Distributed Reduplication

John Frampton

Northeastern University

February 2004

Table of Contents1. Preliminaries and Overview . . . . . . . . . . . . . . . 1

1.1. Moravcik (1978) . . . . . . . . . . . . . . . . . 3

1.2. Wilbur (1973) . . . . . . . . . . . . . . . . . . 6

1.3. A guide to what is to follow . . . . . . . . . . . . 8

2. Transcription and rule application in crossed structures . . . 11

2.1. The No Crossing Constraint (NCC) . . . . . . . . 12

2.2. Rule application in structures with long-distance geminates

16

2.2.1. Geminate inalterability . . . . . . . . . . . 16

2.2.2. Constraints on feature sharing . . . . . . . . 19

2.2.3. Korean consecutive reduplication . . . . . . 25

2.3. The Retraction Condition . . . . . . . . . . . . . 28

3. The morphophonology of reduplication . . . . . . . . . . 30

3.1. Reduplicative affixation is lexically triggered readjustment

35

3.2. Klamath distributive and intensive reduplication . . . 39

3.3. Tagalog (apparent overapplication) . . . . . . . . . 44

3.4. Erromangan:

The interaction of reduplication and readjustment 46

3.5. Nuu-chah-nulth reduplicative readjustment . . . . . 49

3.6. Some closing remarks . . . . . . . . . . . . . . 51

4. Truncated reduplication . . . . . . . . . . . . . . . . . 53

4.1. Truncation at the leading edge (reduplicant truncation) 54

4.1.1. Truncated prefixes and suffixes . . . . . . . 56

4.2. Nested duplicants . . . . . . . . . . . . . . . . 60

ii Contents

4.3. Medial truncation . . . . . . . . . . . . . . . . 62

4.4. Truncation at the trailing edge (remnant truncation) . . 63

4.4.1. Reduplicant truncation combined with remnant

truncation: Permutation, Metathesis, and Infixation 65

4.4.1.1. Varieties of Yaqui reduplication . . . . . 68

4.5. Transcription made easy . . . . . . . . . . . . . 69

5. Sources of variation . . . . . . . . . . . . . . . . . . 72

5.1. Possible juncture insertion rules . . . . . . . . . . 73

5.1.1. Creek plural adjective reduplication . . . . . 76

5.1.2. Some common juncture insertion rules . . . . 76

5.2. Possible domains of reduplicative affixes . . . . . . 77

5.2.1. Ulwa possessive . . . . . . . . . . . . . . 79

5.2.2. Manam final foot reduplication . . . . . . . 80

5.2.3. Exclusion of initial unfooted material . . . . . 81

5.2.3.1. Bella Coola . . . . . . . . . . . . . . 81

5.2.3.2. Orokaiva . . . . . . . . . . . . . . . 82

5.2.3.3. Imdlawn Tashlhiyt Berber . . . . . . . . 83

5.3. Shortcut Repair at the reduplicant-remnant boundary . 84

6. Prosodic Adjustment . . . . . . . . . . . . . . . . . . 88

6.1. Onset incorporation in Kıhehe . . . . . . . . . . . 88

6.2. Augmentation of the duplicant to a heavy syllable . . 92

6.2.1. Mokilese progressive . . . . . . . . . . . . 92

6.2.2. Complications in Mokilese progressive reduplication

96

6.2.2.1. CP-sequences which are not morpheme internal

97

6.2.2.2. Double reduplication . . . . . . . . . . 98

6.2.2.3. Prenazalized geminates . . . . . . . . . 99

6.2.3. Agta . . . . . . . . . . . . . . . . . . . 100

6.2.4. Ilocano . . . . . . . . . . . . . . . . . . 100

6.2.5. Ponapean . . . . . . . . . . . . . . . . . 103

6.3. C-finality as a secondary prosodic desideratum . . . . 107

iii Contents

6.3.1. Lardil iterative reduplication . . . . . . . . 108

6.3.2. Adjustment to C-finality in Nuu-chah-nulth . . 109

6.3.3. C-finality in Yaqui light syllable reduplication . 110

6.3.4. Moravcik’s Generalization reconsidered . . . 112

6.4. Heavy syllable suffixation of left-edge material (Chukchee)

114

6.5. Final syllable prefixation (Madurese) . . . . . . . . 115

6.6. Final syllable suffixation (Kaingang) . . . . . . . . 116

6.7. Korean consecutive syllable reduplication . . . . . . 116

6.8. The onset-coda asymmetry and Moravcik’s Generalization

117

6.9. Towards a parametric theory of prosodic adjustment . 118

7. Case studies . . . . . . . . . . . . . . . . . . . . . . 122

7.1. Ndebele unintensive reduplication . . . . . . . . . 123

7.1.1. Passive/applicative interaction . . . . . . . . 126

7.1.2. Overapplication of passive palatalization . . . 128

7.1.3. Overapplication of perfective imbrication . . . 130

7.1.4. Yi-epenthesis . . . . . . . . . . . . . . . 131

7.2. Morpheme integrity in Kinande . . . . . . . . . . 133

7.2.1. Short causatives . . . . . . . . . . . . . . 136

7.3. Asheninca Campa intensive reduplication . . . . . . 138

7.3.1. Stem augmentation induced by C-initial suffixes 142

7.3.2. Copying a word boundary . . . . . . . . . . 144

7.4. Washo plural reduplication . . . . . . . . . . . . 145

7.4.1. CV-reduplication . . . . . . . . . . . . . 146

7.4.1.1. Bare timing slots in the underlying representation

148

7.4.2. VC-reduplication . . . . . . . . . . . . . 150

7.4.3. The residue . . . . . . . . . . . . . . . . 152

7.5. Tohono O’odham plural reduplication . . . . . . . 154

7.6. Infixing consonant copy: Temiar verbal reduplication and

Levantine Arabic perjorative reduplication 160

iv Contents

7.6.1. Temiar verbal reduplication . . . . . . . . . 160

7.6.2. Levantine Arabic perjorative reduplication . . 163

7.7. Sanskrit intensive and perfective reduplication . . . . 165

7.7.1. Low vowel syncope . . . . . . . . . . . . 166

7.7.2. Perfective reduplication . . . . . . . . . . . 170

7.7.3. Sanskrit specialization of NCC repair . . . . . 171

7.7.4. Intensive reduplication . . . . . . . . . . . 172

7.7.5. Low vowel syncope and intensive reduplication 174

7.8. Cyclicity and double reduplication in Lushootseed . . 175

7.8.1. Diminutive reduplication . . . . . . . . . . 175

7.8.2. Distributive reduplication . . . . . . . . . . 178

7.8.3. “Out of control” (OC) reduplication . . . . . 179

7.8.4. Double reduplication . . . . . . . . . . . . 180

7.9. CVC reduplication in Chumash . . . . . . . . . . 181

7.10. Chaha reduplication . . . . . . . . . . . . . . . 185

7.10.1. Overapplication of impersonal labialization

and palatalization 185

7.10.2. Inherently reduplicated roots . . . . . . . . 188

7.10.3. Chaha Continuant Dissimilation . . . . . . 189

7.10.4. Chaha frequentative reduplication . . . . . . 191

7.10.5. The interaction of frequentative reduplication

and x-dissimilation 193

7.10.6. Kenstowicz and Banksira’s analysis of x-dissimilation

197

A. The NCC and the Retraction Condition . . . . . . . . . . 200

B. Defect driven rule schemata . . . . . . . . . . . . . . . 205

B.1. Late syllabification in Arabic . . . . . . . . . . . 206

C. Raimy’s theory of R-representations . . . . . . . . . . . 210

C.1. R-representations . . . . . . . . . . . . . . . . 210

C.2. Linearization . . . . . . . . . . . . . . . . . . 213

C.3. Prosodic structure and nonlinear R-representations . . 215

v Contents

C.4. Concatenative versus nonconcatenative morphology . 216

Notes . . . . . . . . . . . . . . . . . . . . . . . . . . 219

References . . . . . . . . . . . . . . . . . . . . . . . 225

Glossary of Abbreviations and Rules . . . . . . . . . . . . 230

Subject Index . . . . . . . . . . . . . . . . . . . . . . 232

Language Index . . . . . . . . . . . . . . . . . . . . . 236

Preface

The central goal of theoretical linguistics is understanding the natureof the mental computations which are carried out in producing therepresentations which the articulatory system acts on in producingspeech. Although there are many new ideas in this monograph aboutthe computations which are put to use in reduplicative phonology, somefairly radical, the intent is quite conservative. I hope to provide evidencethat the model of the phonological computation which was developedby Chomsky and Halle in The Sound Patterns of English (1968) isfundamentally correct: surface forms are produced by the successivemodification of underlying forms.

I will call this model derivational phonology. The name is mislead-ing, but in such common use that an attempt to change it might createeven more confusion. It is misleading because the particular feature ofthe SPE model that distinguishes it is not that the computation employsderivations. All computations of any complexity go through a sequenceof intermediate states, hence have intermediate representations andderivations. Rival theories, Optimality Theory (Prince and Smolensky,1993), for example, compute the input-output relations by means ofderivations which consist of progressive filtering of (very large) sets ofphonological representations. What distinguishes derivational phonol-ogy is not that it employs derivations, but the nature of the intermediatestates in the derivational computation. It proposes that intermediatestates in the computation have the same general form as initial states

vii Preface

and final states; the step by step derivation of the output proceedsthrough modification. A better name than derivational phonology mightbe “developmental phonology,” emphasizing that the issue is the pro-gressive change of a single entity. Derivational phonology shares withdevelopmental biology the understanding that certain features of thecurrent state of an organism/representation can only be understood interms of the state of the organism/representation at earlier stages in itsdevelopment.

A thorough and convincing demonstration of the adequacy ofderivational phonology with respect to reduplicative phenomena hasparticular relevance at this time because many researchers still acceptthe claims of McCarthy and Prince (1995) that derivational phonologyis incapable in principle of analyzing reduplication. Inadequacies in theanalyses of reduplication that were available at the time their paper ap-peared gave some grounds for thinking that this might be so. But recentderivational analyses of reduplication, Raimy (2000) in particular, haveshown that McCarthy and Prince’s criticisms of derivational phonology(at least insofar as reduplication is concerned) were only valid criticismsof particular proposals, not demonstrations of insoluble obstacles to aderivational analysis of reduplication. My hope is that this monographwill make this point more thoroughly and convincingly.

Although particular reduplicative processes in many languages arediscussed in Chapters 1–6, which develop the general theory, the bookcloses with an extensive sequence of detailed case studies. The widerange of case studies is not simply intended to show off the accom-plishments of Distributed Reduplication. It is intended to be evidencethat the theory is true. It is a relatively easy task to show that somepurported “general principles” successfully account for some particularphenomena in some particular language. But it is a much more difficulttask to show that these general principles are truly general and can serveas the basis for analyzing widely different phenomena (within the rangethat UG makes possible) in very different languages.

Acknowledgments

There are many people to thank for making this monograph possible.Above all, it would not have been possible (not even close) withoutthe ongoing support and guidance of Morris Halle. It is no accidentthat an impressive series of important books on phonology over the last30 years begin with a tribute to the inspiration and guidance of MorrisHalle. He is a singular teacher.

Thanks also to Sam Gutmann and Sylvain Bromberger for manycomments. Since I devote several pages in what follows to showing thathe did not get it right, I am glad to be able to acknowledge at the startthat it was Eric Raimy’s work which showed that a derivational solutionto the problems posed by reduplication was possible. He reminded usthat most of the crucial advances in phonological theory are advances inunderstanding the structure of phonological representations.

Thanks to audiences at MIT’s “Morph Beer” and Phonology Circle,and the 2002 Graz Reduplication Conference for their interest in andcomments on earlier versions of portions of this material. Thanksto many individuals for helpful discussion: Terry Crowley, StuartDavis, Justin Fitzpatrick, Larry Hagberg, Ken Hale, Sheldon Harrison,Mashudi Bin Kader, Brett Kessler, Alec Marantz, Ngessimo Mutaka,Andrew Nevins, Eric Raimy, Galen Sibanda, Suzanne Urbanczyk, andAlan Yu.

Chapter 1

Preliminariesand Overview

I start from the assumption that humans exercising their language fac-ulty carry out computations on abstract linguistic representations andthat the core project of linguistics is to discover the form of theserepresentations and the computational resources which are brought tobear in manipulating them. Linguists who are interested in this projectbuild models with the hope that comparison of these models with humanlinguistic performance will elucidate some aspects of the languagefaculty and allow the construction of more refined models which willallow further elucidation. In short, they carry out scientific research. Inthis spirit, I will present a theory of reduplication.

Reduplication uses a copying mechanism. The most remarkableexample of a copying-like process in the biological world is the tran-scription of genes to RNA.1 In the simplest case, a gene is a subse-quence of a long DNA sequence. More complex cases involve multiplesubsequences with intrusions that are truncated when transcription takesplace. Transcription is a two step process. First, a marker, called a“transcription factor,” is attached to the DNA at the beginning of thegene to signal the position at which transcription should be initiated.In effect, a transcription factor is a “molecular diacritic.” Second,transcription itself takes place starting at the marker. Marking by theattachment of a transcription factor to the DNA and transcribing markedDNA, are independent mechanisms, carried out by different biologicalsystems. I will attempt to establish in this paper that reduplication is

2 Chapter 1

organized in roughly the same way, as a two-stage process. In placeof the attachment of transcription factors to the DNA, junctures (calledt-junctures, transcription junctures) are inserted into the timing tier bythe morphology. The transcription mechanism, which operates in thephonology, is triggered by these t-junctures to carry out transcription,just as the DNA transcription mechanism is triggered by transcriptionfactors to carry out its biological copying. The t-junctures are inter-preted as instructions which trigger and guide the duplication and/ortruncation of certain material.

Since transcription takes place in the phonology, it is possiblethat phonological operations apply after t-juncture insertion but beforetranscription. It is also possible that phonological rules are sensitiveto the presence of t-junctures. A rich array of interactions is possi-ble. Since several different operations can be involved, rule orderingcan have significant effects. The sometimes surprising complexity ofreduplication processes derives from the intricacy of the interaction ofmultiple simple processes. Since surface patterns are the combinedresult of the morphological insertion of t-junctures and various rulesdistributed throughout the phonology, the theory is dubbed DistributedReduplication (DR).

The decomposition of copying into one stage in which juncturesare inserted into a linear string, and a second stage in which actuallydoubling occurs, is familiar from another source; repetition blocks inmusic. Part of the standard repertoire of signs and symbols used inwriting music scores are repeat marks: and . They are instructions tobe used in the translation of the score from an abstract representationalto a concrete physical performance. There is a strong similarity betweena music score and a phonological representation. Goldsmith (1976)describes an autosegmental representation as an “orchestral score.”Both a music score and an autosegmental representations are abstractrepresentations which can be, but need not be, translated into a physicalperformance.2

Two papers which were published in the 1970s have played apivotal role in the development of theories of reduplication in the periodsince then. Wilbur (1973) discussed several interactions between redu-plicative copying and the general system of phonological rules whichshowed that there was some kind of unexpected influence on the appli-cation of phonological rules to representations involving reduplication

Preliminaries and Overview 3

which had the effect of maintaining the identity of copied material andthe material that it was copied from. She proposed that this requireda revision of the then current view of phonology, with some kind ofan Identity Condition standing outside of the derivation system, oper-ating transderivationally, which produced the desired results. Moravcik(1978), in a broad survey of attested reduplication patterns, found thatcertain kinds of reduplication which linguists might naively think to bepossible or even common, were not attested. In order to account for this,she advanced several proposals about the nature of the computation.

Most of the remainder of this introductory section is devoted todiscussing the relation between DR and the work of Wilbur and Morav-cik and outlining how their empirical findings will be explained inChapters 2-6, starting from the idea that the first step in reduplication isinsertion of t-junctures into the timing tier of a phonological represen-tation.

1.1 Moravcik (1978)

The most striking empirical fact that came out of Moravcik’s survey isthat syllable copy reduplication (Pattern 3 below) appears to be missingfrom the large inventory of reduplication patterns found in the world’slanguages. Patterns 1 and 2 below are widely attested, but Pattern 3 isunattested.

(1) stem Pattern 1 Pattern 2 *Pattern 3gin.dal gi-gin.dal gin-gin.dal gin-gin.dalgi.dal gi-gi.dal gid-gi.dal gi-gi.dalgi:.dal gi-gi:.dal gi:-gi:.dal gi:-gi:.dal

The puzzle that must be addressed by theories of reduplication is whysyllable copy does not exist.

In order to explain the distribution of reduplication patterns shefound, Moravcik proposed that the reduplication mechanism is re-stricted as shown in (2). It is useful for the sake of discussion to separateher hypothesis into a weak version and a strong version.

4 Chapter 1

(2) 1. The mechanism of reduplication makes no reference to theprosodic structure of the stem.

2. The mechanism of reduplication chooses what to copy on thebasis of a CV template.

Although both hypotheses have proved untenable in the form that sheadvanced them, both contain important kernels of truth that have beenincorporated into later analyses.

Marantz (1982) highlighted the example of Yidiny plural redupli-cation (Nash 1979; 1980) as a counterexample to Moravcik’s proposals.It is exemplified below:

(3) stem pluralgin.dal.ba gin.dal-gin.dal.bamu.la.ri mu.la-mu.la.ri

He noted that determining which substring is copied requires referenceto the syllable structure of the stem, contradicting (2.1). Reduplicationcopies the first two syllables to the left. Although Marantz provided ananalysis of Yidiny plural reduplication, by extending Moravcik’s ideaof a CV template to a template containing prosodic units (syllables), hewas then unable to explain why syllable reduplication does not occur.He concluded that it “. . . leaves us with a mystery.” (See Section 5.2 formore on Yidiny plural reduplication.)

Broselow and McCarthy (1983) provided part of the solution tothe mystery. They realized that sometimes an affix applies not to thewhole stem, but to a prosodic subunit of the stem. The computationof the material to be reduplicated therefore naturally bifurcates intotwo computations; the computation of the domain of the affix, and thecomputation of that portion of the domain that is copied. Moravcik’s(2.1) can then be modified to be a hypothesis about the second com-ponent of the computation. Prosodic Morphology (see McCarthy andPrince, 1995, and many references cited there) helped to clarify thatthe kinds of prosodic subunits of the stem which can be the domain ofan affix must be “wordlike” in some sense. Prosodic requirements onwords are weight requirements, bimoraicity or bisyllabicity. Under thisanalysis, domain determination can certainly be sensitive to prosodicconstituent boundaries, but only in a limited way. The sensitivity is to

Preliminaries and Overview 5

foot boundaries, not syllable boundaries. Yidiny plural reduplicationis now straightforward. The domain of the affix which generates thepattern is an initial bisyllabic foot. The entire domain is selected forcopy. Yidiny plural reduplication is total reduplication (of the domain).

DR further subdivides the computation of the material to be tran-scribed. It is factored into three components:

(4) 1. Domain Selection;2. Juncture Insertion; and3. Prosodic Adjustment

Domain Selection and Juncture Insertion are morphological operationswhich establish the initial location of the t-junctures. But these locationscan be modified (i.e. adjusted) by Prosodic Adjustment, which takesplace in the phonology, prior to transcription. Although Prosodic Ad-justment is in the phonology, it is morphologically conditioned. Manyreduplicative morphemes trigger prosodic adjustment, but many do not,so (4.3) may or may not be present in the phonology.

A brief comparison with the architecture of Prosodic Morphology(PM) may be helpful in clarifying the architecture of DR. Prosodic Ad-justment is governed by a target prosodic desideratum (or desiderata),which is roughly equivalent to PM’s notion of a prosodic template. Thefact that some reduplicative affixes trigger prosodic adjustment but somedo not is roughly equivalent to the PM distinction between templaticand atemplatic reduplication. A significant difference between PM andDR is that while atemplatic reduplication is exceptional in PM, it isprosodic adjustment that is exceptional in DR. Pattern 1 in (1), forexample, involves no prosodic adjustment. The morphology simplyinserts a [-juncture before the first timing slot and a ]-juncture after thefirst vowel. Pattern 2 is the result of the same juncture insertion rules,but the material bounded by the duplication junctures is adjusted to aheavy syllable by shifting the ]-juncture to the right.

Moravcik’s (2.1) holds of Juncture Insertion, which uses a very re-stricted inventory of insertion rules. Prosodic Adjustment is driven prin-cipally by a weight desideratum, although secondary prosodic desider-ata (simple onset, C-finality, etc.) are possible. Furthermore, althoughthere are cases in which the rules that Prosodic Adjustment uses toachieve satisfaction of the prosodic desiderata are not sufficient to

6 Chapter 1

achieve satisfaction, several default rules are universally available. Weshall see that these assumptions conspire to make syllable copy redupli-cation impossible.

Chapters 2-4 develop the general theory of juncture insertion andtranscription. Chapters 5 and 6 are devoted to a careful examinationof the space of possible juncture insertion rules, their interaction withprosodic adjustment, and the variety of reduplication patterns which thisgenerates.

1.2 Wilbur (1973)

The kinds of examples that Wilbur pointed out and the hypothesis thatshe advanced to explain the facts was an important factor in the turnaway from generative phonology taken by Optimality Theory and itsfurther development into Correspondence Theory, McCarthy and Prince(1998). Optimality Theory provided a coherent framework in which anIdentity Condition of the kind proposed by Wilbur could be embedded.

A well-known example of the kind of the interaction of redu-plication and the general phonology of a language which led Wilburto propose the Identity Condition comes from Onn’s (1976) work onMalay. There are no underlying nasal vowels in Malay, but nasal-ization spreads to the right (under locality conditions which can beignored in this introductory discussion, but which will be consideredin more detail later) so that underlying aNan, for example, surfacesas aNan. Malay makes extensive use of total reduplication to realizevarious morphemes. Naively, one would expect totally reduplicatedaNan to surface as aNanaNan or aNanaNan, depending on the timing ofreduplication and nasalization. Instead, it surfaces as aNanaNan, with noapparent source of nasalization of the initial vowel. Wilbur concludedthat effects of this kind result from an Identity Condition which holdsbetween the source of reduplication and the reduplicant that persistsafter copying has created the reduplicant. The mysterious nasalizationis then analyzed as the joint result of nasalization of the post-nasal copyof a and the Identity Condition, which somehow makes the two copiesof a identical. Wilbur did not propose an architecture within whichthe Identity Condition could produce the desired effect. It is easy tosee how Wilbur’s Identity Condition evolved into the base-reduplicant

Preliminaries and Overview 7

conditions of Correspondence Theory, which abandoned any attempt tounderstand the mechanism.

Wilbur effects did not receive a coherent account within deriva-tional phonology until Raimy’s (2000) thesis. He proposed a radicalrevision of the structure of the timing tier. In particular, he proposeda structure in which the timing tier is not linearly ordered, so that atiming slot can be preceded or followed by more than one timing slot.The major weakness of Raimy’s proposal is that a nonlinear timing tiermakes it impossible to build syllable and prosodic structure. AlthoughDR rejects Prosodic Morphology’s analysis of reduplication, it is impor-tant that certain of its insights be maintained. This is impossible unlessreduplicative structure and prosodic structure can freely coexist. A fullcritique of Raimy’s proposal is given in Chapter 8.

DR takes a very different approach to understanding Wilbur effects,using ideas of McCarthy (1986). The key step is a correct understandingof the copying operation. It is a two stage process, illustrated for theMalay example in (5).

(5)

a

×N

×a

×n

×copy

−→

a

×

N

×

a

×

n

× × × × ×NCC

Repair−→

a

×N

×a

×n

×a

×N

×a

×n

×

The first step, which will be called Transcription (Trscr) in what follows,copies only timing slots and their phonemic associations, not phonemesthemselves. The second step repairs the violations of the No CrossingConstraint (NCC) which Transcription creates. In (5), the repair iscarried out by what McCarthy called phoneme fission.

Just as the separation of reduplication into juncture insertion andactual doubling allowed for the intervention of prosodic adjustment, theseparation of doubling into Transcription and subsequent NCC repairallows for the intervention of phonological rules. The application ofphonological rules to structures like (5) with long-distance geminatesis the source of many of the puzzling Identity Effects which have beendiscovered. Note that the initial vowel a in (5) occurs in two differentenvironments. One of the timing slots that it is linked to is initial, butthe other timing slot that it is linked to immediately follows a timing

8 Chapter 1

slot that is linked to a nasal consonant. More discussion is obviouslyrequired, and will be provided in Chapter 2, but if standard ideas aboutrule application to geminates are extended to long-distance geminates,and nasalization in Malay is not subject to geminate inalterability, thenthe derivation below results:

(6)

a

×

N

×

a

×

n

× × × × ×Nasalization−→

a

×

N

×

a

×

n

× × × × ×

NCCRepair−→

a

×N

×a

×n

×a

×N

×a

×n

×

The nasalization of the initial vowel in aNan → aNanaNan was posedas an insurmountable problem for derivational phonology by McCarthyand Prince (1998). The decomposition of reduplicative copying intoTranscription and NCC Repair gives a simple derivational solution tothe problem.

1.3 A guide to what is to follow

A closer look at the application of phonological operations on structureswith crossing violations (called crossed structures in what follows as adescriptive convenience) will reveal some subtleties. Nasalization isgenerally thought to be a feature spreading operation. The NCC cannotbe a derivational constraint on Transcription, otherwise (5) would beimpossible. But the NCC (or something like it) does appear to constrainfeature spreading, as a derivational constraint. In order to make theanalysis illustrated in (6) convincing, a thorough discussion of crossedstructures and feature spreading in crossed structures is required. Chap-ter 2 contains a discussion of the No Crossing Constraint, constraints onfeature spreading, and the application of phonological rules in crossedstructures. Some skepticism has been expressed about the existence ofnasalization overapplication in Malay, and about the existence of Wilbureffects in general. The discussion of Chapter 2 meets many of theseobjections by showing that spreading in crossed structures is severely

Preliminaries and Overview 9

constrained. Although aNan → aNanaNan is predicted, for example,aran→ aranaran is not.

A theory of reduplication must fit into a general theory of mor-phophonology. Chapter 3 outlines a theory of morphology and the placeof reduplication in this theory. Juncture insertion is nonconcatenativemorphology. In spite of the fact that reduplication often produces stringsof elements on the surface that look like prefixes or suffixes, redu-plicative inflection derives from the same kind of morphology whichproduces sIN → saN as the realization of the past tense morpheme inEnglish for the verb root sing. Realization of the past tense morphemein this context does not cause the concatenation of affix. But it doestrigger an operation applied to the stem (vowel ablaut).

Steriade (1988) proposed that reduplication consists of total stemreduplication coupled with truncation. DR takes from this the idea thatreduplication consists of the transcription of some substring of the stem(not the whole stem) coupled with truncation. A theory of truncationis presented in Chapter 4 and its complex interactions with copying areexplored. It will turn out that many instances of metathesis as well asinfixation are best analyzed as varieties of reduplication.

Chapter 5 develops Broselow and McCarthy’s (1983) idea that theeffective stem in reduplication processes is sometimes only a subwordof the stem. The notion “subword” must be made precise. Bothfoot-based and morphologically based subwords will be recognized.Chapter 6 investigates prosodic adjustment in detail, with a detaileddiscussion of heavy syllable prefixal reduplication in the related Aus-tronesian languages Mokilese, Ponapean, Agta, and Ilocano.

Chapter 7, by far the longest chapter, is a series of case studies.Almost all of the languages with well-studied complex reduplicativeprocesses are discussed in some detail: Asheninca Campa, Chaha,Kinande, Lushootseed, Ndebele, Sanskrit, and Washo. In the author’sexperience, it is one thing to develop a theory that is designed for oneparticular language, but it is quite another to propose a theory whichholds itself responsible for the full range of empirical data. The test of atheory is not the excellent job it does on its “poster child” language (toborrow Donca Steriade’s colorful phrase), but its ability to make senseof a phenomenon as a computational ability that is put to use in differentways in different languages.

10 Chapter 1

Appendix C is a critique of Raimy (2000). What I have calledDistributed Reduplication is based on ideas of McCarthy, Steriade, andOdden and Odden. Although DR shares very little with Raimy’s ap-proach to reduplication, his 1999 dissertation, was the direct inspirationfor the development of DR. He demonstrated that the key to understand-ing Identity Effects in reduplication is to be found in understanding thespecial structures involved in reduplication. DR follows his lead in this,but comes to different conclusions about the special structure involved.Raimy proposed a nonlinear timing tier. DR rejects that in favor ofMcCarthy’s idea of long-distance geminates. Appendix C is devoted todetailing the weaknesses in Raimy’s approach, indirectly justifying theapproach taken by DR.

Chapter 2

Transcription andrule applicationin crossed structures

A convenient term to refer to the representations with long-distancegeminates produced by Transcription is useful, since I will frequentlyneed to refer to such structures.1 They will be called crossed structures.

Transcription is relatively straightforward, but something needs tobe said about directionality. Transcription can be to the right or the left.

(7) a. [

t

×a

× ]

k

×i

×Right

Transcription−→

t

×a

× × ×k

×i

×

b.

LeftTranscription−→ × ×

t

×a

×k

×i

×

The newly created timing slots in (7) are shaded.If the initial representation had only timing slots and phonemic as-

sociations, as above, there would be no distinction between the resultingstructures. They are both:

×t

×a

× ×k

×i

×

12 Chapter 2

But the structures in (7) are impoverished, not showing morphologicaland prosodic structure. If less impoverished structure is considered, thedistinction between left and right transcription then is clear. Consider:

(8) a. [

σ

t

×a

× ]

l

×σ

k

×i

×Right

Transcription−→

σ

t

×a

× × ×l

×σ

k

×i

×

b.

LeftTranscription−→ × ×

σ

t

×a

×l

×σ

k

×i

×

Right Transcription in (8a) is highly disruptive, requiring syllable reor-ganization, but Left Transcription is not. More for conceptual reasonsand the sake of concreteness than because there is strong evidence oneway or the other, I will assume that the unmarked choice is the onewhich is least disruptive to the structure of the stem. This choice isfixed for the affix, so that if one direction of transcription producesa simpler outcome in most cases, that choice is assigned to the affix.If the duplicant is aligned with one edge of the stem and not theother, the direction of transcription is toward that edge, at least as theunmarked choice. The notion “disruptive”, of course, is not clearlydefined, so this should be taken as a preliminary statement. Therecan be conflicts between disrupting morpheme structure and disruptingprosodic structure. Further discussion will be postponed until the issuepresents itself concretely.

2.1 The No Crossing Constraint (NCC)

Some readers may be reluctant to consider representations like (9)seriously because they massively violate the NCC, the ban on crossedassociation lines.

(9)

a

×

N

×

a

×

n

× × × × ×

Transcription and rule application in crossed structures 13

If the role of the NCC in phonology is considered carefully, however,good grounds can be found for at least entertaining the possibility thatstructures like (9) are present at some underlying level.

Constraints play various roles in derivational phonology. Deriva-tional constraints restrict the output of rules. Rule application thatwould result in a violation of the constraint is blocked. Halle andIdsardi’s (1995) analysis of footing and stress is a good example of theextensive use of derivational constraints to control the application of asimple rule system. Constraints on the distribution of foot delimitersinteract in that theory with general rules inserting foot delimiters toaccount for the stress facts of a large variety of languages. Mostresearchers, however, have not considered the NCC to be a derivationalconstraint.

Goldsmith (1990), in summing up the rule formalism relevant totonal phonology, says:

If a rule is formulated to add a single association line, it can, inprinciple, cause a line-crossing situation. In this case, . . . the linethat the rule adds remains, but the line that formerly existed is takento be the offending line, and is automatically erased.

On this view, the NCC does not block rule application. Rules can applyeven though their output violates it. A subsequent repair operationbrings the structure into compliance with the NCC. For Goldsmith,repair is immediate (i.e. automatic), so that the structure with crossingviolations is highly transitory. Nevertheless, it is clear that Goldsmithdoes not consider the NCC to be a derivational constraint.

Closer to the concerns of this paper is the position taken byMcCarthy in his groundbreaking work on nonconcatenative Semiticmorphology. Example (10) below illustrates the kind of morphologywhich McCarthy proposed for Semitic root-template association. It is(42) from McCarthy (1986).

(10) /sb/root + µ ⇒

root affix

s

×

i

×

b

× ×

e

× ×

14 Chapter 2

The affix µ specifies a template and vocalism. The combination of rootand affix is partitioned into two units: the timing slots derived directlyfrom the root consonants and the timing slots derived from the templatevocalism. The CVCCVC template requires four consonants. Thetwo consonants of the root are associated with the first two consonantpositions, left-to-right, and the last two are filled by multiple associationof the rightmost root consonant.

McCarthy implemented the idea by supposing that the stem andaffix segments are placed on distinct autosegmental tiers. This provideda means to avoid the NCC violation in (10). The idea is that thederivation proceeds as (11), with the root and affix segments initially ondistinct tiers, as shown in (11.1). The phonemes above and below theline should be viewed as being on different tiers. An operation calledTier Conflation applies, eliminating split phoneme tiers, but creatingNCC violations. Violations of the NCC are then removed by theoperation fission, which copies phonemes and their associations withthe timing tier, to eliminate the crossing constraints.

(11) 1. Morphology: /sb/root + template →s

×

i

×b

× ×

e

× ×

2. Tier Conflation: →s

×i

×b

× ×e

× ×

3. Fission: →s

×i

×b

× ×e

×b

×

McCarthy did not propose an explanation for why Tier Conflationapplies in (11.2). It was necessary to assume that it did in order toexplain the fact that at late stages in the morphophonology, nonadjacentconsonants do not act as if they are connected. McCarthy implies thatfission applies more or less immediately to remove the NCC violationscaused by Tier Conflation, taking a position similar to Goldsmith’s.But he does not formally distinguish between output conditions whichdemand repair of violations at some point before the output interface

Transcription and rule application in crossed structures 15

and conditions which demand immediate repair. The picture he paintsis that some phonology takes place while the root and affix tiers aresplit, then Tier Conflation takes place at some point for some reason,with repair following immediately in order to bring the representationinto compliance with the NCC.

It is equivalent, and much more straightforward, to assume thatthere are no split tiers, and that NCC violations may persist in thephonology, repaired eventually but not necessarily immediately. Itseems to me that the more straightforward approach was not takenbecause the importance of the NCC in restricting tone and featurespreading in some environments made it appear to be a strict constrainton derivation. The closest that McCarthy was willing to come toadmitting NCC violations into phonology was to stipulate that the NCCcould be temporarily violated, but violations had to be immediatelyrepaired. Like Goldsmith, McCarthy does not consider the questionof immediate repair versus eventual repair. Certainly, no evidence waspresented that repair must be immediate.

Sagey (1988) argues that the NCC is an output condition on phonol-ogy. Whereas Goldsmith and McCarthy assume that NCC repair isimmediate, undertaken as soon as NCC violations are created, Sageydoes not. DR follows Sagey on this point. NCC violations are repairedat some point in the derivation, but not necessarily immediately afterthey are created. This has the important consequence that representa-tions like the initial representation in (5) can persist long enough forphonological processes to take place while the structure has crossedassociation lines before the NCC violations are repaired by fission.This is crucial in accounting for apparently anomalous nasalization inreduplicated forms in Malay and other similar examples which will beexamined later.

In this paper, I assume that the NCC is an interface condition,satisfied at the phonology-phonetics interface. The relevant point is thatthe NCC violations in (11) are not necessarily repaired immediately,so that phonological rules may apply to (11) before repair. As aninterface condition, the NCC is not a constraint on rule application, but aconstraint on possible grammars. I assume that derivational phonologydoes not employ lookahead. Derivational steps cannot be directlyinfluenced by interface conditions, except insofar as those requirementsare incorporated into the design of the grammar. Grammars that produce

16 Chapter 2

outputs which generate structures at the interface which violate the NCCare rejected, not rule applications which produce outputs which violatethe NCC.

Marantz (1982) briefly entertains a proposal (close to our own)which admits NCC violations, but quickly rejects it as implausible. Hebriefly considers the structure (12), but rejects it almost out of hand.

(12) C C

t

C

k

C

k

C

i

VV

a

V

The structure (12) is just a pre-×-slot version of the structure which hasbeen proposed here. In rejecting (12), Marantz says that “associationlines would cross in violation of the basic condition on autosegmentalphonology [association lines never cross] . . . ” He adds in a footnote(fn. 8) the argument that even if crossed associations were allowed,then special stipulations would be needed to ensure that phenomena likemirror-image reduplication were not permitted.

I already argued above that “the basic condition on autosegmentalphonology” is not a derivational constraint. The second concern carriesno weight. Derivational morphophonology does not operate by freegeneration and filtering.2 Transcription is not free association filteredby a system of well-formedness conditions on possible associations, buta core algorithm of the language faculty. The same mechanism thatensures that dog doesn’t come out of the lexicon as god ensures thattotal reduplication of dog produces dog-dog, not god-dog.3 It is hardto escape the conclusion that the essential obstacle to adopting (12) wasthe presumed inviolability of the constraint against crossed associations.

2.2 Rule application in structures with long-distance geminates

Crossed structures can persist long enough for phonological processesapply to them before NCC repair, but processes applying in crossedstructures are constrained in two important ways by the special char-acteristics of these structures. First, geminate inalterability can preventrule application that would otherwise be expected. Second, constraintson feature spreading are particularly restrictive because multiple linkinggives multiple opportunities for intervention effects.

Transcription and rule application in crossed structures 17

2.2.1 Geminate inalterabilityThe effects of geminate inalterability in Biblical Hebrew (BH) are well-known and provide a good illustration of the issue. See Kenstowicz(1994:410) for a particularly clear discussion. There is a spirantizationprocess in BH which applies to postvocalic stops, but is blocked ongeminate stops. For example, yiktob → yixtov , with both postvocalicstops spirantizing, but the postvocalic geminate in sappir is unaffectedby spirantization (*saffir).

(13) a.

y

×i

×k

×t

×o

×b

×

↓x

↓v

b.

s

×a

× ×p

×i

×r

×

↓*f

In (13), the timing slots involved in spirantization are circled (with a dot-ted perimeter). An arrow points from the the timing slot whose vocalicassociation conditions application of spirantization to the timing slotwhose associated phoneme is the target of spirantization. Obviously,this notation is not part of the phonological representation. It is usedonly in this chapter and is intended to guide the reader through thecomplexities of rule application in crossed structures.

Based on the work of Hayes (1986) and Schein and Steriade (1986),Kenstowicz concludes that the crucial factor is that only one of thetiming slots associated with p in (13b) is in the proper environment forspirantization (i.e. postvocalic). Kenstowicz says:

. . . two outcomes are possible a priori. We might expect the ruleto “overapply” even though just one of the legs of the geminatesatisfies it. Alternatively, the rule might be suspended even thoughone portion of the multiply linked representation does satisfy it.

Kenstowicz does not use the term in this context, but we can say thatthe rule “underapplies” if the rule does not apply even though one legof the geminate does satisfy the condition for rule application.

It is useful to call a timing slot and an associated phoneme anoccurrence of the phoneme. The structural condition of a rule refers tooccurrences of phonemes, not directly to phonemes. This is what gives

18 Chapter 2

rise to an ambiguity and requires additional specification in order to de-termine how the rule applies to phonemes which are linked to multipletiming slots. Some phoneme altering rules require that all occurrencesof the phoneme satisfy the structural condition of the rule, others requireonly that at least one occurrence of the phoneme satisfies the structuralcondition. If there is an occurrence of a phoneme which satisfies thecondition for the application of some rule which has another occurrencewhich does not, then the rule will either overapply or underapply. Ruleswhich underapply in this situation are said to be subject to geminateinalterability.

Based on the facts above, spirantization in Biblical Hebrew requiresall the timing slots associated with the potential spirantizing stop to bepostvocalic (i.e. to follow a timing slot associated with a vowel). Thatis, spirantization in BH is subject to geminate inalterability. Kenstowiczpoints out that the way that spirantization applies to geminates in BH istypical: “Many languages spirantize postvocalic or intervocalic stops.In virtually every case, a geminate consonant resists spirantization.”In Southern Paiute spirantization applies after reduplication, producingpiNwa → piviNwa . There is nothing surprising here. First, piNwa →pipiNwa . Then after fission, spirantization applies.

(14) [

p

×i

×]

N

×w

×a

× → × ×p

×i

×N

×w

×a

× →p

×i

×p

×i

×N

×w

×a

×

↓v

Spirantization cannot apply before fission, because it is subject togeminate inalterability. It cannot apply in:

(15) × ×

p

×

i

×

N

×

w

×

a

×

One of the occurrences of p is postvocalic, but one is not. If the spiran-tization rule were ordered so that it applied in (15), it would underapply.It would fail to spirantize p even though one of its occurrences satisfiesthe condition for spirantization.

Transcription and rule application in crossed structures 19

Kiparsky (2003) calls attention to the Southern Paiute example andthe fact that there are no languages in which spirantization overappliesin similar examples. In the analysis developed here, this is no surprise.It simply reflects Kenstowicz’s observation that sprirantization is quitegenerally subject to geminate inalterability. Kiparsky points out thatthis result is not consistent with Correspondence Theory, which has nomechanism other than stipulation to keep spirantization from overapply-ing to the reduplicant. Note that in DR, the following are structurallyidentical in relevant respects:

(16) a.

s

×

a

×

p

× ×

i

×

r

×

↓*f

b. × ×

p

×

i

×

N

×

w

×

a

×

↓*v

It is worth noting that if rule ordering in Southern Paiute weredifferent and strict cyclicity were observed, there would be no spiranti-zation in pipiNwa. This would be the result if NCC repair were cyclic(as I believe that it always is) and spirantization applied cyclically aftertranscription but before NCC repair and did not apply post-cyclically.Geminate inalterability would block spirantization in the crossed struc-ture and, because of rule ordering and strict cyclicity, spirantizationwould have no opportunity to apply in the structure after NCC repair.

If progressive nasalization in Malay were subject to geminate in-alterability, then nasalization of the initial a would not occur in (17)because only one of the occurences of the initial a is in a nasalizingenvironment.

(17)

a

×

n

×

a

×

m

× × × × ×

However, because Malay progressive nasalization is not subject to gem-inate inalterability, a nasalizes in (17) because one of its occurrences isin a nasalizing environment, even though another of its occurrences isnot.

20 Chapter 2

2.2.2 Constraints on feature sharingIf the NCC is conceived of as a constraint which tolerates temporary vi-olation provided that repair follows (either immediately or eventually),it cannot block undesirable feature spreading. A feature could simplyspread over any arbitrary intervening feature and, if a violation of theNCC results, the resulting structure could be repaired by feature fission.Nevertheless, something like the NCC does appear to be at work inconstraining feature spreading, at least in cases where the interveningfeature has certain special properties. Since a precise understanding ofat least certain aspects of the constraints on feature spreading will benecessary in order to understand how these constraints operate in themultiply linked structures created by reduplication, we consider themhere.

In cases in which features of a certain kind block certain varietiesof feature spreading, the NCC is often held to be responsible. Halle(1995), based on work of Calabrese (1995), takes a nuanced positionwith respect to the application of the NCC in constraining featurespreading. In the case of a contrastive intervening feature, the NCCis taken to be a derivational constraint, blocking rule application.4 Inthe case of a noncontrastive intervening feature, the NCC can be freelyviolated, but violations are repaired. In effect, Halle takes the NCC tobe a derivational constraint with respect to contrastive features and aninterface condition with respect to noncontrastive features. He is notexplicit about the timing of NCC repair (eventual or immediate), sinceit is not relevant to his concerns.

Typically, spreading rules are stated in two parts. The first partspecifies the source, directionality, and target. This is given by state-ments like “spread [+back] to the left from vowels to consonants.” Thesecond part gives the locality conditions on the spreading. One way ofdoing this is to specify the phonemes which are opaque to the spreading.I will assume that the locality conditions on spreading are given byconstraints of the form:

Transcription and rule application in crossed structures 21

(18) * ×◦. . . × . . . ×

◦

[F]

α

if α is opaque to F spreading.

i.e. Sharing F over α is blocked (derivationally) if α is opaqueto F-spread.

The symbols ◦ represent root nodes. Timing slots are not associateddirectly with features, but indirectly via root nodes. There is controversyabout whether root nodes themselves are linked directly to segmentalfeatures or whether there are various intervening nodes. This dependson assumptions about feature geometry which are not directly relevantto the considerations here. I take the association to be direct, forreasons of simplicity of exposition. Nothing here will depend upon thisassumption. If a timing slot is associated with a root node which isassociated with a feature F, then I will say that the timing slot has thefeature F.

This gives a general framework for considering feature spreadingrules. Halle’s specific proposal was that a phoneme α is opaque to F-spreading if it has a contrastive feature on the same tier as F. In the caseof a noncontrastive feature on the same tier as F, spreading is allowed,but the resulting structure is later repaired.

It is convenient to have a verbal statement of (18).

(19) Feature sharing between timing slots is blocked if an interveningtiming slot is associated with a root node which is opaque to F-sharing.

It is implicit in (19) that the intervening timing slot is not also linked tothe shared feature. There is no intention to prevent a feature from beingshared by a string of consecutive timing slots.

