicity explains syllabification patterns and segmental alternations, andrelates prosodic structure and morphological structure. It is thus more strongly andindependently motivated than a Sympathy account. By referring to prosodicstructure, this analysis can straightforwardly derive the right output without the needfor intermediate sympathy candidates, considerably simplifying the grammar.Furthermore, an ambisyllabicity solution is applicable to other phenomena suchSpanish s-aspiration and n-velarization or German dorsal fricative distribution,whereas a Sympathy approach requires a different analysis for each case. Finally, themain argument against a Sympathy approach is simply that the introduction ofambisyllabicity, independently motivated in the theory, and thus the surfacedistinction between final consonants as prosodically ambisyllabic and word-internalconsonants as simple onsets does away with the opacity of final devoicingencountered in a derivational analysis. Whatever the merits of Sympathy theory forgenuine cases of opacity, a Sympathy account is rendered unnecessary in the case ofthe distribution of Catalan obstruents.

5. Conclusion

This paper has presented an extensive analysis of continuancy and voicingalternations in Catalan obstruents. The proposed analysis, which is based on arepresentation of final consonants preceding a vowel-initial word as ambisyllabic,that is, occupying simultaneously a coda and an onset position, captures thetraditional distinction between derived and non-derived onsets and eliminates surfaceopacity, while still working from a monostratal framework that evaluates surfaceoutputs. The benefits of an ambisyllabicity account are many. In addition tosuccessfully explaining segmental alternations and satisfying universal principles ofsyllabification, the proposed ambisyllabic representation encodes the interactionbetween morphological and phonological structure. Framed within the theory ofOptimality Theory, this analysis underlines the significance of syllable structure inthe formulation of relevant constraints, and it stresses the effect of the interaction ofconstraints governing different processes in the evaluation of the right outputs.Finally, the proposal also points to possible lines for further research, such as therelation between morphological boundaries and ambisyllabicity across languages,and the general issue of the relative ranking of constraints referring to differentsyllable positions.

11In fact, as mentioned in section 2.3, Merchant himself advocates for ambisyllabicity in hisanalysis of the distribution of the German dorsal fricative (Merchant 1996).

Page 21: Catalan obstruents and opacity: A case for representational

78 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

References

Avery, P. (1996). The representation of voicing contrasts. Doctoral dissertation,University of Toronto.

Bakovic, E. (1994). Strong onsets and Spanish fortition. MIT Working Papers inLinguistics 23, 21-39.

Beckman, J. (1997). Positional Faithfulness. Doctoral dissertation, University ofMassachusetts at Amherst.

Borowsky, T., J. Itô & R.-A. Mester. (1984). The formal representation ofambisyllabicity: Evidence from Danish. In Proceedings of the FourteenthAnnual Meeting of the North Eastern Linguistic Society, edited by C. Jonesand P. Sells. 34-48.

Butska, L. (1997) Voicing alternations in Ukrainian. Master thesis, University ofToronto.

Causley, T. (1996). Feature correspondence and identity: the Athapaskan case.Proceedings of the 27 meeting of the North Eastern Linguistics Society.Montreal: McGill University.

Cebrian, J. (1997). SV assimilation in Catalan and the implications for anasymmetrical *NC constraint. In Proceedings of the 1997 Montreal-Ottawa-Toronto Phonology Workshop. Toronto Working Papers in Linguistics 16,edited by P. Avery & C. Smallwood, 37-47. Toronto: University of Toronto.

Cebrian, J. (1998). Ambisyllabicity and constraint interaction in Catalan voicingand continuancy alternations. Manuscript, University of Toronto.

Colina, S. (1995). A constraint-based analysis of syllabification in Spanish,Catalan, and Galician. Doctoral dissertation, University of Illinois at Urbana-Champaign.

Cull, N. (1998). Onset augmentation in Kisi. Paper presented the annual meeting ofthe LSA, New York City, January 8-11, 1998.

Harris, J. W. (1969). Spanish Phonology. Cambridge: MIT Press.Harris, J. W. (1993). Integrity of prosodic constituents and the domain of

syllabification rules in Spanish and Catalan. In The View from Building 20:Essays in Linguistics in Honor of Sylvain Bromberger, edited by K. Hale &S.J. Keyser, 177-193. Cambridge, MA: MIT Press.

Hong, S. (1997). Prosodic domains and ambisyllabicity in Optimality Theory.Doctoral dissertation, University of Pennsylvania. (Rutgers Optimality Archive#209-0797).