Principle (19) is obviously related to the NCC, particularly if theopacity is caused by the presence of features on the same tier as theshared feature. But the relation between (19) and the NCC is indirect.The NCC, is an interface condition, a constraint on grammars, whilePrinciple (19) is a constraint on rules. Grammars can adopt variousstrategies for ensuring that they produce output representations which

22 Chapter 2

satisfy interface conditions. One option is to repair defective repre-sentations by operations like Fission. Another is to ensure that rulesdo not produce defective outputs by imposing derivational constraintslike (19). In this, Principle (19) is more like Fission than it is likethe NCC itself. It is one (of many) strategies that can be adopted forensuring that the NCC is satisfied at the interface. We will see later thatalthough Fission is the most common way that the NCC violations thatTranscription produces are repaired, there are other strategies in certaincases.

Now that a precise formulation of the form that constraints onfeature spreading take has been proposed, the analysis of Malay nasal-ization in reduplicated forms can be considered more carefully. I willassume that progressive nasalization spreads [+Nas] to the right, target-ing vowels and nasal consonants. Spreading a [+Nas] to consonantswhich are already [+Nas] has the consequence that a form like anem

will have a single [+Nas] feature, shared between the final three rootnodes, as shown below:

(20) ×◦

×◦

×◦

×◦

[−Nas] [+Nas]{a} {N} {a} {n}

Locality is not an issue in spreading [+Nas] in (21) below becauseall timing slots come to share a single [+Nas] feature.

(21) ×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

[−Nas] [+Nas]{a→ a } {N} {a} {n}

×

Transcription and rule application in crossed structures 23

There are somewhat more complex reduplicated forms with unex-pected nasalization which do raise the issue of locality. Onn (1976)gives the four examples in (22).

(22) root surface

a. hama ham@ ‘germ’ ham@ham@ ‘germs’b. waNi waNı ‘fragrant’ waNıwaNı ‘fragrant (intensified)’c. aNan aNan ‘reverie’ aNanaNan ‘ambition’d. aNin aNen ‘wind’ aNenaNen ‘unconfirmed news’

The analysis of (23d) is the same in relevant details as the analysis of(23c), which was given above. The analysis of (22a) and (22b) dependsupon the particularities of the locality of progressive nasalization inMalay. According to Onn (1976:69), glides (w, y, h, and P) aretransparent to nasal spread, but other consonants are not. He givesderivations like mewah → mewah and mayaN → mayaN to illustratethe transparency of glides to progressive nasalization. There are nounderlying nasal vowels in Malay.

The analysis of waNi→ waNıwaNı is given below:

(23) ×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

[−Nas] [−Nas] [+Nas]{w} {a→ a } {N} {ı}

×

Six timing slots come to share a single [+Nas] feature. The six timingslots are not consecutive. The span is broken up by a timing slotassociated with w. But since glides are transparent to nasal spreading,there is no locality violation.

Notably, in all of the examples provided by Onn in which there isunexpected vowel nasalization in the reduplicated form, there is a singlenasal span broken up at most by intervening glides. The theory outlinedhere predicts this. Unfortunately, Onn did not provide examples toshow that nasalization does not occur if a nonglide nonnasal consonantintervenes between the source of nasalization and the unexpectedly

24 Chapter 2

nasalized vowel. Kiparsky (2003) has tested such examples with anumber of native speakers of the dialect that Onn describes and findsno nasalization.

An analysis of one of Kiparsky’s examples in given in (24). Itshows why harum→ harum-harum (*harum-harum).

(24) ×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

[−Nas] [−Nas] [−Nas] [−Nas] [+Nas]{h} {a→ a } {r} {u} {m}

×

This operation is disallowed by the locality condition on spreading sincethe consonant r intervenes in the disconnected span of 4 timing slotswhich share a single nasal feature.

It should be noted that blocking nasal spreading in the complexstructure created by Transcription raises the possibility of underapplica-tion of nasalization. Why isn’t the result harum → harum-harum , withneither a nasalized at the surface? If nasalization did not apply againafter Transcription, this would be the result. Presumably, the observednasalization in harum→ harum-harum occurs after NCC repair.

Inkelas and Zoll (2004) point out that regressive velar assimilationat the reduplicant-stem boundary is common, but it never overapplies.For example, reduplication of the form kan → kaNkan is common, butkan → kaNkaN , with overapplication of place spreading, is unattested.This is expected under the analysis developed here, since kan→ kaNkaN

would require impossible place spreading in the unrepaired output ofTranscription. The medial vowel blocks the required place assimilation.

(25) ×

◦

×

◦

×

◦

×

◦

×

◦

×

◦

[Velar] [Coronal]{k} {a} n→ N

×

Transcription and rule application in crossed structures 25

In (25), 4 timing slots come to share a velar place feature. But twoinstances of a intervene. Place sharing across a is not possible. Other-wise, we would expect nak→ Nak . Place assimilation as shown in (25)is therefore impossible.

Inkelas and Zoll’s intention is to completely discredit the idea ofoverapplication in reduplication, since the theory of reduplication whichthey propose has no explanation for it. We have seen, however, thatoverapplication in the theory proposed here is highly restricted. Anumber of conditions must be met: 1) NCC repair must be delayed longenough for phonological processes to take place in the complex struc-ture produced by Transcription; 2) whatever phonological processes doapply to the unrepaired output of Transcription must not be subject togeminate inalterability; and 3) because of the peculiar structure of theunrepaired output of Transcription, feature spreading operations canapply in only fairly unique situations. It should be no surprise thatexamples of overapplication are rare. Several additional examples aregiven in Chapter 7. All of them involve feature changing operations,not spreading rules. Malay nasalization is the only example I know ofin which a spreading rule overapplies.

2.2.3 Korean consecutive reduplicationChung (1999) discusses various kinds of reduplication in Korean. Oneparticularly interesting variety (p. 170), which he calls “consecutivereduplication”, provides another good example of the kind of ruleoverapplication which we have already seen in Malay. It is restrictedto the strata of Sino-Korean words, but there are many examples. Bothmonosyllabic and bisyllabic roots reduplicate with intensive semantics.Reduplication of monosyllabic roots is unremarkable. The reduplica-tion of bisyllabic roots is much more interesting:

(26) root reduplicated

a. kikwe ki-ki-kwe-kwe ‘very strange’b. kimyo ki-ki-myo-myo ‘marvelous’c. hy@nsak hy@n-hy@n-sak-s’ak ‘all forms and colors’d. cason ca-ca-son-son ‘generation after generation’e. sipi si-si-pi-pi ‘judgment’f. sikak si-si-kak-k’ak ‘hourly’g. kuc@l ku-ku-c@l-c@l ‘every phrase and sentence’

26 Chapter 2

Each of the two syllables reduplicates independently. Since the ini-tial syllable is reduplicated (see (26c) in particular), this looks like acounterexample to Moravcik’s claim that there is no initial syllablereduplication. Unlike initial syllable reduplication, there are severalattested cases of final syllable reduplication. In Chapter 6, it will beshown how this asymmetry follows from the assumptions of DR. Whatappears to be initial syllable reduplication in (26) is reduplication of theremnant of final syllable reduplication. In Chapter 6, the relevant junc-ture insertion rules will be discussed in detail and derivation leading upto the multiply linked structure produced by Transcription, the lefthandside of (29) below, will be justified. Here I will concentrate on the waythat phonological rules apply in the multiply linked structures producedby transcription. The only phonology at work in the examples in (26) isconsonant tensing, which applies to a syllable initial consonant whichis preceded by an obstruent, as seen in (26c,f).

The phonological changes in the examples in (27) are much moreinteresting. The changes have been boldfaced.

(27) a. hi.lak hi-hi-naN-nak ‘rejoicing’b. yu.lak yu-yu-naN-nak ‘quite willingly’c. u.lyaN u-u-nyaN-nyaN ‘very lonely’d. lwe.lak nwe-rwe-naN-nak ‘broad-minded’

Four different phonological rules are involved in producing the bold-faced changes in (27).

(28) a. Onset l → n following a noncoronal consonant coda.b. Word initially, l deletes / high vowel, else l→ n .c. Onset l → r following an open syllable.d. k→ N / nasal (regressive nasal assimilation).

First, consider (27a). A naive copy theory of reduplication wouldincorrectly predict:

hilak→ hi-hi-lak-lak→ hi-hi- r aN-nak

This assumes that regressive nasalization can apply after onset l → n.For naive copy theory, the onset n in the first conjunct of the reduplicatedsecond syllable is a mystery. The context for l → n is present only

Transcription and rule application in crossed structures 27

in the second conjunct. This is a classic Wilbur Effect. From thepoint of view of naive copy theory, there is no motivation for l → n

in the first conjunct. It is worth noting, by the way, that simplyimposing a reduplicant-base identity condition is not sufficient, becauseit would incorrectly predict hi-hi-naN-naN, with overapplication of nasalassimilation.

The DR account parrallels the account of the overapplication ofnasalization in Malay. Suppose that l → n applies to the output ofTranscription, before crossing violations have been removed, as shownbelow:

(29)

σ

× ×σ

h

×

i

×σ

l

×

a

×

k

×σ

× × ×

l→n−→

σ

× ×σ

h

×

i

×σ

n

×

a

×

k

×σ

× × ×

Note that the operation in (29) is not spreading, so the locality ofspreading is not an issue. I assume that 1) NCC repair is cyclic; 2) onsetl → n following a noncoronal coda consonant is cyclic, not subjectto geminate inalterability, and ordered before NCC repair; and 3) wordinitial l→ r , onset l→ r following a vowel, and regressive nasalizationare all postcyclic. The derivation (29) then continues with cyclic NCCrepair and postcyclic regressive assimilation, producing hi-hi-naN-nak.

These considerations are sufficient to account for (27a,b,c). For(27d), we need to explain why the output is nwe-rwe-naN-nak and noteither nwe-nwe-naN-nak or rwe-rwe-naN-nak. The derivation beginslike (27):

28 Chapter 2

(30)

σ

× × ×σ

l

×

w

×

e

×σ

l

×

a

×

k

×σ

× × ×

l→n−→

σ

× × ×σ

l

×

w

×

e

×σ

n

×

a

×

k

×σ

× × ×

Neither word initial l → n nor post-vocalic l → r applies in the crossedstructure in (30) because both rules are postcyclic, hence follow cyclicNCC repair. The output of cyclic NCC repair is therefore:

(31)

σ

l

×w

×e

×σ

l

×w

×e

×σ

n

×a

×k

×σ

n

×a

×k

×

Post-cyclically, word initial l → n, post-vocalic l → r, and regressivenasalization all apply, producing nwe-rwe-naN-nak, with no overappli-cation of any of these three rules.

2.3 The Retraction Condition

Appendix 1 is devoted to a detailed discussion of the substance of theNCC. It would be a diversion to undertake that discussion here. Theappendix is provided for interested readers. The formulation of the NCCwhich the discussion leads to is given in (32), which is an interfacecondition. A segment is taken to mean an element on a phonemic (asopposed to prosodic) tier.

(32) Retraction Condition (RC): The set of timing slots which areassociated with a segment is connected.

In the remainder of this book, I will assume that the RC is an outputcondition on the phonology.

The RC is based on the proposal that autosegmental representationis the privilege of phonology, not available either in the lexicon orphonetics. Without autosegmental linking, relationships between tierscan only be expressed directly by temporal simultaneity. The RC

Transcription and rule application in crossed structures 29

guarantees that the output of phonology can be transparently translatedinto a phonetic representation. These representations are not autoseg-mental, but composed of a parallel tracks implicitly related by temporalsimultaneity rather than explicitly linked by autosegmental associations.

The RC is both weaker and stronger than the proposal that “asso-ciation lines do not cross.” The representation (33a), with the medialtiming slot unlinked, violates the RC but does not have crossed asso-ciation lines. The representation (33b), on the other hand, has crossedassociation lines, but does not violate the RC.

(33) a. ×

α

× × b. ×

α

×

β

c.

α

×

β

×

In fact, I assume that the phoneme tiers have no intrinsic order, so thatif α and β are phoneme segments the representations (33b) and (33c)are identical. This has various consequences, which seem to me to becorrect: there are no floating segments, and there are no junctures onphoneme tiers. Segments must be associated with the timing tier, andonly segments appear on phoneme tiers.

Since the phoneme tier is unordered, RC repair (which I willcall NCC Repair or NCCR to maintain the familiar terminology) isstraightforward phoneme fission:

(34) ×

a

×

n

×

e

×

m

× × × ×NCCR−→ ×

a

×

n

×

e

×

m

×

a

×

n

×

e

×

m

If the phoneme tiers were ordered, not only fission operations butmultiple metathesis operations would be necessary to carry out RCrepair. Since the phoneme tiers are unordered, the righthand side of(34) is equivalent to:

(35)

a

×n

×e

×m

×a

×n

×e

×m

×

Chapter 3

The morphophonologyof reduplication

Reduplication is one of the ways in which the presence of a morphemein the structure of a word is made manifest. The aim of this sectionis to sketch the theory of morphology that I assume in sufficient detailso that it is clear how reduplication fits into the general theory of wordformation. The major tenants of Distributed Morphology (Halle andMarantz, 1993) are assumed: late lexical insertion and the propositionthat complex word formation, at its core, is lexical insertion into theterminal nodes of a hierarchically organized structure produced by thesyntax.

A morpheme is a bundle of morphological features, without phono-logical content. Morphophonology converts hierarchical structureswhose terminal nodes are morphemes (which I will call m-structures)into morphophonological representations at the interface with postlex-ical phonology. The conversion is derivational, beginning with theinitial m-structure which lexical insertion has not yet provided withphonological content. For the sake of simplicity, I will assume herethat one of the sisters of every node in an m-structure is a terminal nodeand that the computation can determine the root node in some way. Theconversion is carried out recursively, starting at the root morpheme andworking up the tree.

Vocabulary choice at each terminal node determines a lexical itemwhich is inserted into the structure at that node. I ignore here thequestion of how the choice is made. (See Halle and Marantz for the

The morphophonology of reduplication 31

theory.) A phonological string, perhaps null, is inserted at each stepand phonological rules (usually called “the cycle”) apply before thecomputation moves up to the next higher node. After all of the nodeshave been given phonological content in this way, further phonologicalrules (the “postcyclic rules”) apply and the derivation terminates.

A simple example will clarify the core process. Suppose that theinitial m-structure is ((α β) γ), where the root is α. M-structures arehierarchically structured, but not linearly ordered. Linear order, to theextent that it is relevant to m-structures, is induced by the order of thephonological material which they are associated with. The initial m-structure could just as well have been written (γ (β α)) or either of twoother possibilities. Suppose, for example, that α is realized by a rootwith exponent nu, β is realized by a prefix with exponent r-, and γ isrealized by a suffix with exponent -d. Assuming for simplicity that nophonological rules apply, the derivation that results is:

(36) α βγ →

n

×u

×α β

γ →

r

×e

×n

×u

×β α

γ

→

r

×e

×n

×u

×d

×β α γ

The derivation (36) consists of a sequence of lexical insertion opera-tions. The point of this elementary example is to illustrate the recursiveaddition of the exponents of lexical items to the structure and the wayin which the morphemes progressively acquire an induced linear orderbased on the types of lexical items which are used to realize them.

A lexical item can have a null exponent. Such a lexical item will becalled a null lexical item. If β in (36) is realized by a null lexical item,

32 Chapter 3

the derivation would be (37), still assuming that no phonological rulesapply, only lexical insertion.

(37) α βγ LexIns−→

n

×u

×α β

γ LexIns−→

n

×u

×α β

∅

γ

LexIns−→

n

×u

×d

×α β

∅

γ

In (37), β is marked as having been realized by a null morpheme bytagging it with ∅. Something equivalent to this must be available to themorphology so that the fact that insertion has already applied to β isrecorded. Obviously, if β is associated with nonnull phonology, no suchmarking is needed.

Note that the morpheme β does not acquire a linear order withrespect to α in (37). The final representation could just as well havebeen written:

(38)

n

×u

×d

×αβ

∅γ

In (37), the morpheme which is realized by a null lexical item hasno effect on the derivation. Often, however, null lexical items do affectthe derivation. The insertion of a lexical item into the structure can trig-ger the application of morphophonological rules, even if the exponentof the lexical item is null. Furthermore, rules can be sensitive to thepresence of morpheme structure, independently of how this structure isrealized, so that the presence of a morpheme can affect the derivationindependently of the phonological material which it is associated with.

The morphophonology of reduplication 33

There are two blocks of morphophonological rules: the cyclic ruleblock and the postcyclic rule block. By “rule block” I simply mean arule structure of some kind which generates an ordered list of rules.In addition to the rules of the cyclic rule block, there can be rule blocksassociated with lexical items (often consisting of a single rule) which aretriggered by the insertion of the lexical item. After each lexical insertionoperation, the rule block associated with that lexical item applies (ifthere is one) and then the cyclic block of phonological rules applies(unless the lexical item is specifically marked as “non-cyclic”).1 Halle(1990) calls the rules triggered by lexical insertion readjustment rules.See Halle and Marantz (1993) for discussion.

English past tense morphology provides good examples of read-justment rules. First, consider the derivation of sold, the past tense ofsell. In (39) and various other places in this chapter, the initial lexicalinsertion step in which the root is realized is omitted for reasons ofspace.

(39)

s

×E

×l

×[[SELL]] PAST

LexIns−→

s

×E

×l

×d

×[[SELL]] PAST

Readj−→

s

×o

×l

×d

×[[SELL]] PAST

The last operation is the result of a readjustment rule which backs theroot vowel of a class of verb roots. [[SELL]] above should be understoodas an abstract symbol representing a morpheme, which is a bundle ofmorphological features. Morphemes are without phonological content.This will be clearer after allomorphy is considered below.

Realization of PAST in (39) has two aspects, the insertion of anexponent and the application of the readjustment rule which this inser-tion triggers. Morpheme realization in this case has both a concate-native aspect (concatenation of the suffix -d) and a nonconcatenativeaspect (application of the vowel ablaut rule). In the next example, thederivation sang, the past tense of sing, morpheme realization is entirelynonconcatenative.

34 Chapter 3

(40)

s

×I

×N

×[[SING]] PAST

LexIns−→

s

×I

×N

×[[SING]] PAST

∅

Readj−→

s

×a

×N

×[[SING]] PAST

∅

The default past tense suffix is chosen for insertion in PAST in (39),and the null past tense suffix is chosen for insertion in (38). Here, andin what follows, I will use the idiom “inserts a null suffix” to meanthat no suffix is inserted, but that the computation records the fact thatlexical insertion for that morpheme has already been accomplished andthe computation is at the next step, carrying out whatever lexicallytriggered rules are associated with the null lexical item that was chosenfor insertion.

The existence of lexically triggered rules does not exhaust thepossibilities of interaction between the morphological structure of aword and phonology. Cyclic, postcyclic, and lexically triggered rulesthemselves can all be morphologically conditioned. That is, morphemestructure can figure in the structural conditions of applicability of therule.

The two instances of vowel ablaut readjustment in (38) and (39)should be contrasted with allomorphy, in which two different lexicalitems are used to realize the same morpheme, depending upon thecontext that the morpheme finds itself in. [[GO]], for example, is realizedby the root wend in the context of Past, otherwise by the root go. In(38) and (39), there is a single lexical item whose exponent undergoes aminor rule governed modification in certain contexts, but in (41), thelexical item which is chosen for insertion in [[GO]] depends on thepresence or absence of a PAST morpheme.

(41) a. [[GO]] PASTLexIns−→

w

×e

×n

×d

×[[GO]] PAST

LexIns−→

w

×e

×n

×d

×t

×[[GO]] PAST

The morphophonology of reduplication 35

Readj−→

w

×e

×n

×t

×[[GO]] PAST

b. [[GO]] {3,sg}LexIns−→

g

×o

×[[GO]] {3,sg}

LexIns−→

g

×o

×z

×[[GO]] {3,sg}

There is some room for debate, but wend is treated here as an irregularverb of the send-class (send/sent, bend/bent, lend/lent, wend/went).

The sIN/saN alternation and go/wEnd alternations are often lumpedtogether under the heading “allophonic variation.” This leads to confu-sion, since the source and character of the variation in the two cases isvery different. I will avoid use of the the term “allophonic”. Alterna-tions in the choice of lexical item which realizes a particular morphemewill be called allomorphy. The alternation between go and wend asrealizations of [[SING]] is allomorphic. So is the alternation in choice ofthe past tense affix (-t, -d, or null).

With this understanding of the architecture of the morphophono-logical computation, we are ready to consider the place of reduplicationin this architecture.

3.1 Reduplicative affixation is lexically triggered readjustment

An affix which induces reduplication does so by triggering rules whichinsert t-junctures into the timing tier. Later phonological rules, sensitiveto the presence of these junctures, carry out transcription, morphemeassociation, and NCC repair. Hayes and Abad (1989:357) give a goodexample from Ilocano, the paradigm (42). According to them, what theycall “light reduplication” is used with the prefix si- to mean ‘coveredwith, filled with’:

36 Chapter 3

(42)a. buneN (type of knife) si-bu-buneN ‘carry a buneng’b. pandiliN ‘skirt’ si-pa-pandiliN ‘wearing a skirt’c. liNPet ‘perspiration’ si-li-liNPet

‘covered with perspiration’d. roPot ‘leaves, litter’ si-ro-roPot ‘covered with litter’e. yaket ‘jacket’ si-ya-yaket ‘wearing a jacket’

The m-structure is of the form (N COV), with the morpheme COVsome kind of prepositional element which combines with a noun root.The lexical item which realizes COV has a prefixal exponent si- andits insertion triggers t-juncture insertion as shown in the illustrativederivation of (42d), which is given below.

(43) [[ROPOT]] COVLexIns−→

r

×o

×P

×o

×t

×

[[ROPOT]] COVLexIns−→

s

×i

×r

×o

×P

×o

×t

×

COV [[ROPOT]]

Readj−→

s

×i

×[

r

×o

×]

P

×o

×t

×

COV [[ROPOT]]

When the lexical item which realizes COV is inserted, rules which insertt-junctures into the verb root are triggered. We will return later toexamine the structural description of such rules in detail. The presentdiscussion is meant only to examine the place of these rules in themorphophonological derivation.

Transcription (Trscr below) follows at some point.

(44)Trscr−→

s

×i

× × ×r

×o

×P

×o

×t

×

COV [[ROPOT]]

The morphophonology of reduplication 37

Transcription in (44) is to the left, at the morpheme edge. Righttranscription would produce a morpheme internal reduplicant.

The copied timing slots in the terminal representation in (44) arenot associated with morphemes. It could be that they remain withoutmorphemic association. The issue is far from clear, but I will tentativelyassume that this violates basic tendencies of morphophonology and thatrepair is triggered which associates the timing slots of the reduplicantwith a morpheme. There are two obvious possibilities, associationwith the verb root and association with the prepositional morpheme.The repair takes place in the COV-cycle, so it is reasonable to assumethat association with the morpheme which generates the cycle. Thisproposal is also tentative. It could be that there is language particularvariation, so that a similar process in another language might take thematerial added by transcription to be a root extension, rather than part ofthe prefix. If repair does proceed as tentatively suggested, the derivation(44) continues to:

(45)

morphemeassociation−→

s

×i

× × ×r

×o

×P

×o

×t

×

COV [[ROPOT]]Fission−→

s

×i

×r

×o

×r

×o

×P

×o

×t

×

COV [[ROPOT]]

Juncture insertions rules will be examined in great detail in comingsections. The description below of the Ilocano rule above is a startingpoint.

(46) 1. ∅ → ]/

V (leftmost in stem)

2. ∅ → [/

× (leftmost in stem)

The term “stem” here and throughout has only a structural meaning. Atthe point of lexical insertion, the stem is the realization of the sister ofthe morpheme which is currently being realized. In (46.1), “V” is takento be a predicate on timing slots, true if the timing slot is associatedwith a vowel. (This will later be extended so that the predicate is trueof nuclei in general, not just vocalic nuclei.) The structural condition in(46.1) therefore specifies the insertion of a ]-juncture into the timing tier

38 Chapter 3

following a timing slot associated with a vowel. The leftmost instanceof this configuration is chosen as the insertion site.

It will prove to be useful to write the rules (46) in a bipartite form,with the rules on the left and the domain to which they apply on theright.

(47) ∅ → ]/Left V , ∅ → [

/Left × ; stem

In the interest of a compact notation, leftmost (or rightmost) applicationis specified by a subscript on the “/ ” symbol which introduces theenvironment for rule application. It will become clear in Chapter 5that the variation in the rules which reduplicative affixes trigger is betteraccounted for if a domain is factored out and the rules written relative tothe domain, as in (47). The domain is not always the stem. The domainof some reduplicative affixes is a morphological or prosodic subword ofthe stem. The general format is:

(48) juncture insertion rules ; rule domain

Some simplification of (47) is possible on the basis of severalnatural default choices. First, rule application will be assumed to beleftmost, as the default. Second, the domain of juncture insertion rulesis taken to the whole stem, as the default. Unpaired [- and ]-junctureswhich remain after juncture insertion are closed by inserting ] or [ at theedges of the stem. This is called Default Closure (DC).

In view of default leftmost juncture insertion rule application andDefault Closure, (46) simplifies to:

(49) ∅ →/

V ; stem

This rule occurs so commonly that it is useful to have a name for it:C∗V juncture insertion. The Ilocano r-affix above can then be describedsimply by saying that the exponent of COV is prefixal si- and that itsconcatenation triggers stem C∗V juncture insertion. Derivations are not

The morphophonology of reduplication 39

usually shown with the detail that was lavished on (43) through (45)above. Generally, only a sketch will given, as in (50a) or (50b). So:

(50) a. roPotprefix−→ siroPot

JncIns−→ siro]Pot

DC→ si[ro]Pot

Trscr−→ si-ro-roPot

b. roPot⊕COV−→ si[ro]Pot

Trscr−→ si-ro-roPot

The hyphens in the final representations should be understood as aidsto the reader, not as phonological objects which are present in themorphophonological representation. Such hyphens will often be givento make it easier for the reader to recover the underlying process.

3.2 Klamath distributive and intensive reduplication

Klamath distributive and intensive reduplication and their interactionillustrate many of the ideas above. (The examples in this section are alltaken from Zoll, 2002.)

(51) Klamath distributive reduplication

root reduplicated surface

a. qlin ‘choke’ qli-qlin qli-ql@nb. pag-a ‘bark’ pa-pag-a pa-pg-a

Klamath has a rule (Root Vowel Reduction, RVR) which reduces theinitial vowel of the root if that vowel is not the initial vowel of the wordthat the root is embedded in. Typically, prefixes cause RVR to apply.The surface forms in (51) are the result of RVR applied to a reduplicatedform. The vowel deletes in open syllables, as in (51b), and reduces toschwa in closed syllables, as in (51a).

The lexical item which realizes the distributive morpheme andinduces reduplication has a null exponent and triggers C∗V-junctureinsertion. The derivation of (51a) is given in (52), with the distributivemorpheme denoted by DIS. T&R (transcription and repair) denotes the

40 Chapter 3

compound operation consisting of Transcription, morpheme associa-tion, and NCC repair (NCCR).

(52) [[QLIN]] DIS⊕ [[QLIN]]−→

q

×l

×i

×n

×

[[QLIN]] DIS⊕DIS−→

[

q

×l

×i

×]

n

×

[[QLIN]] DIS∅

T&R−→

q

×l

×i

×q

×l

×i

×n

×

[[QLIN]]DISRVR−→

q

×l

×i

×q

×l

×@

×n

×

[[QLIN]]DIS

RVR is a morphologically conditioned rule. It must be sensitive to themorpheme structure so that the initial vowel of the root can be located.This is a typical example of a morphologically conditioned rule. It isa phonological rule in form, but it is neither purely phonological nora readjustment rule (i.e. directly triggered by the insertion of a lexicalitem).

It is worth reviewing the individual steps in the compound opera-tion T&R.

(53)

[

q

×l

×i

× ]

n

×[[QLIN]] DIS

Trscr−→× × ×

q

×

l

×

i

×

n

×[[QLIN]] DIS

morphemeassociation−→

× × ×

q

×

l

×

i

×

n

×[[QLIN]]DIS

phonemefission−→

q

×l

×i

×q

×l

×i

×n

×[[QLIN]]DIS

The morphophonology of reduplication 41

It is valid in this example to consider “transcription and repair” to bea compound operation. This is not always the case. Nothing in thetheory being developed here prevents other operations from interveningbetween the various suboperations of the compound operation. Theintervention of phonological rules between transcription and phonemefission is the source of various effects that have been termed “over-application” in the literature. If other rules do intervene between thesuboperations of the compound operation, it is no longer useful or validto think of it as a compound operation.

It is crucial that RVR applies after fission in the derivation (52).Before transcription, the context for root vowel reduction is not present.If RVR applied after transcription, but before fission, the result in (51a),for example, would be ql@ql@n or qliqlin. If RVR were subject togeminate inalterability, underapplication would result since only one ofthe occurrences of the vowel is a root vowel, so RVR would not apply. IfRVR were not subject to geminate inalterability, would result becausethe reduplicant vowel is simultaneously the root vowel before fission,so ql@ql@n would be produced. It is also crucial that transcription copiestiming slots to the left, not the right. If transcription were to the right,the condition for application of RVR would not be met since the rootvowel would remain the initial vowel of the word.

Transcription to the left in this example is not an arbitrary stipula-tion. As noted earlier, the default direction of transcription of a dupli-cant which is at the left edge of a morpheme and not simultaneously atits right edge, is to the left. The reason is straightforward. Transcriptionto the right would produce a the complex structure in (54), with thecopied material intruding into the morpheme. This is probably notexcluded by general principles and may in fact occur in some cases. Butall other things being equal, simplicity favors maintaining morphemecontiguity.

(54)

q

×

l

×

i

× × × ×

n

×

[[QLIN]] DIS

42 Chapter 3

Klamath has another reduplicative affix which realizes a morphemewhich has intensive semantics (denoted by INT below). A few examplesare given below.

(55) Klamath intensive reduplication

root reduplicated surfacea. Wit ‘flop (as a fish)’ Wit-Wit Wit’-Wit’b. kesp ‘pant for breath’ kesp-kesp kesp-kesp’

W, according to Barker (1964), Zoll’s source, is a voiceless sonorant.The changes t→ t’ and p→ p’ are late glottalization processes, unrelatedto reduplication. Two things stand out in (55): 1) the entire root isdoubled; and 2) there is no vowel reduction. We will see that thesethings are related.

The lexical item which realizes INT has a null exponent and triggersthe readjustment rule:

(56) ∅ → [/

× ; stem

(Recall that there is default leftmost application and default closture.)The morphology of (55a) is straightforward:

(57) INT[[WIC]]⊕ [[WIC]]−→

W

×i

×t’

×

[[WIC]] INT⊕ INT−→

[

W

×i

×t’

×]

[[WIC]] INT∅

In distributive reduplication, discussed above, transcription wasto the left in order to avoid splitting the root. Total reduplicationdoes not dictate a direction of transcription, since either left or righttranscription is to the morpheme boundary. In Klamath, the absence ofvowel reduction in the totally reduplicated forms in (55) indicate thatcopying is to the right.

The morphophonology of reduplication 43

(58)Trscr−→

W

×

i

×

t’

× × × ×

[[WIC]] INT∅

repair−→

W

×i

×t’

×W

×i

×t’

×

[[WIC]] INT

RVR does not apply because the root vowel remains word initial.Zoll observes that the vowel of the first conjunct of the intensive

form does reduce when it is not word initial.2 She gives variousexamples, of which the following is particularly interesting because itcombines intensive and distributive reduplication.

(59) Wic ‘be stiff’Wic-Wic-l’i ‘stiff’Wi-W@c-Wic-l’i ‘stiff-DISTRIBUTIVE’

The derivation of the distributive intensive form is given in somedetail below, simplified in the interests of brevity by ignoring the -l’i

suffix. The affix which realizes INT is assumed to be cyclic. I know ofno cases in which there is a strong argument that a reduplicative affix isnot cyclic (in the framework discussed in this paper) and will thereforeassume throughout this paper that reduplicative affixes are cyclic. Thereare three cycles of lexical insertion.

44 Chapter 3

(60)[[WIC]] INT

DIS

⊕ [[WIC]]−→

W

×i

×c

×[[WIC]] INT

DIS

⊕ INT−→[

W

×i

×c

×]

[[WIC]] INT∅

DIS T&R−→

W

×i

×c

×W

×i

×c

×[[WIC]] INT

DIS

⊕DIS−→[

W

×i

×]

c

×W

×i

×c

×[[WIC]] INT

DIS∅

T&R−→

W

×i

×W

×i

×c

×W

×i

×c

×[[WIC]] INTDIS

RVR−→

W

×i

×W

×@

×c

×W

×i

×c

×[[WIC]] INTDIS

An alternative approach to the analysis in (60) is given in a footnote.3

3.3 Tagalog (apparent overapplication)

Tagalog has a nominalizing prefix which interacts with a phonologicalprocess of nasal assimilation and an intensive reduplicative affix in aninteresting way. Lieber (1992:179) gives the following examples whichillustrate the interaction of paN-prefixation and nasal coalesence.

(61) noun prefixed noun

a. atip ‘roofing’ paN-atip ‘that used for roofing’b. pu:tul ‘cut’ pa-mu:tul ‘that used for cutting’

The nasal coda of the prefix combines with the stop onset of the root in(61b) to produce a single phoneme which has the nasality of the codaand the place of articulation of the onset. I assume that nasality firstspreads to the right from N, then the timing slot associated with N deletes:

paN-pu:tul→ paN-mu:tul→ pa-mu:tul

The morphophonology of reduplication 45

Tagalog has an intensive affix which has a null exponent and trig-gers C∗V-reduplication. paN-prefixation can combine with the intensiveaffix. The interaction that has provoked much discussion in the literatureis exemplified below:

(62) pa-mu-mu:tal ‘a cutting in quantity’

If it assumed that C∗V-reduplication applies to the root before paN-prefixation, (62) is mysterious. The attempted derivation would be:

(63) pu:tal rootpu-pu:tal C∗V-reduplication (before prefixation)paN-pu-pu:tal paN-prefixationpa-mu-pu:tal nasal coalesence

p→ m for the root initial consonant is unexplained. This has sometimesbeen interpreted as a Wilbur Identity Effect.

Aronoff (1988) and Lieber (1992) realized that (62) presents a puz-zle only if it is assumed that prefixation follows reduplication. Underthe assumption that the reduplicative affix combines with the stem afterprefixation, the only problem is to explain the mechanism by whichreduplication can reach inside the stem and double the initial C∗V ofthe root. Aronoff takes an approach which permits reduplication to“see” at least certain aspects of the morphological structure of the stem.Lieber rejects this and assumes that the prosodic structure of the stem issufficient to locate the material which is to be doubled.

Many examples in the coming pages will show that juncture inser-tion can be sensitive to either the morphological and prosodic structureof the stem it combines with. Among reduplicative affixes whosejuncture insertion rules make reference to morphological structure,sensitivity to an embedded root is very common. Since sensitivityto root boundaries is common, the simplest analysis in the presentframework is to suppose that intensive reduplication in Tagalog is C∗Vroot reduplication. It is generated by the readjustment rule:

(64) ∅ → ]/

V ; root

This is identical to (47) except that the domain of the rules is the root,not the stem.

46 Chapter 3

The derivation of (63) follows, beginning at the point just after theprefix has been inserted. NOM is the nominalizing morpheme and INTis the intensive morpheme.

(65)

p

×a

×N

×p

×u

× ×t

×u

×l

×[[PUUTUL]]NOM

INTNasal

Coalescence−→

p

×a

×m

×u

× ×t

×u

×l

×[[PUUTUL]]NOM

INT

⊕ INT−→

p

×a

×[

m

×u

× ×]t

×u

×l

×[[PUUTUL]]NOM

INT∅ T&R−→

p

×u

×m

×u

×m

×u

× ×t

×u

×l

×NOM• [[PUUTUL]]•

INT•

Note that DC is also sensitive to the domain of the affix.The final representation in (65) is virtually the same as the repre-

sentation that Lieber’s analysis produces. The derivation, however, isentirely different. In her theory, the intensive affix is a C∗V-template,which is infixed. In (65), the intensive affix has a null exponent. In thefinal representation, INT has acquired associated phonological material,but this association is derived. INT itself is purely nonconcatenative. Inthe final representation in (65), mu could be accurately described as a“derived infix” or a “surface infix.”

3.4 Erromangan:The interaction of reduplication and readjustment

The Oceanic Austronesian language Erromangan is the subject of anexcellent study by Crowley (1998). Verbal inflectional morphologyis particularly interesting since there is an extensive system of rootreadjustment. Since there is also total verb root reduplication associ-ated with intensive semantics, the interaction of root readjustment andreduplication can be examined.

The morphophonology of reduplication 47

Tense and agreement morphology is largely prefixal. In the exam-ples below, IMP is imperative and DISTPAST is the “distant past” tense.The orthography is Erromangan “practical orthography.” It should poseno problem, except that g represents IPA N.

(66) w-aruvo2PL:IMP-sing‘you all sing’

y-epm-aruvo3SG:DISTPAST-PRIOR-sing‘he/she had sung’

y-etu-velom3SG:DISTPAST-NEG-come‘he/she did not come’

y-em-aruvo3SG:DISTPAST-EM-sing‘(while) he/she was singing’

The prefix em- does not appear to be associated with independentsemantics and plays only a formal role in the morphology. Certainprefixes require its presence. Crowley glosses it as “EM”.

There is another morphological element (which I will gloss as“AN”), which is present in certain environments and whose exponentappears immediately before the root. It is realized by the prefix n- formost roots, but as an- for a large class of roots which Crowley callsstrong verbs. The AN-prefix/root combination is called the modifiedroot.

(67) co-n-aruvo3SG:FUT-AN-sing‘he/she will sing’

c-am-n-aruvo3SG-PRES-AN-sing‘he/she is singing’

c-an-vag > campag3SG:FUT-AN-eat‘he/she will eat’

Both allomorphs of AN trigger readjustment. an- triggers deletionof a root initial vowel, if there is one. The nasal n of both prefixes deletesbefore nasals (m, n, g), glides (y,w), the voiceless obstruents s and hand the lateral l. The readjustment rules are ordered, and V-deletionfeeds n-deletion. Various examples of the derivation of modified rootforms is given in (68). Various secondary morphophonological rulesapply after readjustment to derive the modified root. They are respon-sible for np → mp , ntn → tn , and nr → nd below. To some extent,these secondary rules appear to be morphological conditioned (applying

48 Chapter 3

in some, but not all, morphological environments). Since the preciseconditions are not important to what follows, discussion is omitted.

(68) Modified root formationroot initialV deletion

n deletion modifiedroot

an-pat ampat (< anpat) ‘block’an-mah a�n-mah amah ‘die’an-oruc an-�oruc anduc (< anruc) ‘bathe’an-etni an-�etni atni (< antni) ‘cook’an-elwo an-�elwo a�n-�elwo alwo ‘vomit’an-omol an-�omol a�n-�omol amol ‘fall’

Crowley takes a different approach to the formation of the modifiedroot. He analyzes it as pure root mutation, with no prefix. The additionof fixed phonetic material to the root suggests, however, that the expo-nent of some morpheme is involved in modified root formation. Theaddition of the nasal n, subject to the readjustment rules discussed, anda for strong verbs does not have the character of simple root mutation.It is therefore preferable to take the added phonetic material to be theexponent of a morpheme, rather than material added by morphophono-logical rule. Crowley’s main argument against the presence of a prefix isthe absence of a semantic content to the morpheme. Since he considersem- to be prefixal, and not realizing a morpheme with semantic content,this argument does not have much force.

Having sketched the application of readjustment in deriving certainverb forms, we are now in a position to examine its interaction withreduplication. Erromangan realizes an intensive morpheme via totalroot reduplication. Crowley notes that in intensive forms which call forthe modified root, only one of the reduplicative conjuncts is modified.The 3PL:FUT intensive form of omol ‘fall’, which has the modifiedroot amol, is cw-amol-omol (*cw-amol-amol). Under the theory ofmorphophonology adopted in this paper, which incorporates Halle’sidea of readjustment rules triggered by lexical insertion, this is expectedif the morpheme which induces intensive reduplication is closest to theroot, a plausible assumption. The derivation is given below:

The morphophonology of reduplication 49

(69) ∅ [[OMOL]] ⊕ INT ⊕ AN ⊕ 3SG:FUT

omol INT ⊕ AN ⊕ 3SG:FUT

omolomol AN ⊕ 3SG:FUT

anomolomol 3SG:FUT

a�n�omolomol 3SG:FUT (readjustment)cwamolomol ∅

Various theories of morphology (Anderson, 1992, for example)do not recognize readjustment rules as part of the architecture ofmorphophonology. Past tense sold, for example, is analyzed as theselection of the sol allomorph of sell from the lexicon, coupled withconcatenation of the -d suffix. Such theories are criticized by Halle andMarantz (1993) for failing to make a qualitative distinction between al-ternations of the sell/sold type and alternations of the go/went type. Theinteraction of intensive reduplication and readjustment in Erromanganprovides another powerful argument against “allomorphy only” theoriesof morphology. If there is only allomorphy, either the allomorph omolmust be chosen, incorrectly yielding *cwomolomol, or the allomorphamol must be chosen, incorrectly yielding *swamolamol.

Above I argued that the modified root should be analyzed as re-sulting from prefixation plus readjustment. It is worth noting that theincompatibility of the Erromangan data with “allomorphy only” theo-ries of morphology is independent of the claim that the modified rootcontains prefixal material. If Crowley’s proposal that the modified rootis formed by pure root mutation had been accepted, the incompatibilitywould be equally clear.

3.5 Nuu-chah-nulth reduplicative readjustment

This book takes the position that reduplication is the consequence ofjuncture insertion rules which are triggered by lexical insertion. Lexicalinsertion, in general, has two aspects; the concatenation of the exponentof the lexical item and the application of the readjustment rules whichthe lexical item triggers. The fact that reduplicative readjustment isaccompanied by lexical insertion is often obscured by the fact that theexponent is null, so that only nonconcatenative morphology is induced.

Nuu-chah-nulth, a Wakashan language spoken on Vancouver Island(off the Southwestern coast of Canada), provides many examples ofmorphemes which have both robust nonnull affixal morphology and

50 Chapter 3

simultaneous robust reduplicative morphology. Kim (2002) identifies 8different classes of verb suffixes of this type in Nuu-chah-nulth. Suffixesin each affix class have a nonnull exponent, but also induce a pattern ofroot reduplication characteristic of the suffix class. The habitual suffix(Class II in Kim’s classification), for example, has the exponent -Pi:k

and induces the following pattern of reduplication:

(70) a. c’usc + Pi:k → c’u:-c’usc-Pi:ksuspicious habitual

b. Pu:wa + Pi:k → Pu:-Pu:wa-Pi:kit-to-say habitual

A full list of the 8 suffix classes is given below, along with anexample suffix from each class and examples of its reduplicative effecton typical verb roots. See Kim for glosses and further details. In eachcase one verb root whose initial vowel is short and one verb root whoseinitial vowel is long is given. For classes I-1 and II-2, the length ofthe reduplicant vowel mirrors the vowel length of the initial vowel ofthe underlying verb root. For the other suffix classes, the length of thereduplicant vowel is determined by the suffix class, not the underlyingverb root. For the last four suffix classes, the length of the initial vowelof the verb root is adjusted to a fixed length characteristic of the suffixclass.

(71) Nuu-chah-nulth reduplicative suffix classes

Class I-1 (e.g. -ń∼a): C1V1C2 /C1V1:C2 reduplicant.yacmil∼ → yac-yacmil∼-ń∼a Pu:s → Pu:s-Pu:s-ń∼a

Class 1-2 (e.g. -Pal∼uk): C1V1 /C1V1: reduplicant.capx. → ca-capx. -Pal∼uk Pu:kw → Pu:-Pu:kw-Pal∼uk

Class II (e.g. -Pi:k): C1V1: reduplicant.c’us → c’u:-c’us-Pi:k Pu:wa → Pu:-Pu:wa-Pi:k

Class III (e.g. -yukw): C1V1 reduplicant.qwi → qwi-qwi-yukw Pu:s → Pu-Pu:s-yukw

Class IV (e.g. -a): C1V1:C2 reduplicant, long surface root vowel.cuc → cu:c-cu:c-a t’i:ck → t’i:c-t’i:ck-a

Class V (e.g. -k’uk): C1V1 reduplicant, short surface root vowel.ń∼ ’ic → ń∼ ’i-ń∼ ’ic-k’uk ń∼ ’i:xw(a)a → ń∼ ’i-ń∼ ’ixwa-k’uk

The morphophonology of reduplication 51

Class VI (e.g. -sis): C1V1: reduplicant, short surface root vowel.wikity’ak → wi:-wikity’ak-sis si:city’ak → si:-sicity’ak-sis

Class VII (e.g. -Pis): C1V1 reduplicant, long surface root vowel.wiksapi → wi-wi:ksapi-Pis Pu:ssapi → Pu-Pu:ssapi-Pis

The theory necessary to account for precisely how these patternsare generated will be developed in later chapters. See in particular thediscussion of Tohono O’odham reduplication in Section 7.5.

3.6 Some closing remarks

1. Showing the full morpheme structure is a typographic burden thatwe have borne in this section in order to emphasize the cyclic mor-phophonology. But it will be impossible to maintain this level ofmorphological detail in the coming sections. In general, the morphemestructure will not be represented in the sections that follow. The readershould keep in mind that this structure is always present and that themorphophonology, unsurprisingly, routinely makes reference to themorpheme structure.

2. UG must constrain the possible transformations of the stem which af-fixes can induce. Concatenation of an exponent is inescapable. Since itappeared from McCarthy’s work on Semitic verb morphology that tem-platic operations had to be admitted into morphophonology, Marantzattempted (with considerable success) to analyze reduplication withoutexpanding the power of affixes beyond concatenation and templaticmorphology. The cost was an expansion of the array of possibletemplates and possibilities for associating phonemes with the template.Marantz needed to introduce CV-templates, whole morpheme tem-plates, and others, as well as propose new varieties of association withtemplates. DR proposes a new kind of operation on stems, the insertionof duplication junctures (soon to be joined by truncation junctures).Aside from questions of empirical adequacy, it is fair to question thewisdom of expanding the powers of affixal stem modification. Thecost is the introduction of a new stem modification mechanism. Thebenefit, as I hope to demonstrate, is a conceptually and empiricallysatisfactory theory of reduplication. It has the added benefit of freeingtemplatic morphology from the burden of accounting for reduplication,so that a much more highly constrained templatic theory is possible.4

52 Chapter 3

Constraining the variety of possible reduplicative affixes will receiveongoing attention in the sections which follow. I will show that entirelyplausible restrictions on the array of possible juncture insertion rulesyields a theory which is empirically adequate, on the one hand, andadequately constrained, on the other.

Chapter 4

Truncated reduplication

We first consider string copy as a general process. One way thecomputation might be organized is illustrated by the successive stages ofthe computation in (72). The string which is to be copied is marked with{ and } symbols, and the location to which it is to be copied is markedwith a � symbol. Markers 〈 and 〉 are inserted in (72.2) in order to keeptrack of which elements have already been copied. The actual copyingis done in (72.3) through (72.5). At each step, the 〉 symbol is shifted tothe right of an element to be copied, and that element is copied to theimmediate left of the � symbol. Finally, 〉-Shift is no longer possible,and the computation terminates by removing the bookkeeping symbols.

(72) 1. k a t{ } �

2. k a t{〈〉 } �

3. k 〉 a t{〈 } k �

4. k a 〉 t{〈 } k a �

5. k a t{〈 〉} k a t �

6. k a t k a t �

I will suppose that (left) transcription uses this mechanism, withthe duplication junctures [ and ] used as end markers in place of { and }and the target location of copy immediately to the left of [, so thatno � symbol is needed. The full reduplication of kat to kat-kat, with

54 Chapter 4

copying to the left, is then computed derivationally as shown in (73),with names given to the three suboperations which are used in carryingout transcription.

(73) [

k

×a

×t

× ]Init−→

k

×a

×t

× ][〈〉CopyShift−→ ×[〈

k

×〉a

×t

× ]

CopyShift−→ × ×[〈

k

×a

×〉t

× ]CopyShift−→ × × ×[〈

k

×

a

×

t

×〉]

Reduce−→ × × ×

k

×

a

×

t

×

A mirror image version is used for copying to the right, with Initinserting 〈〉 in the context

/× ] and 〈 shifting to the left as copyingis carried out from right to left.

4.1 Truncation at the leading edge (reduplicant truncation)

The 〈 and 〉 symbols above are inserted by the transcription rules andused for keeping track of the progress of the computation. They arebookkeeping symbols, nevertheless they are real phonological objects,called truncation junctures. (Their relation to truncation will soonbe apparent.) A juncture which is either a duplication juncture or atruncation juncture will be called a transcription juncture or t-juncture.Suppose that morphology takes it upon itself to insert truncation junc-tures in addition to duplication junctures. For example:

k

×a

×t

×morphology−→ [〈

k

×〉a

×t

× ]

As far as phonology is concerned, this looks like an intermediate repre-sentation in the derivation (73), and it is treated as such by transcription.The derivation below results:

Truncated reduplication 55

CopyShift−→ ×[〈

k

×a

×〉t

×]CopyShift−→ × ×[〈

k

×a

×t

×〉]Reduce−→ × ×

k

×a

×t

×

So, under left transcription, [kat] → kat kat , but [〈k〉at] → at kat .To aid readability, the reduplicant is shaded and the remnant is boxed. Icall this leading edge truncation, or reduplicant truncation. The generalrule is [〈u〉v] → v uv under left transcription and [u〈v〉] → uv vunder right transcription, where u and v are strings. A 〈 . . . 〉 block atthe leading edge of the duplicant is “truncated” from the copy. In fact,of course, it is not actually truncated. It is simply not copied.

Madurese plural reduplication (Stevens, 1968:34) was discussed byMarantz (1982). Some examples follow:

(74) root pluralbuwaP waP-buwaP-an ‘fruits’maen en-maen-an ‘toys’estre tre-estre-an ‘wives’

Assuming Left Transcription, the pretranscription structures have lead-ing edge truncation.

(75) [〈bu〉waP] → waP buwaP

[〈ma〉en] → en maen

[〈es〉tre] → tre estre

It remains to show how these pretranscription structures comeabout as the result of morphological rules which insert t-junctures andrules which might affect the configuration before transcription applies.The focus of this chapter is on transcription, so these questions will beput off until Chapters 5 and 6.

Marantz (1982:451) also discussed a Chukchee reduplicative pat-tern which is similar to the Madurese pattern, but in which the partialcopy of the stem is to the right. Some examples are given in (76). The

56 Chapter 4

meaning associated with the morpheme which induces the reduplicationis the absolutive singular.

(76) root absolutive singularnute nute-nut ‘earth, ground’inu inu-in (part of a reindeer leg)jilPe jilPe-jil ‘gopher’

The reduplicated forms are the result of right transcription and thepretranscription forms with leading edge truncation in (77).

(77) [nut〈e〉] → nute nut

[in〈u〉] → inu in

[jil〈Pe〉] → jilPe jil

The derivation of the pretranscription forms will be given in Chapter 6.

4.1.1 Truncated prefixes and suffixesThere are many examples of leading edge truncation coupled with pre-fixation or suffixation. Yoruba has a nominalizing affix which producesthe following pattern:

(78) root nominal formlo.d li-lo.d ‘to go’dun di-dun ‘to be tasty, sweet’

The data is from Pulleybank and given in Marantz (1982:449), whoanalyzed Yoruba nominalizing reduplication as “prefixing a CV redu-plication skeleton whose V is fixed to i.”

The considerations above suggest a rather different analysis: i isa prefix, but juncture insertion rules apply after the prefix has beenconcatenated with the stem, as shown below:

Truncated reduplication 57

(79)

d

×u

×n

×prefix−→

i

×d

×u

×n

×JncIns−→ [〈

i

×〉d

× ]

u

×n

×

Trscr−→ ×

i

×d

×u

×n

× (didun)

The statement of the juncture insertion rules is significantly simplifiedif the exponents of lexical items can contain t-junctures and the domainof juncture insertion rules is taken to be the stem.1 If the Yoruba NOM-prefix is taken to be [〈i〉- , juncture insertion reduces to:

∅ → ]/

V

The resulting simplification justifies the assumption that t-junctures canenter the computation not only because they are inserted in the stemby juncture insertion rules, but also because they are embedded in theexponents of lexical items. I will henceforth make this assumption. Inthe discussion of Chaha in Chapter 7, it will be proposed that t-juncturescan be embedded not only in the exponents of inflectional affixes, butalso in roots.

Affixes similar to the Yoruba nominalizing affix are widespread.Healy (1960) reports the following data from Agta.

(80) root diminutivewer ‘creek’ wala-wer ‘small creek’bag ‘g-string’ bala-bag ‘small g-string’pesuk ‘peso’ pala-pesuk ‘a mere peso’PassaN ‘small’ Pala-PassaN ‘very small’

The similarity between the diminutive affix responsible for (80) andthe nominalizig affix responsible for (78) should be clear. The exponentis [〈ala〉- and it triggers ∅ → ] /Left V.2 An illustrative derivationfollows in (81). Recall that the domain of the juncture insertion rule isthe stem wer, not [〈ala〉wer .

58 Chapter 4

(81)

w

×e

×r

×prefix−→ [〈

a

×l

×a

×〉w

×e

×r

×]-insert−→ [〈

a

×l

×a

×〉w

×]

e

×r

×

Trscr−→ ×

a

×l

×a

×w

×e

×r

× (w-ala-wer)

What is called “tasty-shmasty reduplication” is the only produc-tive reduplication process in English and has therefore received aninordinate amount of attention. Examples are well-known: oil-smoil,money-shmoney, Bush-Shmush, etc. An interesting point is that thereduplicated form has the prosody of a two word sequence. The suffix -

〈#shm〉], coupled with the juncture insertion rule ∅ → [ / V, where #is an embedded word boundary, produces derivations like the following:

(82)

t

×a

×s

×t

×i

× →t

×[

a

×s

×t

×i

×〈#sh

×m

×〉]

Trscr−→

t

×

a

×

s

×

t

×

i

× #

sh

×

m

× × × × × (tasti-#shm-asti)

Penalfini and Breen (1999) discuss three reduplicative processesin Arrernte, an Arandic language of Central Australia, which furnishgood examples of reduplicative prefixes and suffixes. Penalfini andBreen fit them into the mold of Prosodic Morphology by claiming thatsyllables in Arrernte are necessarily onsetless, but an analysis in theterms developed here is much more straightforward. The affixes areeasily described:

(83) exponent juncture insertion

Habitative: −e”n] ∅ → [ /Right V

Frequentative: −〈ep〉] ∅ → [ /Right V

Attenuative: [〈elp〉− ∅ → ] / Vsecond

The specification of the context / Vsecond , “before the secondvowel”, is provisional. The vocabulary for writing juncture insertion

Truncated reduplication 59

rules will be severely constrained in Chapter 5 so that / Vsecond willno longer be a possible structural description for a juncture insertionrule. Writing the desired rule in Arrernte will depend on the assumptionthat juncture insertion rules can be relativized to a subword (eitherprosodic or morphological) of the stem. That is, the domain of thejuncture insertion rules associated with an affix can be specified to be asubword of the stem. In Arrernte attenuative reduplication, the domainis taken to be the initial bisyllabic foot and the rule is ∅ → ] /Right V.

A few examples of frequentative reduplication are given below,along with derivations. The semantics are clear from the glosses. Theportion of the root which is copied, along with the copy, are boldfacedto make it easier for the reader to parse the reduplicated forms.

(84) Frequentative ( −〈ep〉] ; ∅ → [ /Right V )

a. eN-em ‘is standing’, eNepeN-em ‘keeps standing’eN → [eN〈ep〉] → eNep eN

b. ater-em ‘is fighting’, atereper-em ‘keeps fighting’ater → at[er〈ep〉] → at erep er

c. anentelil-em ‘is putting together’,anentelilepil-em ‘keeps putting together’

anentelil → anentel[il〈ep〉] → anentel ilep il

A few examples of habitative reduplication, along with derivations,are given in (85). The morpheme which is realized by the habitativeaffix is nominalizing. The semantics are variable, under the generalrubric of “object habitually associated with the action denoted by root.”

(85) Habitative ( -e”n] ; ∅ → [ /Right V )

a. aNk ‘eat’, aNke”naNke”n ‘food’aNk → [aNk-e”n] → aNk-e”n aNke”n

b. atwer ‘fight’, atwere”nere”n

atwer → atw[er-e”n] → atw ere”n ere”n

60 Chapter 4

A few examples of attenuative reduplication follow in (86). The se-mantics are generally “beginning to do something” or “doing somethingunintensively.”

(86) Attenuative ( [〈elp〉− ; ∅ → ] / Vsecond )

a. itir-em ‘thinking’, itelpitir-em ‘half thinking’itir → [〈elp〉it]ir → it elpit ir

b. empwar.-em ‘making’,empwelpempwar.-em ‘starting to make’

empwar. → [〈elp〉empw]ar. → empw elpempw ar.

Although the juncture insertion rule cannot apply to a monosyllabicroot, because there is no second vowel, Default Closure applies.

ar.prefix−→ [〈elp〉ar.

DC→ [〈elp〉ar.]

Trscr−→ ar. elpar.

So the root ar. gives ar.-em ‘looking’ and ar.elpar.-em ‘starting to look’.

4.2 Nested duplicants

There are a number of examples in which the morphology creates aduplicant which is nested inside another one, as in the form below.

[[

e

× ]

m

× ]

The core copying mechanism has little difficulty in transcribing such aform. Consider transcription to the right, for example. Transcriptionproceeds as follows:

(87)Init+CopyShift−→ [[

e

×]〈

m

×〉]×Init+CopyShift−→

e

×

m

× 〉] × ×〉]〈[[〈

Reduce−→ [

e

×〈

m

×〉]× ×CopyShift−→ [〈

e

×

m

×〉]× × ×Reduce−→

e

×

m

× × × ×

Truncated reduplication 61

Two points should be noted. First, Init can apply to either the inneror outer duplicant. I assume that transcription is always from the edgein. In right transcription, this means that the transcription rules operatefrom right to left. Consequently, the initial application of Init targets therightmost ]-juncture. Second, note that CopyShift always deposits thecopy it makes outside of the outer duplicant. This is as it should be. Allof the steps in (87) are steps in the transcription of the outer duplicant,even when the inner duplicant is being processed.

Wilbur (1973:10) called attention to two varieties of verbal redu-plication in Takelma, based on the work of Sapir (1922), one of whichuses (87a).

(88) root aorist stem frequentative stem3

hemg hemeg heme:mg ‘take out’masg masag masa:sg ‘put’baxm baxam baxa:xm ‘come’wism wisim wisi:sm ‘move’

The juncture insertion rules for both aorist and frequentative redu-plication begin with initial VC reduplication: hemg → h [em]g, forexample. In the aorist, the duplicant consonant is truncated, and inthe frequentative, the vowel is duplicated again. Transcription is to theright.

(89) root aorist frequentativehemg h[em]eg h[e〈m〉]eg h[[e]m]eg

Right transcription of the aorist produces hemeg, as shown in (90a),and right transcription of the frequentative produces heme:mg, as shownin (90b), using (87).

(90) a.

h

× [

e

× 〈

m

×〉]

g

×CopyShift+Reduce

−→

h

×

e

×

m

× ×

g

×

b.

h

×

e

× ]

m

× ]

g

×[[ →

h

×

e

×

m

× × × ×

g

×

62 Chapter 4

Several more examples of nested duplicants will be consideredbelow. Before that, some reduplication specific extensions of the corecopying mechanism are considered.

4.3 Medial truncation

Consider left transcription of

[〈s

×〉t

×〈a

×〉u

× ]

Using the rules discussed above leads to a deadend.

CopyShift−→ ×[〈

s

×t

×〉〈a

×〉u

× ]CopyShift−→ × ×[〈

s

×t

×〉〈a

×u

×〉]

The transcription rules which reduplication uses are, in fact, extendedto include (91), which allows the computation to proceed.

(91) Combine: 〉〈→ ∅The full derivation is then:

(92) [〈s

×〉t

×〈a

×〉u

× ]CopyShift−→ ×[〈

s

×t

×〉〈a

×〉u

× ]

Combine−→ ×[〈

s

×t

×a

×〉u

× ]CopyShift+Reduce

−→ × ×s

×t

×a

×u

×

Medial truncation of this kind is used extensively in Sanskrit verbalreduplication, from which (92) is taken. It is part of the computationof the perfect stem tustau of the verb stau ‘praise’. Medial truncationis also used in Kinande unintensive reduplication. Both Sanskrit andKinande are discussed in detail in Chapter 7.

Truncated reduplication 63

4.4 Truncation at the trailing edge (remnant truncation)

Now consider left transcription of

[

l

×i

×m

×〈a

×〉]

The rules developed so far, including Combine, produce

(93) [

l

×i

×m

×〈a

×〉]Init+CopyShift∗−→ × × ×[〈

l

×i

×m

×〉〈a

×〉]

Combine+Reduce−→ × × ×

l

×i

×m

×a

×

CopyShift∗ indicates multiple applications of CopyShift. The result isidentical to the result of left transcribing

[

l

×i

×m

× ]

a

×

A further extension of the transcription rules takes advantage ofthis redundancy to provide more flexability in transcription. The ruleCombine is barred from applying to nonempty 〈 . . . 〉 blocks whichare directly followed by ]. The rule CopyDelete, (94) below, appliesinstead.

(94) CopyDelete: 〉〈× → 〉〈, with × copied in the usual way

This rule applies only at the left edge of a far edge truncate and, as aspecial rule, bleeds Combine.

So, for example, rather than (93), the derivation (95) results.

64 Chapter 4

(95) [

l

×i

×m

×〈a

×〉]Init+Shift∗−→ × × ×[〈

l

×i

×m

×〉〈a

×〉]

CopyDelete−→ × × × ×

a

[〈l

×i

×m

×〉〈 〉]Combine+Reduce−→ × × × ×

al

×i

×m

×

Some descriptive terminology is useful to describe the effects oftranscription. What remains of the duplicant is called the remnant andthe copied material is called the reduplicant. Since in some situationsthe direction of transcription is not known, it is also useful to havenames which are independent of the direction of transcription. Theleftmost of the reduplicant and remnant is called the first conjunct andthe rightmost is called the second conjunct.

Trailing edge truncation provides a simple mechanism wherebymaterial can be inserted into the duplicant in such a way that thereduplicant is affected, but not the remnant. We will see in Chap-ter 6 that many reduplicative processes follow juncture insertion withadjustment of the reduplicant to a characteristic prosodic shape beforetranscription applies. Material added to satisfy the prosodic demands isoften truncated, so that it appears in only one of the post-transcriptionconjuncts. In Ndebele, for example, A bisyllabic reduplicant is obtainedfrom [lim] in Ndebele by

[lim]→ [lim〈a〉]The effect is lim → lima lim . This is called either Truncated TrailingEdge Epenthesis, which emphasizes the mechanism, or First ConjunctVowel Epenthesis (FCVE), which emphasizes the effect.

In Mokilese, a bimoraic reduplicant is obtained from [wi] by trun-cated timing slot epenthesis at the right edge, coupled with spreading.Again, transcription is to the left.

(96) [σµ

w

×i

× ]

TruncatedLengthening−→ [

σµµ

w

×i

×〈×〉] → × × ×

w

×

i

× (wi:-wi)

Truncated reduplication 65

This is called First Conjunct Vowel Lengthening (FCVL).The Ndebele and Mokilese examples above will be considered in

detail in later sections with particular attention to how the morphologypositions the t-junctures. The intention here is only to illustrate how thetranscription process responds to t-junctures in the representation.

4.4.1 Reduplicant truncation combined with remnant truncation:Permutation, Metathesis, and Infixation

now how left transcription responds to a form like [〈k〉a〈t〉], with bothleading edge truncation and trailing edge truncation. The result is[〈k〉a〈t〉] → at ka , with k missing from the copy and t missing fromthe remnant. Leading edge truncates do not appear in the copy andtrailing edge truncates do not appear in the remnant.

(97) [〈k〉a〈t〉]1. a[〈ka〉〈t〉] Copy/Shift2. at[〈ka〈t〉〉] Copy/Shift3. at-ka〈t〉 Reduce4. at-ka Truncate

Surprisingly, juncture insertion and transcription can accomplishpermutation: [〈k〉〈at〉]→ at-k .

(98) [〈k〉〈at〉]1. a[〈k〈a〉t〉] Copy/Shift2. at[〈k〈at〉〉] Copy/Shift3. at-k〈at〉 Reduce4. at-k Truncate

This is an important result. It means that both metathesis and infixationare special cases of reduplication.

Arrernte furnished several good examples of reduplicative prefix-ation and suffixation. It also furnishes an excellent example of howjuncture insertion and transcription accomplish metathesis. Accordingto Pensalfini and Breen (1999), from whom the data is taken, “RabbitTalk is a language game that involves transposing the initial portion of

66 Chapter 4

a word to the end of the word, not unlike the Pig Latin of English.”Examples follow:

(99) Arrernte Rabbit Talkemen enem ‘plant food’ekwen.et’ek en.et’ekekw ‘to put in’araNkw aNkwar ‘no’itirem iremit ‘thinking’ulket. et.ulk ‘perentie lizard’

For polysyllabic words, the translation from Arrernte to Rabbit Talkis accomplished by two rules:

∅ → [/

× , ∅ → 〉〈/

Vsecond

The derivation (100) assumes right transcription, but there is noevidence which favors right over left transcription.

(100)

i

×t

×i

×r

×e

×m

×Juncture Insertion−→ [

i

×t

×〉〈i

×r

×e

×m

×

DC→ [〈

i

×t

×〉〈i

×r

×e

×m

×〉]Trscr−→

i

×

r

×

e

×

m

× ×

i

×

t

There is allomorphy, with a different affix (prefixal y-) used for mono-syllabic stems. This gives iNk→ yiNk , emp→ yemp , etc.

Permutation can be used to accomplish infixation. We demon-strate this with an analysis of Choktaw passive morphology, which wasdiscussed by McCarthy and Prince (1995). They give the followingexamples:

(101) stem infixed surfaceabani ‘to barbecue’ albani albani

apisa ‘to set a date’ alpisa aìpisa

hokci ‘to plant’ holkci holokci

takci ‘to plant’ talkci talakci

Truncated reduplication 67

Certain late operations (epenthesis, vowel spreading, and l-mutation)produce the surface forms.

Transparently, l is infixed after the initial vowel. A simple way toaccomplish this is to prefix l, then permute the prefix and the initialC∗V of the stem. This can be realized very simply by means of areduplicative prefix. The exponent is [〈l〉〈- and it triggers ∅ → ] /V .

takciprefix−→ [〈l〉〈takci

JncIns−→ [〈l〉〈ta]kci

DC→ [〈l〉〈ta〉]kci

Trscr−→ ta l ci

An illustrative derivation follows in (102). It is assumed for thesake of concreteness that transcription is to the right, but there is noevidence for this.

(102)

t

×a

×k

×c

×i

×prefix−→ [〈

l

×〉〈t

×a

×k

×c

×i

×

]-insert−→ [〈

l

×〉〈t

×a

× ]

k

×c

×i

×DC→ [〈

l

×〉〈t

×a

×〉]k

×c

×i

×

Trscr−→

t

×

a

× ×

l k

×

c

×

i

× (ta-l-kci)

Prosodic considerations are virtually absent from this analysis.McCarthy and Prince, on the other hand, are forced to fit their analysisinto the mold of Prosodic Morphology, which obscures the simplicityof the morphology.

Formally, l infixation is actually prefixation under negative prosodiccircumscription of an initial light syllable σµ, requiring Law-of-Parsing mediated restructuring of an initial heavy σ (Urbanczyk1992). The morphological rule, restricted in this way, is expressedby O /Φ(σµ,Left), where O = “Prefix l”. (McCarthy and Prince,1995:347)

The tortuous account of infixing passive morphology in Chocktawwhich Prosodic Morphology is forced into is evidence for an importantconclusion we hope to establish: Prosodic Morphology misidentifies

68 Chapter 4

instances of juncture insertion in the context V as instances ofprosodic specification via syllable weight. Earlier, the specification ofC∗V-reduplication was made without explicit appeal to prosody. TheChoctaw passive is another example of the same kind. Choctaw passiveis exactly what it seems to be, “put l after the first vowel.” Using thetools at its disposal, there is only one method for the morphophonologyto accomplish this: “put” l at the edge of the stem by concatenation(as the exponent of a morpheme), then position it by reduplicativemetathesis. A few secondary parameters must be fixed, either bylanguage particularities or markedness. Is concatenation to the rightor left? Is transcription to the right or left?

4.4.1.1. Varieties of Yaqui reduplicationHaugen (2003) discusses four different reduplicative verbal affixes inYaqui, an Uto-Aztecan language spoken in northern Mexico, with adiaspora in Arizona in the US. One of the affixes is usually associatedwith iterative aspect, but is also used with a continuative meaning aswell as various idiosyncratic meanings. I will simply identify it as“iterative”. The other three are associated with a habitual meaning.Comparison of the morphology is revealing.

(103) prefix juncture insertion name

[〈×〉− ∅ → ] /V iterative

[〈×〉〈− ∅ → ] /V geminating (habitual)

(none) ∅ → ] /V light syllable (habitual)

(none) ∅ → ] /Vsecond disyllabic (habitual)

× is a bare timing slot.The effect of these affixes is illustrated below in some simple cases:

Truncated reduplication 69

(104) iterative geminatinglight

syllabledisyllabic

yena hine vusa kupikteprefix [〈×〉yena [〈×〉〈hineJncIns [〈×〉ye]na [〈×〉〈hi]ne vu]sa kupi]kteDC [〈×〉〈hi〉]ne [vu]sa [kupi]kteTrscr ye ×ye na hi × ne vu vu sa kupi kupi kte

spreading yeyyena hinneyey-yena hi-n-ne vu-vusa kupi-kupikte

There are some minor complications for iterative and light syllable redu-plication involving the interaction of long vowels and accent placement.There is a major complication in light syllable reduplication in case theinitial syllable of the stem is closed. The entire syllable copies. Thiswill be discussed in detail in Section 6.3.3 along with the interactionof vowel length and stress placement. It is a consequence of prosodicadjustment which applies after the affix combines with the stem, not toany modification of the specification of the prefix. Disyllabic redupli-cation is identical to Diyari plural reduplication, which is discussed inSection 5.2.

According to Haugen, the choice of affix between light syllablereduplication and geminating reduplication to realize habitual variesfrom speaker to speaker for some verbs and is sometimes unstable fora single speaker. He cites hinne/hi-hine and himmaPako/hi-himaPako

alternations. Viewed at the surface level, the connection between lightsyllable reduplication and geminating reduplication is obscure at best.But the analysis of geminating reduplication given above makes theconnection clear. If the affix which generates geminating reduplicationsimplifies by losing its exponent, the affix which produces light syllablereduplication results.

4.5 Transcription made easy

To this point, the computation of transcription has been broken downinto the basic low level operations which carry it out. It is possibleto describe transcription at a higher level. simple duplicantSimple

70 Chapter 4

duplicants, ones not containing embedded duplicants, are easy to tran-scribe. 〈 . . . 〉 blocks at the trailing edge of the duplicant appear in thereduplicant, but not the remnant. Other 〈 . . . 〉 blocks appear in theremnant, but not the reduplicant. Other material appears in both thereduplicant and remnant.

For a simple duplicant [α], we introduce the notations Copy([α])

and Remnant([α]), whose meaning should be clear from (105). On the

right side, the copy is shaded and the remnant is boxed. It is “whatremains in the box after transcription.”

(105) a. [ l o 〈w a c 〉 ] → l o w a c l o

Copy([lo〈wac〉]) = lo, Remnant

([lo〈wac〉]) = lowac

b. [ l i m 〈 a 〉 ] → l i m a l i m

Copy([lim〈a〉]) = lima, Remnant

([lim〈a〉]) = lim

c. [ 〈 s 〉 t 〈 a 〉u ] → t u s t a u

Copy([〈s〉t〈a〉u]

)= tu, Remnant

([〈s〉t〈a〉u]

)= stau

Note that Copy([α])

and Remnant([α])

can depend on the direction oftranscription.

Transcription of complex duplicants can be simplified by using thereduction rule (106), combined with the rule for transcribing simpleduplicants. The rule applies only if there are no other rules which canapply nearer to the active edge.

(106) [α]→ Remnant([α]), with Copy

([α])

copied.

Broselow and McCarthy (1983:82) discuss a German languagegame in which words are broken up into syllables and subject to theoperation exemplified by kat → kathatlefat . The operation is suffix-ation of -〈h〉]〈lef 〉], coupled with ∅ → [[ / V. So, for example,kat → kat〈h〉]〈lef 〉] → k [[at〈h〉]〈lef 〉]. Using (106), transcription iscarried out in two steps as shown below.

(107) 1. k [ [ a t 〈h 〉 ] 〈 l e f 〉 ] → k a t h [ a t 〈 l e f 〉 ]2. k a t h [ a t 〈 l e f 〉 ] → k a t h a t l e f a t

Truncated reduplication 71

Creek pluralization uses the following for plural reduplication ofstems which end in k followed by m, n, l, or s. The stem is caniks,‘sideways’.

(108) 1. [ c a 〈n i [ 〈k 〉 〈 s 〉 ] 〉 ] → [ c a 〈n i s 〉 ] k

2. [ c a 〈n i s 〉 ] k → c a n i s c a k

Stems with other terminations do not have this embedded metathesis.For the stem lowack ‘soft’, for example:

[ l o 〈 w a c 〉 ] k → l o w a c l o k

Mokilese progressive reduplication uses double reduplication formonosyllabic stems. For example, ca:k ‘bend’→ ca:ca:ca:k.

(109) 1. [ [ c a a ] ] k → c a a [ c a a ] k

2. c a a [ c a a ] k → c a a c a a c a a k

Chapter 5

Sources of variation

A wide variety of reduplicative surface patterns are found in the world’slanguages. In part, the differences are due to different affix specifica-tions; different exponents and different juncture insertion rules. Butthese differences do not account for the full range of variation. Thesurface pattern that is associated with a reduplicative affix is the resultof a number of different factors, whose effect is felt at various stages inthe derivation.

(110) 1. the exponent of the affix;2. the juncture insertion rules that the affix triggers;3. pretranscription modification of the duplicant;4. the direction of transcription; and5. post-transcription operations.

This chapter will focus primarily on the possibilities for (110.2) andsecondarily on (110.5), particularly those aspects of post-transcriptionmodification which are special to reduplication. Discussion of (110.3)is the focus of the next chapter. Pretranscription modification of theduplicant is very important in some languages, playing a major rolein determining the surface pattern. This requires extensive discussion,which Chapter 6 is devoted to.

There is not a great deal to say about either (110.1) or (110.4).Obviously, different exponents produce different patterns, particularly

Sources of variation 73

since the exponents of reduplicative affixes can contain embedded em-bedded t-junctures. But since there is little that can be said of a generalnature about the range of possibile exponents other than to describewhat occurs, (110.1) will be left without further comment. We saw inthe discussion of Klamath in Chapter 3 that the possibility of root vowelsyncope, which affects the surface pattern, depends on the direction oftranscription. Other than noting that the direction of transcription canaffect the surface pattern, nothing further will be said here about thevariation which can be attributed to (110.4). There are only two choices.

5.1 Possible juncture insertion rules

We begin by putting aside the question of possible domains of junctureinsertion rules, which is taken up in the next section, and simply assumethat juncture insertion rules apply in some domain. The goal is to showthat a very small inventory of juncture insertion rules suffices for anempirically adequate theory. Insertion sites for junctures are locatedwith respect to the syllable nuclei of the domain, and the edges ofthe domain. In order to make this idea precise, we need the notionof the “nuclear skeleton” of the domain. Consider, for example, therepresentation in (111a), where the nuclei (ν) are explicitly representedon their own plane. Phonologically, the nuclear skeleton of (111a) is thestructure (111b), which is embedded in (111a).

(111) a.

p

×O

×d

×o

×k

×σ σ

ν ν

b. ×ν

× ×ν

× ×

Morphological associations are not shown in (111a), but (111a) and itsnuclear skeleton are assumed to share the same morphological associa-tions.

The starting point for establishing the inventory of juncture inser-tion rules is the proposal that juncture insertion rules are operationson the nuclear skeleton. Rules like ∅ → ] / ν must be possi-ble juncture insertion rules, but there is an ambiguity which must beclarified. Insertion is into the timing tier, but ν is an autosegment,not on the timing tier. What does the environment ν mean? If ν

74 Chapter 5

is linked to a single timing slot, there is no ambiguity; the meaningis “to the right of the timing slot linked to ν”. In the case of a longvowel or diphthongal nucleus, an ambiguity of the kind familiar fromconsiderations of geminate inalterability arises. Consider:

×ν

× × ×ν

× ×The environment ν can be interpreted narrowly to mean “immedi-ately to the right of the set of timing slots linked to a ν”, or interpretedmore broadly to mean “immediately to the right of one of the timingslots linked to a ν”. The broad interpretation specifies the locations in(112a) and the narrow interpretation specifies the locations in (112b).

(112) a. ×ν

× × ×ν

× ×↑ ↑ ↑ b. ×ν

× × ×ν

× ×↑ ↑I assume that both interpretations are available, with the broad interpre-tation the default and the narrow interpretation (112b) a marked option.

In Chapter 7, the marked option plays a central role in the analy-sis of Sanskrit and Tohono O’odham, an indigenous North Americanlanguage. Nuu-chah-nulth reduplication triggered by Class I-2 affixes,introduced in Section 3.5, provides another straightforward examples.The Class I-2 suffix -Pal∼uk produces the following pattern:

(113) capx. → ca-capx. -Pal∼ukPu:kw → Pu:-Pu:kw-Pal∼uk

This is straightforward juncture insertion:

(114) ∅ → ] / ν (narrow interpretation)

A ]-juncture is inserted to the right of the nucleus (i.e. to the right ofall of the timing slots associated with the nucleus).1 (Omission of thesuffix below is simply to save space.)

P

×ν

u

× ×kw

×JncIns−→

P

×ν

u

× ×]

kw

×DC→ [

P

×ν

u

× ×]

kw

×Trscr−→ × × ×

P

×ν

u

× ×kw

×

Sources of variation 75

The notation / ν will be used in the future for the narrowinterpretation and /V for the broad interpretation. More generally,the symbol V appearing in the environment of a juncture insertion ruleswill stand for the predicate which is true of timing slots linked to anucleus, regardless of the phonemic association of the timing slot. Inthe same way, the symbol C will stand for the predicate which is true oftiming slots which are not linked to a nucleus.

We are now in a position to propose that juncture insertion rules arealways drawn from the array in (115).

(115) Possible juncture insertion rules

Insert

[]〈〉

{

beforeafter

} V×νC

,{

leftmostrightmost

}in the domain of ρ

There is one other general issue of rule application that is importantfor some reduplicative processes. I assume:

(116) Balancing an unpaired duplication juncture, forming a non-trivial duplicant if possible, takes precedence over leftmost orrightmost application.

The significance of (116) is illustrated by the reduplicative processused to realize plurality in Mangarayi, which has been of interest for along time. The data is from Merlan (1982).

(117) root plurala. gal.ugu gal.al.ugu ‘poor things’b. waNgij waNgaNgij ‘children’c. jimgam jimgimgam ‘knowledgeable people’

The affix has a null exponent and triggers readjustment:

∅ → [/

V ; ∅ → ]/

V

Take (117) to be an ordered list of rules. Consider, for example, itsapplication to (117c), jimgam. First, ∅ → [ / V applies to the stemand produces j [imgam. Then ∅ → ] / V applies. The leftmostapplication which forms a nontrivial duplicant is chosen. The result

76 Chapter 5

is therefore j [img]am, rather than j []imgam. Right transcription thenproduces jimg-img-am.

5.1.1 Creek plural adjective reduplicationRiggle (2004) discusses the unusual pattern of reduplication which isused to realize adjective plurality in Creek, based on data from Haas(1977), Booker (1980), and Martin and Mauldin (2000).

(118) singular pluralfayatk-i: fayatfak-ı: ‘crooked’camp-i: camcap-ı: ‘sweet’lowack-i: lowaclok-i: ‘soft’polo:k-i: polo:pok-ı: ‘round’

This is a particularly clear example of the usefulness of Default Closurein simplifying the specification of complex reduplication processes.There are two juncture insertion rules:

∅ → ]/Right ×, ∅ → 〈

/V

Derivations follow:

(119) JncIns DC TrScr

a. fayatk fa〈yat]k [fa〈yat〉]k fayat fa k

b. camp ca〈m]p [ca〈m〉]p cam ca p

c. lowack lo〈wac]k [lo〈wac〉]k lowac lo k

d. polo:k po〈lo:]k [po〈lo:〉]k polo: po k

Creek uses a special rule if the root ends in certain consonantclusters, as discussed in Section 4.5 (p. 70).

5.1.2 Some common juncture insertion rulesIt is useful to have names for some of the common juncture insertionrules and the kind of reduplication they generate (if there are no othert-junctures except those inserted by default closure).

Sources of variation 77

(120) C∗V-rule: ∅ → ] /V (C∗V-reduplication)C∗ν-rule: ∅ → ] / ν (C∗ν-reduplication)×-rule: ∅ → ] /× (initial timing slot reduplication)×∗-rule: ∅ → [ / × (total reduplication)

5.2 Possible domains of reduplicative affixes

The juncture insertion rules which a reduplicative affix specifies aredetermined in a composite fashion. Rules drawn from the possibilitiesenumerated in (115) are specified, but a domain of application of theserules which differs from the stem can also be specified. A rule like

∅ → ]/Right V

for example, has a very different effect if its domain of application isthe initial foot of the stem or it is the entire stem. If an affix doesnot designate a special domain, the rules are assumed to apply to thestem. In other words, the stem is taken to be the default domain forjuncture insertion rules. The designated (or default) domain is calledthe domain of the affix. The domain of a reduplicative affix can bespecified on either the basis of the prosodic structure of the stem or itsmorphological structure. In all the examples that I am aware of in whichthe morphological structure of the stem is used in domain specification,a morphological constituent of the stem is designated as the domain ofthe affix. Most commonly, morphologically based stem designation isthe root, embedded in an inflected form.

The theory of prosodically based contraction that I adopt derivesfrom Broselow and McCarthy (1983). They argue that several varietiesof reduplication are most insightfully analyzed (in the template con-catenation framework that their theory is framed in) by supposing that“reduplicative morphemes may be prefixed not only to morphologicalbut to phonological constituents, as is argued for Warlpiri prefixingreduplication by Nash (1980).” This idea was further developed byMcCarthy and Prince within the framework of Prosodic Morphologyas the notion of “positive prosodic circumscription.” See the discus-sion in McCarthy and Prince (1995:340). A key insight of ProsodicMorphology was that only special kinds of prosodic constituents couldserve as the locus of affixation; those that had wordlike characteristics.

78 Chapter 5

It was established that if a language imposed a prosodic minimalitycondition on words, the minimality requirement was that there be suffi-cient material to form a well-formed foot. In other words, the prosodicconstituents that are “wordlike” in the appropriate sense are feet. Since Ido not know of any counterexamples, I will assume that specification ofa prosodic constituent as the domain of a reduplicative affix is restrictedto specifying it to be the first foot; the initial foot in left to right footing,or the final foot in right to left footing.

The Yidiny reduplicative plural affix (Dixon, 1977; Nash, 1979–80)specifies the initial bisyllabic foot as its domain, has a null exponent,and triggers the ×∗-rule. This leads to derivations like (121).

(121)

( σ

g

×i

×n

×σ

d

×a

×)

l

×σ

b

×a

××∗-rule−→

( σ

[

g

×i

×n

×σ

d

×a

×)

l

×σ

b

×a

×

DC−→

( σ

[

g

×i

×n

×σ

d

×a

×)

l

×]σ

b

×a

× → gindal gindalba

Crucially, when DC applies, it must insert the ]-juncture in the domain,which is the initial foot. Insertion at the right edge of the domainproduces a duplicant with maximal extent.

The Diyari plural affix (Austin (1981)), like the Yidiny plural affix,specifies the initial bisyllabic foot as its domain and a null exponent.Unlike Yidiny, it triggers the rule.

∅ → ] /Right V

This leads to derivations like:

(122)

( σ

N

×a

×n

×σ

k

×a

×)

n

×σ

t

×i

× →( σ

N

×a

×n

×σ

k

×a

×])

n

×σ

t

×i

×

DC−→

( σ

[

N

×a

×n

×σ

k

×a

×])

n

×σ

t

×i

× → Nanka Nankanti

Sources of variation 79

Note that ∅ → ] /Right V applies rightmost in the domain of the affix,the initial foot.

Both the Yidiny and Diyari reduplicative plural affixes are dis-cussed by McCarthy and Prince (1986).

5.2.1 Ulwa possessiveThe original analysis and source of subsequent data is Hale and LacayoBlanco (1989). Other analyses include Bromberger and Halle (1988)and McCarthy and Prince (1995). In Ulwa, various possessive mor-phemes are realized by affixes which, descriptively, induce infixation ofan exponent characteristic of the morpheme. The position of the infixis determined by the foot structure of the stem. Footing is mora basedand left to right binary. The paradigm for the 3sg possessive is given in(123). The initial foot is parenthesized.

(123) Ulwa possessive infixation

stem possessed

(bas) bas-ka ‘hair’(sa.na) sana-ka ‘deer’(su:).lu su:-ka-lu ‘dog’(bas).kar.na bas-ka-karna ‘comb’(si.wa).nak siwa-ka-nak ‘root’ka.(ras).mak karas-ka-mak ‘knee’

Reasonable people can differ over whether the initial foot in the lastexample, karasmak, is as shown or includes the first syllable. All that isrelevant to the analysis here is the location of the right boundary of theinitial foot.

The affix designates the initial foot as its domain, has the suffixalexponent -〉〈ka〉] and triggers:

(124) ∅ → [/Right ×

The juncture insertion rule is relativized to the designated domainof the affix, but suffixation is to the stem. Suffixation and prefixationare strictly concatenative in DR. There is no infixation per se, althoughfrom time to time the term will be used descriptively to refer to a processwith a certain surface form.

80 Chapter 5

A typical derivation is shown in (125). Transcription is taken to beto the left. Only the initial foot is shown.

(125)

( σµ

s

×i

×σµ

w

×)

a

×σµµ

n

×a

×k

×suffix−→

( σ

s

×i

×σ

w

×)

a

×n

×a

×k

×〉〈k

×a

×〉]

[-Ins−→

( σ

s

×i

×σ

w

×)

a

×[

n

×a

×k

×〉〈k

×a

×〉]DC→

s

×i

×w

×a

×[〈n

×a

×k

×〉〈k

×a

×〉]

Trscr−→

s

×i

×w

×a

× ×k

×an

×a

×k

× (siwa-ka-nak)

Note that exponent of the affix is a suffix, not an “infix” of any kind.It is the t-junctures contained in the prefix and the juncture insertion rulewhich its insertion triggers that cause ka to show up on the surface asan “infix.” Note also that (124) inserts a [-juncture after the rightmosttiming slot of the domain.

5.2.2 Manam final foot reduplicationManam has an affix which induces total reduplication of the final foot.Footing is quantity sensitive and right to left. The affix takes the finalbimoraic foot as its domain, has a null exponent, and triggers the ×∗-rule. Two derivations follow, one with a bisyllabic final foot and onewith a monosyllabic final foot.

(126) a.

σµ

s

×a

×σµ(

l

×a

×σµ

g

×a

×) ×∗-rule−→

σµ

s

×a

×σµ(

[

l

×a

×σµ

g

×a

×)

DC−→

σµ

s

×a

×σµ(

[

l

×a

×σµ

g

×a

×])

→ salaga laga

Sources of variation 81

b.

σµ

m

×a

×σµ

l

×a

×σµµ(

b

×o

×N

×) ×∗-rule−→

σµ

m

×a

×σµ

l

×a

×σµµ(

[

b

×o

×N

×)

DC−→

σµ

m

×a

×σµ

l

×a

×σµµ(

[

b

×o

×N

×])

→ malaboN boN

5.2.3 Exclusion of initial unfooted materialThere are many examples in which prosodically defective material atthe edges of a word is excluded from the foot structure. It can thereforebe excluded from the domain which a reduplicative affix designates.

5.2.3.1. Bella CoolaBagemihl (1991: 598, 609), gives the following examples of what hecalls CV reduplication in Bella Coola:

(127) a. qayt qa-qaytb. tqn­ k t-qn­ -qn­ kc. stn­ s-tn­ -tn­d. t’ksn­ t’k-sn­ -sn­e. st’xwm­ st’-xwm­ -xwm­

He argues convincingly that Bella Coola syllable structure admits atmost one obstruent into the onset and that the infixing pattern (127) isa consequence of this syllable structure. Clearly, nonvocalic sonorantscan be syllable nuclei.

The examples (127) illustrate two features of the discussion above.First, V in the juncture insertion rule ∅ → ] /V must be interpretedas being true of syllable nuclei in general, not just vowels. Second,there is domain selection. The domain of the affix which generates(127) excludes the unsyllabified initial obstruents. Derivations of someof the examples in (127) are given in (128). The domain is the initialfoot, which excludes the unsyllabified initial obstruents. There is nominimal word condition in Bella Coola. The crucial point is that

82 Chapter 5

Juncture Insertion (JncIns) and Default Closure (DC) are relativizedto the selected domain.