Hualde, J. I. (1991). Basque Phonology. London/New York: Routledge.Hulst, H. van der. (1984). Syllable Structure and Stress in Dutch. Dordrecht,

Holland: Foris.Itô, J. (1989). A prosodic theory of epenthesis. Natural Language and Linguistic

Theory 7, 217-259.Kahn, D. (1976). Syllable-based generalizations in English phonology.

Bloomington: Indiana University Linguistics Club.Kang, S. K. (1991). Moraic representation of ambisyllabicity: Evidence from

Korean. Studies in the Linguistic Sciences 21, 89-100.Kreidler, C. (1995). The Pronunciation of English, A Coursebook in Phonology.

Oxford: Blackwell.

Page 22: Catalan obstruents and opacity: A case for representational

J U L I C E B R I A N 79

Lamontagne, G. & K. Rice. (1995). A correspondence account of coalescence. InUniversity of Massachusetts Occasional Papers in linguistics 18, Papers inOptimality Theory, edited by J. Beckman, L. Walsh Dickey & D. Urbancyk,211-224. Amherst: GSLA.

Lombardi, L. (1995). Positional faithfulness and the phonology of voicing inOptimality Theory. Manuscript, University of Maryland, College Park.

Lombardi, L. (1996). Positional faithfulness and voicing assimilation in OT.Manuscript, University of Maryland, College Park.

Lozano, M. del C. (1979). Stop and spirant alternation: Fortition and spirantizationin Spanish phonology. Doctoral dissertation, Ohio State University.Distributed by IULC, Bloomington, Indiana.

Mascaró, J. (1983). La fonologia i el cicle fonològic. Barcelona: UniversitatAutònoma de Barcelona.

Mascaró, J. (1984). Continuant spreading in Basque, Catalan and Spanish. InStudies in Phonology presented to Morris Halle by his teacher and students,edited by M. Aronoff & R. T. Oehrle. Cambridge, MA: MIT Press.

Mascaró, J. (1987). Syllable-final processes in Catalan. In Studies in RomanceLanguages. Publications in Language Sciences 25, edited by C. Neidle & R.A.Nuñez Cedeño, 163-180. Dordrecht: Foris.

Mascaró, J. (1991). Iberian spirantization and continuant spreading, CatalanWorking Papers in Linguistics, 167-179.

McCarthy, J. & A. Prince. (1993). Prosodic Morphology I: Constraint interactionand satisfaction. Manuscript, University of Massachusetts at Amherst andRutgers University. RuCC-TR-3.

Merchant, J. (1996). Alignment and fricative assimilation in German. LinguisticInquiry 27, 709-719.

Merchant, J. (1997). Sympathetic Devoicing and Continuancy in Catalan.Phonology at Santa Cruz 5, 57-62.

Palmada, B. (1994). La fonologia del català. Els principis generals i la variació.Barcelona: Universitat Autònoma de Barcelona.

Palmada, B. (1997). Continuant spreading and feature organization. In C. Neidle &R.A. Nuñez Cedeño (eds.), Issues in the Phonology and Morphology of theMajor Iberian Languages, edited by C. Neidle & R.A. Nuñez Cedeño (eds.),151-172. Washington, D.C.: Georgetown University Press.

Piggott, G. L. (1992). Variability in feature dependency: The case of nasality.Natural Language and Linguistic Theory 10, 33-77.

Recasens, D. (1993). Fonètica i fonologia. Barcelona: Enciclopèdia Catalana.Rice, K. (1990). Predicting rule domains in the phrasal phonology. In The

Phonology-Syntax Connection, edited by S. Inkelas and D. Zec, 298-312.Chicago: University of Chicago Press.

Rice, K. (1993). A reexamination of the feature [sonorant]: The status of 'sonorantobstruents'. Language 69, 308-344.

Rice, K. and P. Avery. (1989). On the interaction between sonorancy and voicing.Toronto Working Papers in Linguistics 10, 65-82.

Rice, K. and P. Avery. (1991). On the relationship between laterality and coronality.In Phonetics and Phonology Volume 2. The Special Status of Coronals:Internal and External Evidence, edited by C. Paradis & J-F. Prunet. San Diego:Academic Press, Inc.

Page 23: Catalan obstruents and opacity: A case for representational

80 T O R O N T O W O R K I N G P A P E R S I N L I N G U I S T I C S

Selkirk, E. (1982). The syllable. In The Structure of Phonological Representations(Part II), edited by H. van der Hulst and N. Smith, 337-383. Dordrecht: ForisPublications.

Steriade, D. (1988). Greek accent: A case for preserving structure. Linguistic Inquiry19, 271-314.

Wheeler , M. (1979). Phonology of Catalan. Oxford: Blackwell.