(128) domain JncIns DC Trscr

qayt qayt qa]yt [qa]yt qa-qayt

tqn­ k t qn­ k t qn­ ]k t [qn­ ]k t qn­ -qn­ k

st’xwm­ st’ xwm­ st’ xwm­ ] st’ [xwm­ ] st’ xwm­ -xwm­

5.2.3.2. OrokaivaHalle and Vergnaud’s (1987:48) discussion of Western Aranda stressshowed that excluding initial onsetless syllables from the computationof foot structure is an option that UG makes available. Western Arandahas a bimoraic word minimum, so this exclusion is only available inthat language when at least two moras are not excluded. In a languagewhose metrical structure employs this device, initial onsetless syllablescan be excluded from the domain of a reduplicative affix.

Orokaiva (Healy, Isoroembo, and Chittleborough, 1969; McCarthyand Prince, 1999) has a reduplicative affix whose domain is the ini-tial foot, whose exponent is null, and which triggers the C∗V-rule.It realizes repetitive verbal aspect. Like Western Aranda, the initialfoot excludes an initial onsetless syllable, if there is one. Illustrativederivations follow:

(129) a.

σ(

h

×i

×σ

r

×i

×) σ

k

×e

×C∗V-rule−→

σ(

h

×i

×]σ

r

×i

×) σ

k

×e

×

DC→

σ(

[

h

×i

×]σ

r

×i

×) σ

k

×e

× → hi hirike

b.

σ

u

×σ(

h

×u

×σ

k

×e

×)

C∗V-rule−→

σ

u

×σ(

h

×u

×]σ

k

×e

×)

DC→σ

u

×σ(

[

h

×u

×]σ

k

×e

×)

→ u hu huke

Sources of variation 83

5.2.3.3. Imdlawn Tashlhiyt BerberDell and Elmedlaoui (1988) discuss the following paradigm from Imd-lawn Tashlhiyt Berber.

(130) Imperfective geminationroot imperfective

a. mrz m:rz ‘wound in the head’b. frn f:rn ‘sift’c. rkm rk:m ‘rot’d. kSm kS:m ‘enter’

They argue that the contrast between initial gemination and second slotgemination is a precise reflection of syllable structure, with (130a,b)having initial syllables with onsets, (130c) having an initial onsetlesssyllable, and (130d) having an initial unsyllabified obstruent. Obstru-ents can be nuclei word internally, but not in word initial or word finalposition. The syllable structures are:

(131) a.

σ

m

×r

×σ

z

× b.

σ

f

×r

×σ

n

× c.

σ

r

×σ

k

×m

× d.

k

×σ

S

×m

×

Descriptively, the first onset geminates.Under the assumption that these stems have metrical structure

(there is no word minimality) and that the leftmost left foot boundaryis aligned with the leftmost syllable which has an onset, this pattern isgenerated by an null affix which designates the initial foot as its domainand triggers the ×-rule, ∅ → ] /× .

σ

r

×σ(

k

×)

m

××-rule−→

σ

r

×σ(

k

×)

]

m

×DC→σ

r

×σ(

[

k

×)

]

m

×Trscr−→

r

× ×k

×m

×

Transcription is to the left, the default for left edge duplicants.2

This is one of the few examples in which a multiply associatedphoneme in the output of transcription does not violate the NCC. Theimperfective geminate is the direct product of duplication, with noNCC repair. Note also that the example demonstrates fairly clearly theinadequacies of templatic approaches to reduplication. A template can

84 Chapter 5

neither specify what is to be duplicated nor predict that the output is ageminate rather than simply a doubled consonant.

One final point should be noted. Ideally, the correlation betweenthe reduplication patterns that are found in the world’s languages andthe reduplication patterns that are easy to describe would be transparent.To a considerable extent, there is such a correlation. The two mostwidespread patterns are total reduplication and C∗V-reduplication, bothof which have very simple rules. But the rule which specifies initialgemination is just as simple and it is rare. Why should this be so?Juncture insertion must mesh with the rest of the phonology of thelanguage. If transcription produces a representation which violatesthe syllable well-formedness conditions of a language, then substantialphonological support is needed to transform the output of transcriptioninto an acceptable form. The implicit phonological support whichmay be necessary means that the complexity of a reduplication processcannot be evaluated simply on the basis of the simplicity of the junctureinsertion rules involved. The reason that initial gemination is rare isnot because the juncture insertion rules are complex, but because it israre to find a language in which initial geminate consonants are allowed.Imdlawn Tashlhiyt Berber happens to be such a language.

5.3 Shortcut Repair at the reduplicant-remnant boundary

So far, this chapter has discussed the variation in reduplicative processesdue to the variation in juncture insertion, factored into two components,domain selection and rule specification relativized to that domain. Inmost cases, NCC repair is simple phoneme fission. There are a numberof languages, however, in which special rules are used to repair thetranscribed structure, adding further possibilities to the range of possiblereduplicative processes. Sanskrit, discussed in detail in Chapter 7,has several specialized rules of repair that contribute to the distinctivecharacter of Sanskrit reduplication. What I call “shortcut repair” is usedby several languages.

Echo reduplication in Kolami (Emeneau, 1955, McCarthy andPrince, 1995) is exemplified below:

Sources of variation 85

(132) a. pal pal-gilb. kota kota-gitac. sa: sa:-gi:d. ma:sur ma:sur-gi:sur

The pattern is generated by the suffix -〈gi 〉] coupled with the junc-ture insertion rule ∅ → [ /V and a special mechanism that somelanguages (including Kolami) use for eliminating crossing violations atthe reduplicant-remnant boundary.

Examples (132a, b), with a short final vowel, are straightforward.

(133)

k

×o

×t

×a

×suffix−→

k

×o

×t

×a

×〈g

×i

×〉]

JncIns−→

k

×o

× [

t

×a

×〈g

×i

×〉]Trscr−→

k

×

o

×

t

×

a

×

g

×

i

× × ×

Now consider the case of a long final vowel, (133c, d). For sa:, forexample:

(134)

s

×a

× × →s

×a

× ×[ 〈g

×i

×〉]Trscr−→

s

×a

× ×g

×i

× ×

Fission would yield sa:gia, with vowel hiatus, not the desired sa:gi:.I suppose, in the case of Kolami, that crossing violations can

be removed not only by fission, but also by (135), which not onlyeliminates a crossing violation but also eliminates vowel hiatus.

(135)

V1

× . . .

V2

× ××

86 Chapter 5

If (134) continues by eliminating the crossing violation using (135)the result is the desired:

s

×a

× ×g

×i

× ×

Note that no vowel is lost in the final representation. At most anoccurrence of a vowel is lost, but one occurrence remains. Informally,the operation can be looked at as hiatus elimination licensed by a kindof recoverability of deletion.

The repair mechanism used by Kolami at the reduplicant-remnantboundary occurs often enough to merit a designation (Shortcut NCCRepair) and some discussion. The usual context in which it occursis given in (136), where a CVC reduplicant has been formed to theleft of the remnant. In order to make the shortcut stand out moreclearly, the first step has carried out partial NCC repair, leaving only thecrossing violation at the boundary between the reduplicant and remnantunresolved. This is simply to make the shortcut operation clearer.The operation removes a crossing violation whether the other crossingviolations have been removed or not.

(136) × × ×

C1

×

V

×

C2

× . . . →

C1

×

V

× ×

C1

×

V

×

C2

× . . .

Shortcut Repair−→

C1

×

V

× ×

C1

×

V

×

C2

× . . .

Hausa pluractional reduplication (Newman, 2000) is the clearestillustration of (136) since in many contexts either Fission or ShortcutRepair can be used, resulting in surface variation. The process copiesa CVC to the left.3 If C2 is nonsonorant and noncoronal, ShortcutRepair is obligatory, otherwise there is free variation between Fissionand Shortcut Repair. In the examples below, the surface forms, in which

Sources of variation 87

standard Hausa phonological rules apply after transcription, are shownin parentheses.

(137) Hausa pluractional reduplication

Fission Shortcut Repairfita fit-fita fif-fita ‘go out’

(firf ıta) (fiff ıta)gasa gas-gasa gag-gasa ‘roast’

(gargasa) (gaggasakira)saya say-saya sas-saya ‘buy’

(sais.ay`a) (sassay`a)tambaya tam-tambaya tat-tambaya ‘ask’

(tantambay`a) (tattambay`a)

In Ponapean heavy syllable reduplication, which is analyzed indetail in Chapter 6, Shortcut Repair applies if C1 and C2 are sonorantand C1 is coronal, otherwise Fission applies.

(138) Ponapean durative reduplication

Fission Shortcut Repairliro:ro ‘contract’ *lir-liro:ro lil-lirlironur ‘protective’ *nur-nur nun-nurtep ‘begin’ tep-tep (>tepitep) *tet-tep

Chapter 6

Prosodic Adjustment

After t-junctures are inserted, operations can apply to the duplicantbefore transcription applies. In many languages, prosodic desiderata areimposed on the duplicant and rules specified for adjusting the duplicantso that the desiderata are satisfied. It will be important to constrain theset of possible prosodic targets and set of possible adjustment rules, sothat the options made available by the theory corresponds to the range ofoptions that are attested. Before we address that issue, however, severalexamples will be discussed.

6.1 Onset incorporation in Kıhehe

In (139), ita, lima and i:mbila are verb roots, and ku- is an infinitivalprefix. The examples are from Odden and Odden (1985).

(139) reduplicated

a. kuceenga ‘to build’ ku-ceenga-ceenga ‘to build a bit’b. kulima ‘to cultivate ku-lima-lima ‘to cultivate

a bit’c. kwi:ta ‘to pour kwi:ta-kwi:ta ‘to pour a bit’d. kwi:mbila ‘to sing kwi:mbila-kwi:mbila ‘to sing a bit’

According to O&O, “any high vowel before another vowel becomes aglide, with compensatory lengthening of the following vowel.” This

Prosodic Adjustment 89

is responsible for the fusion of the infinitival prefix with the rootin (139c, d).

In the theoretical framework which was available to them at thetime, Marantz’s template concatenation theory, (139a, b) forced theconclusion that the template is concatenated before the prefix applies.O&O pointed out the problem raised by the duplication of the prefix in(139c, d) if the reduplicative affix is inside the infinitival prefix. Oddenand Odden’s solution was to separate the operation of template concate-nation from the operation of association with the template. They pro-posed that association with the template occurred after syllabificationfused the prefix with the root and that the rules of association were for-mulated in such a way that material that was attached syllabically to theroot was associated to the template along with the root itself. AlthoughI do not think this analysis withstands detailed scrutiny, particularly onthe question of precisely what “concatenation of a template” means, theidea that the immediate changes in the representation induced by themorphology are only a prelude to the ultimate effects that these changeslead to is an important insight which DR has incorporated.

We suppose that there is an unintensive morpheme (UNINT) whichis realized after the infinitival prefix has been combined with the stem.It is realized by a reduplicative affix ρ with a null exponent, whichspecifies the root as its domain and triggers the ×∗-rule. Nothing furtherneeds to be said in the case of C-initial roots. I assume that transcriptionis to the right.

(140)

[[ku-]]

k

×u

×[[lima]]

l

×i

×m

×a

×⊕UNINT−→

[[ku-]]

k

×u

×[[lima]]

[

l

×i

×m

×a

×

DC→

[[ku-]]

k

×u

×[[lima]]

[

l

×i

×m

×a

×]Trscr−→

[[ku-]]

k

×u

×[[lima]]

l

×i

×m

×a

× × × × ×

With the V-initial root ita, for example, juncture insertion (andsyllabification of the duplicant) produces (141). There is room forargument about the morphemic association of the first timing slot of

90 Chapter 6

the geminate. Since it does not bear on the issue at hand, I foregodiscussion.

(141)

k

×w

×i

× ×[t

×a

× ]

∗σ σ[[ku-]] [[ita]]

The initial syllable of the duplicant is is onsetless. Crucially, Kıheherequires well-formed duplicant syllable structure and specifies [-Left asthe mechanism to achieve it. There are two potential repairs:

(142) a.

k

× [

w

×i

× ×t

×a

× ]

σ σ[[ku-]] [[ita]]

b. [

k

×w

×i

× ×t

×a

× ]

σ σ[[ku-]] [[ita]]

The first repair yields kwi:ta-wi:ta and the second kwi:ta-kwi:ta. Iassume that Kıhehe chooses to align the duplicant with a morphemeboundary. In the discussion in Chapter 7 of Kinande, another Bantu lan-guage, very clear morpheme boundary effects in duplicant adjustmentwill be evident.

Alternatively, one might propose that Kıhehe repair chooses toalign the duplicant syllable structure with the stem syllable structure.Juncture insertion itself does not have access to the location of syllableboundaries in the stem in which the duplicant is embedded. But thediscussion of Yaqui reduplication, later in this section, will show thatprosodic adjustment can be sensitive to the prosodic structure of thewhole stem. So it could be that (142b) is the result of a prosodicadjustment process which aligns the duplicant syllable structure withthe stem syllable structure.

Duplicant syllabification does not necessarily supplant or obliteratestem syllable structure. They can coexist on separate tiers:

Prosodic Adjustment 91

(143)

k

×w

×i

× ×[t

×a

× ]

∗σ σ

σ σ

Prosodic adjustment rules are not universal. Different languagesuse different prosodic adjustment rules to accomplish the same prosodicobjectives. Kıhehe used [-Right to eliminate an onsetless syllable. InAsheninca Campa, discussed in detail later, [-Left is used in someenvironments.

(144)

n

×∗σ[

o

×σ

s

×a

×m

×σ

p

×i

×]PrAdj−→

n

×o

×[σ

s

×a

×m

×σ

p

×i

×] → nosampi sampi

The prefix n- is a first person marker, osampi ‘paddle’ is a verb root,and reduplication realizes a kind of intensive prefix, usually glossed as‘more and more’. Note that the result is not nosampinosampi, which iswhat [-Left would produce.

Asheninca Campa uses [-Right, as above, rather than [-Left, exceptwhen it conflicts with satisfaction of a second prosodic desideratum,the desirability of a bisyllabic duplicant. In (144), the duplicant iscontracted, but in similar V-initial cases in Asheninca Campa with ashort root, [-Left is used, so that the duplicant is augmented. Forexample, with the same prefix as (144) and the verb root asi ‘cover’:

(145)

n

×∗σ[

a

×σ

s

×i

×]PrAdj−→ [

σ

n

×a

×σ

s

×i

×] → nasi nasi

In Kıhehe, the duplicant is always augmented, even with long roots, asshown by (139d).

92 Chapter 6

6.2 Augmentation of the duplicant to a heavy syllable

The prototypical heavy syllable reduplication pattern is:

(146) a. lan.tu → lan.lan.tub. la:.nu → la:.la:.nuc. la.nu → lan.la.nu

Heavy syllable reduplication has the effect of concatenating a heavysyllable surface prefix to the stem, regardless of the weight of the initialsyllable of the stem. The term “surface prefix” is used in order todescribe the surface effect, but to emphasize that the vocabulary itemwhich results in the surface effect is not itself a prefix.

Heavy syllable reduplication attracted a great deal of attentionwhen reduplication was first examined in a generative framework be-cause it is clear from (146c) that the portion of the stem which is copiedis not necessarily a natural constituent of the base, consisting in (146c)of the initial syllable of the stem plus the onset of the following syllable.This played an important role in Marantz’s CV template theory (1982).Subsequently, Levin’s (1985) analysis of heavy syllable reduplicationin Mokilese and Hayes and Abad’s (1989) analysis of heavy syllablereduplication in Ilocano were important in establishing the inadequacyof CV templates for analyzing reduplication and supporting McCarthyand Prince’s (1986) theory of prosodic morphology.

Heavy syllable reduplication in four closely related Austronesianlanguages is analyzed in this section: Ilocano, Agta, Mokilese (twodialects), and Ponapean. A comparison of the differences betweenthese closely related languages is both instructive about the sourcesof variation from the core pattern given in (146) and, since there hasbeen extensive work on reduplication in these languages, useful inilluminating the special features of the DR theory of reduplication.In addition to Levin’s work on Mokilese and Hayes and Abad’s workon Ilocano, McCarthy’s (1984) and McCarthy and Prince’s (1986)work on Ponapean were important steps in the evolution of ProsodicMorphology. We begin with one of the Mokilese dialects, because thereduplication process in this dialect is the simplest. The source for adescription of Mokilese is Harrison (1973, 1976).

Prosodic Adjustment 93

6.2.1 Mokilese progressiveWe begin with relatively simple cases before considering some compli-cations. The analysis here takes the important analyses of Levin (1985)and McCarthy and Prince (1986) as a starting point.

(147) root progressive sourcepOdok pOd-pOdok ‘is planting’ [pOd]okkasO kas-kasO ‘is throwing’ [kas]Owadek wad-wadek ‘is reading’ [wad]ekwia wi:-wia ‘is doing’ [wi〈i〉]apouce po:-pouce ‘is connecting’ [po〈o〉]ucedauli da:-dauli ‘is passing by’ [da〈a〉]u

Assuming that the progressive forms are the result of transcribing repre-sentations with embedded t-junctures, the pretranscription sources canbe deduced without difficulty and are given in the last column.

The exponent of the progressive affix is null and and that it triggersthe C∗V-rule. Default Closure follows. The duplicant is then adjustedbefore transcription applies. The mechanics of the adjustment aresimply stated. ]-Right applies in the context C, otherwise FCVL(First Conjunct Vowel Lengthening) applies (see Section 4.4, p. 65).Two illustrative derivations follow:

(148) a.

k

×a

×s

×O

×C∗V-rule−→

k

×a

× ]

s

×o

×DC−→ [

σµ

k

×a

× ]

s

×o

×

]-Right−→ [

σµµ

k

×a

×s

× ]

o

×Transcription−→ × × ×

k

×a

×s

×o

×

b.

w

×i

×a

×C∗V-rule−→

w

×i

× ]

a

×DC−→ [

σµ

w

×i

× ]

a

×

FCVL−→ [

σµµ

w

×i

×〈×〉]a

×Transcription−→ × × ×

w

×i

×

94 Chapter 6

With an understanding of the mechanics, it is possible to under-stand the coherence of the adjustment process as being driven by aprosodic desideratum: the first conjunct of the duplicant must be abimoraic syllable. Using the notion of a defect driven rule (DDR)introduced in Frampton (2001), it is possible to incorporate rules drivenby prosodic desiderata into a derivational framework. The general formof such a rule is:

desideratum :: repair rule ; higher desiderata list

The desideratum and the higher desiderata are conditions which may ormay not hold of a particular representation. Informally, the DDR

d :: r ; (e1, . . . , en)

applies to a representation if the representation is is defective withrespect to the condition d and r can transform it into one that is notwithout introducing defects with respect to the specified more highlyranked desiderata. More formally, the DDR applies to α to produce β if:1) α does not satisfy d; 2) r applies to α to produce β; 3) β does satisfyd; and 4) β satisfies (e1, . . . , en) at least as well as α. The last conditionmeans that for any k such that α satisfies ek and β does not, there is aj < k such that β satisfies e j and α does not.

This is an extension of Sommerstein’s (1974) notion of a “phono-tactically driven rule”, shorn of the surface orientation of his phonology.It can be generalized to the notion of a desideratum driven rule schema(DDRS):

desiderata list :: repair rule list ; higher desiderata list

The rules for expanding a DDRS into an ordered list of DDRs aregiven in Appendix 2. The examples in this paper are simple and aninformal approach to the application of a DDRS will suffice. Choosethe rule application which makes the best possible repair. If two ruleapplications make equally good repairs, choose the highest ranked rule.DDRSs apply iteratively, with application stopping when no furtherrepair is possible.

It is often most natural to write a desideratum in bipartite form:

substructure ; condition

Prosodic Adjustment 95

The first term singles out some substructure of the representation and thesecond term specifies a condition which this substructure may or maynot satisfy. In these terms, the Mokilese rule for adjusting the duplicantcan be written:

(149) first conjunct ; bimoraic syllable ::[ ]-Right

FCVL

]

There is an imprecision in the terminology which needs to beclarified. The first conjunct, as it has been defined, is a subpart ofthe post-transcription structure. The schema (149) applies to the pre-transcription structure. The crucial point is that conjuncts can be easilyidentified and manipulated in the pretranscription structure. The firstconjunct is the residue in the duplicant of the left-edge truncate, if thereis one, otherwise it coincides with the duplicant. The second conjunctis the residue in the duplicant of the right-edge truncate, if there is one,otherwise it coincides with the duplicant. A nuanced terminology couldbe introduced, with “first conjunct” distinguished from “first conjunctof the duplicant.” I will refrain from this and rely on the reader todetermine the correct referents of uses of the term “first conjunct.”

The rule ordering in (149) is intrinsic. FCVL is always applicableto the result of C∗V-reduplication, since it always has a V-final du-plicant and application of FCVL always brings the representation intosatisfaction of the desideratum. ]-Right applies only in special cases.Before contrasting the formulation (149) with the formulation (148.3),we first must establish that (149) is empirically adequate. In the case ofbisyllabic wia, consider:

[σµ

w

×i

×]

a

×]-Right−→ [

σµ

w

×i

×σµ

a

×]

This is not a repair, because the output does not satisfy the desideratum.The alternative is:

[σµ

w

×i

×]

a

×FCVL−→ [

σµµ

w

×i

×〈×〉]a

× ,

This is a repair since the desired prosodic shape results.

96 Chapter 6

The examples with surface diphthongs (pouce, dauli) are lessclearcut. McCarthy and Prince speculate that at the relevant level ofsyllabification the surface vowel-glide sequence is in fact a bisyllabicvowel-vowel sequence. If so, then the same considerations that appliedto wia also apply to examples like pouce, which appear at the surfacewith a diphthong. The “relevant level” here is the syllabification of theduplicant, considered independently from the syllabification of the wordit is embedded in. Alternatively, it could be that the vowel sequences arediphthongs at the relevant level, but a somewhat modified DDR itselfis responsible for excluding diphthongs from the surface prefix. Themodified DDR would be:

first conjunct ; bimoraic syllable ::[ ]-Right

FCVL

]; *Diphthong

In this formulation, the exclusion of diphthongs from the duplicantis built into the DDR which establishes the material to be copied bytranscription.

The use of ]-Right in prosodic adjustment to bimoraicity is com-mon, but a language can choose not to use it. Consider, in this context,how reduplication in Mokilese would change if the rule ]-Right in (149)were lost. Then, FCVL would always apply, and the surface patterngenerated would be:

(150) stem progressivea. pOdok pO:-pOdokb. kasO ka:-kasOc. wadek wa:-wadekd. wia wi:-wiae. pouce po:-ouce

This pattern does in fact occur as a recent innovative dialect of Mok-ilese discussed by Blevins (1996).1 It is a straightforward example ofdiachronic change by rule loss. Prosodic adjustment has lost one of itsrepair options.

6.2.2 Complications in Mokilese progressive reduplicationThere are a few complications. Only one of them requires any amend-ment to the analysis above. The others simply require an explication

Prosodic Adjustment 97

of the interaction of some particularities of Mokilese phonology withreduplication. We need to start with one of these particularities, how-ever, because an understanding of it is needed before we can discuss themore serious complication.

6.2.2.1. CP-sequences which are not morpheme internalAlthough the Austronesian languages studied in this section have occa-sional morpheme internal CP sequences, such sequences are generallyeliminated otherwise. This affects reduplication of P-initial roots, sincethe surface prefix that reduplication creates will often be C-final, cre-ating a CP sequence which is not morpheme internal. Several differentmechanisms are used for repairing CP sequences that are created deriva-tionally. In Mokilese, the phoneme P is deleted (not its timing slot) andthe consonant spreads to the bare timing slot which remains, creating ageminate. The rule is:

C

× ××

P

In Agta (obligatorily) and Ilocano (optionally), the timing slot asso-ciated with the glottal stop is simply deleted. Resyllabification thentakes place. In Ponapean, the glottal stop is replaced by a y-glide.Resyllabification then takes place, vocalizing the glide.

The three different strategies that are employed are illustrated in(151), the starting point in each case is the product of heavy syllablereduplication of Palu:

σ

P

×a

×l

×σ

P

×a

×σ

l

×u

×

(151) Varieties of CP-elimination

Mokilese-A Agta, Ilocano Ponapean

Pal.Pa.lu Pal.Pa.lu Pal.Pa.luCP-elimination Pal.la.lu Pal.Pa.lu Pal.ya.luresyllabification Pa.la.lu Pa.li.a.lu

98 Chapter 6

Note the opacity of heavy syllable reduplication process in Agta, Ilo-cano, and Ponapean. All have light initial syllables on the surface.

As a consequence of CP→ C: in Mokilese, we have:

(152) stem progressive glossa. Ponop Pon-nonop *Pon-Ponop ‘is preparing’b. Palu Pal-lalu *Pal-Palu ‘is walking’

I will use “CP-reduction” as a general term for the process, whichtakes different forms in the different languages.

6.2.2.2. Double reduplicationThe surface pattern of the progressive forms of monosyllabic roots isstrikingly different than the pattern for bisyllabic roots.

(153) root progressivea. pa ‘weave’ pa:pa:-pa *pa:-pab. Pir ‘laugh’ Pirrir-rir (<PirPir-Pir) *Pir-rirc. ca:k‘bend’ ca:ca:-ca:k *ca:-ca:k

It is clear that reduplication is iterated. The issue is how this is enforcedby the morphology. I assume that there is allomorphy, with lexicalinsertion sensitive to the prosodic characteristics of the stem. In thecase of a monosyllabic root, juncture insertion ∅ → ]] /V appliesin place of ∅ → ] /V . Example (152a) is more subtle. It shows thatprosodic adjustment can be interwoven with transcription.

(154) JncIns pa → pa]]DC pa]] → [[pa]]PrAdj (FCVL) [[pa]] → [[pa〈a〉]]Inner Trscr [ [pa〈a〉] ] → paa [ pa ]PrAdj (FCVL) paa[pa] → paa[pa〈a〉]Outer Trscr paa [pa〈a〉] → paa paa pa

As in the illustrations of nested duplicant transcription in Chapter 4,the duplicant being transcribed is boxed on the left. On the right, thereduplicant is shaded and the remnant (what is left in the box) is boxed.

Prosodic Adjustment 99

The interweaving of prosodic adjustment and transcription in (154)suggests that prosodic adjustment should be viewed as a preconditionfor transcription and that the cyclic structure of the transcription ofnested duplicants induces a cycle of prosodic adjustment. Transcriptionis cyclic, working from the inside out. Inner duplicants are transcribedfirst. Each cycle of transcription is preceded by prosodic adjustment, ifnecessary.

This explains the mechanics of double reduplication in Mokilese.There is allomorphy, with a different juncture insertion rule used formonosyllabic roots. Finally, it is worth noting that Mokilese doublereduplication sharply illuminates the limitations of OT approaches tomorphophonology. It is clear that what is happening in Mokilese doublereduplication is that an operation is applying twice. OT has no meansfor even talking about something happening twice, because it rejects theidea of things happening in sequence.

6.2.2.3. Prenazalized geminatesThere is one last example which requires discussion.

(155) stem progressivePandip Pan-dandip *Pan-nandip (< Pan-Pandip) ‘is spitting’

If nd is taken to be a consonant sequence, the duplicant would extendto the second vowel and the prosodic adjustment rule proposed abovewould be inadequate. Raimy (2000), however, argued that nd is aprenasalized geminate. We adopt this proposal. The derivation is thenstraightforward.

(156) [σ

P

×a

×n

×d

× ×]i

×p

×T&R−→

σ

P

×a

× ×ð

×d

σ

P

×a

×n

×σ

d

× ×i

×p

×

CP-reduction−→

σ

P

×a

×n

×σ

d

× ×a

×n

×σ

d

× ×i

×p

×

100 Chapter 6

Order on the phoneme tier is not covertly used in the last step. ReplacingP with a link to d is the only way to satisfy the syllable conditions inMokilese.

6.2.3 AgtaIn Agta, CP-reduction is simply deletion of the P-slot. C becomes anonset after resyllabification.

(157) Agta plural and diminutive

root gloss reduplicated glossa. Patu ‘dog’ Pat-atu ‘puppy’b. balataN ‘girl’ bal-balataN ‘little girl’c. Pulu ‘head’ Pul-ulu (da) ‘(their) heads’d. Puffu ‘thigh’ Puf-uffu ‘thighs’

A representative derivation follows:

(158)

P

×a

×t

×u

× → [

P

×a

×]

t

×u

×PrAdj−→ [

σ

P

×a

×t

×]

u

×

T&R−→

P

×a

×t

×P

×a

×t

×u

×Syllab−→

σ

P

×a

×σ

t

×a

×σ

t

×u

×

Note the opacity in this derivation. The heavy syllable which is tran-scribed loses a mora in resyllabification and the output consists of lightsyllables. The prosodic condition which drives prosodic adjustment isnot satisfied at the surface.

6.2.4 IlocanoGiven the discussions above, most of the facts of Ilocano heavy syl-lable reduplication follow directly. We begin with the straightforwardexamples.

Prosodic Adjustment 101

(159) Ilocano plural and verbal progressive

a. kal diN ‘goat’ kal kal diN ‘goats’b. jyanitor ‘janitor’ jyan-jyanitor ‘janitors’c. pusa ‘cat’ pus-pusa ‘cats’d. trabaho ‘to work’ trab-trabaho ‘is working’e. saNgit ‘to cry’ saN-saNgit ‘is crying’f. Paso ‘dog’ Pas-Paso, Pas-aso ‘dogs’g. dait ‘to sew’ da:-daPit ‘is sewing’

Note the variant forms in (159f). CP sequences arising from tran-scription are only optionally eliminated. When they are, the derivationproceeds just as in Agta. Note also that ]-Right does not apply in (159g).This is the same restriction on ]-Right that was seen in Mokilese-A.Juncture shift is prohibited from bringing a vowel sequence into theduplicant. Instead, FCVL is used to augment the duplicant to a heavysyllable. The root internal glottal stop in (159g) is inserted by a late rule,distinct from cyclic glottal stop insertion at the left edge of vowel-initialstems.

There are two particularities of Ilocano heavy syllable reduplicationwhich need to be discussed. First, as in Mokilese, there is a quirk inprosodic adjustment to a bimoraic syllable duplicant when the stem ismonosyllabic.

(160) a. trak ‘truck’ tra:-trak (*trak-trak) ‘trucks’b. bas ‘bus’ ba:-bas (*bas-bas) ‘buses’c. nars ‘nurse’ na:-nars (*nar-nars) ‘nurses’

In these examples, ]-Right does not apply for some reason, so FCVLis used instead to augment the duplicant to a heavy syllable. Theconditions under which ]-Right is blocked are clear, but the motivationfor blocking ]-Right with monosyllabic stems is not.

Second, there is variation for polysyllabic stems with a consonant-glide complex onset.

(161) a. bwaya ‘crocodiles’ bu:-bwaya, bway-bwaya, bu-bwayab. pyano ‘pianos’ pi:-pyano, pyan-pyano, pi-pyanoc. dwa ‘two’ du:-dwa (no variation)

102 Chapter 6

The first two examples, with a three way variation in the surface form,are from Boersma and Hayes (2001), who undertake a major study ofhow variation of this kind can be analyzed in OT terms. Hayes andAbad (1989), from which (161c) is taken, note that there is no variationfor monosyllabic stems.

First, we consider the three-way variation in (161a,b). FollowingHayes and Abad, I assume that the y-glides in (161) are melodicallyidentical to i-vowels. Some of the variation in (161a,b) can be attributedto varying underlying syllabification. Consider (161b), for example,where y/i orthography will be used to make the syllabic role clearer.

(162) a.

σ

p

×i

×σ

a

×σ

n

×o

× b.

σ

p

×y

×a

×σ

n

×o

×

The two possibilities lead to the derivations below.

(163) a.

σ

p

×i

×σ

a

×σ

n

×o

× → [σ

p

×i

×]

a

×n

×o

×FCVL−→ [

σ

p

×i

× ×〈 〉]a

×n

×o

×

→ pii pi ano → pi:pyano

b.

σ

p

×y

×a

×σ

n

×o

× →σ

[

p

×y

×a

×]

n

×o

×]-Right−→

σ

[

p

×y

×a

×n

×]

o

×

→ pyan pyan o → pyanpyano

Note that ]-Right is blocked in (163a) because its application wouldproduce vowel hiatus in the duplicant.

The third variation, pipyano in (161b), is the result of complex onsetreduction which is independent of reduplication; possible after a heavyopen syllable.

Prosodic Adjustment 103

(164)

σ

p

×i

× ×σ

p

×y

×a

×σ

n

×o

× →σ

p

×i

× ×p

×σ

y

×a

×σ

n

×o

×

→σ

p

×i

×p

×σ

y

×a

×σ

n

×o

×

Compensatory shortening (like keep + t → kep-t in English verbalmorphophonology) follows resyllabification.

Finally, we can address the invariability of the reduplicated formof a monosyllabic stem with a consonantal-glide onset, (161c) above.Ilocano has a bimoraic minimal word phonotactic. This rules out un-derlying monomoraic (165a), so the reduplicative morpheme combinesonly with bimoraic (165b). The result is that only du:.dwa surfaces.

(165) a. *

σµ

d

×w

×a

× b.

σµ

d

×u

×σµ

a

×

6.2.5 PonapeanWe now turn to the complexities of durative reduplication in Ponapean,which were unraveled by McCarthy (1984) and McCarthy and Prince(1986), based on the data and analysis of Rehg and Sohl (1981). Again,we temporarily bypass the complexities of V-initial roots (P-initial at thesurface) and restrict our attention to C-initial roots. The puzzle is thatthe duplicant is sometimes bimoraic, sometimes monomoraic.

(166) a. pa pa:-pa ‘weave’b. mi mi:-ma ‘exist’c. mem mem-mem ‘sweet’d. kaN kaN-kaN ‘eat’e. du:p du-du:p ‘dive’ *du:-du:pf. mi:k mi-mi:k ‘suck’ *mi:-mi:kg. mand ma-mand ‘tame’ *man-mandh. leNk le-leNk ‘acrophobic’ * leN-leNk, *len-leNk

104 Chapter 6

The generalization which McCarthy extracted from this was that lightmonosyllable stems undergo heavy syllable reduplication, and thatheavy monosyllable stems undergo light syllable reduplication, with thecrucial insight that word final consonants do not count in determiningsyllable weight. He called this phenomenon “complementarity.”

In DR terms, the difference between (166a–d) and (166e–h) isthat in the later examples there is no prosodic adjustment. Specifyingthe class of stems for which prosodic adjustment is annulled is notstraightforward, as the following examples show:

(167) a. du:pek du:-du:pek ‘starved’b. nO:rok nO:-nO:rok ‘greedy’c. ma:sa:s ma-ma:sa:s ‘cleared of vegetation’d. to:ro:r to-to:ro:r ‘be independent’e. wa:ntu:ke wa-wa:ntu:ke ‘count’f. lia:n li:-lia:n ‘outgoing’

McCarthy deduced that the conditions under which the duplicantis monomoraic are related to the foot structure. We do not follow hisanalysis, but exploit its core idea. If footing is moraic, right to left, andfinal C is taken to be extraprosodic, as suggested by McCarthy, it isplausible that the foot structures of the examples (168) are as given inthe second column. Suppose also that the initial foot is taken to be thedomain of the r-affix. The derived prefix which is actually produced byheavy syllable reduplication is given in the last column.

(168) root reduplicated footstructure

initialfoot

surfaceprefix

kaN kaN-kaN (ka)N ka kaNmand ma-mand (man)d man madu:pek du:-du:pe (du:pe)k du:.pe du:ma:sa:s ma-ma:sa:s (ma:)(sa:)s ma: mawa:ntu:ke wa-wa:ntu:ke (wa:n)(tu:ke) wa:n walia:n li:-li:an (lia:)n li.a: li:

When the data is organized in this way, it is easy to see that thereis no prosodic adjustment when the initial foot is a heavy syllable. Thisaccounts for all of the examples above. What this accomplishes is to

Prosodic Adjustment 105

avoid reduplicated forms in which the two initial feet are prosodicallyidentical. The motivation for this is obscure. It stress were iambic, itmight be seen as a way to avoid stress clash. But footing patterns whichadmit heavy-light syllable feet rarely go along with iambic stress. Sincethe Ponapean footing and stress system is not well-known, I will leaveunresolved the question of why Ponapean reduplication is organized sothat sequences of two heavy syllable feet at the left edge are avoided.Note that sequences of two heavy syllables cab be generated (du:-du:pek

for example), but only when the second heavy syllable is footed with alight syllable to its right.

Ponapean has a number of verb roots with an initial nasal geminateor prenasalized geminate. The first portion of the geminate surfacesas a syllable nucleus. Two derivations follow, with DUR the durativemorpheme:

(169) a.

σ

P

×m

× ×σ

e

×d

×⊕DUR−→

σµ

[

P

×m

× ×]e

×d

×PrAdj−→

σµµ

[

P

×m

× ×]

e

×d

×

T&R−→

σ

P

×m

× ×σ

P

×σ

m

× ×e

×d

×CP-Repair−→

σ

P

×m

× ×σ

y

×m

× ×σ

e

×d

×

resyllab.−→

σ

P

×m

× ×σ

i

×m

× ×σ

e

×d

× (Pm.mim.med)

b.

σ

P

×Nd

× ×σ

a

×⊕DUR−→

σµ

[

P

×Nd

× ×]a

×PrAdj−→

σµµ

[

P

×Nd

× × ]

a

×

T&R−→

σ

P

×Nd

× ×σ

P

×Nd

× ×σ

a

×CP-Repair−→

σ

P

×Nd

× ×σ

y

×Nd

× ×σ

a

×

resyllab.−→

σ

P

×Nd

× ×σ

i

×Nd

× ×σ

a

× (Pn.din.da)

106 Chapter 6

Finally, there are processes at the reduplicant-remnant boundarywhich must be accounted for. In the examples below, taken from Davis(2002), the root has the form C1VC2 . . . and, according to the analysisdeveloped up to this point, we expect a reduplicated form:

C1VC2.C1VC2 . . .

The expectation is fulfilled if C1 = C2 and they are both sonorant.

(170) root expected actual

rer ‘tremble’ rer-rer rer-rerlal ‘make a sound’ lal-lal lal-lal

If C1 and C2 are not identical, or if they are not both sonorant, amodification occurs:

(171) root expected actual

2. If C1 is a coronal sonorant and C2 is a coronal sonorantliro:ro ‘protective’ lir-liro:ro lil-liro:ro

nur ‘contract’ nur-nur nun-nur

3. else if C1 = C2

tit ‘build a wall’ tit-tit tin-tit

kak ‘able’ kak-kak kan-kak

4. else if C1 is a coronal obstruent and C2 is a coronal sonoranttar ‘strike (of fish)’ tar-tar tan-tar

tilep ‘mend a root’ til-tilep tin-tilep

5. elsetep ‘kick’ tep-tep tepe-tep

tep ‘begin’ tep-tep tepi-tep

ker ‘flow’ ker-ker kere-ker

Rehg and Sohl (1981) give an account of (171) that depends cru-cially on two reduplication specific rules operating at the reduplicant-remnant boundary.2 The reduplication specificity suggests that theanomalies may be an aspect of NCC repair. Indeed, if we assumethat Shortcut Repair (see Section 5.3) occurs at the reduplicant-remnantboundary in (171.1–4), we can account for all the data in (171). Thisrequires Shortcut Repair to apply if C1 and C2 are identical, or if theyare both coronal and C2 is sonorant.

Prosodic Adjustment 107

We examine the cases in (171) in turn. First, note that in (171.1),NCC repair produces a geminate coda-onset sequence, rather than asequence of identical consonants. The examples in (171.2) require nocomment, since they are the expected result of Shortcut Repair. Inthe remaining cases, Ponapean syllable structure repair rules come intoplay. Ponapean requires codas which are not word final to be sonorantswhich share place with a following onset. If a coda-onset sequence isan sonorant geminate, as it is in the examples in (171.2), no repair isneeded. If it is an obstruent geminate, the syllable structure is repairedby nasalizing the geminate, producing a prenasalized geminate..

α

× × →n

×α

× ×

Shortcut repair produces such a geminate in the examples in (170.3)and (170.4). Syllable repair accounts for the surface form. A typicalderivation is given below.

(172) × × ×t

×i

×t

× →σ

t

×i

× ×σ

t

×i

×t

× →σ

t

×i

× ×n

×σ

t

×i

×t

×

In the remaining cases, although there is a violation of the syllablewell-formedness conditions, Shortcut Repair cannot apply. Instead,repair is vowel epenthesis, as illustrated below:

(173) × × ×t

×e

×p

× →σ

t

×e

×p

×σ

t

×e

×p

× →σ

t

×e

×σ

p

×V

×σ

t

×e

×p

×

Determining the quality of the inserted vowel is a complex issue, subjectto some lexical specification. It is usually, but not always, a copy of thefollowing vowel. See Rehg and Sohl, p. 91. The different outcomes fortep ‘kick’ and tep ‘begin’ in (171.5) demonstrate the need for at leastsome lexical specification.

108 Chapter 6

6.3 C-finality as a secondary prosodic desideratum

The core of weight based stress theory is built around mora count.But secondary weight distinctions play a role in some stress systems.3

The same is true of reduplicative prosodic adjustment. This sectiondiscusses C-finality as a secondary prosodic desideratum in Lardil,Yaqui, and Nuu-chah-nulth. It all three cases, coda consonants do notcontribute to syllable weight, but reduplicants are C-final when possible.

6.3.1 Lardil iterative reduplicationThe key data is given below:

(174) Lardil iterative reduplication

root 4 iterative glossa. kele kele-kele ‘cut’b. la la:-la ‘spear’c. Na:li Na:l-Na:li ‘thirst’d. pareli parel-pareli ‘gather’

On the basis of minimal word and stress considerations, Wilkinson(1988) showed that Lardil footing is mora based, with CV(C) syllablesmonomoraic and CVV(C) syllables bimoraic. The affix which realizesiterative has a null exponent, specifies the initial moraic foot as itsdomain, and triggers the ×∗-rule. After juncture insertion and defaultclosure, but before prosodic adjustment, the structures of the examplesin (174) are:

(175) a.

σµ

[

k

×e

×σµ

l

×e

×] b.

σµ

[

p

×a

×σµ

r

×e

×]σµ

l

×i

×

c. [σµµ

n

×a

× ×]

l

×i

× d.

σµ

[

l

×a

×]

The duplicant is then adjusted to bimoraicity by a prosodic adjust-ment rule very similar to the prosodic adjustment rule used in the Aus-tronesian languages analyzed above. The primary prosodic desideratum

Prosodic Adjustment 109

is the same and the repair rules are the same. The one difference is thatthere is a secondary desideratum, C-finality. Transcription is to the left,the default for left-edge duplicants. The prosodic adjustment rule is:

(176) First Conjunct ;

[bimoraic

C-final

]::

[]-Right

FCVL

]

Prosodic adjustment in (175) produces:

(177) a.

σµ

[

k

×e

×σµ

l

×e

×] b.

σµ

[

p

×a

×σµ

r

×e

×l

×]

i

×

c. [σµµ

n

×a

× ×l

×]

i

× d.

σµµ

[

l

×a

× ×〈 〉]

In (175a), the duplicant is bimoraic, but not C-final. No prosodicadjustment is possible, however, because neither of the repair rulesproduces a C-final first conjunct. The duplicant satisfies the bimoraicityweight desideratum in (175b) and (175c), but is not C-final in eithercase. ]-Right produces the desired C-final reduplicant, as shown in(177b) and (177c), without violating the primary bimoraicity desidera-tum. The representation (175d) satisfies neither the bimoraicity desider-atum nor the C-final desideratum. First Conjunct Vowel Lengtheningproduces (177d), which satisfies the primary desideratum, but no repairis available which could produce C-finality.

6.3.2 Adjustment to C-finality in Nuu-chah-nulthThe reduplication pattern triggered by Class I-1 suffixes in Nuu-chah-nulth (repeated from Section 3.5; see also (113) in Chapter 5) is illus-trated below, with the suffix -ń∼a.

(178) yacmil∼ → yac-yacmil∼-ń∼a , Pu:s→ Pu:s-Pu:s-ń∼a

110 Chapter 6

The suffix triggers the C∗ν-rule and, after Default Closure, the duplicantis adjusted to C-finality using ]-Right.

(179) C∗ν-rule ya]cmil∼-ń∼a Pu:]s-ń∼aDC [ya]cmil∼-ń∼a [Pu:]s-ń∼aPrAdj [yac]mil∼-ń∼a [Pu:s]-ń∼aTrscr yac yacmil∼-ń∼a Pu:s Pu:s-ń∼a

6.3.3 C-finality in Yaqui light syllable reduplicationIn Section 4.4.1.1, various Yaqui reduplicative verbal affixes were in-troduced. Consideration of one variety, light syllable reduplication, wasrestricted to stems whose initial syllable is CV, which is doubled by thereduplication process. Some examples are repeated here:

(180) stem reduplicateda. vu.sa vu.vu.sa ‘awaken’b. he.wi.te he.he.wi.te ‘agree’c. ko.a.rek ko.ko.a.rek ‘wear a skirt’d. chi.ke chi.chi.ke ‘comb one’s hair’

If a more complete range of stems is considered, there are veryinteresting complications.

(181) stem reduplicatedCVC initial syllablea. chuk.ta chuk.chul.ta ‘cut with a knife’b. chep.ta chep.chep.ta ‘jump over’c. bwal.ko.te bwal.bwal.ko.te ‘soften, smooth’d. vui.te vui.vui.te ‘run’

CVV initial syllablee. waa.te wa-waate ‘want’f. kaa.te ka-kaate ‘walk’

CVV initial syllableg. kaa.te ka-kate ‘build a house’h. waa.te wa-wate ‘remember’

We first peel away the aspects of (181) that are the consequence ofgeneral features of Yaqui phonology and unrelated to the reduplication

Prosodic Adjustment 111

process itself. Accent is on the second mora in Yaqui unless the rootis lexically specified as initial accenting, in which case it is on theinitial mora. Reduplicated forms inherit the accentual type of the stem.This is responsible for the accent placement in (181e-h). Long vowelsshorten unless one of their legs is accented, producing a short vowelin the remnant in (181g-h). Finally, the glide status of i in (181d) wasconfirmed by Demers, Escalante, and Jelinik (1999), on the basis ofspectrographic analysis.

What remains for the theory of reduplication to account for is thecore paradigm:

(182) stem reduplicateda. chi.ke chi-chi.keb. chuk.ta chuk-chuk.tac. kaa.te ka-kaa.te

On the basis of (182a) and (182b), we can surmise (along with Yaquichildren learning their language) that juncture insertion is the C∗V-ruleand that (182b) must be the result of prosodic adjustment. On the basisof their analysis of the accentual system, Demers, Escalante, and Jelinik(1999) concluded that CV(C) were light and CVV(C) syllables wereheavy. Prosodic adjustment is therefore to C-finality, just as in Lardiland Nuu-chah-nulth Class-1 reduplication. The problem is to accountfor why ]-Right does not apply in (182a).

Prosodic adjustment can be sensitive to the prosodic structure of thestem that the duplicant is embedded in. Examples are rare, but Ponapeanprovides one clear example. Yaqui provides a second example. Prosodicadjustment is constrained so that duplicant is contained in the initialsyllable of the stem. Call the relevant constraint the Initial SyllableDuplicant Condition (ISDC): the duplicant must be contained in theinitial syllable of the stem. We can then write the prosodic adjustmentrule as:

(183) First Conjunct ; C-final :: ]-Right ; ISDC

C-finality as a secondary prosodic desideratum: All the examples of C-finality as a first conjunct prosodic desideratum which were consideredabove are from languages in which coda consonants are nonmoraic. I

112 Chapter 6

will assume that this is not accidental and follows from the primacy ofweight in prosodic adjustment. Specifically, I assume that:

(184) Primacy of weight desiderata in prosodic adjustment

a. prosodic adjustment never alters prosodic weight except tomeet a prosodic weight desideratum; and

b. prosodic weight desiderata always take precedence overother prosodic desiderata.

Taken together, these assumptions prevent adjustment to C-finality fromplaying a role in prosodic adjustment in languages in which codaconsonants add syllable weight.

6.3.4 Moravcik’s Generalization reconsideredConsider a hypothetical “Yaqui-A”, which differs only from Yaqui inthat the reduplication process corresponding to Yaqui light syllablereduplication is based on the C∗ν-rule rather than the C∗V-rule. Theresult is first syllable reduplication.

(185) Yaqui-A (hypothetical variant of Yaqui)C∗V-rule PrAdj

a. [chi].ke chi chike

b. [chu]k.ta [chuk]ta chuk chukta

c. [kaa].te ka: ka:te

Prosodic adjustment is blocked in (185a,c) by the ISDC.No examples of first syllable reduplication have been attested, as

far as I know. Does the fact that DR allows Yaqui-A present a problem?Yaqui light syllable reduplication (built around the C∗V-rule) is unusualin two respects. In the first place, it employs adjustment of the redupli-cant to C-finality, one of only a few languages to do so. In the secondplace, prosodic adjustment is sensitive to the prosodic structure of thestem that the duplicant is embedded in, one of only a few languagesin which prosodic adjustment has this property. Normally, prosodicadjustment incorporates phonemic material into the duplicant withoutregard for the prosodic structure which that material is found in. Yaquiand Ponapean are the only two examples of this kind that I am aware

Prosodic Adjustment 113

of. Use of the C∗ν-rule is also rare. Nuu-chah-nulth, Tohono O’odhamand Sanskrit are the only examples in this book. Bear in mind that, forobvious reasons, unusual reduplication processes are significantly overrepresented among the languages considered in this book. A languagelike hypothetical Yaqui-A would therefore have to combine an array ofhighly marked options. It is therefore not a surprise that no reduplicationprocess like the one considered in hypothetical Yaqui-A has, as yet, beenuncovered.

It may have occurred to some readers that the unusual FSDCconstraint might also permit first syllable reduplication in a languagein which coda consonants are moraic. Consider now the possibility of ahypothetical Yaqui-B, which differs from Yaqui in that coda consonantsare moraic. Suppose that reduplication is based on the C∗V-rule, andthat a bimoraic syllable desideratum is imposed on the duplicant, with]-Right the only repair rule. These assumption would lead to theparadigm:

(186) Yaqui-B (hypothetical variant of Yaqui)C∗V-rule PrAdj

a. [chi].ke chi chike

b. [chu]k.ta [chuk]ta chuk chukta

c. [ka]a.te [kaa].te ka: ka:te

The duplicant does not satisfy the bimoraicity desideratum in (186a),but prosodic adjustment (limited to ]-Right) cannot apply because ofthe ISDC. As far as I know, no such reduplication system is attested.The reason, I believe, is that FCVL is universally available as a repairoption. If so then [ta]ku in (186a) would be adjusted to [ta〈a〉]ku, whichwould produce ta: taku, in compliance with the FSDC.

In place of (186), the paradigm would be:

(187) Yaqui-C (hypothetical variant of Yaqui)C∗V-rule PrAdj

a. [chi].ke [chi〈i〉].ke chi: chike

b. [chu]k.ta [chuk]ta chuk chukta

c. [ka]a.te [kaa].te ka: ka:te

114 Chapter 6

If more languages are uncovered which subject duplicant prosodicadjustment to the ISDC, and in which coda consonants are moraic, weexpect that (187) might be found.

These considerations go a long way towards clarifying the status ofMoravcik’s Generalization. Moravcik was clear about the fact that shewas proposing a hypothesis about mechanism, not an empirical gener-alization. She realized that in certain cases, reduplicants were alwaysidentical to prosodic constituents, CV reduplication in a language inwhich all syllables were CV, for example. The issue for her was thecharacteristics of the stem that must be available in the description ofthe rules which generate the reduplicant. Moravcik concluded that itwas possible to specify reduplicants without recourse to stem prosodicconstituency. The conclusion reached here is that Moravcik was par-tially correct. The complication that DR introduces is that the surfaceshape of the reduplicant is determined by multiple factors: mainly rulesof juncture insertion and rules of prosodic adjustment. The former ruleshave the characteristic that Moravcik pointed to; they do not refer toprosodic constituency, although the domain in which they apply can bedetermined by foot boundaries. As Yaqui and Ponapean demonstrate,prosodic adjustment does not.5

6.4 Heavy syllable suffixation of left-edge material (Chukchee)

It should be unnecessary to say at this point (but will be repeated so thatthere are no grounds for confusion) that the term “suffixation” is used inthe title of this subsection only descriptively. Morphologically, there isno suffix, which is a term relevant to the concatenation of the exponentof an affix. The surface affix in this case is the product of reduplication.

It is straightforward to express the Chukchee pattern (76) fromChapter 4 in terms of t-junctures.

(188) Chukchee absolutive singularroot reduplicated pretranscriptionjilPe jilPe-jil [ jil〈Pe〉]nute nute-nut [nut〈e〉]

Transcription is to the right.Simply writing the pretranscription representation is not sufficient.

An account is needed for how it is derived using the mechanisms which

Prosodic Adjustment 115

the theories makes available. Prosodic adjustment is necessary. Thejuncture insertion rules are given in (189a) and the prosodic adjustmentrule is given in (189b). The exponent of the reduplicative morpheme isnull. Transcription is to the right.

(189) a. ∅ → 〈 /V ; ×∗-rule

b. Second Conjunct ; Bimoraic :: 〈-Right

Two derivations follow:

(190) jilPe nuteJncIns [ji〈lPe [nu〈teDC [ji〈lPe〉] [nu〈te〉]PrAdj [jil〈Pe〉] [nut〈e〉]Trscr jilPe jil nute nut

6.5 Final syllable prefixation (Madurese)

It is remarkable that although initial syllable prefixation is not attested,final syllable prefixation is. If the Madurese pattern (74) from Chapter 4is expressed in terms of t-junctures, we have:

(191) Madurese pluralroot reduplicated pretranscription

bu.waP waP-buwaP [〈bu〉waP]ga.ra.dus dus-garadus [〈gara〉dus]a.bit bit-abit [〈a〉bit]

Transcription is to the left.The reduplicative affix has a null exponent and triggers ×∗-rule and

∅ → 〉 /Right V. Prosodic adjustment follows, with a well-formedsyllable structure (WFSS) desideratum imposed on the first conjunct.The single repair rule is 〉-Left. The derivation in (192) is typical.

116 Chapter 6

(192)

b

×u

×w

×a

×P

×JncIns−→ [

b

×u

×w

×〉a

×P

×DC→ [〈

b

×u

×w

×〉∗σ

a

×P

×]

〉-Left−→ [〈

b

×u

×〉σ

w

×a

×P

×] → waP buwaP

Although the effect is duplication of the final syllable, the content of theduplicant is not determined on that basis. Juncture insertion isolates therime of the final syllable. The first conjunct is augmented in order toeliminate the onsetless syllable in the first conjunct of the duplicant.

6.6 Final syllable suffixation (Kaingang)

Whereas Madurese has final syllable prefixation, Steriade (1988), dis-cusses Kaingang final syllable suffixation, based on Wiesemann (1972).

(193) stem reduplicated pretranscriptionje.mı je.mı-mı je[mı]jen.gag jen.gag-gag jen[gag]

The affix which induces this pattern has a null exponent and triggers∅ → [ /Right V. Prosodic adjustment to well-formed duplicantsyllable structure occurs, using the repair rule [-Right.

An illustrative derivation follows:

(194)

j

×e

×n

×g

×a

×g

×JncIns−→

j

×e

×n

×g

×[

a

×g

×DC→

j

×e

×n

×g

×∗σ

[

a

×g

×]

[-Left−→

j

×e

×n

×[σ

g

×a

×g

×] → jengag gag

Just as in Madurese, the final syllable is isolated. But the content of theduplicant is not determined on that basis.

Prosodic Adjustment 117

6.7 Korean consecutive syllable reduplication

The following reduplication pattern was discussed in Section 2.2.3.

(195) stem reduplicated pretranscriptiona. cason ca-ca-son-son [ca][son]b. kikwe ki-ki-kwe-kwe [ki][kwe]c. hy@nsak hy@n-hy@n-sak-s’ak [hy@n][sak]

Note in (195c) that the internal junctures fall at the syllable boundary.It is possible that these Sino-Korean words are bimorphemic in a sensethat would allow the juncture insertion rules to correctly position thejunctures. Juncture insertion can be sensitive to morpheme boundaries,but not syllable boundaries. We shall see in a future section on Kinandethat juncture insertion can be sensitive to “frozen morpheme bound-aries” in roots that are not obviously formed by combining morphemesthat enter into the usual word formation processes. This is not unusual.It is likely that English speakers parse the word cranberry with aninternal morpheme boundary in spite of the fact that they have no otheraccess to a morpheme whose exponent is cran.

It is interesting, nevertheless, to consider whether prosodic ad-justment can account for (195), without an appeal to polymorphemicstems. Suppose that the specification of the affix which induces Koreanconsecutive syllable reduplication is that it has a null exponent andinduces ∅ → ][ /Right V. Prosodic adjustment of the duplicants towell-formed syllable structure follows, using juncture shift.

The crucial derivational step is:

(196) [σ

h

×y

×@

×n

×s

×∗σ

a

×k

× ]][PrAdj−→ [

σ

h

×y

×@

×n

×σ

s

×a

×k

× ]][

→ hy@n hy@nsak sak

6.8 The onset-coda asymmetry and Moravcik’s Generalization

Systematically examining the various examples of prosodic adjustmentabove has had a dual purpose. First, of course, was to show that

118 Chapter 6

reduplication processes discussed in earlier chapters are consistent withthe restricted framework of juncture insertion and prosodic adjustmentthat has been proposed. Second was to show that the possibility ofprosodic adjustment did not inadvertently predict the occurrence ofunattested reduplication patterns.

In the Kaingang, Madurese, and Korean examples above, the finalsyllable was isolated by first isolating the final VC∗, then adjusting itto well-formed syllable structure. The initial syllable cannot be isolatedin this way. The initial CV∗ can be isolated, but it has perfectly well-formed syllable structure. A coda consonant can be attached to CV∗only on the basis of weight. This yields the heavy syllable (prefixal)reduplication we analyzed in Mokilese and related languages, on theone hand, or heavy syllable (suffixal) reduplication we analyzed inChukchee. Heavy syllable prefixal reduplication (Mokilese) comesfrom isolating initial C∗V as the first conjunct of the duplicant and ad-justing it to be a bimoraic syllable. Heavy syllable suffixal reduplication(Chukchee) comes from isolating initial C∗V as the second conjunct ofthe duplicant and adjusting it to be a bimoraic syllable.

6.9 Towards a parametric theory of prosodic adjustment

The discussion of prosodic adjustment has assumed implicitly thatprosodic adjustment is a specific process, not simply a descriptive namefor whatever rules happen to alter the prosodic structure of the duplicantor one of its conjuncts. A specific form for the rule which carries outprosodic adjustment was proposed, a defect driven rule schema. Thepurpose of this section is to make these assumption explicit, refine them,and provide some justification for them.

The main justification for thinking that prosodic adjustment ismore than merely a name for what happens, is the narrow range ofpossibilities that is encountered. Consider again the rule which wasinitially identified as empirically adequate to account for the Mokilesereduplication paradigm.

(197) ]-Right/

C, otherwise FCVL

Prosodic Adjustment 119

Suppose that the otherwise clause was dropped, and the rule reduced to]-Right / C. The surface paradigm that would result would be:

(198) root reduplicatedpOdok pOd-pOdokkasO kas-kasOwia wi-wiako:kO ko-ko:kO

As far as I can determine, such a paradigm is unknown in a languagewhich, like Mokilese, distinguishes the syllable weight of C∗V andC∗VC syllables. This demands some account.

A reasonable conjecture is that over and above the specific ruleswhich alter the prosodic structure of the target of adjustment (theduplicant or one of its conjuncts), there are prosodic conditions whichthose rules subserve. Given some assumptions about adequate defaultadjustment rules, the paradigm (198) would then be ruled out becausethere are no coherent prosodic conditions which govern the adjustment.The matter is entirely analogous to the process which builds syllablestructure. Various rules (projecting nuclei, attaching codas, etc.) arebound together in a way that serves a collection of well-formednessconditions imposed on syllable structure. In Frampton (2001) I showhow this can be accomplished for syllable structure in a general wayby extending ideas of rule schemata in Chomsky and Halle (1968) tothe notion of a defect driven rule schema. It will be clear in the casestudies that follow that prosodic adjustment must employ an array ofrules and that it must be iterative. Defect driven rule schemata offer ameans to combine various subrules into a single complex rule which canbe iterated.

The specific rules which are designated for prosodic adjustmentsubserve the prosodic desiderata, but their particular form has a majorimpact on the form that adjustment takes. We saw this already in theMokilese example. If juncture shift is available to build a heavy syllablefirst conjunct, the standard surface pattern results. If it is not, a differentsurface pattern results: the pattern of the innovative dialect of Mokilesethat was illustrated in (150) above.

If we assume that C-finality, in languages in which coda consonantsdo not contribute to syllable weight, can be taken to be a syllable

120 Chapter 6

structure condition for reduplication, then all the examples of prosodicadjustment considered up to this point have the general form:

(199) target ;

[syllable structure conditions(

weight condition) ]

:: adjustment rule list ; constraints on adjustment

The various case studies in the next chapter will not require any revisionof (199).

It does little to constrain the system to maintain that there is anoption of carrying out prosodic adjustment using an operation of theform (199) unless other rules cannot carry out similar operations. I willassume that the prosodic adjustment schema (199) is uniquely privilegedto manipulate t-junctures (insert them and shift their position). Prosodicadjustment is, in some sense, an extension of the morphology of t-juncture insertion into the phonology. Manipulating t-junctures is not anoption for arbitrary phonological rules. The situation is very much thesame with, for example, parameterized theories of footing such as Halleand Idsardi (1995). If other rules were capable of manipulating footingdelimiters, outside the context of the tightly parameterized theory whichis developed there, the force of the theory would be lost.

The hardest aspect of (199) to constrain is the list of constraints onrepair rules. In Yaqui, the duplicant was required to be contained in thefirst syllable of the stem. In Ponapean, the duplicant can not be heavyif the initial foot of the stem consists of a heavy syllable. We will see inKinande that t-junctures must be placed at morpheme boundaries. Thisis a heterogeneous list. All the restrictions are plausible, but it is noteasy to find the common thread. A condition like “the duplicant cannotbe heavy if the initial syllable of the stem is light” would immediatelyallow syllable copy reduplication. While it is not hard to draw somedistinction between this condition and the condition which contrainsPonapean prosodic adjustment, the distinction remains ad hoc withouta definition of “possible condition on repair”. I am unable to providethis. Ultimately, this is what is needed, both from the theoretical pointof view and from the point of view of the language learner.

If the proposal (199) about the form of prosodic adjustment iscorrect, then it is a major residue of Prosodic Morphology in DR.

Prosodic Adjustment 121

The restriction of the domain of a reduplicative affix to a foot is thesecond major residue. It is important to emphasize, however, that theDR analysis of most reduplicative processes involves neither domaindesignation nor prosodic adjustment.

Chapter 7

Case studies

At this point, all the tools are in hand. We proceed to analyze anumber of complex examples. The first three, Ndebele and Kinande un-intensive reduplication and Asheninca Campa intensive reduplication,were chosen not only because of their intrinsic interest, but becausethey have been extensively analyzed in very different frameworks.McCarthy and Prince (1998) claim that derivational phonology is aninadequate framework for understanding the complexities of AshenincaCampa intensive reduplication. They rank it with Malay reduplication,which we have already analyzed, as a demonstration of the need forrepresentational theories of phonology. Inkelas and Zoll (2000) claimthat Ndebele unintensive reduplication demonstrates that apparentlyduplicated phonological material is not in fact the result of copying inthe phonology. We will show that both of these claims are unfounded.

Three prominent examples from Raimy (2000) are included (fromTohono O’odham plural reduplication, Temiar continuative reduplca-tion, and Chaha intensive reduplication) so that the reader can assessthe differences with and similarities to his treatment. A thoroughtreatment of Sanskrit verbal reduplication is included so that readers canmake a similar assessment with respect to Steriade’s (1988) well-knowndiscussion of perfect and intensive reduplication in Sanskrit.

Case Studies: Ndebele 123

7.1 Ndebele unintensive reduplication

This section relies almost completely on the data and description inHyman, Inkelas, and Sibanda (1999).1 The strong point of their analysisis the careful attention they pay to the cyclic morphology and its impor-tance in an analysis of reduplication in Ndebele. But their contentionthat the Ndebele facts show that reduplication is something other thanphonological copying is unsupported.

In Ndebele, verbal reduplication has unintensive semantics. It istranslated as ‘a bit’ or ‘here and there.’ Roughly, it is the differencebetween ‘he ate his food’ and ‘he picked at his food.’ Some preliminaryexamples are given in (200). The suffix -el is an applicative affix.

(200) a. uku-zi-nambith-el-a ‘to taste for’uku-zi-nambi-nambith-el-a ‘to taste for, a bit’

b. uku-zi-lim-el-a ‘to cultivate for’uku-zi-lime-lim-el-auku-zi-lima-lim-el-a

}‘to cultivate for, a bit’

The structure of the part of Ndbele verb forms which is relevant toreduplication is given in (201a). The possible extension suffixes arecausative, applicative, reciprocal, and passive. The parse of (200a) isgiven in (201b). The verb root in (200b) is lim ‘cultivate’.

(201) a. . . . + object agreement + root + extension suffixes + . . .

b. uku zi nambith el a

NONFIN CLASS 10 taste APP FV

The status of the final vowel (FV) in (202) is not clear. It could bethe default realization of a morpheme, or perhaps simply an epentheticvowel inserted for syllabification reasons. Resolution of this questionwill not be necessary for what follows.

The forms in (200) suggest that juncture insertion is total root redu-plication. Subsequently, there is prosodic adjustment to first conjunctbisyllabicity, with variant adjustment rules being responsible for the

124 Chapter 7

variation in (200b). The reduplicated forms in (200) are derived asfollows:

(202) zi-nambith-el zi-lim-el zi-lim-el⊕UNINT zi-[nambith]-el zi-[lim]-el zi-[lim]-elPrAdj zi-[nambi]th-el zi-[lim-e]l zi-[lim〈a〉]-elTrscr zi- nambi nambith-el zi- lime lim-el zi- lima lim-el

If the duplicant is subbisyllabic, Ndebele has the option carrying outprosodic adjustment using either ]-juncture shift or First ConjunctVowel Epenthesis (FCVE). The latter operation is very similar to FirstConjunct Vowel Lengthening in Mokilese and the other Austronesianlanguages analyzed in Chapter 6. Mokilese and Ndebele are comparedbelow:

(203) a. Mokilese

[σµ

p

×O

× ]

d

×o

×k

×PrAdj−→ [

σµµ

p

×O

×d

× ]

o

×k

× → pOd pOdok

b. Mokilese (innovative dialect, see p. 96)

[σµ

p

×O

× ]

d

×o

×k

×PrAdj−→ [

σµµ

p

×O

×〈×〉]d

×o

×k

× → pOO pOdok

c. Ndebele

[σ

l

×i

×m

× ]

e

×l

×PrAdj−→ [

σ

l

×i

×σ

m

×e

× ]

l

× → lime limel

[σ

l

×i

×m

× ]

e

×l

×PrAdj−→ [

σ

l

×i

×σ

m

×〈a

×〉]e

×l

× → lima limel

In Mokilese, the prosodic target is bimoraicity; in Ndebele, it is bisyl-labicity. The target prosodic shape in both cases can be achieved eitherby ]-Right or epenthesis. In the innovative dialect of Mokilese, a timing

Case Studies: Ndebele 125

slot is epenthesized, truncated, and its phonemic value is filled in byspreading. In Ndebele, a timing slot is epenthesized, truncated, and itsphonemic value is filled in with an epenthetic a. Both repair operationsare available in standard Mokilese, but there is rule ordering, so that]-Right is used where possible. In the innovative dialect, ]-Right is notavailable, so FCVE is forced. These repair rules are not ordered inNdebele, so either one can be used for prosodic adjustment.

It is implicit in (202) that UNINT is realized after the the applicativemorpheme has been realized by -es. This assumption is necessary toget [lim]-el → [lim-e]l since we have been assuming that prosodicadjustment and transcription takes place cyclically. We will shortlymake the assumptions about the hierarchical position of UNINT moreprecise.

There is variation in the reduplicated forms in (200b) becausethe root is sub-bisyllabic. For consonantal roots there is even morevariation.

(204) uku-zi-dl-is-el-a ‘to cause to eat for’uku-zi-dlise-dl-is-el-auku-zi-dlisa-dl-is-el-auku-zidli-zi-dl-is-el-auku-zidla-zi-dl-is-el-a

‘to cause to eat for, a bit’

The four-way variation in (204) is easily accounted for if theinitial juncture insertion rule is somewhat more general than total rootreduplication. Assume that the initial duplicant can include a prefix orsuffix which is adjacent to the root, provided that the duplicant is notthereby made super-bisyllabic. Then both zi-[dl-is]-el and [zi-dl]-is-el

are possible initial configurations, and there are two ways of augmentingeach duplicant to bisyllabicity, ]-Right and FCVE, producing the four-way variation.

Since it will be important in what follows, we need to make theassumptions about the hierarchical position of UNINT precise. Thesyntactic structure relevant to the inner portion of an unreduplicatedverb form, the boxed part of (201a), is (205).

126 Chapter 7

(205)

root E1

. . .En

Agr

The morphemes Ej are those which are realized by extension suffixesand Agr is realized by an object agreement prefix (which might benull). The representation (205) does not yet contain the morphemeUNINT whose realization triggers unintensive reduplication. The factthat the object agreement prefix can be brought into the duplicant showsthat UNINT can occur higher than AGR. The interaction of passive andapplicative will show that a low position of UNINT must be possibleas well. I will assume that UNINT can occur at any level in the tree(205a). It is not unusual for a morpheme which has adverbial seman-tics, as UNINT does, to have considerable freedom in its hierarchicalposition. Since UNINT can occur no higher than directly above AGR,only material associated with the morphemes in (205a) can undergoreduplication, explaining why only the material in the box in (201a)is involved in reduplication.

The tree structure (205) will usually be represented as

root ⊕ E1 ⊕ · · · ⊕ En ⊕ AGR

The operator ⊕ is binary and associates left to right.

7.1.1 Passive/applicative interactionCertain examples of reduplicated verb forms with both passive andapplicative suffixes appear to violate the idea that the initial duplicantconsists of contiguous morpheme exponents.

In (206), -w is the passive suffix and -el the applicative suffix. Bothare extension suffixes.

(206) phek-el-w-a ‘be cooked for/at’pheke.phek-el-w-apheka.phek-el-w-aphekwa.phek-el-w-a

‘be cooked for/at, a bit’

Case Studies: Ndebele 127

The reduplicants pheke and pheka are expected, following the patternin (200b). The reduplicant phekwa, however, is unexpected. It appearsto combine material from nonadjacent morphemes. The reduplicantphekwa seems to somehow be generated from the stem phek-el-w.

It is a characteristic of Bantu verbal morphology that the surfacepositions of the exponents is only partly determined by the syntacticstructure (205). Various processes can rearrange the linear order ofthe exponents of the Ei so that the linear order of the extension suf-fixes does not mirror the hierarchical position of the morphemes theyrealize, but adapts itself to conditions on surface morpheme exponentorder. In particular, HI&S give examples to show that both underlying[[phek]]⊕APP⊕PASS and [[phek]]⊕PASS⊕APP produce the surface linearorder phek-el-w ‘were cooked for’. The Bantu languages generallyrequire the surface order applicative-passive. There are various potentialanalyses of this phenomenon. HI&S assume that there is a process ofmorpheme metathesis that converts the structure [[phek]] ⊕ PASS ⊕ APPto [[phek]] ⊕ APP ⊕ PASS at the morphological level. I will make a lessradical assumption and assume that the applicative suffix is infixed tothe left of the passive suffix, if the latter is present.

The following derivation reveals the source of the problematic verbform phekwa.phek-el-w-a. At the point in the derivation that UNINT isrealized, phek and w are adjacent.

(207) ⊕PASS phek-w⊕UNINT [phek-w]

Pradj [phekw〈a〉]Trscr phekwa phekw

⊕APP phekwaphek-el-w (infixation)

I will not give the details, which are easily established followingthe model in (207), but it is interesting to see the results for all of thepossible orders of combination of APP, PASS, and UNINT. The formsmarked with † involve infixation of the applicative affix. Note that infor some orders, there is variation in the output.

128 Chapter 7

(208) [[phek]] ⊕ UNINT ⊕ PASS ⊕ APP pheka-phek-el-w †[[phek]] ⊕ UNINT ⊕ APP ⊕ PASS pheka-phek-el-w

[[phek]] ⊕ APP ⊕ UNINT ⊕ PASS pheke/pheka-phek-el-w

[[phek]] ⊕ APP ⊕ PASS ⊕ UNINT pheke/pheka-phek-el-w

[[phek]] ⊕ PASS ⊕ UNINT ⊕ APP phekwa-phek-el-w †[[phek]] ⊕ PASS ⊕ APP ⊕ UNINT pheke/pheka-phek-el-w †

According to (208) the reduplicant phekwa is impossible if PASSis inside APP. HI&S provide the evidence that this is indeed thecase, distinguishing between passivized applicative constructions andapplicativized passives on a semantic basis. The reduplicant phekwa

is not possible with passivized applicatives. In closing this section,it should be made clear that, in spite of the fact that HI&S cast theirdiscussion in a framework in which cyclic morphology has no role,most of the basic morphophonological insights in the discussion of thepassive/applicative interaction come from their analysis.

7.1.2 Overapplication of passive palatalizationI have assumed that duplication juncture insertion, duplicant adjust-ment, and transcription apply in the UNINT-cycle. The issue of NCCrepair was intentional left open. A palatalization phenomenon triggeredby the passive suffix shows that NCC repair does not take place in theUNINT-cycle.

HI&S show that the passive suffix palatalizes a preceding labial.The result is shown boxed below. The labial can be several syllablesaway, but only the nearest labial is palatalized.

(209) a. boph-a ‘tie’ bo tsh -w-a ‘be tied’

b. bumb-a ‘mould’ bun j -w-a ‘be molded’

c. gombolozel-a ‘encircle’ gon j olozel-w-a ‘be encircled’

Combining PASS and UNINT, in the two possible orders, producesthe following.

Case Studies: Ndebele 129

(210) (boph) botsha-bothsh-w-a, botshwa-bothsh-w-a*bopha.botsh-w-a

(bumb) bunja-bunj-w-a, bunjwa-bunj-w-a*bumba.bunj-w-a

(gombolozel) gonjo-gonjolozel-w-a*gombo-gonjolozel-w-a

(fumbath) funja-funjath-w-a*fumba-funjath-w-a

The labial of the reduplicant is always palatalized. This is expectedif PASS is inside UNINT and palatalization occurs before combinationwith the reduplicative affix. But it is not expected if reduplicationoccurs prior to combination with PASS, since only the closest labial ispalatalized.

Consider, for example, the derivation corresponding to

[[fumbath]] ⊕ UNINT ⊕ PASS .

If NCC repair takes place in the UNINT-cycle, then we expect:

fumbafumbath⊕PASS−→ fumbafun j ath-w

The reduplicant labial is not palatalized. If, on the other hand, NCCrepair has not been carried out at the point that PASS is realized, andpassive palatalization is not restricted by geminate inalterability, weexpect:

× × × × ×

f

×

u

×

m

×

b

×

a

×

th

×⊕PASS−→ × × × × ×

f

×

u

×

n

×

j

×

a

×

th

×

w

×

The reduplicant labial is palatized, in agreement with the empirical data.For the sake of concreteness, I will assume that NCC repair occurs

no earlier than the AGR-cycle and that passive palatalization is notsubject to geminate inalterability. Since PASS is inside AGR, this willensure that the in case UNINT applies before PASS, the crossed structurewhich produces overapplication will still be present. We now turn toevidence that NCC repair can (optionally) occur even later than theAGR-cycle.

130 Chapter 7

7.1.3 Overapplication of perfective imbricationThe default perfective suffix in Ndebele is -ile. It is not an extensionsuffix and therefore outside the scope of AGR. If UNINT and PERF areboth present, and UNINT is as high as possible (i.e. directly above AGR),the structure is:

root . . .Agr

UnintPerf

As predicted by this structure, material from the perfective suffix cannotbe incorporated in the duplicant in unintensive reduplication: uku-zi-

lima-lim-ile, *uku-zi-limi-lim-ile. Contrast this with the causative suffix-is, which produces both uku-zi-lima.lim-is-a and uku-zi-limi-lim-is-a.

There is affix allomorphy, with a second potential realization ofthe perfective morpheme, -e. For the purposes of this section, call theroots which take the short suffix Type 1 roots. Most Type 1 roots cantake either the long and short perfective suffixes. But there is a subclassof the Type 1 roots which does not allow the long perfective suffix incertain environments (detailed below). Call these roots Type 2 roots.The short perfective suffix induces stem ablaut. If the final vowel of thestem is a, it raises to e. This is called imbrication in the Bantu literature.

(211) Type perfectivelim ‘cultivate’ lim-ile *lim-edabul 1 ‘tear’ dabul-ile dabul-edumal 2 ‘become depressed’ *dumal-ile dumel-e

(The perfective forms will actually occur with various prefixes, whichare irrelevant to the present point.)

The interaction of perfective suffix choice, imbrication, and unin-tensive reduplication, is unexpected. There are three different realiza-tions of [[dumal]] ⊕ AGR ⊕ UNINT ⊕ PERF, apparently in free variation:(realization of agreement is omitted for reasons of space):

Case Studies: Ndebele 131

(212) a. duma-dumel-e b. dume-dumel-e c. duma-dumal-ile

The variation between (212a) and (212b) can be undertood as vari-ation in the timing of NCC repair. If NCC repair follows transcriptionin the UNINT-cycle, we expect:

UNINT

d

×u

×m

×a

×[[DUMAL]]

d

×u

×m

×a

×l

×⊕PERF−→

UNINT

d

×u

×m

×a

×[[DUMAL]]

d

×u

×m

×e

×l

×PERF

e

×

But if NCC repair is delayed until the PERF-cycle, then we expect:

UNINT

× × × ×[[DUMAL]]

d

×

u

×

m

×

a

×

l

×⊕PERF−→

UNINT

× × × ×[[DUMAL]]

d

×

u

×

m

×

e

×

l

×PERF

e

×

The result is overapplication of imbrication. Of course, it must also beassumed (quite plausibly) that imbrication is not blocked by geminateinalterability.

The reason for the appearance of the long perfective suffix in (212c)is much less clear. It is not an explanation, but we can attribute it to themorpheme realization rules (i.e. the lexicon).

(213) 1. PERF→ -ile, except/α

(Type2)⊕ . . . ⊕ AGR

2. PERF→ -e/α

(Type1)⊕ . . .

This yields the desired outcome because the environment that produces(212c) is . . . ⊕ UNINT ⊕ , which does not block -ile according to(213.1). Since there is no evidence that the grammaticality of (212c) isother than an ideosyncratic fact about morpheme realization in Ndebele,the issue will not be pursued.2

7.1.4 Yi-epenthesisIndependently of reduplication, Ndebele imposes a bisyllabic minimalword condition. Because verbal prefixation and suffixation is so preva-lent, augmentation to minimal word weight occurs only in a narrow

132 Chapter 7

range of examples: imperatives of a consonantal root (dl ‘eat’, m

‘stand’, etc.).

(214) uku-dl-a ‘to eat’yi-dl-a (*dl-a) ‘eat! (imperative)’

uku-lim-a ‘to cultivate’lim-a (*yi-lim-a) ‘cultivate! (imperative)

Left edge yi-epenthesis is used for augmentation to minimal wordweight.

The most straightforward examples of yi-epenthesis used for du-plicant augmentation are unintensive imperatives.

(215) yidla-yidl-a, dlayi-dl-a ‘eat!’ (a bit)

There is variation, with yi-epenthesis at either the left or right edge ofthe duplicant.

(216) dl dl⊕UNINT [dl] [dl]PrAdj (WFSS) [dl〈a〉] [dl〈a〉]PrAdj (weight) [yidl〈a〉] [dl〈ayi〉]Trscr yidla yidl dlayi dl

WFSS is well-formed syllable structure.Earlier, it was proposed that the choice of the initial duplicant

only optionally included prefixes or affixes, assuming that this waspossible without making a super-bisyllabic duplicant. This was anoversimplification. There are two dialects. In the conservative dialect,the duplicant is not well-formed if there is a suffix which could augmentthe duplicant without pushing it over the bisyllabic weight limit. In thatdialect, given the stem zi-dl-is, juncture insertion can produce [zi-dl]-isor zi-[dl-is], but not zi-[dl]-is. In the innovative dialect, all are possible.But even the innovative dialect does not allow zi-[dl]-is-el. If there aretwo extension suffixes, the innovative dialect does not allow total rootreduplication of a consonantal root. The inner suffix must be included

Case Studies: Kinande 133

in the duplicant. With the variation discussed earlier and yi-epenthesis,the innovative dialect produces 5 different outputs from zi-dl-el !

(217) JncIns PrAdj (WFSS) PrAdj (weight) Trscr

[zi-dl]-el [zi-dl-e]l zidle zi-dl-el

[zi-dl]-el [zi-dl〈a〉]-el zidla zi-dl-el

zi-[dl-el] zi-[dl-el〈a〉] zi- dlela dl-el

zi-[dl]-el zi-[dl-e]l zi-[yidl-e]l zi- yidle yidl-e

zi-[dl]-el zi-[dl〈a〉]-el zi-[dl〈ayi〉]-el zi- dlayi dl-el

7.2 Morpheme integrity in Kinande

Kinande and Ndebele are related Bantu languages whose inflected verbshave a similar morphological structure. They also have similar unin-tensive verbal reduplication processes. The semantics of the Kinandeunintensive reduplication is usually given a meaning of ‘hurriedly’ or‘here and there’. The most significant difference between the Kinandeand Ndebele systems was uncovered and analyzed by Hyman and Mu-taka (1990), which this section is based on. They discovered that thereis a whole morpheme effect in Kinande reduplication, which blockspartial morpheme reduplication. They called it the Morpheme IntegrityConstraint (MIC) and viewed it as a restriction on templatic association:“If the whole of a morpheme cannot be successfully mapped into thebisyllabic reduplicative template, then none of the morpheme may bemapped.” I take it to be a restriction on where duplication junctures canappear: only at morpheme boundaries.

Besides the MIC, we can identify the following differences betweenKinande and Ndebele unintensive reduplication:

(218) In Kinande, as opposed to Ndebele:a. there is no analog of yi-epenthesis;b. bare consonantal roots are doubly reduplicated; andc. UNINT is always high in the morphological structure.

134 Chapter 7

First consider:

(219) e-ri-sw-a ‘to grind’e-ri-swaswa-sw-a ‘to grind (hurriedly)’

Kinande does not use syllable epenthesis for prosodic adjustment tobisyllabicity. Instead, Kinande uses double reduplication just as it isused in Mokilese in the case of monosyllabic roots (see Section 6.2.2.2).Readjustment rule allomorphy inserts a pair of ]-junctures at the rightedge of the stem. The derivation of swaswa-sw is almost identical to(154) in Section 6.2.2.2.

(220) [[sw]]→ [[sw〈a〉]]→ swa-[sw]→ swa-[sw〈a〉]→ swa-swa-sw

The evidence that UNINT must be situated high in the hierarchicalstructure comes from the fact that extension suffixes, if present, arealways used to add weight to the reduplicant.

(221) a. e-ri-sw-er-a ‘to grind for’b. e-ri-swera-sw-er-a ‘to grind for (hurriedly)’c. *e-ri-swaswa-sw-er-a

If UNINT could combine with the root before the applicative suffix, asit can in Ndebele, we would expect

sw → swaswa-sw → swaswa-sw-es → e-ri-swaswa-sw-es-a

Since we know that that extension suffixes can reduplicate, (221b)for example, the effect of the MIC is easy to see.

(222) a. e-ri-hum-a ‘to cultivate’b. e-ri-huma-hum-a ‘to cultivate (hurriedly)’c. e-ri-hum-er-a ‘to cultivate for’d. e-ri-huma-hum-er-a ‘to cultivate for (hurriedly)’e. *e-ri-hume-hum-er-a

Under the Ndebele rules, (222e) would be good. But it requires theprosodic adjustment [hum-er] → [hum-e]r , which is blocked by theMIC in Kinande.

Case Studies: Kinande 135

The high position of UNINT in the morphological hierarchy is alsoresponsible for the ungrammaticality of (223c). The suffix -w is thepassive affix.

(223) a. e-ri-hum-w-a ‘to be cultivated’b. e-ri-humwa-hum-w-a ‘to be cultivated (hurriedly)’c. *e-ri-huma-hum-w-a

In Ndebele, (223c) would be grammatical, because UNINT can berealized before the passive morpheme. In Kinande, the passive suffixhas already been combined with the root at the point that UNINT isrealized. Prosodic adjustment does not allow [hum-w] → [hum]-wbecause prosodic adjustment is driven by prosodic desiderata and ]-Left does not contribute to satisfying the desiderata. Consequently, thepassive suffix is always reduplicated in Kinande.

Overweight (more than bisyllabic) roots pose a problem for theKinande reduplication system. In Chapter 6, it was proposed thatit was highly marked state of affairs for the repair rules available toProsodic Adjustment to be inadequate to achieve the prosodic desideratawhich drive the adjustment rule. Since the MIC does not permit partialreduplication of a verb root, Kinande does not have any way to producea bisyllabic reduplicant if the root is overweight. Some languages, whenfaced with a similar problem, simply exclude a prosodically definedclass of stems from reduplication. Kinande does not fully solve theproblem. Many overweight roots are excluded from the reduplicationprocess, but many are not. This is specified lexically as a property ofsome roots. About half the overweight roots do not reduplicate. I willcall these non-reduplicating roots. Furthermore, no overweight rootwith an extension suffix reduplicates. Call this the NOEXT constrainton juncture insertion.

NOEXT: *[α - extension suffix . . . ]

Somewhat more than half the overweight roots can reduplicate, subjectto NOEXT. In (224), bindul ‘change’ is non-reduplicating and bugul

‘find’ is reduplicating. The starred forms would be good under theNdebele rules.

136 Chapter 7

(224) Overweight roots

Kinandebindul bindula-bindul *binda-bindulbigul (no reduplication) *biga-bigulbindul-er (no reduplication) *binda-bindul-erbigul-er (no reduplication) *biga-bindul-er

Kinande has a class of bimorphemic verb forms which act in someways like roots. Their meaning is idiomatic, not compositional. Forconvenience, I will call them pseudo-roots. For reasons that are notclear, pseudo-roots are subject to NOEXT. The reduplicating pseudo-root gund-ul, for example, reduplicates as gunda-gund-ul, with prosodicadjustment to bisyllabicity responsible for [gund-ul ]→ [gund ]-ul→[gund〈a〉]-ul→ gunda-gund-ul. But gund-ul-er, with an extension suffix,because of NOEXT, does not reduplicate.

7.2.1 Short causativesThe normal causative inflection is bimorphemic -es-y (or -is-y , depend-ing on vowel height harmony).3

(225) e-ri-hum-is-y-a ‘to cause to beat’e-ri-gend-es-y-a ‘to cause to go’

There is a subclass of roots which can appear with only the outer suffix-y, with no semantic difference between the forms with the two suffixes.The root gend is such a root, but hum is not.

(226) *e-ri-hum-y-a ‘to cause to beat’e-ri-gend-y-a ‘to cause to go’ (= e-ri-gend-es-y-a)

So that a terminology is available, call this subclass of roots the SC-class(Short Causative).

It will turn out to be most consistent with certain reduplicationfacts that will soon be discussed if we assume that there is an optionalreadjustment rule associated with -y realization that deletes the innersuffix in the context of certain roots.4

(227) es/is→ ∅/α y

(CAUSE), α an SC-root

Case Studies: Kinande 137

The reduplication data of particular interest is the following three-way variation.

(228) a. e-ri-gendya-gend-y-a ‘to cause to go (hurriedly)’b. e-ri-genda-gend-es-y-ac. e-ri-gendya-gend-es-y-a

The derivations of (228a) and (228b) are straightforward:

(229) gend-y gend-es-yUNINT [gend-y] [gend-es-y]PrAdj (weight) [gend]-es-yPrAdj (WFSS) [gend-y〈a〉] [gend〈a〉]-es-y

gendya gend-y genda gend-es-y

The simplest explanation for the occurrence of (228c) is that a variantof the rule (227) is used as a prosodic adjustment rule.

(230) es→ 〈es〉/α y

(CAUSE), α an SC-root

Recall from Section 4.3 that medial truncation has the effect of redupli-cant truncation. The truncated material is not copied to the reduplicant.The effect is first-conjunct deletion. The derivation of (228c) is:

(231) gend-es-y⊕UNINT [gend-es-y]PrAdj (weight) [gend-〈es〉-y]PrAdj (WFSS) [gend-〈es〉-y〈a〉]Trscr gendya gend-es-y

Further examples which explore the interaction of other extensionsuffixes with the short causative -y show that the prosodic adjustmentrule (230) is further extended.

(232) α→ 〈α〉/

SC-root y(CAUSE)

; MIC

138 Chapter 7

The MIC ensures that t-junctures are only inserted at morpheme bound-aries. Truncation (232) is responsible for (233c) and (233d).

(233) reduplicateda. e-ri-bul-y-a e-ri-bulya-bul-y-a ‘to ask’b. e-ri-bul-ir-y-a e-ri-bula-bul-ir-y-a ‘to ask for’c. e-ri-bulya-bul-ir-y-ad. e-ri-bul-ir-an-y-a e-ri-bula-bul-ir-an-y-a ‘to ask fore. e-ri-bulya-bul-ir-an-y-a each other’

The derivation of (233e), for example, is:

(234) bul-ir-an-y⊕UNINT [bul-ir-an-y]PrAdj [bul-〈ir-an〉-y]PrAdj [bul-〈ir-an〉-y〈a〉]Trscr bulya bul-ir-an-y

Subsequent morphophonology adds the prefixex e-ri- and the finalvowel to produce e-ri-bulya-bul-ir-an-y-a.

In summary: Juncture insertion is double stem reduplication forconsonantal stems, null for certain lexically specified super-bisyllabicstems, and total stem reduplication otherwise. The syntax specifieswhere UNINT occurs in the hierarchical structure of the words in whichit occurs. Prosodic adjustment is carried out according to:

(235) first conjunct ; bisyllabic ::]-Left

FCVE

α→ 〈α〉/

SC-root y(CAUSE)

; MIC

7.3 Asheninca Campa intensive reduplication

The version of “Asheninca Campa” that is analyzed here is the hypo-thetical language analyzed in McCarthy and Prince (1998). It is basedlargely on Payne (1981), an impressive analysis of a single speaker’slanguage. Secondarily, Spring (1990) is used to fill in many details

Case Studies: Asheninca Campa 139

of the treatment of unprefixed monosyllabic roots. Her data comesfrom field work by her and Payne with a second speaker. Since thesecond speaker’s reduplication data showed significant instability, thereare grounds for wondering if a coherent language possibility has beenhypothesized.5 Since an important goal of this paper is to show thatMcCarthy and Prince’s (1998) claim that a satisfactory analysis ofAsheninca Campa is beyond the reach of derivational morphophonol-ogy, I put these concerns aside and (for the purposes of this paper)accept the hypothesized language as a valid subject of inquiry.6

Asheninca Campa has a productive verbal reduplication processused to signal “excess” and usually glossed ‘more and more’. It fre-quently occurs in conjunction with the continuative suffix -wai. Verbsusually appear with at least one prefix (commonly a person marker orfuture tense marker) and one or more suffixes.

Before we consider reduplication in Asheninca Campa, we needto describe some of the basic morphophonology. Syllable structureis straightforward. Codas must be nasal consonants sharing point ofarticulation with a following consonant. Onsetless syllables are per-mitted only initially. Syllabification at stem/suffix junctures and wordfinally is carried out by familiar epenthetic operations. The epentheticvowel a breaks up impermissable CC sequences and syllabifies finalconsonants. The epenthetic consonant t breaks up impermissable VVsequences. A few diphthongs are permitted, as in the continuative suffix-wai, but hiatus is generally avoided. In (236), epenthetic segments areunderlined.

(236) a and t epenthesis at stem-suffix boundaries

kowwant

+ a:n /chiNONFIN

kowa:n /chi‘to want’

komapaddle

+ a:n /chiNONFIN

komata:n /chi‘to paddle’

kowwant

+ waiCONT

+ akPERF

+ iNONFUT

kowawaitaki‘has continually wanted’

Prefix attachment operates on different principles than suffix at-tachment, with no epenthesis. Instead, there prefix readjustment, withsuffix-final syncope used to maintain well-formed syllable structure.

140 Chapter 7

See Payne (1981) for the details. The examples in (237) are typical.At the surface, the first person prefix alternates between no- and n-. Theexponent of the future morpheme is a placeless nasal (-ð below) whichmust share place of articulation with a consonant which directly followsit. It deletes if it is not followed by an appropriate consonant.

(237) Prefix readjustment

no1st+ asi

cover→ na.si

no1st+ ð

FUT+ asi

cover→ na.si

no1st+ ð

FUT+ koma

paddle→ noN.ko.ma

With the basics of prefixal and suffixal morphophonology in hand,we can consider reduplication, which is straightforward if the verb rootis polysyllabic, C-initial, and V-final. The root is totally reduplicated.This is illustrated in (238) for the trisyllablic root kawosi ‘bathe’ andthe bisyllabic root koma ‘paddle’.

(238) a. no-kawosi-wai ‘I bathe’no-kawosi-kawosi-wai ‘I bathe more and more’

b. no-koma-wai ‘I paddle’no-koma-koma-wai ‘I paddle more and more’

Now consider V-initial roots with a C-initial prefix.

(239) a. n-asi-wai ‘I cover’n-asi-nasi-wai ‘I cover more and more’

b. n-osampi-wai ‘I ask’n-osampi-sampi-wai ‘I ask more and more’

First note that (239b) provides evidence that transcription is to theright, as we have been assuming. Left transcription would producen-o-sampi-sampi-wai, with the target of copying inside the root.

Onset incorporation occurs in (239a), just as in the Kıhehe examplediscussed in Section 6.1. The account is the same. The reduplicativemorpheme is realized after the prefix has already combined with thestem. It triggers the ×∗-rule, ∅ → [ / ×, with the root designated as

Case Studies: Asheninca Campa 141

the domain of the affix. Rule application is leftmost, by default, and isfollowed by Default Closure. Prosodic adjustment follows, with well-formed duplicant syllable structure a prosodic desideratum. Exactly asin Kıhehe, [-Left is available as a repair rule, so n-[asi]→ [n-asi] .

The duplicant WFSS desideratum drives adjustment in (239b) aswell, but [-Right applies, not [-Left. What accounts for the difference?Plausibly, because [-Left in (239a) would result in an underweightduplicant. The next examples bear on this:

(240) Prefixed monosyllabic C-initial roots

a. no-na-wai ‘I carry’no-na-nona-wai ‘I carry more and more’

b. no-na:-wai ‘I chew’no-na:-nona:-wai ‘I chew more and more’

In both cases, the prefix is used to augment the duplicant, even thoughthe duplicant syllable structure is well-formed. In (240b) the duplicantis initially bimoraic but is augmented, so one might speculate thataugmentation is to bisyllabicity. But (239b), and (238a), together, showthat this cannot be correct. The first shows that [-Right is availablefor prosodic adjustment. The second shows that it is not used, evenwhen the duplicant is trisyllabic. The conclusion is that the weightdesideratum is polysyllabicity, not bisyllabicity.

Summarizing: The reduplicative affix which realizes intensive hasa null exponent, specifies the root as its domain and triggers the ×∗-rule.Prosodic adjustment applies:

duplicant ;[

WFSSpolysyllabic

]::[ [-Right

[-Left

]

The rule ordering above is necessary to ensure that the [-juncture shiftsto the right when it is allowed to do so by the weight desideratum.Otherwise, nosampi-nosampi would result in (239b).

The speaker also needs to know the position of INT in the morpho-logical hierarchy of the verb form. I will assume that INT is realizedafter all the prefixes and before all the suffixes. Since there are nosuffixes at the point that juncture insertion applies, Default Closurewill close the duplicant at the right edge of the stem. In the examplesconsidered to this point, the right edge of the stem has coincided with

142 Chapter 7

the right edge of the root. This is not always the case. We will shortlysee some examples in which an epenthetic vowel appears at the rightedge of the stem.

Nothing further needs to be said about the morphology of Ashen-inca Campa intensive reduplication. In order to see this, however,it is first necessary to understand a different prosodic augmentationprocess in Asheninca Campa verbal morphology. It is independent ofreduplication.

7.3.1 Stem augmentation induced by C-initial suffixesC-initial suffixes induce augmentation of light verb stems. The rootsp ‘feed’, na ‘carry’, and na: ‘chew’, along with the 1st person agreementprefix no-, are sufficient to demonstrate the effect. The augmentation isshown boxed.

(241) stem before C-initial suffixes

a. p p a: ‘feed’

b. na na ta ‘carry’

c. pa: pa: ‘chew’

d. no-p nop a ‘I feed’e. no-na nona ‘I carry’f. no-pa: nopa: ‘I chew’

Some examples are given below. The C-initial suffixes are -wai

(continuative) and -piro, which is glossed as ‘verity’. The V-initialinfinitival suffix a:n /chi is given for comparison. The epenthetic materialadded to augment the stem has been boxed, as above, and the epentheticmaterial (t) used to break up stem/affix hiatus is underlined.

(242) a. p-a:n /chi ‘to feed’

b. p a: -piro-ta:n /chi ‘to truely feed’

c. no-p a -wai ‘I feed’

d. na-ta:n /chi ‘to carry’

e. na ta -piro-ta:n /chi ‘to truely carry’

f. no-na-wai ‘I carry’g. na:-ta:n /chi ‘to chew’

Case Studies: Asheninca Campa 143

h. na:-piro-ta:n /chi ‘to truely chew’i. no-na:-wai ‘I chew’

Note that the epenthetic vowel in (242c) is attributed to prosodic ad-justment of the stem, not syllable structure repair at the root-suffixboundary.

The one subtlety in stating the augmentation rules is to ensurethat p augments with a long vowel, while pa is augmented by syllableepenthesis. The following augmentation rule, which is applied to thestem before concatenating a C-initial suffix, accomplishes this:

(243) Stem ; Polymoraic ::

Final Vowel Epenthesis (a)Vowel LengtheningSyllable Epenthesis (ta) / root

It is implicit in the prosodic weight desideratum that the stem must besyllabified and is therefore subject to syllable well-formedness condi-tions. Syllable epenthesis is restricted to the right edge of the rootin order to prevent p→ p a → p ata . According to (243), syllableepenthesis applies only at the morpheme boundary.

We can now consider the interaction of stem augmentation withreduplication. Intensive reduplication, although it does not have asuffixal exponent, does almost invariably produce a C-initial derivedsuffix at the surface. It is therefore not surprising that Asheninca Campaassigns the intensive affix to the class of affixes which trigger stemaugmentation.

A number of illustrations are given below. Recall that we areassuming that INT realization takes place before any suffixes enter thecomputation.

(244) stem augment JncIns DC PrAdj Trscr

a. p p a: [pa: [pa:] pa: pa:

b. no-p no-p a no-[pa no[pa] [nopa] nopa nopa

c. na na ta [nata [nata] nata nata

d. no-na: no-[na: no[na:] [nona:] nona: nona:

e. /chik /chik a [ /chika [ /chika] /chika /chika

144 Chapter 7

A few examples of fully inflected verbs are given in (245). Asidefrom the root kow ‘paddle’, the roots and suffixes are as in (242).The material used to augment the stem to bimoraicity is boxed, thereduplicant is shaded, and epenthetic t used to join the V-initial suffix tothe V-final stem is underlined.

(245) a. /chik /chik-a:n /chi /chik a -piro-ta:n /chi /chik a /chika ta:n /chi

b. na na-ta:n /chi na ta -piro-ta:n /chi na ta nata ta:n /chi

c. p p-a:n /chi p a: -piro-ta:n /chi p a: pa: ta:n /chi

d. na: na:-ta:n /chi na:-piro-ta:n /chi na: na: ta:n /chi

Note that in a few cases the repair rules specified for prosodicaugmentation of the duplicant to polysyllabicity are not always adequateto achieve the desired weight. But there is only a very narrow range ofcases for which this is true, unprefixed C and CV: roots, (245c) and(245d) above, and only in the unusual case where there is no prefix.

7.3.2 Copying a word boundaryThere is one last feature of Asheninca Campa intensive reduplication toconsider. Consider an unprefixed VCV root like asi ‘cover’.

(246) asi-ta:n /chi ‘to cover’asi asi-ta:n /chi ‘to cover more and more’

Surprisingly, the reduplicated form has an internal word break. Thereare two kinds of evidence for this: 1) the phonetic judgments of linguistfield workers; and 2) the stress pattern is that of separate words. SeeSpring (1990, fn. 3, p. 148).

Payne (1981:146), in his careful and insightful study of AshenincaCampa, says:

One of the most striking elements of Asheninca reduplication isthat it reduplicates a word boundary into some forms, but not intoothers. Preferring two-syllable verbs with initial consonants toreduplicate, it appears that in the absence of the initial consonantin VCV-type verbs, the rule actually copies the word boundary.(italics added)

Case Studies: Washo 145

Payne’s insight can be formalized in the following way. Supposethat word breaks are represented by timing slots which are associatedwith some kind of special marker, # is used in (247). Then (246) isthe result of the regular application of prosodic adjustment—the ruleactually copies the word boundary. Recall as well that the idea thatword breaks can occur inside the duplicant was already introduced inthe discussion of tasty-shmasty reduplication in Section 4.1.1.

(247)

#

×[

∗σ

a

×σ

s

×i

×]

#

×[-Left−→ [

#

×σ

a

×σ

s

×i

×]

#

× →#

×σ

a

×σ

s

×i

× × × × ×#

×

If a vowel is associated with a timing slot which follows a timing slotassociated with a word boundary, it satisfies the conditions on WFSS, so[-Left is motivated by the WFSS desideratum imposed on the duplicant.

I take it to be evidence in favor of the theory advocated here that this“striking element” of Asheninca Campa reduplication is so immediatelycomprehensible in terms of the computational mechanisms which thetheory makes available. It is hard to see how a theory which viewsreduplication as templatic prefixation or suffixation could account for(246) in such a natural way.

7.4 Washo plural reduplication

Washo plural reduplication was analyzed by Broselow and McCarthy(1983) in their important study of internal reduplication. Their start-ing point was the analysis of Winter (1970), based on the data inJacobsen (1964). It is set in the melody copying framework inheritedfrom Marantz (1982) and depends heavily on the manipulation of longvowels in ways that have since proved untenable.7 Washo reduplicativepluralization is also the subject of a recent study by Yu (2001), whoadds much useful data and discussion of various approaches to Washoreduplication. I rely on Broselow and McCarthy and Yu for the data.

There is allomorphy in Washo reduplicative pluralization, with aspecial affix used to realize plural on monosyllabic stems and somelexically specified polysyllabic stems. This fits a pattern that we haveencountered several times before, with special treatment of monosyl-labic stems. I will consider the two affixes in turn, beginning with

146 Chapter 7

straightforward examples of the effects of the default affix, which willbe called ρCV, for reasons that will soon be clear.

7.4.1 CV-reduplicationFirst, some examples:

(248) root plurala. hanakmuwe hanakmuwewe ‘elks’b. baloxat baloxaxat ‘bows’c. daPa daPaPa ‘mother’s brother’d. duweP duweweP ‘to try to, to want to’e. moya moyaya ‘shoulder’f. Pelel Pelelel ‘mother’s father’g. PewsiP PesiwsiP ‘father’s brothers’h. nent’us net’unt’us ‘old women’i. mokgo mogokgo ‘shoes’

The task for both the linguist and the Washo language is to deducethe pretranscription representations, and then the rules for generatingthese pretranscription representations. Examples (248g–i) are the mostcomplex, and therefore the most revealing. The simplest way to gen-erate this pattern is left transcription with leading edge truncation. Thecopy is shaded (as usual) and the remnant is boxed.

(249) ne[〈n〉t’u]s → ne t’u nt’u s

mo[〈k〉go] → mo go kgo

Pe[〈w〉si]P → Pe si wsi P

We can go on to speculate that the juncture insertion rules are

∅ → ]/Right V , ∅ → [

/Right V

and that prosodic adjustment of the first conjunct results in truncation.Since an increasing sonority requirement on onset consonants is acommon well-formedness condition, we can further speculate that thetruncation is driven by a well-formed syllable structure desideratumfor the first conjunct. These considerations successfully derive all the

Case Studies: Washo 147

examples in (248). Note that the juncture insertion rules are the mirrorimage of the Mangarayi juncture insertion rules.

Jacobsen (1964:117) observes that the only initial two-consonantclusters that occur in indigenous Washo words are of the type P fol-lowed by a voiced sonorant. If we assume that this is a syllable well-formedness condition on onsets, the following examples support theidea that the truncation in (249) is driven by first conjunct syllable well-formedness.

(250) root plurala. baPlew baPlePlew ‘Paiutes’b. inPy1n iPy1nPy1n ‘varicolored’c. iPdeb idePdeb ‘wrinkled’

The derivations are:

(251) JncIns PrAdj Trscr

baPlew ba[Ple]w ba Ple Ple w

inPy1n i[nPy1]n i[〈n〉Py1]n i Py1 nPy1 n

iPdeb i[Pde]b i[〈P〉de]b i de Pde

In each case, the onset is truncated minimally to a permissible onsetusing left edge truncation. These examples make the designation “CV-reduplication” for this kind of Washo plural reduplication problematic.I will maintain it in spite of its problematic aspect because it does de-scribe the reduplicant accurately in the large majority of cases and doescontrast nicely with the other type of reduplication, “VC-reduplication”,which will be discussed shortly.

V-initial roots display a superficially different surface pattern.

(252) root plural

a. ileg leleg ‘red’b. ipes pepes ‘black’c. aNkas kaNkas ‘hollow’d. emc’i c’imc’i ‘to wake up’

Winter (1970) realized that the examples in (252) were the result ofvowel syncope of initial unstressed vowels in Washo. Stress is usually

148 Chapter 7

penultimate. Once this is taken into account, the analysis of the exam-ples in (252) is no different than the analysis of the examples in (251).

(253) Trscr stress syncope

a. i[le]g i le le g ileleg leleg

b. i[pe]s i pe pe s ipepes pepes

c. a[〈N〉ka]s a ka Nka s akaNkas kaNkas

d. e[〈m〉c’i] e c’i mc’i ec’ımc’i c’ımc’i

There is stem ablaut associated with CV-reduplication, which isevident in the following examples:

(254) root plural pretranscription

a. t’anu t’anono ‘person’ ta[no]b. asun soson ‘red’ a[so]nc. albul bolbol ‘spherical’ a[〈l〉bo]ld. alPmul PolPmol ‘big and round’ a[〈l〉Pmo]le. amk’um k’omk’om ‘arched’ a[〈m〉k’o]m

Under ρCV-reduplication, a high round vowel in the duplicant is loweredif the preceding vowel is low. I will call this ρCV-ablaut. This ablautoccurs after juncture insertion, before transcription.

7.4.1.1. Bare timing slots in the underlying representationRoots with a long penultimate vowel appear to be exceptions to thereduplication rule as developed above.

(255) root plurala. wa:siw wasi:siw ‘Washo’b. me:hu mehu:hu ‘to be a boy’c. t’e:liw t’eli:liw ‘to be a man’d. memde:wi memdewi:wi ‘deer’e. Pa:t’u Pat’o:t’o ‘older brother’f. ma:gu mago:go ‘sister’s child’g. 1:l1l il1:l1l ‘pure white’

Case Studies: Washo 149

Halle (p.c.) suggests that long vowels can be represented in un-derlying forms as a single occurrence of the vowel followed by a baretiming slot. This immediately explains the examples in (255). Thederivation (256) is typical.

(256)

w

×a

× ×s

×i

×w

× →w

×a

×[〈×〉s

×i

×]

w

× →w

×a

× × × ×s

×i

×w

×

→w

×a

×s

×i

× ×s

×i

×w

× →w

×a

×s

×i

× ×s

×i

×w

×

ρCV-ablaut applies in (255e) and (255f), as expected.Finally, there is one last puzzling group of examples of ρCV-

reduplication, exemplified below:

(257) root plurala. ayam ya:m ‘to hit with an instrument’b. ayuk yo:k ‘parent-in-law’c. iyeb ye:b ‘to copulate’

We can account for these examples if we slightly modify theproposal that bare timing slots which are postvocalic are associated withthe vowel, producing a long vowel. Suppose that this rule is blockedif the timing slot is also prevocalic, as a hiatus avoidance mechanism.Suppose also that if this rule does not provide phonemic association to abare timing slot, then phonemic association is provided by epenthetic y.

Assuming that the y-glides in (257) come from underlying baretiming slots, which induce epenthetic y because long vowel formationis blocked, the derivation of (257a), for example, begins:

(258)

a

× ×a

×m

×JncIns−→

a

× [×a

× ]

m

×Trscr−→

a

× × × ×a

×m

×

If we can find a reason why the second bare timing slot deletes, the othersteps in the derivation are clear. The first bare timing slot is realizedas y intervocalically, and the unstressed initial vowel syncopates asdiscussed above. Note that there is a “crossing violation” in the finalrepresentation in (258) in the sense discussed in Chapter 2. There

150 Chapter 7

are two timing slots associated with a particular phoneme, but thereis an intervening timing slot which is not associated with that phoneme.This provides a motivation for deleting the the second bare timing slot.I assume that NCC repair, as a simpler alternative to fission, simplydeletes the intrusive timing slot.

All of the examples in (257) are explained in the same way. ρCV-ablaut applies as expected in (257b). This completes the discussion ofthe mechanics of Washo ρCV-reduplication. We now turn to illustratingVC-reduplication and discussing how the choice between ρVC and ρCV

is made.

7.4.2 VC-reduplicationThe juncture insertion associated with ρCV-reduplication does not pro-duce a nontrivial reduplicant if the stem is monosyllabic. We haveseen already that it is not unusual for reduplication rules to breakdown on monosyllabic stems. Recall that the Mangarayi reduplicativeplural, which also needs a pair of nuclei in order to demarcate theduplicant, does not apply to monosyllables. The Arrernte Rabbit Talktransformation similarly needs a pair of nuclei. In that case, there isallomorphy, with a different affix combining with monosyllabic roots.Monosyllabic roots in Washo are therefore of particular interest. Someexamples follow:

(259) root plural

a. ips peps ‘up from a surface’b. isl sesl ‘to give’c. iPb PePb ‘cry, weep’d. im mem ‘out from’e. iw wew ‘in a certain direction’f. akd kakd ‘slowly’g. sesm sesesm ‘to vomit’

In general, polysyllabic words are stressed on the penultimatesyllable (saksag, baloxat, malosan, etc.). Note, however, that in poly-syllabic (259g) there is final stress. If it is the case that all reduplicatedmonosyllables are have final stress, then a simple account of (259) ispossible. Assume for the moment that reduplicated monosyllabic roots

Case Studies: Washo 151

have final stress. This will be justified below. The pre-transcriptionrepresentations of the forms in (259) are given below.

(260) root JncIns ablaut TrScr stress syncope

a. ips [ip]s [ep]s ep ep s epeps peps

b. isl [is]l [es]l es es l esesl sesl

c. iPb [iP]s [eP]s eP eP s ePePs PePs

d. im [im] [em] em em emem mem

e. iw [iw] [ew] ew ew ewew wew

f. akd [ak]d ak ak d akakd kakd

g. sesm s[es]m s es es m sesesm

Like CV-reduplication, VC-reduplication also triggers stem ablaut, butof a different variety. A high vowel in the duplicant is lowered to a midvowel. In the examples above, i→ e .

Why should the bisyllabic output of reduplicated monosyllablesbe stressed on the final syllable? It cannot be the case that stress isassigned before duplication, necessarily final for a monosyllabic form,and simply persists. Many examples like moya → moyaya , (248e)above, show that reduplication can shift the location of stress. If stresspersisted, we would expect moyaya, not moyaya. The desired resultfollows, however, if foot delimiters are inserted cyclically and persist,assuming that main stress is assigned to the leftmost element of therightmost foot.

The persistence of penultimate stress in moya → moyaya , followsbecause cyclic binary footing can insert another foot delimiter afterreduplication.

(261)

m

×( ∗

o

×y

×∗

a

× →m

×( ∗

o

× ×∗×

y

×∗

a

× →m

×( ∗

o

× ×( ∗×

y

×∗

a

×

152 Chapter 7

With a monosyllabic root, like (261g), foot delimiter insertion cannotshift main stress off the final syllable.

(262)

s

×( ∗

e

×s

×m

× →s

×∗× ×

( ∗

e

×s

×m

×

We conclude therefore that in the context of monosyllabic stems,plural is realized by an affix which has a null exponent and triggers therules:

∅ → [/

V, ∅ → ]/

C

This affix will be called ρVC.There are a number of polysyllabic roots which are lexically stipu-

lated to take the ρVC plural affix.

(263) root plural pretranscription

a. maPsaP maPaPsaP ‘brother’s child’ m[aP]saPb. mayNa mayayNa ‘fawn’ m[ay]Nac. sawlam sawawlam ‘to be a girl’ s[aw]lamd. helme helelme ‘three’ h[el]mee. hesge hesesge ‘two’ h[es]gef. aPsam PaPsam ‘to lie’ [aP]samg. baliP balaliP ‘to shoot’ b[al]iPh. asiw sasiw ‘clear’ [as]iwi. ac’im c’ac’im ‘green, yellow’ [ac’]imj. asd1m sasd1m ‘to hide’ [as]d1mk. ap’1l p’ap’1l ‘tail’ [ap’]1l

The roots in (263) are of two types. The first type has the form(C1)VC2C3V(C4), where the two vowels are identical. The second typeis of the form (C1)aC2C3VC4, where V is a high unround vowel, i or 1.This is unlikely to be an accident. Some speculation is relegated to afootnote.8

Case Studies: Washo 153

7.4.3 The residueThe only examples from Broselow and McCarthy (1983) and Yu (2001)which are still unexplained are given in (264). The results of VC- andCV-reduplication (if applicable) are given in each case.

(264) actual VC CVa. Na:m Namim *Na:ma:m ‘son’b. els1m sels1m *lels1m *s1ls1m ‘to sleep’c. emlu mumlu *memlu *lumlu ‘to eat; food’

d. a:s da:s *sa:s ‘in, into’

e. iPib PePb *PePib *PiPib ‘cry, weep’f. iPis PePs *PePes *PiPis (empty stem)g. iPiw PwePw *PePew *PiPiw ‘to eat’

h. a:t’i t’a:t’i ‘upwards, uphill’i. a:m ma:m ‘happy’

I will leave the first group without comment. They are presentedin the interest of full disclosure. (264d) may be analogous to ArrernteRabbit Talk use of a consonantal prefix (y) in case the reduplicationrule for polysyllabic stems fails to apply. Yu (2001) argues that theroots in (264e-g) are underlying monosyllables and the i which appearsin the singular form is epenthetic. If so, they would be subject to VC-reduplication, explaining why the vowel lowers.9

The most interesting examples are (264h-i). There are two potentialexplanations, both of which are plausible, but neither of which hassupporting evidence. The problem is the assumption which was madeabove that long vowels in Washo are represented as vowel plus baretiming slot at the point that ρCV juncture insertion applies. The analysisof (264h-i) would go through easily if the long vowel is doubly linkedand if unstressed long initial vowels syncopate. The derivation of (264i)would be:

154 Chapter 7

(265)

a

× ×m

×JncIns−→ [

a

× ×m

×]Trscr−→ × × ×

a

× ×

m

×NCCrepair−→

a

× ×m

×a

× ×m

×

Stress−→ a:ma:m

Syncope−→ ma:m

It could be that initial long vowels are treated differently than mediallong vowels, or that long vowels in ρVC-reduplication are treated differ-ently than long vowels in ρCV-reduplication. While this is plausible, itis not an explanation.

Another potential solution revolves around the ambiguity of thestructural condition /C in the VC-reduplication juncture insertionrule ∅ → ] /C . In Chapter 4, C was taken to be the contradiction ofV. However, if C is taken to be false of a bare timing slot, a:m→ ma:m

can be derived.

(266)

a

× ×m

×JncIns−→ [

a

× ×m

× ]Trscr−→ × × ×

a

× ×m

×

The result is a:ma:m, with the usual final stress pattern of reduplicatedmonosyllables. If unstressed initial long vowels delete just like un-stressed initial short vowels, the desired result is obtained.

7.5 Tohono O’odham plural reduplication

In Washo, there is a class of polysyllabic roots which are lexicallyspecified to reduplicate according to the rules for monosyllabic roots.There is extensive lexically specified irregularity in Tohono O’odhamreduplication and developing an analysis in which lexically specifi-cation is straightforward is the primary concern. Although differingin many significant respects from the analysis of Raimy (2000), theanalysis presented here is indebted both to his approach to the problemand organization of the data. The examples are from Saxton, Saxton,and Enos (1983) and Zepeda (1983).

Case Studies: Tohono O’odham 155

The impressive variation in Tohono O’odham reduplication is evi-dent in the forms below:

(267) to:n to:-ton ‘knee’gimai gi-gimai ‘braggart’bo:l bo-bol ‘a ball’da:k da:-dk ‘nose’gaso ga-gso ‘a fox’

pad paPa-pad ‘badly’sikol siPi-skol ‘circular’lo:ba lo-lba ‘dry goods’go:k goPo-gok ‘(a) two’ban ba:-ban ‘coyote’

This includes all the varieties of reduplication which were classified byRaimy, except for a group of forms with an initial consonant clusterwhich will be discussed below. See his book for many more examplesof each type

As we have done several times already, we attack the problem fromthe standpoint of the language learner trying to deduce the rules ofjuncture insertion which generate these forms. The natural conjectureabout underlying forms (and derivations) is:

(268) a. toon → [to〈o〉]n → too to-n

b. gimai → [gi]mai → gi gi-mai

c. daak → [d〈aa〉]k → daa d-k

d. gaso → [g〈a〉]so → ga g-so

e. ban → [ba〈a〉]n → baa ba-n

f. bool → [bo]l → bo bo-l

g. looba → [l〈o〉]ba → lo l-ba

h. gook → [go〈Po〉]k → goPo go-k

i. pad → [pa〈Pa〉]d → paPa pa-d

j. sikol → [s〈iPi〉]kol → siPi s-kol

The starting point in all cases is left transcription and C∗ν-reduplicationwith a null exponent. The second conjunct of the reduplicant is alwaysreduced to CV or C. I assume that the default is reduction to CV. Callthis operation Default Truncation (DefTrunc). Some roots are lexicallyspecified to undergo Exceptional Duplicant Truncation (XTrunc). This

156 Chapter 7

generates (268a-d), with daak and gaso lexically specified as XTruncroots.

(269) C∗ν-junctures XTrunc DefTrunc TrScr[gi]mai gi gimai

[too]n [to〈o〉]n too ton

[ga]so [g〈a〉]so ga gso

[daa]k [d〈aa〉]k daa dk

Aside from XTrunc, there are three other lexically specified adjust-ments of the duplicant: Exceptional Shortening (XShort), ExceptionalLengthening (XLong), and Exceptional Degemination (XDegem). Theeffect of each should be clear from the derivations below. I assume thatdegemination automatically induces P-insertion to break the resultinghiatus.

(270) a. [boo]lXShort−→ [bo]l

Trscr−→ bo bol

b. [ba]nXLong−→ [baa]n

DefTrunc−→ [ba〈a〉]n Trscr−→ baa ban

c. [goo]kXDegem−→ [goPo]k

DefTrunc−→ [go〈Po〉]k Trscr−→ goPo gok

Not surprisingly, some roots are specified to undergo more than oneexceptional rule.

(271) a. [loo]baXShort−→ [lo]ba

XTrunc−→ [l〈o〉]baTrscr−→ lo lba

b. [pa]dXLong−→ [paa]d

XDegem−→ [paPa]dDefTrunc−→ [pa〈Pa〉]d

Trscr−→ paPa pad

c. [si]kolXLong−→ [sii]kol

XDegem−→ [siPi]kolXTrunc−→ [s〈iPi〉]kol

Trscr−→ siPi skol

Some combinations of lexically specified operations are eitherincoherent (duplicant shortening and degemination, for example) or

Case Studies: Tohono O’odham 157

unlearnable (duplicant shortening followed by duplicant lengthening).The only plausible combinations that are not found in the data areillustrated in (272), with fictitious examples. Some ordering is assumed,with XShort or XLong applying first and XTrunc applying last.

(272) a. [ka]nolXLong−→ [kaa]nol

XTrunc−→ [k〈aa〉]nolTrscr−→ kaa knol

b. [noo]kolXDegem−→ [noPo]kol

XTrunc−→ [n〈oPo〉]kolTrscr−→ noPo nkol

It could be the case that compensatory shortening would apply in (272a)to produce kaknol. If so, the combination of XLong and XTrunc mightnot be learnable. I have no explanation for why examples like (272b)do not occur.

Fitzgerald (2000) claims that the reduplication pattern

CV1V2 . . .→ CV1CV2 . . .

is obligatory if the vowel sequence V1V2 is one of {io, 1o, oa, ua}. Forconvenience in the discussion, call these vowel sequences Type F vowelsequences and call a root with an initial Type F vowel sequence a Type Froot.10 The following examples are from Fitzgerald (p. 715), whoshould be consulted for more examples and glosses:

(273) a. CV1V2 . . .→ CV1CV1V2 . . .

kui ku-kuihoiki ho-hoikiñ1a ñ1-ñ1añ1id ñ1-ñ1id

b. CV1V2 . . .→ CV1CV2 . . .

hiopcig hi-hopcigc1o c1-codoa do-dacuama cu-cama

Raimy (2000:125), however, gives examples to show that not all Type Fverb roots behave as the ones in (273b) do.

Assuming that the CVV sequence in the roots in (273) is bisyllabic,the reduplication pattern in (273b) is the result of exceptional truncation.

(274) [cu]amaXTrunc−→ [c〈u〉]ama

Trscr−→ cu c ama

158 Chapter 7

In terms of the analysis above, the observations of Fitzgerald and Raimycan be summarized by saying that the default for Type F roots is theclass of XTrunc verbs, but there are exceptions.

Fitzgerald develops an OT account of the reduplication propertiesof Type F roots in terms of hiatus avoidance for these sequences in theoutput of reduplication. In the theory developed here, avoidance of suchsequences may well be the reason that certain verb roots tend to be put inthe XTrunc class. But this does not mean that hiatus avoidance must bereflected directly in the form of the rule which achieves it. To the extentthat the hiatus avoidance account is correct, it is an account of whythese particular roots tend to be put in the XTrunc class. Exceptionaltruncation itself is quite independent of hiatus avoidance and many rootsin the XTrunc class have long initial vowels, for which the issue ofhiatus avoidance is moot. There is no imperative to include a mentionof hiatus avoidance in the grammar. See Frampton (2002) for somediscussion of the distinction between understanding the pressures thatphonotactics may exert for including a particular rule in the grammarand understanding what the rule is and how it interacts with the rulesystem it is embedded in.

Finally, there is one more set of forms to consider. There are somestop-liquid initial consonant clusters in Tohono O’odham. They areassociated with a characteristic reduplication pattern which is not sofar accounted for.

(275) tla:mba tlalamba t-la-la-mbapla:mba plalamba p-la-la-mbakla:wao klalawao k-la-la-waoplo:mo plolomo p-lo-lo-motlo:gi: tlologi: t-lo-lo-gi:

This is (99) from Raimy.The simplest account is to suppose that the initial stop is excluded

from the domain of the reduplicative plural affix. Presumably, it isunsyllabified, although I have no other evidence for this. All of the verbsin (275) must also be lexically specified as XShort verbs. A typicalderivation is:

Case Studies: Tohono O’odham 159

(276) ploomoJncIns−→ ploo]mo

DC−→ p[loo]moXShort−→ p[lo]mo

Trscr−→ p-lo-lomo

I have no explanation for why shortening obligatorily applies inthese forms. The verb roots in question form a small class withrelatively similar form, so it is perhaps not surprising that they all havethe same irregularity. If there were no lexically specified adjustmentrules (i.e. no irregularities), we would expect, for example, tla:mba →t-la:-la-mba , with regular prosodic truncation of the second conjunctto CV.

In closing the discussion of Tohono O’odham, it should be notedthat it is extremely simple to code the irregularities in the system pre-sented above. There are four exceptional subclasses of verbs: shorten-ing, lengthening, degeminating, and exceptionally truncating.11 Mem-bership in each of these verb classes has a simple effect, the applicationof a special duplicant adjustment rule. The Tohono O’odham learnermust carry out such coding in some way. A complex system ofirregularities like this presents a serious challenge to OT. It is not atall clear that OT approaches to reduplication can carry out this codingin a plausible fashion. If it were the case, for example, that a hierarchyof penalty functions could be developed so that the irregularities weredetermined by turning off and on particular penalty functions, a systemof comparable complexity might be possible. But it is extremelydubious that this can be done. If each class of irregularities requiresa reranking of the penalty functions, the complexity of such a systemargues strongly against it. Recall that the irregularity classes overlap anda root can be in as many as three different classes. This complicates anattempt to account for the irregularities by reranking. No OT analysis ofthe complexities of Tohono O’odham reduplication has been attempted,so no comparison can be made at this point.

160 Chapter 7

7.6 Infixing consonant copy: Temiar verbal reduplication andLevantine Arabic perjorative reduplication

Temiar is a Mon-Khmer language spoken in Thailand and Malaysia.Both Levantine Arabic and Temiar realize certain morphemes as infix-ation of a copy of an edge consonant. The similarity between the twomechanisms was noted by Broselow and McCarthy (1983).

7.6.1 Temiar verbal reduplicationReduplication in Temiar has attracted a great deal of attention in theliterature on reduplication. See in particular Gafos (1999) and Raimy(2000) and the references cited there. The interest stems from the factthat Temiar reduplication patterns are resistant to a plausible treatmentin prosodic terms.

A variety of reduplicative affixes in Temiar will be considered inthis section. Verb roots in Temiar have the form (C)CV(V)C. Variousverbal inflections appear as infixation at the left edge of the CVC portionof the root. The infixes below have been boldfaced.

(277) simulfactive causative continuativeslOg (root) salOg srlOg sglOgg@l (root) gag@l trg@l glg@lsrlOg (causative stem) sralOg srglOgtrg@l (causative stem) trag@l trlg@l

The simulfactive and causative affixes have exponents, a and r respec-tively. CVC roots (g@l above) are augmented with an initial consonantprior to infixation. In the causative, epenthetic t extends the root.Otherwise, the root is extended by geminating the initial consonant.The continuative suffix has a null exponent, but infixes a copy of thefinal consonant of the stem.

We first address the question of the infixation site, the left edge ofthe CVC portion of the stem. How is the left edge of the CVC portionof the stem located with the limited resources available in DR? I assumethat verbal stems have a natural bipartite prosodic structure:

consonantal extension + CV(V)C kernal

Case Studies: Temiar 161

Opinions have varied about the prosodic structure of the consonantalextension in the literature. Some have proposed a category of “minorsyllables” for the consonantal extension. Others have proposed that theextension consonants remain unsyllabified. It is consistent with bothproposals to assume that the extension is prosodically inferior in someway and that the stem kernel is singled out as the domain of verbalaffixation.

We can now address the question of the special treatment of CVCstems. If it is a requirement for affixation that the stem have a nonemptyconsonantal extension, presumably because there must be a suitableinfixation site, roots like g@l could not be inflected if no supplementarymechanism is provided. Prosodic requirements on the stem which areimposed by affixes are sometimes met by modification of the stem priorto affixation. CVC stem augmentation in Temiar is such a modification.The modification which is made is t epenthesis at the left edge of thestem in the causative and initial consonant gemination in the simulfac-tive and continuative.

Traditional grammars of Senoic, the language family to whichTemiar belongs, describe the morphophonology in just this way.

. . . the most common morphological process in Senoic is infixationof the final consonant between two initial consonants; e.g., cPu:l ‘tochoke’ → clPu:l ‘to be choking’. In roots having only a simpleinitial, e.g. ci:p ‘walk’, the final is inserted between the initialconsonant and its reduplication: ci:p → cci:p → cpci:p, ‘to bewalking’. This is a productive process in verbs which I calledelsewhere the Indeterminate mode. Diffloth (1976)

The continuative affix has a null exponent, designates the CVCkernel of the stem as its domain, extends the stem if necessary, andtriggers the juncture insertion rules:

∅ → [/

×, ∅ → 〉/Right ×

For example:

162 Chapter 7

(278) a.

s

×l

×o

×g

×JncIns−→

s

×[

l

×o

×〉g

×DC→

s

×[〈l

×o

×〉g

×]Trscr−→

s

× ×l

×o

×g

×

b.

g

×@

×l

×gemination−→

g

× ×@

×l

×JncIns−→

g

× ×[@

×〉l

×DC→

g

× ×[〈@

×〉l

×]

Trscr−→ × × ×

g @

×l

×

The simulfactive affix is -〉〈a〉] and the causative affix is -〉〈r〉], bothwith the same domain as the continuative affix and both triggering stemextension, if required. For example:

(279)

s

×l

×o

×g

×domain︷︸︸︷ ⊕CAUSE−→

s

×l

×o

×g

×〉〈r

×〉]domain︷︸︸︷ DC

→s

×[〈l

×o

×g

×〉〈r

×〉]

Trscr−→

s

× ×rl

×o

×g

×

Note that DC depends heavily on the specification of the domain (theCVC kernel of the root) boundaries to locate the insertion point for thejunctures which close the duplicant and truncate.

The continuative pattern is somewhat different in Semai, a closelyrelated language.

(280) root continuativebP@l blbP@l ‘painful embarrassment’dNOh dhdNOh ‘appearance of nodding’kmrPE:c kckmrPE:c ‘short, fat arms’

Raimy (2000:149) makes the valid point that an adequate theoryof reduplication in Temiar should make clear its connection with Semaireduplication. Two straightforward modifications of the Temiar specifi-cations above give the correct results for Semai. First, there is no specialdomain specification. Second, all stems are extended by geminating the

Case Studies: Levantine Arabic 163

initial consonant, not only CVC stems, as in Temiar. The continuativeaffixes in Temiar and Semai trigger the same juncture insertion; the ×∗-rule and ∅ → 〉 /Right ×.

(281)

b

×P

×@

×l

×gemination−→

b

× ×P

×@

×l

×domain︷ ︸︸ ︷ JncIns

−→b

× ×[P

×@

×〉l

×

DC→

b

× ×[〈P

×@

×〉l

×]Trscr−→ × ×

l

×b P

×@

×l

×

Note that the default domain is the stem, before it is readjusted bygeminating the initial consonant.

7.6.2 Levantine Arabic perjorative reduplicationBroselow and McCarthy (1983) called attention to the similarity be-tween continuative reduplication of CCVC (triconsonantal) roots inTemiar and intensive/pejorative reduplication of triconsonantal stems inLevantine Arabic. They give the following examples (and many more):

(282) stem reduplicatedmarat. marmat. ‘cut unevenly’barad barbad ‘shaved unevenly’sah.at. sah. sat. ‘dragged unevenly’

Descriptively, a copy of the initial consonant is postposed to the positionfollowing the second consonant. The semantics above are perjorative,but sometimes the semantics are essential intensive. I will call themorpheme INT in what follows, simply for convenience.

The affix can be described as having a null exponent and triggeringthe juncture insertion rules ∅ → 〈 /× and ∅ → ] /Right V. Theeffect of these rules is illustrated below:

164 Chapter 7

(283)

b

×a

×r

×a

×d

×⊕ INT−→

b

×〈a

×r

× ]

a

×d

×DC→ [

b

×〈a

×r

×〉]a

×d

×

Trscr−→

b

×a

×r

× ×a

×d

×

Transcription in (283) is taken to be to the right. The transcriptionoperation is considerably simpler if transcription is to the right, sincetruncation is then carried out by noncopying. This should be comparedwith the Temiar continuative in (278a). The output forms are close tobeing mirror images of each other.

Although this analysis does produce the correct results, it has oneproblematic aspect. No direct reference to the “second consonant”is made in the juncture insertion rules. The formalism developed tothis point does not allow it. The insertion point of the ]-juncture,immediately after the second consonant, is identified as “before the finalvowel” ( / V). Given the consonant oriented morphology of theSemitic languages, a more direct way of locating this insertion site isdesirable. The simplest analysis is that reduplication actually applies tothe consonantal root, before vowels are inserted into the representation.

(284) brd⊕ INT−→ b〈r]d

DC→ [b〈r〉]d

Trscr−→ brbd

In order for this analysis to go through, the domain of the affix mustconsist of the initial consonant pair. It stretches usual notions of footstructure, but it is plausible that in a consonant oriented language thatfooting can take place in the absence of syllable structure, with eachconsonant taken to be a footable element. The initial consonant pairis an initial binary foot under this interpretation. Assuming that this ispossible and that the domain of the intensive affix is this initial foot, thenthe juncture insertion rules are simply the ×∗-rule and ∅ → 〈 /× .

Intensive reduplication applies to biconsonantal roots. The bicon-sonantal root lf, for example, has the simple verb form lafaf and theintensive/pejorative form laflaf. It is suggested in Section 7.10.2 thatin Chaha a biconsonantal root like lf is stored in the lexicon as l[f,with an embedded duplication juncture. This then triggers transcription

Case Studies: Sanskrit 165

after the root is inserted into the morphophonological computation. Ifthis is true in Levantine Arabic as well, it could be that there is notranscription in the root cycle and intensive/pejorative reduplicationapplies directly to l[f, producing [l〈[f. Default Closure then yields[l〈[f ]〉], with nested duplicants. This is then transcribed as discussed inSection 4.2, producing lflf. Duplicants are transcribed from the insideout.

(285) [

l

×f

×〈[ ]〉]inner

transcription−→ [

l

× 〈f

× 〉] ×outer

transcription−→

l

×f

× × ×

The resulting representation is identical to the representation producedby totally reduplicating lf.

7.7 Sanskrit intensive and perfective reduplication

This section should be read as a revision of Steriade (1988), whichcarefully unravels the relevant Sanskrit phonology. The major differ-ences are over the reduplication process itself. Most of the phonologicalinsights, organization of the empirical data, and highlighting of crucialexamples are due to Steriade.

Various Sanskrit verbal inflectional morphemes trigger reduplica-tion. There are five varieties of prefixal reduplication: perfect stem,desiderative, present stem, aorist, and intensive, and one variety ofpresent stem infixation. This section focuses on perfect and intensivereduplication, both of which are quite productive and reveal all themechanisms which are employed in Sanskrit reduplication. The par-ticular interest of Sanskrit for the general theory is the extent to whichSanskrit specialization of the NCC repair process shapes the surfaceform of the reduplicant. The discussion of “Shortcut Repair” in Kolami,Ponapean, and Hausa in Section 5.3 has already introduced the ideaof language particular specialization of the repair process. Sanskritspecializations have a much more pervasive effect.

Descriptively, perfective reduplication prefixes a (C)V syllable tothe verb root to create what is called the “perfect stem.” Intensivereduplication prefixes a CaX syllable to the verb root. Both kinds ofreduplication interact extensively with a widespread process of low

166 Chapter 7

vowel syncope which many roots are subject to in certain prosodicenvironments. The syncopated root is called the zero grade form ofthe root in the Sanskrit literature, contrasting with the full grade form.

7.7.1 Low vowel syncopeThe underlying vowel inventory in Sanskrit is {a, i, u}. Underlying verbroots are heavy monosyllables. Except for a few roots with long highvowels, the nucleus of underlying roots is the low vowel a. For verbroots which are subject to low vowel syncope, and whose reduced formhas a vocalic nucleus, the vowel of the perfect prefix is the nuclear vowelof the reduced form. This is shown in (286). In those roots which do notsyncopate, the vowel of the perfect prefix is a. Explaining this relationis the major challenge in analyzing Sanskrit reduplication.

(286) full-grade 0-grade perfect stema. suap ‘sleep’ svap sup su.svapb. suaj ‘embrace’ svaj sa.svajc. miaks. ‘be situated’ myaks. miks. mi.myaks.d. tiaj ‘forsake’ tyaj ta.tyaje. uas ‘shine’ vas us u.vasf. uas ‘clothe’ vas va.vas

Underlying u surfaces as a v aspirant in syllable initial onset position(as in German) and as a round labial glide in noninitial nonnuclearposition. Sanskrit orthography uses “v” in both cases. A “y” is used fornonnuclear occurrences of underlying i. All sonorants can be nuclearin Sanskrit and syncopated roots frequently have a nonvocalic nucleus.Roots of this kind invariably have the vowel a in the perfect prefix.

The puzzle is to explain why, for those roots which syncopate andwhose syncopated root has a vocalic nucleus (286a,c,e), the vowel ofthe perfect prefix is the same as the nuclear vowel of the reduced root,and for those roots which do not syncopate (286b,d,f), the vowel ofthe perfect prefix is invariably a. Solving this puzzle led Steriade toconclude that the computation of the perfect prefix was accomplished byfirst copying the entire root, then computing its 0-grade reduction, thenidentifying the eventual prefix as a subpart of this 0-grade reduction anddeleting the superfluous material. For roots whose syncopated form hasa vocalic nucleus, this gives the correct results. For roots whose reduced

Case Studies: Sanskrit 167

form has a syllabic liquid, the proposal makes the wrong prediction.Steriade predicts derivations such as those in (287). For roots with asyllabic nasal in the reduced form, tan in (287), the correct surface formis obtained because syllabic nasals in Sanskrit surface as the low vowela. For roots with a syllabic liquid in the reduced form, mard in (287c),Steriade’s analysis produces the incorrect outcome, as she notes. She isforced to stipulate a rule converting such syllabic liquids to a in certainreduplicative contexts.

(287) root tan mardfull copy tan.tan mard.mard0-grade tn­ .tan mr­d.mardtruncation mr­ �d.mardsurface form ta-tan mr­-mard

Even more serious than this empirical problem is the fact that theanalysis relies on an incorrect assumption about the accentual condi-tions under which low vowel syncope applies. Steriade assumes that lowvowel syncope applies in all unstressed syllables. Studies of Sanskritstress, however, conclude that the environment for low vowel syncoperequires an unstressed syllable which is followed directly by a stressedsyllable. See Halle (1997:292), for example. We can see why thisshould be so from one of Halle’s examples. The noun svasar ‘sister’bears inherent stress on the first syllable. When it combines with astressed suffix, the singular dative suffix -e, for example, low vowelsyncope applies and svas.re (svas�are) results. (Only the leftmost stresssurfaces.) When it combines with an unstressed suffix, the singularaccusative suffix -am, for example, low vowel syncope does not apply.Instead, sva.sar.am → sva.sa.ram . (Compensatory lengthening accom-panies the recruitment of the root final consonant to be the onset of thefollowing syllable). The deciding factor excluding svas.ram (parallel tosvas.re) must be the absence of inherent stress on the suffix, not simplythe absence of stress on the final syllable of the root.

Without the assumption that low vowel syncope applies in allunstressed syllables, the derivations which Steriade proposes do not

168 Chapter 7

go through. According to Steriade (p. 123), the derivation of the 3sgperfect form bu-budhus, from the root baudh ‘know’, is:

(288) full copy baudh-baudh-uslow vowel syncope bud -bud-ustruncation bu-bud-us

As we have seen, however, there is no justification for the boxedoutcome since the following syllable is unstressed.

An account of the interaction of low vowel syncope and reduplica-tion in the DR framework requires an underlying structural differencewhich leads, on the one hand, to the syncopated root or, on the other,to the correct reduplicant. The starting point is the observation thatalthough a certain accentual environment is certainly required, somecondition on the ease of resyllabification is also in force. The rootpat ‘fly’, for example, syncopates if it is an environment in whichthe onset and coda can be incorporated into adjacent syllables, butnot otherwise. So pat has the inflected perfect form pa-pt-us, withsyncopation, but the root does not syncopate in the form pat-ı-ta. Thesame is generally true of CaC roots. For another example, the root nau

‘praise’ syncopates to nu before C-initial stressed suffixes, but not informs like nau-anti → na.vanti , in which u is recruited as an onset tothe following syllable.

The root suap syncopates to sup, but the root suaj does not synco-pate. How can this difference be related the possibilities of resyllabifi-cation? I assume that the difference is in the nuclear structure of the tworoots. suap has a long nucleus (the diphthong ua) and suaj has a shortnucleus. Furthermore:

(289) Resyllabification Condition: Syncope requires the possibility ofresyllabification without altering nuclear structure.

Since suap has a long nucleus, syncope simply shortens the nucleus.No resyllabification is necessary. The root suaj, on the other hand, has ashort nucleus. Syncope removes the nucleus, leaving three unsyllabifiedconsonants. There is no context which can absorb all three consonantsinto the existing syllable structure. The only possibility for resyllab-ification is promotion of u to be a new nucleus, but this alteration ofnuclear structure is not allowed by (289).

Case Studies: Sanskrit 169

What accounts for the difference in nuclear structure? One mightspeculate that the phoneme features of u in suaj differ from the phonemefeatures of the u in suaj, somehow preventing it from becoming nuclear.Steriade argues convincingly that this cannot be the case. The crucialfact is that for roots like duais, with both a pre-a vocoid and a post-avocoid, it is invariably the post-a vocoid which becomes the nucleus ofthe syncopated root. It is impossible to explain this if vocoids can bearan underlying feature which prevents them from becoming nuclear.

I assume that roots with a potential falling diphthong syllabify withlong nucleus, unless there is lexical marking to the contrary. Other rootssyllabify with a short nucleus, again unless there is lexical marking tothe contrary. Crucially, the lexical marking [±long nucleus] (or simply[±LN]) makes no mention of particular phonemes which should beincluded in the nucleus. In these terms, suap is marked [+LN] andsuaj is unmarked for [LN]. The root duais is unmarked for [LN] andtherefore syllabifies with a long falling nucleus, dves. If it were marked[−LN], it would syllabify with a short nucleus. There is no markingwhich causes it to syllabify with a rising nucleus.12

In terms of lexical marking for [LN], we analyze the examples in(286) as:

(290) nucleus full-grade 0-gradea. suap [+LN] ‘sleep’ suap svap sup

b. suaj ‘embrace’ suaj svaj

c. miaks. [+LN] ‘be situated’ miaks. myaks. miks.d. tiaj ‘forsake’ tiaj tyaj

e. uas [+LN] ‘shine’ uas vas us

f. uas ‘clothe’ uas vas

Long nuclei have been underlined.Several additional examples follow:

(291) nucleus full-grade 0-gradea. mard ‘crush’ mard mard mr­db. nard [−LN] ‘bellow’ nard nard

c. skand ‘leap’ skand skand skn­ dd. siand ‘move on’ siand syand syn­ de. mand [−LN] ‘exhilarate’ mand mand

170 Chapter 7

With an analysis of low vowel syncope in hand, we are now in aposition to analyze perfective and intensive reduplication.

7.7.2 Perfective reduplicationThe morpheme which induces perfect stem formation triggers C∗ν-reduplication. Prosodic adjustment applies after juncture insertion, witha CV reduplicant (first conjunct) target.

(292) First conjunct :

CV

vocalic nucleusnonlow nucleus

obstruent C

::

]-Left

medial truncationedge truncation

First, a vocalic nucleus is preferred to a nonvocalic nucleus, so]-Left is used in (293a) to shorten the nucleus, but medial truncationis used in (293b).

(293) a. [marν

]d → [maν

]rd → ma ma rd

b. [graν

]bh → [g〈r〉aν

]bh → ga gra bh

The final forms above are supposed to represent the post-transcriptionstructure. We will see below that some further reduplication specificmodification of the reduplicant takes place in the course of NCC repair.The surface form on (293b) is ja-grabh, for example.

A nonlow vocalic nucleus is preferred to a low vocalic nucleus, so]-Left is used in (294a) to shorten the nucleus, but medial truncation isused in (294b).

(294) a. [suaν

]p → [suν

]ap → su su ap

b. [bauν

]dh → [b〈a〉uν

]dh → bu bau dh

Case Studies: Sanskrit 171

An obstruent onset is preferred to a non-obstruent onset, so medialtruncation trims the onset in (295a), but edge truncation is used in(295b).

(295) a. [kranν

]d → [kraν

]nd → [k〈r〉aν

]nd → ka kra nd

b. [skanν

]d → [skaν

]nd → [〈s〉kaν

]nd → ka ska nd

One final example shows that medial truncation is preferred to edgetruncation.

(296) [mnaaν

] → [mnaν

]a → [m〈n〉aν

]a → ma mna a

The repair rule ordering in (292) is crucial in (296). Medial truncationis ordered before edge truncation. If it were not, it would be possible toproduce na-mnaa.

7.7.3 Sanskrit specialization of NCC repairWhen copies of root consonants appear in the reduplicant, velar conso-nants are palatalized and aspirated consonants are deaspirated. Steriade(p. 106) argues convincingly that this should be seen as dissimilation.How and when in the derivation is it carried out? The simplest assump-tion is that it is carried out at the point that fission creates a separatereduplicant phoneme in the course of NCC repair. The repair of theform produced by transcription after C∗ν juncture insertion and perfectprosodic adjustment is illustrated in (297) for the roots skand ‘leap’,dhar ‘hold’, and khad ‘chew’.

(297) NCC repair, with dissimilation

a. × ×s

×k

×a

×n

×d

× →c

× ×s

×k

×a

×n

×d

× →c

×a

×s

×k

×a

×n

×d

×

b. × ×dh

×a

×r

× →d

× ×dh

×a

×r

× →d

×a

×dh

×a

×r

×

c. × ×kh

×a

× ×d

× →c

× ×kh

×a

× ×d

× →c

×a

×kh

×a

×a

×d

×

172 Chapter 7

In (297c) there is both palatalization and deaspiration.

7.7.4 Intensive reduplicationLike perfect reduplication, the process of intensive reduplication beginswith C∗ν juncture insertion. The reduplicant onset is shaped by prosodicadjustment and NCC repair exactly as in perfect reduplication. For rootswith a falling diphthong or long a, this is all that needs to be said.

(298) transcription surfacea. baudh bau-baudh bo-bodh

b. mard mar-mard mar-mard

c. duais dai-duais de-dves

d. sa:d sa:-sa:d sa:-sa:d

e. gaup gau-gaup jo-gop

f. sphaut phau-sphaut po-sphot

The onset is truncated in (298c,f) just as in perfective reduplication.NCC repair, which follows transcription, also carries out dissimilationand deaspiration.

Consideration of roots with a short nucleus show that there isprosodic adjustment of the reduplicant to a bimoraic syllable. The rootsin (299b,c) are lexically specified [−LN], so they syllabify with a shortnucleus in spite of the fact that a post-a sonorant is present.

(299) C∗ν-junctures intensivea. pat [pa]t pa:-pat

b. nard [na]rd na:-nard

c. tan [ta]n tan-tan

The first two examples use FCVL (First Conjunct Vowel Epenthesis, asin Mokilese) to achieve a bimoraic duplicant, but the last uses ]-Right.Illustrative derivations are given in (300). The full first conjunct syllablestructure is given.

Case Studies: Sanskrit 173

(300) a. [

n

×a

×]

r

×d

×ν σµ

FCVL−→ [

n

×a

×〈×〉]r

×d

×ν σµµ

Trscr−→ × × ×

n

×a

×r

×d

×

b. [

t

×a

×]

n

×ν σµ

]-Right−→ [

t

×a

×n

×]

ν σµµTrscr−→ × × ×

t

×a

×n

×

The choice between FCVL and ]-Right is not completely pre-dictable, although there are some clear tendencies. Obstruents are neverincorporated into the duplicant. Since post-a vocoids (u or i) almostalways syllabify into a long nucleus, the issue of ]-Right vs. FCVLarises only in the case of post-a liquids and nasals. Nonnuclear post-anasals are much more likely to be drawn into the intensive reduplicantthan nonnuclear post-a liquids, (299c) vs. (299b), for example. Fur-thermore, root nonnuclear post-a sonorants are more likely to be drawninto the intensive reduplicant if they are root final, as in (300c), so thatthe reduplicant and root are coextensive. I will not explore the issuefurther, but note that some lexical specification is needed to cover theexamples which counter the general tendencies noted above. It is alsoworth noting that this variation, which in DR is a variation betweenchoice of prosodic adjustment strategies, has no natural explanation inSteriade’s theory of whole stem copying and reduction. It is not easyto understand why na-nard should result in that theory rather than nar-

nard.The prosodic adjustment rule, which has been discussed somewhat

informally to this point, can be now be stated precisely:

(301) First conjunct :[CVX, X sonorantobstruent onset

]::

]-Right, FCVL

medial truncationedge truncation

Medial and edge truncation are relevant to determining the onset, justas in perfect reduplication.

174 Chapter 7

We have not yet considered roots with rising diphthongs. Thepredictions made by the theory as developed to this point are incorrect.The roots are the left have long nuclei, which are underlined.

(302) prediction actuala. suap sua-suap saa-suap

b. grabh jra-grabh jaa-grabh

c. kriid cii-kriid cai-kriid

It is relatively straightforward to describe the effect: the nucleus of thereduplicant is always a-initial. It is less straightforward to specify themechanism by which this comes about.

I assume that this is part of the NCC repair process; a furtherSanskrit particular interpenetration of the repair process and ordinaryphonological processes. Unlike dissimilation, it is morpheme specific,triggered only by realization of the intensive morpheme. Illustrativederivations are given below:

(303) Intensive NCC repair (dissimilation, deaspiration, lowering)

a. ×σ

× ×k

×r

×σ

i

× ×d

× →c

×σ

× ×k

×r

×σ

i

× ×d

×

→c

×σ

a

× ×k

×r

×σ

i

× ×d

× →c

×σ

a

×i

×k

×r

×σ

i

× ×d

×

b. ×σ

× ×s

×σ

u

×a

×p

× →s

×σ

× ×s

×σ

u

×a

×p

×

→s

×σ

a

× ×s

×σ

u

×a

×p

× →s

×σ

a

× ×s

×σ

u

×a

×p

×

Case Studies: Lushootseed 175

7.7.5 Low vowel syncope and intensive reduplicationAlthough there are no examples in the literary record, the Sanskritgrammarians claimed that the intensive prefix of those roots which wehave identified as having a long rising nucleus (ia, ua, ra, etc.) dependedon whether or not the root appeared in the full grade or zero grade. Theroot svap/sup, for example, has present active intensive forms s ´asvapmi(1sg) and sosupmas (1pl). This indicates that the intensive morpheme isrealized late, after the morpheme whose realization induces low vowelsyncope. The domain of the intensive reduplicative affix is the root.Two illustrations of intensive reduplication applying after syncope aregiven below.

(304) a.

s

×u

×p

× . . .JncIns−→ [

s

×σ

u

×]

p

× . . .PrAdj−→ [

s

×σ

u

×〈×〉]p

× . . .

Trscr−→ × × ×

s

×u

×p

× . . .NCCR−→

s

×a

×u

×s

×u

×p

× . . . (so-sup. . . )

b.

g

×σ

r

×b

×h

× . . .JncIns−→ [

g

×σ

r

×]

bh

× . . .PrAdj−→ [

g

×σ

r

×〈×〉]bh

× . . .

Trscr−→ × × ×

g

×r

×bh

× . . .NCCR−→

j

×a

×r

×g

×r

×bh

× . . . (jar-gr­bh. . . )

7.8 Cyclicity and double reduplication in Lushootseed

This section is based on Urbanczyk (2000). The examples and basicgeneralizations are all hers. The theoretical framework, hence the anal-ysis, is different. Lushootseed has three kinds of verbal reduplication.The particular interest of Lushootseed for the general theory is thatthere are many examples of verb forms in which two different kindsof reduplication have applied.

176 Chapter 7

7.8.1 Diminutive reduplicationIn surface form, diminutive inflection is a CV-prefix whose C is theinitial C of the stem and whose vowel is either the initial V of the stemor, in certain contexts, the vowel i. In several examples below, irregularP coda epenthesis in a stressed open syllable results in a CP prefix at thesurface. I assume this is part of post-reduplication phonology.

The examples below are from Urbanczyk (1996).

(305) CV-inital stems (V �= @)a. cal@s hand ca ca l@s little hand

b. s-dukw bad s- du P du kw riff-raff

C@-initial stemsc. t@law-il run tı t @law’-il jog

d. x@c-bid afraid xı P x @c-bid a little afraid of it

CCV-inital stemse. c’ń∼’aP rock cı P- c ’ń∼’aP little rock

f. c’qays@b flower s- c’ı c’ qays@b flower

CV:-initial stemsg. s-du:kw knife s- dı - d u:kw small knife

h. bu:s four bı P b u:s four little items

Particularly straightforward examples have been chosen above. Theoverall corpus is not as clearcut as these examples might indicatebecause there are a significant number of irregularities and extensivevowel syncope. See Urbanczyk for a careful discussion of all ofthese issues. There are only a few examples in the data of diminutivereduplication of the later two varieties, but many examples of diminutivereduplication of C@-initial stems.

Urbanczyk gives arguments that initial consonant clusters are bro-ken up by what she terms a “voiceless schwa.” Something along thoselines is forced by the Optimality Theory theoretical framework sheadopts since that theory does not admit intermediate levels. But herarguments translate into arguments that at some intermediate level, theapparent CC cluster is actually C@C and the vowel later syncopates. Iassume that the CCV-initial stems in (305) are C@CV-initial at the point

Case Studies: Lushootseed 177

of juncture insertion. The examples categorized as CCV-initial abovetherefore fall together with those classified as C@C-initial.

Pretranscription forms can be easily identified, assuming left tran-scription.

(306) Exceptional cases pretranscription formC@ t@law → ti-t@law [t〈i〉]@lawCV: bu:s → bi-bu:s [b〈i〉]u:s

Otherwisedukw → du-dukw [du]kw

The pretranscription forms suggest that juncture insertion is initial con-sonant reduplication, since the initial consonant is the element commonto the duplicant in all cases. Additionally, there must be prosodicadjustment with a CV target prosodic shape and both ]-Right and FCVE(First Conjunct Vowel Epenthesis) prosodic adjustment repair rules,with epenthetic vowel i.

(307) a. [t]@law → [t〈i〉]@law FCVEb. [b]u:s → [b〈i〉]u:s FCVEc. [d]ukw → [du]kw ]-Right

An explanation is needed for the choice of FCVE in (307a) and (307b).Stress in Lushootseed diminutive forms is on the reduplicant.

Stressed schwa, while it occasionally occurs in Lushootseed, is un-common. The prosodic adjustment rule for diminutive reduplicationis designed to avoid avoid stressed schwa by employing FCVE ratherthan [-Right when the stem-initial vowel is schwa. Technically, a *C@constraint is imposed on the duplicant in prosodic adjustment. The ideathat a rule is “designed” in a certain way merits some discussion becausethere is widespread confusion about this idea. If one views stressedschwa as something that the Lushootseed phonological system tries toavoid, we can ask at what level the solution is found. It could be that inthose instances in which stressed schwa occurs, some corrective actionis taken. Alternatively, it could be that operations which produce astressed schwa in an environment in which they are liable to be stressedare disfavored (in the space of possible operations) and therefore modi-fied so that they avoid producing schwas in those environments.13 Thestatement that the prosodic adjustment rule is designed to avoid stressed

178 Chapter 7

schwa should be taken to mean that Lushootseed solves the stressedschwa problem in reduplication by choosing its reduplication rules in acertain way.

It is harder to understand why diminutive reduplication treats CV:stems as a special case: du:k → di-du:k , using FCVE, not du-du:k,using ]-Right. It could be that dissimilation with distributive redupli-cation (at the level of design) is responsible. Distributive reduplicationgenerally produces a CVC prefix, but in cases where there is initial CV:,a CV prefix results, du:k → du-du:k , for example. Since there are onlytwo examples of distributive reduplication in the data with a long vowel,this matter will be left without further comment.

Note the similarity to Mokilese reduplication. Juncture shift isused where possible and First Conjunct Vowel Epenthesis is used wherejuncture shift does not obtain the desired form. The same is true inMokilese, except that only a timing slot is epenthesized, yielding a longvowel in the first conjunct.

7.8.2 Distributive reduplicationGenerally, distributive inflection is accomplished by prefixing a copy ofthe initial stem CVC.

(308) a. saqw’ fly saqw’ saqw’ fly here and there

b. g@lk’ entangle P@s- gel g@l k’ all tangled up

c. past@d Caucasian pas pas t@d many white folks

Note the modification of @ under stress in (308b).There are two environments in which the distributive prefix is not

CVC. If the stem is CV:-initial, the prefix is CV. This suggests thatjuncture insertion is ∅ → ] /V and prosodic adjustment is drivenby C-finality, with repair limited to ]-Right. In the case of CV:-initialstems, C-finality via ]-Right is impossible, so there is no adjustment.Urbanczyk (1996:180) makes the important observation that in thespecial case that the stem is C1VC2 . . ., with C1 = C2, then the surfacereduplicant is C1V, not C1VC2.

Case Studies: Lushootseed 179

(309) a. t’it’-@b bathe t’i-t’it’-@b bathe for a whileb. wıw’su little wıwıw’su little (plural)c. l@lwaPs@d sleeping l@-l@lwaPs@d sleeping

platform platforms

Urbanczyk also observes that sequences of identical phonemes (eithergeminate or not) are quite rare in Lushootseed and attributes the depar-ture from the CVC shape requirement to avoidance of this configuration.

The simplest way to incorporate this idea into the present analysisis to suppose that elimination of the sequence of identical phonemes iscarried out as part of the NCC repair process. For (309c), for example:

(310) [

l

×@

×l

×] . . .Trscr−→ × × ×

l

×@

×l

× . . . →l

× × ×l

×@

×l

× . . .

→l

×@

× ×l

×@

×l

× . . . →l

×@

×l

×@

×l

× . . .

Distributive reduplication of stems which (on the surface) beginwith a consonant cluster support the idea that there is an interveningschwa at some level of representation.

7.8.3 “Out of control” (OC) reduplicationFirst, some examples:

(311) Pıb@s walk, travel or journey over land by any means

P ıb ib @s pace back and forth, walk without achieving

a destinationsaqw’ fly

s aqw’ aqw’ fly around, wheeling in the sky

dzz@kw’ fall, topple over

dz @kw’ @kw’ totter, teeter back and forth, stagger

The initial VC of the stem is copied. Right copy avoids internal copyingin the many examples in which the stem is CVC, so I assume that this

180 Chapter 7

is the direction of copy. Above, the copy is shaded and the remnant isboxed.

7.8.4 Double reduplicationGiven the basics of diminutive, distributive, and OC reduplication, weare in a position to examine their interactions. It is anticlimactic.Once Urbanczyk’s insights into the phonology are incorporated into theanalysis, the interaction of diminutive and distributive reduplication isexactly what we expect. Both orders of application are possible.

(312) Lushootseed (Central Coast Salish)a. b@daP child,offspringb. bı-b@daP young child DIMc. b@d-b@daP children DISTd. bı-b@d-b@daP young children DIM after DISTe. bı-bi-b@daP litter (of animals) DIST after DIM

Derivations are given below:

(313) b@daP b@daP1. [b]@daP ⊕DIM [b@]daP ⊕DIST2. [b〈i〉]@daP PrAdj1 [b@d]aP PrAdj3. bi b@ daP Trscr b@d b@d aP Trscr4. [bi]b@daP ⊕DIST [b]@db@daP ⊕DIM5. [bib]@daP PrAdj [b〈i〉]@db@daP PrAdj1

6. bib bib @daP Trscr bi b @db@daP Trscr

7. bibib@daP NCCR2 bib@db@daP NCCR

Note 1: *C@ blocks ]-Right, which forces FCVE.Note 2: Exceptional NCC repair to avoid bb.

There are a few examples in which OC reduplication combines withdistributive or diminutive reduplication.

(314) bali forgetbalal’bali suddenly [. . . ] forgettingcal(a) chase, pursue, catchcaPc@lal’, cic@lal’ almost caught

Case Studies: Chumash 181

The derivations are given below:

(315) bali cal1. [ba]li ⊕DIST c[al] ⊕OC2. [bal]i PrAdj c al al Trscr

3. bal bal i Trscr [c]alal ⊕DIM4. b[al]bali ⊕OC [ca]lal PdAdj5. b al al bali Trscr ca ca lal Trscr

Discussion of the various processes which yield caPc@lal’ or cic@lal’

from ca-cal-al are beyond the scope of this discussion. See Urbanczykfor discussion.

7.9 CVC reduplication in Chumash

The analysis of CVC reduplication in Chumash has a prominent placein McCarthy and Prince’s (1995) “Faithfulness and Reduplicative Iden-tity.” Since McCarthy and Prince claim that Chumash CVC redupli-cation gives evidence for overapplication and Inkelas and Zoll’s (2004)“Reduplication as Morphological Doubling” denies that overapplicationexists, I&Z go to some lengths to show that M&P’s analysis of Chumashis incorrect. I follow Inkelas and Zoll, up to a point, accepting theirconclusions about prefixal morphology and the hierarchical positionof the morpheme which induces CVC-reduplication, as well as theirconclusion that Chumash does not provide evidence for overapplication.But the mechanism which they propose for the computation of redupli-cated forms is unconvincing, as I will try to show. A DR analysis of thedata will be given.

CVC-reduplication is associated with intensive, distributive, repet-itive, or continuative semantic force. We begin with a few typical forms.In the reduplicated form, the remnant is boxed.

(316) reduplicated

a. s-kitwon s- kit kit won ‘it is coming out’

b. k-ni-c’eq k-ni- c’eq c’eq ‘I’m tearing it up’

c. s-ikuk sik s-ik uk ‘he is chopping, hacking’

d. s-is-expec s-i sex s-ex pec ‘they two are singing’

182 Chapter 7

The verb root is on the right and s- and k- are agreement prefixes. It iscommon for various other prefixes (ni- and is- above) to appear betweenthe root and the agreement prefixes.

M&P conclude on the basis of (316a, b) that the morpheme respon-sible for the reduplication (ρ in what follows) is hierarchically adjacentto the root. This leads to a mystery for M&P, since it implies that thefinal consonant of the exponent of a hierarchically dominating mor-pheme in (316c) and (316d) is copied by reduplication. M&P advocatea theory of base-reduplicant correspondence to solve the mystery. Quitesensibly, I&K fail to see a mystery and conclude on the basis of (316c, d)that ρ combines with the stem after the agreement prefix. ChumashCVC-reduplication is then a variety of internal reduplication, in which astem internal substring is duplicated. From this perspective, the problemis to specify a principled mechanism by which the CVC subsequence ofthe stem is singled out for copying, not to explain apparent counter-cyclicity.

I&K examine a wider range of examples than (316) and show thatthe surface structure of unreduplicated verb forms is:

outer prefixes + agreement prefixes + inner prefixes + root + suffixes

Only agreement prefixes and inner prefixes can be involved in redupli-cation. The inner prefixes split into two lexical classes, reduplicatinginner prefixes and non-reduplicating inner prefixes. They can occur inarbitrary linear order. The descriptive generalization about the contentof the duplicant is that the V of the CVC reduplicant is the leftmostvowel which is in a reduplicating inner prefix or the root, along withthe flanking consonants. They can be drawn from any inner prefix,agreement prefix, or the root. Some of the examples that Inkelas andZoll give to illustrate this, taken from Applegate (1972), are given in(317). In the pre-reduplication form, the reduplicating inner prefixesand root are boldfaced. In the reduplicated form, the remnant is boxed.

Case Studies: Chumash 183

(317) reduplicated

k-su-towic k- sut su-t owic I’m doing it fast

s-wati-k’ot s-wati- k’ot k’ot . . . is broken to pieces

k-sili-pi-wayan k-sili- piw pi-w ayan I want to swing

Pal-aqsa-n Pal Pal -aqsa-n the dead

k-xu-ni-y1w k- xun xu-n i-y1w I am looking all

over for its-is-expec s-i sex s-ex pec they two are singing

The generalization translates directly into a DR mechanism. Junc-ture insertion is straightforward, using the common /V (left-most, by default) insertion site, complicated only by avoiding non-reduplicating (NR) prefixes.

(318) 1. ∅ → ]/

V (V /∈ NR) ; ∅ → [/Right C

2. Prosodic adjustment of the duplicant to CVC using ]-Right.

The default is to close the duplicant. The second juncture insertionrule therefore applies rightmost among those positions which close theduplicant.

Inkelas and Zoll’s story is considerably more complicated. Amorphological doubling operation is followed by three cycles of rules.Cophonology1 derives the reduplicant and Cophonology2 (the “standardconcatenative morphophonology”) derives the stem. The reduplicant isthen infixed into the stem by the third cycle of rules, Cophonology3,which operates on the results of the first two cycles. The computation issketched below:

184 Chapter 7

(319)expeciss

MorphologicalDoubling−→

expeciss expeciss

phonology−→ expecis

sexpecis

s

↓Cophonology1 ↓Cophonology2

sex sisexpec

↓Cophonology3

si-sex-sexpec

This is an considerable expansion in the machinery available to mor-phophonology. Putting that aside and accepting the machinery, wecan ask if it manages to give a satisfactory analysis of the issues athand, which is a minimal requirement. There are two core issues; theextraction of the CVC reduplicant in Cophonology1 and the positioningof reduplicant in Cophonology3.

For both tasks, I&Z introduce the notion of an extended prosodicroot. The idea is that reduplicating inner prefixes “cohere” (theirterminology) prosodically to the root and non-reduplicating prefixes donot. The root and its cohering inner prefixes are supposed to form aprosodic constituent. Presumably, agreement prefixes which provideonsets for syllables anchored in the root or cohering prefixes also cohereand are incorporated into the extended prosodic root. Reduplication isthen simply CVC-prefixal reduplication of the extended prosodic root.In DR terms, one would say that the domain of the reduplicative affixis the root, augmented by the pre-root material which coheres to itprosodically. The fatal flaw in this idea is that non-cohering prefixescan occur inside cohering prefixes, preventing a cohering prefix fromcohering either to the root or a prefix which coheres with the root. One isthen forced to simply stipulate that the extended prosodic root consistsof the minimal string of syllables which contains the root and all thereduplicating prefixes. That stipulation is sufficient for I&Z’s analysis,

Case Studies: Chaha 185

but it is hard to escape the conclusion that it is simply an ad-hoc deviceto abide by the rules of the Prosodic Morphology game, which mustgive a prosodic identity to the locus of infixation.

Even though the stipulation yields the correct result in Chumash,the separation of the extraction of the reduplicant in Cophonology1 andits placement in Cophonology3 makes it an accident that the reduplicantis placed adjacent to material that it happens to be identical with phono-logically. It is equally simple to specify that the initial C∗VC of the stem(boldfaced below) is infixed at the left edge of the prosodic stem (boxedbelow). The infix is underlined in the output.

(320) a. s-ik uk sik s-ikuk

b. s-i s -expec si sis s -expec

c. k-ni- c’eq k-ni- knic’ c’eq

In (320b), sis is infixed inside itself, as is knic in (320c). As far as Iknow, this behavior is unattested. It is impossible in DR. The problemstems from the disconnect between the process which generates thereduplicant and the process which positions the reduplicant in the stem.This disconnect is not only possible in I&Z’s theory, it is part of the corearchitecture of the theory.

The conclusion that can be drawn from this is not novel. When themachinery available for computation is expanded in order to computesome recalcitrant example, the likely result is that not only can therecalcitrant example be computed, but a great many other undesirablethings can be computed as well. The big problem in linguistics is not afew recalcitrant examples, but overgeneration.

7.10 Chaha reduplication

Chaha, one of the Semitic languages of Ethiopia, has been a rich sourceof phenomena which provide insights into autosegmental phonology.

7.10.1 Overapplication of impersonal labialization andpalatalizationChaha has impersonal verb forms alongside verb forms which are in-flected for person, number, and gender. Impersonal inflection is realized

186 Chapter 7

by labialization of the rightmost labial or velar consonant in the rootand palatalization of the rightmost consonant in the root, provided thatit is a coronal obstruent. The following examples (taken from Hudson,1995:789 and Banksira, 2000:207) are illustrative:

(321) 2sg masc impersonala. f1r@x f1r@xw (*fw1r@xw) ‘Tolerate!’b. k1ft k1fwc (*kw1fwc) ‘Open!’c. g1r@z gw1r@Z ‘Age!’d. n1t’1r n1t’1r (*n1c’1r) ‘Separate (from the teats)!’

Rounding is indicated by, for example, f → f w and palatalization isindicated by, for example, s→ s . (321a-c) show that initial, medial,and final consonants can be rounded. (321a) and (321b) show thatrounding applies only to the rightmost consonant which is subject torounding. (321b) shows that both palatalization and labialization canapply. Finally, (321d) shows that palatalization is restricted to the finalconsonant.

In certain cases, labialization or palatalization appears to apply to aconsonant which is not the rightmost potential target.

(322) Personal Impersonala. s@k@k s@kw@kw ‘plant in the ground’b. g@m@m g@mw@mw ‘chip the rim’c. b@t@t bw@c@c ‘be wide’d. s@k@k s@kw@kw ‘place a peg’

Insightfully, McCarthy realized that overapplication in (322) is relatedto the C1C2C2 consonant structure and proposed a theory in whichthe identical consonants were two occurrences of the same phoneme,a long-distance geminate, as such nonadjacent paired phonemes cameto be called. This structure was achieved by supposing adjacent con-sonants were barred from consonantal roots and the particular wayin which a biconsonantal root associates with a CVCVC-template. Ifassociation is right to left, then we expect:

Case Studies: Chaha 187

(323) gm →g

× ×@

m

× ×@

×

McCarthy placed the vocalism on a separate tier in order to avoid anNCC violation.

Combination with the affix IMP which realizes the impersonalmorpheme takes place while multiple association still persists, so thatmutation occurs while the final consonant is still spread over multipletiming slots. The result is apparent “overapplication” of labial mutation.

(324)

g

× ×@

m

× ×@

×⊕ IMP−→

g

× ×@

mw

× ×@

× → g@mw@mw

It is assumed that the labialization rule is not subject to geminateinalterability. Only one of the timing slots linked to m is in the properenvironment for labialization. If the labialization rule were subjectto geminate inalterability, in place of the overapplication illustrated in(324), m would resist labialization and underapplication would result.

There is another class of biconsonantal roots, the so-called 1212roots, which exhibit similar overapplication. The following examplesare from Hudson (1995:794).

(325) Personal Impersonala. b1t@b@t bw1c@bw@c ‘dissolve’b. s1b@s@b s1bw@s@bw ‘gather’c. n1k’@n@k’ n1k′w@n@k′w ‘shake’

Imitating the analysis of 122 roots would lead to a derivation like:

(326) sb →s

××1

b

××@

××@

×⊕ IMP−→

s

××1

bw

××@

××@

× → s1bw@s@bw

188 Chapter 7

In (326), there is no escape from crossing violations by putting theoffending elements on a separate tier. Since the NCC was consideredsacrosanct, McCarthy considered a representation like (326) to be im-possible and proposed that 1212 roots were reduplicative, so that theoverapplication in (325) was an effect of Wilbur’s Identity Constraint,rather than the presence of long-distance geminates (LDGs). It wasclearly untenable to claim that 122 overapplication was due to the pres-ence of an LDG, but that 1212 overapplication was due to reduplication.There was resistance, however, to a uniform account, because 122reduplication does not fit easily into the mold of Prosodic Morphology,which was held to govern reduplication. Eventually, however, opinionshifted to favor a uniform account of 122 and 1212 roots, with overap-plication phenomena attributed to a Wilbur type Identity Condition (ina modern Correspondence Theory form) operating between the sourceand target of reduplication. See Gafos (1999), for example. The long-distance geminates in (324) and (326) were abandoned. I think this wasa mistake. From the standpoint of DR, the debate pitting LDGs againstreduplication is a false debate rooted in the Correspondence Theoryof reduplication which OT has adopted. Representations with LDGslike (326) and (324) are exactly what DR predicts as a consequence ofreduplication.

7.10.2 Inherently reduplicated rootsThere has been ongoing discussion among experts in Semitic mor-phology about whether reduplication is somehow encoded in the root,or the root is simply biconsonantal (consider 122 roots for example)and reduplication is coded in (somehow) or triggered by (somehow) atemplate. Hudson (1995) proposed that reduplication was coded in theroot as, for example:

g [m]α α or [s b]α α

The notation for coding reduplication into the root proposed by Hudsonis remarkably similar to the idea of inserted t-junctures that DR proposesas the trigger for reduplication. Exploiting default closure, we can writethe roots in the DR formalism as:

g[m or [sb

Case Studies: Chaha 189

If we suppose that roots consist of consonants, with an optionalembedded [-juncture, we have a uniform account of the full rangeof reduplicative roots which are found in various Semitic languages.McCarthy’s insight that identical adjacent consonants are barred fromthe root is maintained.

(327) a. C1[C2 → C1C2C2 b. [C1C2 → C1C2C1C2

c. C1C2[C3 → C1C2C3C3 d. C1[C2C3 → C1C2C3C2C3

I suppose that the vocalism is epenthetic, inserted in the process ofsyllabification, subject to morphophonological conditions on the choiceof vowel. By morphophonological, I mean that both the morphologi-cal and phonological environments can influence the choice of vowel.These assumptions lead to derivations roughly like the following:

(328) [sb

(lexicon)→ [

s

×b

×DC→ [

s

×b

×]Trscr−→

s

×b

× × ×⊕ IMP−→

s

×bw

× × ×

NCCR−→

s

×bw

×s

×bw

×Syllabification−→

σ

s

×1

×σ

bw

×@

×σ

s

×@

×bw

×

There is some uncertainty in the order of the various operations.NCCR (NCC Repair) must follow combination with IMP, but syllabifi-cation could be much earlier. It is difficult to find evidence in Chaha,but there is evidence from other Semitic languages that syllabification islate. So-called Classical Arabic Metathesis has a relatively straightfor-ward explanation under the assumption of late syllabification. It wouldbe a distraction from the main argument to go through an analysis here,but one is provided in Appendix 3.

7.10.3 Chaha Continuant DissimilationThere is a dissimilation process in Chaha whose interaction with redu-plicative roots gives further evidence of the presence of long-distancegeminates. Banksira (2000) gives extensive evidence that k does notoccur in underlying representations and that surface k is the result of

190 Chapter 7

either devoicing of underlying g or the dissimilation of x to k becauseof the rule:

(329) x → k/

. . . [−sonorant,+continuant]

The [−sonorant,+continuant] phonemes are f, s, z, x, and Q. For thepresent discussion only, call such a phoneme a “triggering phoneme.”There is no adjacency requirement in (329), so dissimilation can belong-distance. Later, we will see that dissimilation is blocked by atleast some morpheme boundaries, so that (329) must be amended. Itcertainly applies morpheme internally.

Rule (329) is not subject to geminate interpretability. x → k

requires only that one timing slot associated with x be followed by atiming slot associated with a triggering phoneme. It is not required thatall timing slots linked to x be followed by a timing slot linked to atriggering phoneme. Consequently, geminate x will always dissimilateto geminate k. The first timing slot of

x

× ×

is followed by a timing slot linked to x, a [−sonorant,+continuant]phoneme. Since adjacency is not required in (329), long-distancegeminate x will also dissimilate to k.

Kenstowicz and Banksira (1999) analyze the interaction of x → k

dissimilation with reduplication in Chaha and show that it overappliesin very much the same way that impersonal labialization/palatalizationoverapplies. A few representative illustrations are given below. SeeKenstowicz and Banksira for further examples, most of which are lesstransparent than the ones given here because of various interactingmorphophonological processes.

(330) a. /sx/ s1k1k ‘drive a peg’b. /xt/ k@tk1t ‘crush’

Naively, one might expect /sx/ → s1k1x , with the medial x dis-similating to k because there is a [−sonorant,+continuant] phoneme toits right, the final x. But the fact is that the final x also changes to k

Case Studies: Chaha 191

in (330a), even though there is no [−sonorant,+continuant] phoneme toits right.

The analysis of 122 and 1212 roots given above makes it clear whythere is overapplication.

(331) a.

s

× [

x

× ]Trscr−→

s

×x

× ×x-Dissim−→

s

×k

× × → s1k1k

b. [

x

×t

× ]Trscr−→

x

×t

× × ×x-Dissim−→

k

×t

× × × → k@tk1t

7.10.4 Chaha frequentative reduplicationWe begin with frequentative reduplication in Tigre, another EthiopianSemitic language which is closely related to Chaha. The morphophonol-ogy of Tigre is more transparent than the morphophonology of Chaha,making it easier to see the basic pattern. Frequentative reduplication isfound in all the Ethiopian Semitic languages. The facts are from Rose(2001).

(332) Root type Regular FrequentativeA d@gm-a: d@ga:g@m-a: ‘tell’B w@ll@b-a: w@la:l@b-a: ‘look both ways’C ba:r@k-a: b@ra:r@k-a: ‘bless’122 m@zz-a: m@za:z@z-a: ‘give responsibility’Quadriliteral d@ng@s’-a: d@n@ga:g@s’-a: ‘be scared’1212 n@kn@k-a: n@k@na:n@k-a: ‘shake in hysterics’1233 d@nz@z-a: d@n@za:z@z-a: ‘be numb’

Rose notes that the vocalism of the frequentative form is invariant.The same observation is made about Chaha by Banksira (2000:37),who states that “vowels of the unreduplicated base . . . are not retainedwhen reduplicated.” This supports the proposal that syllabification andvowel insertion are late, after the affix which generates frequentativereduplication combines with the stem.

192 Chapter 7

The morphology of the stem modification in (332) is clear. Thepenultimate consonant is geminated and a: is infixed between the legsof the geminate. Formally, there are two ways in which this can beimplemented. One possibility, is that gemination applies first, then-〉〈a:〉] is suffixed, with the final consonant pair designated as the domainof the affix.

(333)

d

×g

×m

×gemination−→

d

×domain︷︸︸︷

g

× ×m

×affixation−→

d

×domain︷︸︸︷

g

× ×m

×〉〈a

× ×〉]

DC→

d

×domain︷︸︸︷

[〈g

× ×m

×〉〈a

× ×〉]Trscr−→

d

× ×

g

×

a

× ×

m

× (dga:gm)

Alternately, there is no separate gemination operation. Instead,-〉〈a:〉] is suffixed, with the final consonant pair designated as the domainof the affix, and juncture insertion ∅ → ] /× applies. Nestedduplicants result and transcription is carried out in two cycles, as inSection 4.5.

(334)

domain︷︸︸︷

d

×g

×m

× →domain︷︸︸︷

d

×g

× ]

m

×〉〈a

× ×〉]DC→

d

×g

× ]

m

×〉〈a

× ×〉][〈[

innertranscription−→

d

× ×[〈g

×m

×〉〈a

× ×〉]outer

transcription−→

d

× ×

g

×

a

×

g

×

m

×

The derivation (333) or (334) continues to:

(335)syllabification−→

σ

d

×@

×σ

×a

× ×σ

g

×@

×m

× (d@ga:g@m)

Case Studies: Chaha 193

Finally, NCC repair takes place. Alternatively, NCC repair takes placebefore syllabification.

Recall that the devices for specifying the domain are limited. Thepenultimate consonant must be located by designating a domain, whichrequires constructing a foot at the right edge of the root. I assume thatconsonants are moraic in the root cycle, so that the domain is simply afinal binary foot.

(336)

∗

d

×∗

g

×∗

m

×footing−→

∗

d

×( ∗

g

×∗

m

×

7.10.5 The interaction of frequentative reduplication andx-dissimilationThere are some subtleties in the interaction of frequentative reduplica-tion and x-dissimilation in Chaha which are discussed by Kenstowiczand Banksira (1999). Some preliminaries are necessary. Banksira(2000) carefully establishes a number of relevant facts about Chahamorphophonology: 1) the morpheme realizing perfect aspect readjuststhe root by geminating its penultimate consonant; 2) geminates devoiceif they are followed in the root by a sonorant; and 3) geminates inChaha systematically degeminate (by deletion of one of the twins). Thissynchronic phonology of Chaha is made clear by comparing Chaha withsome of its sister Ethiopian Semitic languages.

(337) Gemination and devoicing in the perfect (root /Bdr/, ‘be first’)

Imperative Imperfect PerfectEzha y@-Bd@r y1-B@d1r b@dd@r-@mEndegen y@-Bd@r y1-B@d1r b@tt@r-@mChaha y@-Bd@r y1-B@d1r b@t@r-@m

Ezha has transparent penultimate consonant gemination in the perfect.Endegen adds the geminate devoicing rule. Chaha renders the processopaque by adding a late rule of degemination.

Penultimate gemination, which is triggered in all roots by theperfective morpheme, is also triggered by the imperfective morpheme,

194 Chapter 7

but only for quadriliteral roots. Banksira (2000:32) gives the template-C1C2@C3C3(1)C4 for the imperfect form of quadriliterals. Triliteralroots are not subject to imperfect penultimate gemination. With re-spect to their perfective and imperfective morphology, forms producedby reduplication are treated in the morphophonology as if they wereroots. Roots which become quadriliteral as the result of either inherentreduplication (called 1212 roots earlier) or frequentative reduplicationare therefore subject to imperfect penultimate gemination. To illustrate,frequentative forms of the triliteral root /sBr/ are given in (338). De-voicing, B → p , is the surface sign of underlying gemination in Chaha.The final sonorant r licenses geminate devoicing. In Tigre, the infixedexponent of the frequentative morpheme is a:. In Chaha it is a, whichgenerally shows up on the surface as @.

(338) root imperative imperfect perfect

/sBr/ s1B@B1r t1-sB@p1r s1B@p@r-x@-m

These forms are derived as follows:

(339) Imperative Imperfect Perfect

s

× ×a

×B

×r

×s

× ×a

×B

× ×r

×s

× ×a

×B

× ×r

×

fission

s

×B

×a

×B

×r

×s

×B

×a

×B

× ×r

×s

×B

×a

×B

× ×r

×

devoicing

s

×B

×a

×B

×r

×s

×B

×a

×p

× ×r

×s

×B

×a

×p

× ×r

×

Syllabification follows. The penultimate consonant devoices (B → p )in the context of the final sonorant. Since the initial B does not devoice,it must be that devoicing follows NCC repair, as assumed in (339), sothat B has already undergone fission into separate phonemes by the timethat devoicing applies.

We now are in a position to consider the interaction of frequentativereduplication and x-dissimilation. Kenstowicz and Banksira (1999) givethe examples in (340). For each root, the first line gives the regularforms and the second line gives the frequentative forms. (I assume that

Case Studies: Chaha 195

the vowel quality and syllable structure variation in the forms below issecondary to the issue at hand.) The forms in which x-dissimilationapplies are marked with †.(340) Imperative Imperfect Perfect

/sxr/s1x@r y1-sx@r s@k@r † ‘get drunk’

t@-sxax@r y1-t-s1kak@r † t@-skak@r † ‘act naughtily’

/mxr/m1x1r y1-m@x1r m@k@r † ‘advise’

t@-mxax@r y1-t1-mkak@r † t@-mkak@r † ‘adviseeach other’

/rxB/n1x@B y1-r@x1B n@k@B † ‘find’

t@-rx@x@B y1-t1-rk@k@B † t@-rk@k@B † ‘show up’

The consonantal structure of the roots involved above is illustratedbelow for the root /mxr/.

(341) Imperative Imperfect Perfect

regularm

×x

×r

×m

×x

×r

×m

× ×x

×r

×

m1x1r m@x1r m@k@r †

freq.m

× ×a

×x

×r

×m

× ×a

× ×x

×r

×m

× ×a

× ×x

×r

×

t@-mxax@r y1-t1-mkak@r † t@-mkak@r †If x-dissimilation were restricted by adjacency, the explanation

for the dissimilation pattern in (340) would be trivial. The forms inwhich x-dissimilation applies are precisely the forms with a geminatewhose two occurrences are adjacent. The dissimilation of the initialoccurrence of x in the frequentative perfect and imperfect forms is anoverapplication effect, a side-effect of the dissimilation of the medialoccurrence of x.

The issue is more complex, however, because x-dissimilation doesnot require adjacency. B&K give /xBs/ → kBaBs ‘make dirty’ and/xtf/ → kt1f ‘hash’, for example. In both cases, the dissimilation

196 Chapter 7

trigger is the final nonsonorant continuant consonant (s in one case andf in the other). Consequently, if the failure of x-dissimilation is to beattributed to the intervention of a, the argument requires some subtlety.One obvious difference between potential long distance dissimilation inthe frequentative imperative form in (341) and the examples given byB&K is that a is a copy of the exponent of the frequentative morpheme.

Banksira (2000:33) provides evidence that x-dissimilation doesnot apply across a root boundary. He gives two relevant examples.In (342a), x-dissimilation applies between occurrences of x in theexponents of two different morphemes. But neither is the root. Theyare closely related agreement morphemes, one a subject agreementmorpheme and the other an object agreement morpheme. In (342b),on the other hand, an occurrence of x in the exponent of an agreementmorpheme does not trigger dissimilation of an occurrence of x in theroot, even though the two occurrences are adjacent.

(342) a. s@p@rk@m ‘I have broken you (2sg. m.)’s@p@r x x@ m

break 1sg subject 2sg. m. object perfect

b. naxx1nam (*naknam) ‘I sent her’nax x1 n a m

send 1sg subject object case 3sg. f. object perfect

We can account for (342) by constraining the dissimilation rule

x→ k/

(×)∗ ×[−sonorant+continuant

]

so that it applies to root phonemes only if it applies root internally.That is, if the target is root internal, the trigger and any intermediatetiming slots must also be root internal. Of course, it must also part ofthe description of the rule that it not be subject to geminate inalterability.

This brings us closer to an explanation for why there is no x-dissimilation in the imperative frequentative form, which is repeatedbelow:

Case Studies: Chaha 197

(343)

m

× ×a

×x

×r

×

Plausibly, the target of dissimilation is a root phoneme, but a is not.In order for this to be true, two assumptions are sufficient.

(344) 1. The morphemic association of the exponent of a reduplicativeaffix is maintained under transcription.2. Other root internal timing slots which result from morpho-phonological operations associate with the root morpheme. Thisincludes timing slots produced by root internal epenthesis andgemination.

This leads to the derivation:

(345)

[[MXR]] FREQ

m

×x

×r

× 〉〈a

× 〉]gemination−→

[[MXR]] FREQ

m

× ×x

×r

× 〉〈a

× 〉]

DC→

[[MXR]] FREQ

m

× × [〈x

×r

× 〉〈a

× 〉]Trscr−→

[[MXR]] FREQ

m

× × ×ax

×r

×

This completes the discussion of the x-dissimilation pattern in(341). The fact that the initial x does not dissimilate in the frequentativeimperative is due to the intervention of the exponent of the frequentativemorpheme, as shown in (345). Dissimilation of the initial x in thefrequentative imperfect and perfect is due to the overapplication ofdissimilation of the medial occurrence of x.

7.10.6 Kenstowicz and Banksira’s analysis of x-dissimilationB&K is valuable in presenting the basic facts of x-dissimilation andits relation to reduplication since clear examples of overapplicationare fairly rare. Their analysis, however, is untenable. Since Chahaphonology is particularly opaque and B&K attempt an analysis in theframework of Correspondence Theory, which is known to break down

198 Chapter 7

in analyzing opaque processes, it is not surprising that CorrespondenceTheory proves inadequate to the task.

Two illustrations will suffice to make clear the problem.

1. B&K claim that x-dissimilation is due to a phonotactic constraint:

(346) *x . . . [+continuant,−sonorant]

This cannot be correct. The 3sg Jussive form of triliteral root fxQ ‘es-cape’ is y@-fka (B&K, ex. 5a). In the surface representation, there is notrigger for x-dissimilation. x-dissimilation cannot be the consequenceof a surface phonotactic.

Derivationally, the surface form is no mystery. The Q-trigger fordissimilation deletes after it triggers dissimilation.14

(347) fxQx-dissimilation−→ fkQ

syllabification−→ fk@Q

Q-assimilation−→ fka

2. One of the objectives of the analysis is to account for the interactionof frequentative reduplication and penultimate devoicing in the perfec-tive and imperfective. This is impossible unless there is a coherentaccount of penultimate devoicing (imperative s1B@r versus perfect s@p@r

‘break’). B&K’s account is not. Banksira (2000) has a carefully workedout derivational analysis of the origin of penultimate p which relieson morphologically conditioned gemination (template satisfaction), fol-lowed by geminate devoicing, followed by degemination. The geminatewhich triggers devoicing is opaque, present only at an intermediatestage in the derivation. This analysis is not available to an OT account.B&K appear to forget this, saying that “the template for the perfectiveof the derived frequentative verbs requires the penultimate radical to bea geminate “which surfaces [my emphasis] as a stop if nonstrident.”

In addition to providing an additional example of overapplicationof the Malay variety, B&K intend their analysis to support McCarthyand Prince’s contention that generative phonology is descriptively inad-equate. They say:

As observed originally by Wilbur (1973), the traditional deriva-tional model in which reduplication is expressed by a copy ruleapplying at some fixed point in the derivation is unable to describethis phenomenon adequately.

Case Studies: Chaha 199

We have shown this claim to be false by providing a detailed deriva-tional analysis of the interaction of x-dissimilation and reduplication. Acopy rule (transcription) applies at some fixed point in the derivation.Crucially, timing slots and their associations with phonemes are copied,but not phonemes. In addition to showing this claim to be false, wehave shown that the Optimality Theory model which B&K’s analysisis intended to confirm is not even capable of adequately describing thecore facts of x-dissimilation and penultimate devoicing. Instead of sup-porting the idea that derivational phonology is descriptively inadequate,B&K succeed only in giving further evidence that opaque processesin phonology are pervasive and that Optimality Theory is unable toproperly account for them.

Appendix A

The NCC and theRetraction Condition

Sagey (1988) argues that it follows from the logic of temporal simul-taneity that the NCC is an interface condition. This is not convincing.Phonological representations, even at the interface with phonetics, arenot representations of physical events. They are instructions for creatingphysical events. More accurately, they are once removed from suchinstructions. Phonological representations are blueprints for creatingmotor instructions. The explanation for the NCC as an interface con-dition must be found in the structure of representations employed bythe motor system, not directly in the structure of the physical world.Output interface conditions reflect the demands of the system whichis being fed, in this case the motor system leading to articulation. Inwhat way might the motor system cause the imposition of the NCCas an output condition on phonology? Under the plausible assumptionthat the lexicon is optimized for space efficiency, it is implausible tosuppose that exponents are stored as autosegmental representation. Itherefore suppose that exponents are stored as simple phoneme/juncturesequences and that timing slots are supplied in the process of tran-scription from the lexicon to the phonological workspace. I supposethat geminates are recorded by some diacritic which is removed inthe process of transcription: perhaps a [+long] feature in the phonemeitself which is translated into autosegmental terms and removed in theprocess. I suppose that autosegmental representation is the privilege

The NCC and the Retraction Condition 201

of phonology, used in its internal computation, but that autosegmentalrepresentation is unintelligible to both the input and output systems.

Suppose further that the interface mapping from phonology to themotor system, which translates the output of phonology into the lan-guage of the motor system, has limited resources. In particular, supposethat copying is not available to the interface mapping. Autosegmentalrepresentations contain tiers of three types: the timing tier, phonemictiers, and prosodic tiers. The point is controversial, but I assume thatphonemic tiers and the prosodic tiers have rather different properties,with the timing tier serving as a kind of interface, and the only interface,between phonemic tiers and prosodic tiers.

Phonemic tiers in autosegmental representations are, in many cases,easily converted into sequential representations by a simple retractionoperation, with the help of a few special symbols (diacritics).

(348)

t

×i

×k

× ×a

× × → t i k : a :

This is just the reverse of the process which mapped lexical items toautosegmental representations.

Multiple segmental tiers translate into multiple parallel (aligned)retractions onto the timing tier. See Eisner (1997), who works out sucha representational theory for autosegmental phonology as a way to pro-vide a representational foundation for Optimality Theory. The pictureof the output of phonology that this gives is multiple streams of motorinstructions, aligned in time (representational temporal simultaneity).The mapping (349), in which a separate nasal tier is assumed, suggestswhat is involved.

(349)

t

×i

×N

k

× ×a

× × → N

t i k : a :(tiNka:)

Retraction onto the timing tier is not sensitive to linear order on thephoneme tier. Consider the abani→albani Choctaw passive example

202 Appendix A

(102a). The structure that is produced by transcription retracts onto thetiming tier without accessing phoneme tier order.

(350)

a

×

l

× ×

b

×

a

×

n

×

i

→ a l b a n i

Phoneme tiers are unordered. Some implications of this are that therecan be no floating elements on phonemic tiers and no junctures onphonemic tiers. Both of these predictions are correct, as far as I can see.Note that if phoneme tiers are unordered, many standard formulationsof the NCC are inadequate since they rely of comparison of phonemeorder with the order of associated timing slots.

In general, reduplication produces autosegmental representationswhich cannot be retracted. The retraction of the representation (78)produced in Yoruba nominalization, for example, fails:

(351)

d

×

i

× × ×

u

×

n

→ d i u n

Two copies of d are needed for retraction to be successful, but theinterface mapping cannot carry out copying. The requisite copying in(351) must be done in phonology, prior to the output interface. Thiscopying is what we have been calling fission.

(352)

d

×

i

× × ×

u

×

n

phonology

Fission−→ ×

d i

× ×

d

×

u

×

n

interface mapping

Retraction−→ d i d u n

I will suppose that the output of phonology must be retractable,with the actual retraction part of the interface between phonology and

The NCC and the Retraction Condition 203

phonetics. This imposes the output condition (353) on autosegmentalrepresentations.

(353) Retractability Condition (RC): The set of timing slots associ-ated with a segment is connected.

A few examples will clarify the import of the RC. First, note that (354a)and (354b) are equivalent with respect to the RC, since the sets oftiming slots associated with the two phonemes are identical in the twoexamples. Both satisfy the RC.

(354) a. ×a

× ×b

b.

a

× ×b

×

In (355a), on the other hand, the set of timing slots associated witha is not a connected sequence of timing slots, so (355a) violates the RCin exactly the same way that (355b) does.

(355) a. *

a

×b

× × b. *

a

× × ×

The RC and the NCC impose different conditions. There are severaladvantages to the RC. Most importantly, it can be derived from the as-sumption that autosegmental representations are the privilege of phonol-ogy and plausible assumptions about the representations employed bythe motor system and the mapping from phonological representationsto motor system representations. With respect to reduplication it alsohas the advantage that then melodic tiers can be considered to beunordered, with the order of segments in motor system representationsderived entirely from order on the timing tier. This has the importantconsequence that no RC repair is needed in representations like (102)and the output of infixation-type reduplication in general. Minimal RCrepair (fission) is needed in examples like (352) and similar products oftranscription. Note that if phonemic tiers are considered to be orderedand NCC repair were needed in examples like (352), not only fissionbut multiple metathesis operations would be required as well.

204 Appendix A

Finally, note that the RC makes metathesis comprehensible asan RC violating preposing operation followed by timing slot deletionwhich removes the RC violation.

(356)

a

×b

×prepose−→ ×

a

×b

×delete−→ ×

ba

×

Appendix B

Defect driven rule schemata

The purpose of this appendix is to specify when and how a schema ofthe form (357) applies to a representation α.

(357) desiderata-list :: rule-list ; higher-desiderata

We first need to the notion of a desideratum driven rule which isconstrained by a list of higher desiderata (CDDR, Constrained Desider-atum Driven Rule):

desideratum :: rule ; higher-desiderata

If α and β are representations and K is a condition on represen-tations, we say α K β if K(α) is true and K(β) is false; and we sayα =K β if neither α K β nor β K α is true. If (K1, . . . ,Kn) is a list ofrepresentations, we say that α (K1,...,Kn) β if α K1 β, or if α =K1 β andα (K2,...,Kn) β; and we say α =(K1,...,Kn) β if neither α (K1,...,Kn) β norβ (K1,...,Kn) α.

Now, suppose d :: r ; (K1, . . . ,Kk) is a CDDR. We say that it applies

to a representation α to produce the representation β if:

1. r applies to α to produce β;

2. α d β; and

3. α �(K1,...,Kn) β.

206 Appendix B

Informally, a rule applies if it can bring an undesirable representationinto satisfaction of the desideratum without degrading its satisfaction ofthe more highly ranked desiderata.

Application of (1) is specified by expanding the schema into a listof CDDRs by recursively applying the expansions (358).

(358) 1. (d1, d2, . . . , dn) :: R ; (e1, . . . , ek)

→(d1&d2, . . . , d1&dn) :: R ; (e1, . . . , ek)

d1 :: R ; (e1, . . . , ek)(d2, . . . , dn) :: R ; (d1, e1, . . . , ek)

2. d :: (r1, . . . , rn) ; E →

[d :: r1 ; E

d :: (r2, . . . , rn) ; E

]

It is implicit in the formulation that expansion is done on desideratabefore expansion on rules.

For example:

(d1, d2) :: (r1, r2) ; ∅ →

d1&d2 :: r1 ; ∅d1&d2 :: r2 ; ∅

d1 :: r1 ; ∅d1 :: r2 ; ∅

d2 :: r1 ; (d1)d2 :: r2 ; (d1)

B.1 Late syllabification in Arabic

It is difficult to find evidence in Chaha, but there is evidence fromother Semitic languages that syllabification is late. So-called ClassicalArabic Metathesis is easily accounted for under the assumption oflate syllabification. In Classical Arabic, there is an affix dependentalternation in syllabification patterns for 122 roots in Arabic. Contrastthe syllabification of an ordinary triconsonantal root in (359a,b) and a122 reduplicative root in (359c,d).

Defect driven rule schemata 207

(359) /ktb/ a. ya-ktud-na ‘they (f) write’b. ya-ktud-a ‘he writes’

/m[d/ c. ya-mdud-na ‘they (f) extend’d. ya-mudd-a ‘he extends’

The examples are from Gafos (2001), with the roots expressed accord-ing to the analysis developed in Section 7.10.

The two forms of the root /m[d/ which are embedded in (359c)and (359d), mdud and mudd, are the puzzle that must be accounted for.Theories which assume that the affixes are attaching to a form which hasalready undergone syllabification are more or less forced to take one orthe other to be the preaffixation form and have needed some variety ofmetathesis to derive one of the outputs. The question has an extensivehistory and Gafos should be consulted for the extensive background.

Under the assumption that syllabification does not take place untilafter the affixes and root have been concatenated, the syllabification pat-tern (359) can be understood as a syllabification/geminate interaction,not as affix induced metathesis of an already syllabified root.1 Althoughthe intuition is clear, how it should be implemented is not entirely so.The reason that epenthesis follows the initial root consonant in (359d),but the penultimate root consonant in (359b), cannot be the consequenceof an outright prohibition on epenthesis into a geminate, because thereis just such epenthesis in (359c).

Two factors are at work, avoidance of epenthesis into a geminateand avoidance of epenthesis at the root-suffix boundary. The formalismof DDRs allows for just such competition between avoidance con-straints. Frampton (2001) uses this formalism to develop a theory ofsyllabification and footing in which the structure starts with multipleunsyllabified elements or unfooted stressable elements. The order inwhich the defects are targeted for removal is determined by an orderedlist of preferences. For syllabification, sonority and directionality arethe key preferences. For a language like Arabic, vowels are syllabifiedbefore consonants, and elements to the right syllabified before elementsto left, with the vowel/consonant distinction more important. The otheringredient is a list of repair rules and an ordered list of constraints,avoidance of which is to be avoided, to the extent possible. The highest

208 Appendix B

ranked repair rule which incurs the minimal violation of the constraintsis used. For Arabic, the repair rule schema would be:

(360)

1C V

σ

2C

σ

V C

3σ

C V C

4C Vep

σ

5Vep C

σ

;[*Root/Affix Epenthesis

*Split Geminate

]

The derivations of the forms in (359) are given below. A hyphenindicates an affix/root boundaries and, at each point in the derivation,the most prominent defect, determined by sonority (reduced to thevowel/consonant distinction in Arabic) and “most right,” is boxed.

(361) (a) (b) (c) (d)

y a-k t b-n a y a-m d d-n a y a-k t b-a y a-m d d-a

1. y a-k t b-n aσ

y a-m d d-n aσ

y a-k t b-aσ

y a-m d d-aσ

2. y aσ

-k t b-n aσ †

y aσ

-m d d-n aσ †

y aσ

-k t b-aσ

y aσ

-m d d-aσ †

3. y aσ

-k t u bσ

-n aσ

y aσ

-m d u dσ

-n aσ

y aσ

-k t uσ

b-aσ

y aσ

-m u dσ

d-aσ

4. y aσ

-k t u bσ

-n aσ

y aσ

-m d u dσ

-n aσ

y a-kσ

t uσ

b-aσ

y aσ

-m u dσ

d-aσ

5. y a-kσ

t u bσ

-n aσ

y a-mσ

d u dσ

-n aσ

Defect driven rule schemata 209

Except for the representations marked with †, the choice of repairrule is determined by repair rule ordering, since in all other casesthe highest ranked rule whose structural conditions are met does notproduce a violation of any of the avoidance constraints. For the boxedrepresentations, the highest ranked rule which applies is Rule 4, but inthe † instances, it produces a violation of one or the other avoidanceconstraint. Rule 4 is passed over in the first two columns in favorof Rule 5 because application of Rule 5 does violate either of theconstraints. Rule 4 is passed over in the last column in favor of Rule 5because, although its application violates the constraint against splittinga geminate, there is no better alternative. Application of Rule 4 violatesa more highly ranked constraint.

For the sake of comparison, perfect forms are given below. Thereis no prefix and the epenthetic vowel is a. The boxed representationsare the ones to which the highest ranked rule whose conditions ofapplication are satisfied does not apply, because of one or the otheravoidance constraint. These are the forms which force epenthesis tothe left of the unsyllabified consonant instead of the more highly rankedepenthesis to the right. Avoiding constraint violations takes precedenceover rule ordering.

(362) (a) (b) (c) (d)

k t b-t u m d d-t u k t b-a m d d-a

1. k t b-t uσ

m d d-t uσ

k t b-aσ

m d d-aσ

2. k t a bσ

-t uσ †

m d a dσ

-t uσ †

k t aσ

b-aσ

m a dσ

d-aσ †

3. k t a bσ

-t uσ

m d a dσ

-t uσ

k aσ

t aσ

b-aσ

m a dσ

d-aσ

4. k aσ

t a bσ

-t uσ

m aσ

d a dσ

-t uσ

Appendix C

Raimy’s theory ofR-representations

The theory of juncture insertion and transcription which DR is based onwas developed as a means of overcoming certain conceptual and empiri-cal problems with the theory proposed by Raimy (2000). This section isdevoted to outlining Raimy’s theory, explaining what its problems are,and contrasting it with DR. Raimy proposes that over/underapplicationeffects (many of them, at least) in reduplication are a consequence ofphonological rules applying to a representation with a nonlinear timingtier. Since the remainder of this appendix is devoted to explainingwhy I think Raimy’s proposal should be rejected, it is important tofirst acknowledge my debt to Raimy for reconnecting the theory ofreduplication with the core questions of phonology: What phonologicalstructures does UG make available? What is their architecture andvocabulary? What operations on these structures are possible and howis their relative complexity evaluated?

C.1 R-representations

Raimy proposes that phonological representations are structures inwhich phonological adjacency is represented explicitly by linking ratherthan implicitly by adjacency in a linearly ordered structure. The pro-posed representation of kat ‘cat’, for example, is (363a), rather than(363b).

Raimy’s theory of R-representations 211

(363) a.

#

k

at

%

b.

k

×a

×t

×

I will call representations like (363a) R-representations. It will be clearshortly why explicit start (#) and stop (%) symbols are necessary.

In the interests of typographical convenience, Raimy writes R-representations on a line, to the extent possible. So (363a) would bewritten as:

(364) # k æ t %

Raimy’s intention, however, is to divorce the structure from any implicitlinearity. The reader should be careful in what follows not to investrepresentations like (364) with more structure than is justified by theexplicit links which are given.

Since the links between phonemes are represented explicitly bypointers, there is no formal obstacle to structures with loops. Raimyproposes that Malay example from Section 1, for example, has therepresentation (365) at an early stage in the derivation.

(365) # a n e m %

Although Raimy pictures R-representations without timing slots,as above, this is also a typographical convenience and (365) should betaken to represent:

(366) #

a

×n

×e

×m

× %

212 Appendix C

The loop structure in (366) makes it clear why R-representationsmust have explicit start (#) and (%) symbols. Without them, the struc-ture would be (367), with no indication of either where to enter or whereto exit the loop.

(367) a

n

e

m

Raimy proposes that phonological rules can apply in R-structures,with predecessor and successor contexts for rule application establishedby the pointer relations on the timing tier. Under this assumption, if thestructural conditions for nasalization require only that the target havesome nasal predecessor, then nasalization applies in (366) and produces:

(368) # a n e m %

a nasalizes because one of its predecessors is nasal.Finally, a linearization operation maps the R-representation (368)

to the R-representation:

(369) # a n e m a n e m %

There are two serious problems with this approach to reduplication.First, a linearization operation which is adequate for the full range ofR-representations which must be considered turns out to quite complex.In fact, Raimy does not even attempt to provide a linearization algorithmwhich proceeds by modifying the input representation. Instead, Raimyresorts to a Correspondence Theory account of linearization to select alinear representation which best corresponds to the input representation.Even with this peculiar intrusion of nontransformational phonology intoan otherwise transformational theory, the linearization algorithm whichRaimy provides is empirically inadequate.1 Second, although segmentalphonology extends easily to R-representations, prosodic phonologydoes not. Raimy does not even attempt it.2 This cuts an R-representationanalysis of reduplication off from some prosodic considerations whichseem to be essential in the theory of reduplication.

Raimy’s theory of R-representations 213

C.2 Linearization

Raimy presents the linearization algorithm informally, so we need tomake it somewhat more precise before it can be discussed. This runsthe usual risk of misconstruing the author’s intention, but it cannot beavoided. Raimy views the linearization as an optimization process. Theranked list of desiderata given in (371) was culled from various places inRaimy (2000). Two different kinds of pointers are recognized: pointersthat are already present in the lexicon and pointers that are introducedeither in attaching the exponent of an affix to a stem or by readjustmentrules acting on the stem. For convenience, call the former l-pointers(lexical pointers) and the latter m-pointers (morphological pointers). Inpictorial representations, heavy arrowed lines are used to represent m-pointers. The representation

(370) # t a p %

for example, underlies tap→ tatap .

(371) Optimal linearization

1. Spell out as many phonemes as possible.2. Use as many of the m-pointers as possible.3. Use m-pointers as early in linearization as possible.4. Use as many of the l-pointers as possible.5. Use as few steps as possible.

Although Raimy says “use as many of the m-pointers as possible”,this does not appear to be intended to mean that the new representationshould contain as many of the m-pointers of the old representation aspossible. If the linearization (370) contained the m-pointer, that wouldmean that t and a are metathesized in the process of linearization.Rather, what Raimy appears to intend is that identifying which elementsof the input representation persist and appear in the output representa-tion is a mute point; the output representation is constructed ab ovo.Conditions like “use as many m-pointers as possible” is interpreted tomean that as many m-pointers as possible should have a correspondentin the output.

214 Appendix C

Consider, for example, the input below:

(372) #1

t2

a3

p4

%

5

The pointers are indexed so that correspondence to the input representa-tion can be represented by coindexing. The list of candidates in (373) isthen considered. It is not hard to write down the conditions that a linearrepresentation in correspondence with (372) must satisfy in order to beincluded in (373).

(373) 1. #↑ 1

t↑ 2

a↑ 3

p↑ 4

%

2. #↑ 1

t↑ 2

a↑ 5

t↑ 2

a↑ 3

p↑ 4

%

3. #↑ 1

t↑ 2

a↑ 5

t↑ 2

a↑ 5

t↑ 2

a↑ 3

p↑ 4

%...

Finally, the optimal candidate is chosen on the basis of (371), with “use”suitably reinterpreted.

The conceptual problem that is posed by a linearization processof this sort is that it introduces a completely new kind of rule intoderivational phonology. The problem is not that ranked desiderata areused to choose the output from some candidate set. The problem isthat the candidate outputs are not modifications of the input, so thatcorrespondence relations must be introduced into the phonology.

Since the list of disambiguating criteria used to pick out the optimallinearization grows as one progresses through Raimy’s book, it is notsurprising that most of the nonlinear representations he proposes aresuccessfully linearized. But gaps remain. (374) is (31c) from Raimy(p. 33). All the pointers are lexical.

(374) #1

k2

13

t4

%

@5 6

7

Raimy’s theory of R-representations 215

Linearization is ambiguous. Either (375a) or (375b) is possible.

(375) a. #↑ 1

k↑ 2

1↑ 3

t↑ 7

k↑ 5

@↑ 6

t↑ 4

%

b. #↑ 1

k↑ 5

@↑ 6

t↑ 7

k↑ 3

1↑ 3

t↑ 4

%

In all likelihood, the optimality criteria can be patched up by theadditions of a suitable “*Do-the-wrong-thing” constraint of some kind,or justification can be found for stipulating that a suitable pointer in(374) is lexical. (In some other examples, Raimy needs to stipulate thatpointers added for template satisfaction are l-pointers, not m-pointers,so it certainly is not obvious how the required stipulation can be made inthis case.) The intent here is not to provide a refutation of the proposalon the basis of empirical inadequacy but to illustrate its complexity.This is important because Raimy claims a conceptual advantage forhis theory of R-representations based on the claim that the theoryrequires no reduplication specific devices. A reasonable definition ofa reduplicative structure in Raimy’s theory is one which has a nontriviallinearization. Since the linearization algorithm has no other use thanlinearizing such representations, it is hard to avoid the conclusion thatthe linearization algorithm is reduplication specific. If the linearizationalgorithm were more or less straightforward and conceptually obvious,it would be fair to claim that nothing new is introduced into the theory.But, as we have seen, linearization in Raimy’s theory is far from beingeither obvious or straightforward.

C.3 Prosodic structure and nonlinear R-representations

Recall that the analysis of Asheninca Campa in Section 7.3 relied on theill-formed syllable structure of forms like:

(376) [

∗σ

a

×σ

m

×i

×n

× ]

216 Appendix C

The unsyllabified final consonant triggered final vowel epenthesis andthe ill-formed onsetless initial syllable triggered [-Right, resulting in aduplicant with well-formed syllable structure.

[

∗σ

a

×σ

m

×i

×n

× ] → [

∗σ

a

×σ

m

×i

×σ

n

×a

× ] →a

× [σ

m

×i

×σ

n

×a

× ]

This produces a contrast between amin-a:n /chi (‘look’, nonfinite) and thecorresponding continuative form a-mina-mina-t-a:n /chi, with t epenthe-sized to join the V-final stem and the V-initial suffix.

It is not at all clear how Raimy can deal with phenomena ofthis kind. Although segmental phonology can be extended to R-representations, the prosodic phonology of R-representations is obscureat best. The R-representation analog of (375) is:

(377) # a m i n %

The backpointer indicates actual precedence, not simply an instructionto repeat certain material. What then is the syllable structure of (377)?Does the “final” n syllabify as the onset of the “initial” a? Raimy doesnot have anything to say about the syllabification of R-representations.

In Chaha, for example, Raimy proposes the structure on the rightbelow, but does not have anything to say about how it is derived fromthe consonantal root.

(378) # k t %?−→ # k 1 t %

@

Since templatic syllable structure conditions are generally assumed tobe involved in inserting the vowels into the consonantal root, there isa gap in Raimy’s analysis. Some new theory of template satisfaction isrequired, since it is implausible that there is a theory of R-representationsyllabification adequate to the task.

C.4 Concatenative versus nonconcatenative morphology

In the R-representation theory, there is no difference in kind betweenprefixation and suffixation, on the one hand, and infixation, on the other.

Raimy’s theory of R-representations 217

In Distributed Reduplication, there is a difference in kind. Prefixationand infixation are concatenative, while infixation involves both concate-nation and t-juncture insertion. Although there are no crucial exampleswhich give evidence one way or the other, the difference is noteworthyenough that is worth reviewing.

Consider, for example, the combination of the stem taki and an affixwith the exponent ga in Raimy’s theory. Prefixation, forming gataki, isshown in (379a); suffixation, forming takiga, is shown in (379b); andinfixation before the final syllable, forming tagaki, is shown in (379c).The three R-representations that result are:

(379) a. # t a k i % g a

b. # t a k i % g a

c. # t a k i % g a

Given the fact that there is no difference in kind in his theorybetween infixation, suffixation, and prefixation, Raimy must accountfor the relative rarity of infixation. He attributes it to the greatercomplexity in specifying the source and target positions in the stem forthe attachment of links between the exponent of the affix and the stem.

The corresponding DR structures are:

(380) a.

g

×a

×t

×a

×k

×i

×

b.

t

×a

×k

×i

×g

×a

×

c.

t

×a

× [〈k

×i

×〉〈g

×a

×〉] or [〈g

×a

×〉〈t

×a

×〉]k

×i

×

218 Appendix C

A prefix or a suffix is joined to the stem by concatenation. An infix isan underlying concatenated prefix or suffix, coupled with the insertionof t-junctures whose ultimate effect is to prepose a suffix or postpose aprefix into the stem.

Notes

Chapter 11. I call it “copying-like” because it combines copying with translation. Bases aremapped to complementary bases.2. The translation of a phonological representation/score into a physical performanceis not direct. A phonological representation is first translated into a phonetic score,which can undergo further processing before it in turn is translated into a physicalsound signal.

Chapter 21. Transcription generally produces crossed structures, but there are circumstances inwhich it does not. If an initial consonant is reduplicated, for example, a geminate isproduced, with no crossing. This occurs in Imdlawn Tashlhiyt Berber:

[

σ

f

× ]

r

×σ

n

× → ×σ

f

×

r

×σ

n

×

2. See Frampton and Gutmann (1999, 2000) for a development of essentially the samepoint of view in syntax. We argue against Chomsky’s early Minimalist Program idea ofmassive overgeneration and filtering by economy conditions.3. Certainly, one of the more bizarre aspects of OT’s conception of the languagefaculty is that a complex calculation is needed to move dog out of the lexicon andinto a phonological computation.4. This oversimplifies the extensive discussion in Halle (1995). In some cases, onlymarked features block feature spreading.

220 Notes

Chapter 31. It is possible to entertain the idea of a more complex theory, in which triggeredrules are interspersed among the cyclic rules, but this possibility is not considered here.2. Zoll uses this fact to demonstrate that McCarthy and Prince’s (1995) Correspon-dence Theory analysis of the absence of root vowel reduction in intensive reduplicationis flawed. While I believe that Zoll’s counteranalysis is needlessly stipulative, thisdoes not detract from the insightfulness of her criticism of the Correspondence Theoryaccount.3. Associating the timing slots generated by transcribing a reduplicant with a mor-pheme has been taken to be a repair operation. It was proposed above that theassociation is with the morpheme which initiates the cycle in which the transcriptionoccurs. It could be that there is language or even morpheme particular variation on howthe repair is carried out. If Klamath transcribes both the intensive and the distributivereduplicants to the left, but takes the reduplicant generated by intensive reduplicationto be a root extension, rather than derived affixation. Repair would proceed as in (i).If this is so, root vowel reduction would not apply because the initial vowel of the rootremains the initial vowel of the word after reduplication.

(i)

× × ×

W

×

i

×

c

×[[WIC]]INT →

× × ×

W

×

i

×

c

×[[WIC]]INT

4. It may be the case that templates can be shown to be only descriptive and can beremoved altogether from the ontology of phonology.

Chapter 41. This is slightly inaccurate, since I assume that t-junctures are timing tier junc-tures and assume that lexical entries are not autosegmental. More accurately, lexicalentries contain symbols which translate into t-junctures (in an obvious way) whenthe lexical entry is translated into an autosegmental representation upon entering themorphophonological computation.2. Healy doesn’t give any examples with initial consonant clusters, so it is possiblethat the stem adjustment rule is ∅ → ]

/Left × .

3. The frequentative stems are simplified somewhat. The length the second vowel isinconsistently long in Wilbur’s data.

Notes 221

Chapter 51. McCarthy and Prince (2001:126) note the reduplication pattern in Nuu-chah-nulth(which they call Nootka) and a similar one in Nitinaht, a related language, based onStonham 1990 and Shaw 1992. They account for patterns by assuming that a “heavysyllable template” is imposed, but that penalties against syllables with codas outrankthe penalties against nonmaximal copying. This account is implausible, even on OTterms, since the possibility of a penalty on the presence of a coda outranking penaltiesagainst nonmaximal copying can easily produce unattested reduplication patterns. Thefollowing unlikely variant of total reduplication results if NoCoda is elevated over theconstraint which penalizes nonmaximal copying.

(i) a. da ti kun → da ti ku - da ti kunb. da tin ku → da ti - da tin kuc. dan ti ku → da - dan ti ku

Max(B,R) forces as much material to be reduplicated as possible, but is stoppedby the more prominent NoCoda. NoCoda is, in turn, outranked by the usual arrayof correspondence constraints which ensure that BR-correspondence imitates faithfulcopying. As far as I know, such a pattern is completely unattested.2. I assume here that it is sufficient that left transcription to be simpler for manystems (i.e. those which begin with a syllable with an onset) for the grammar to chooseuniform left transcription for all roots. It would be a complication for the direction oftranscription to depend on the particular stem.3. It is worth noting that a CVC is copied to the left even if the inital vowel of the stemis long, so ka:ma: → kamka:ma:/kakka:ma: . The source is [k〈a〉am ]maa. An accountis given in Chapter 6.

Chapter 61. Blevins analyzes the transition from Mokilese-A to Mokilese-B as a diachronicinstance of “the emergence of the unmarked.” In particular, the elevation of NoCodain the penalty function hierarchy. This account is implausible for the reasons given inChapter 5 fn. 1.2. Davis (2002) avoids an appeal to reduplication specificity by using SympathyTheory and proposals concerning the moraicity of geminates.3. The discussion of syllable weight in Hayes (1995:270) is useful. He suggeststhat the notion weight should be restricted to mora count, and that it is more usefulto analyze what I call secondary weight distinctions as prominence distinctions. Thatmay be correct.4. Lardil roots are often written with a final laminal consonant T. This consonant issupposed to surface when a V-initial suffix is concatenated, otherwise to delete becausethe syllable well-formedness conditions in Lardil do not allow this consonant as a coda.But it could be that roots are minimized to reduce redundancy and that T is added byrule to eliminate vowel hiatus. I assume that this is the case.5. Haugen (2003:79) takes Yaqui to demonstrate that Moravcik was simply wrong.This is an overly harsh judgment. There remains a kernal of Moravcik’s insight that anadequate theory of reduplication must account for. Syllable copy is still an unattestedform of reduplication, which still demands an account. Haugen is incorrect in claiming

222 Notes

that “Yaqui illustrates a pattern of reduplication which has been repeatedly claimed notto exist: the pattern of so-called ‘syllable-copy’.” This is not entirely true; if the initialsyllable is bimoraic, it is not copied.

Chapter 71. It is supplemented by very helpful personal communication with Galen Sibanda.2. Another approach might be to relate the occurrence of -ile to the fact that multiplyassociated phonemes may be present at the point that PERF combines with the stem.3. Inkelas and Zoll (2004) claim that -esy is monomorphemic, but Mutaka (2000)provides strong evidence that -esy is in fact bimorphemic -es-y. He gives examples inwhich the es of -esy reduplicates independently.

(i) root causative reduplicationot mo-tu-a-ot-es-es-y-a ‘we caused someone to light’sw eri-sw-es-es-y-a ‘to cause to grind’hi eri-hi-is-is-y-a ‘to cause to burn’kohol eri-kohol-es-es-y-a ‘to cause someone to cough’

The meaning is emphasis of causation rather than the unintensive meaning of bisyllabicreduplication. The straightforward explanation is that -esy is actually bimorphemic-es-y and causative reduplication targets the exponent of its inner morpheme. If -esywere monomorphemic, the MIC would prevent partial reduplication of the suffix.4. Inkelas and Zoll (2004) claim (p. 12) that the short causative is not produced bytruncation in the phonology because “such an analysis would predict the Short causativeto be possible whenever the Long causative is possible, but it is not.” Implicit is theassumption that phonological rules cannot be morphologically conditioned. But it isimpossible to get very far in understanding morphophonology under this assumption.Inkelas and Zoll themselves only manage to get three pages before they claim on p. 15that the fact that causative -y floats to the right past extension suffixes is due to a ruleoperating in the phonology. This rule must know that the phoneme y is the exponentof the causative morpheme. The phonological rule is conditioned by the morphologicalstructure in which it operates, just as short causative formation by truncation is sensitiveto whether or not the root is a SC-root. (In fact, we claimed that es-deletion was areadjustment rule, not a rule of the phonology. But the point remains.)5. See footnote 2 (p. 147) in Spring’s thesis for some discussion of the instability inher informant’s reduplication rules.6. Needless to say, this carries some risk. If the hypothesized language turns out not toa possible human language, the fact that DR can model it is evidence of the inadequacyof DR.7. In DR, melody copying is incoherent, because copying requires linear order andthe melodic tiers have no intrinsic order.8. For roots of the first type, the difference between VC-reduplication and CV-reduplication is minimal: (C1)VC2VC2C3V(C4) versus (C1)VC3VC2C3V(C4). This isunlikely to be an accident. Unfortunately, without a theory of how verb irregularities ofthis kind are learned (the irregularity being affix selection), it is impossible to transformthis observation into a theoretical proposal. Roots of the second type might reflectan earlier stage in the language in which the roots were monosyllabic, ending in aconsonant cluster. Later, after i or 1 broke up the consonant cluster, irregular assignment

Notes 223

to the VC-reduplicating class might have been made in order to minimize the change inthe reduplicated forms.9. If true, this would be related to the existence of a sizable number of rootsof the form (C)aCiC or (C)aC1C which undergo VC-reduplication rather than CV-reduplication, which is the expected variety of reduplication for polysyllabic roots.10. The vowel inventory is {i, 1, u, o, a}. There are nine V1V2 sequences which occurin Tohono O’odham which are not Type F: {ia, iu, 1a, 1i, 1u, ui, oi, ai, au}.11. Some combinations of class membership are unlearnable under almost any plau-sible assumptions about the role that formal simplicity plays in learning. There are noverbs which are both shortening and lengthening, for example. But it would be incorrectto state this as part of the grammar. No plausible learning mechanism will assign a rootto both of these classes simultaneously. 10 out of the 16 possibilities are learnable and8 of these are attested. As far as I know, there are no verbs which are lengthening andexceptionally truncating, although there are exceptionally truncating verbs with a longinitial vowel (i.e. da:k → da:-d-k); and the only verbs which are both exceptionallytruncating and degeminating are also lengthening (i.e. sikol→ siPi-s-kol).12. Alongside the preference for falling diphthongs is a secondary preference fordiphthongs with smaller sonority differences, so au is preferred to an, for example. Ifthe preference for falling diphthongs is combined with the preference for low sonoritydifference diphthongs, with the first criterion given priority, the following markednessscale on potential nuclear diphthongs is produced.

(381) Diphthong Preference Hierarchy (DPH):

{ai, au} > {ar, al} > {am, an} > {ia, ua} > {ra, la} > {na,ma}The DPH precisely reproduces the relative likelihood that a root with a sequence

of sonorants syllabifies with a long nucleus. The following table uses the figures whichSteriade gives. The “possible column” gives the count of the number of roots whosemost preferred diphthong is of the indicated kind. The “actual column” gives the countof the number of these that have a zero grade reduction.

(382) possible actual percentage{ai, au} 157 156 99%{ar, al} 80 71 89%{an, am} 48 36 75%{ia, ua} 28 17 61%{ra, la} 29 5 17%{na,ma} 6 0 0%

13. The “modification”, it should go without saying, is not under the conscious controlof the language speaker, but under the automatic control of the language learningapparatus which works in the complex environment which results in language changeover time.14. There is an oversimplification here. The vowel quality that results is always a.Various accounts are possible.

224 Notes

Appendix B1. Lowenstamm (1996) also rejects the idea that metathesis is involved and developsan analysis along those lines.

Appendix C1. The criticism is directed at the intrusion of Correspondence Theory into the analy-sis, not Optimality Theory per se. The early OT theory of Prince and Smolensky (1993)views the output as a modification of the input. In that theory, the input is actuallyexplicitly embedded in the output. That theory, however, proved to be inadequate.2. Although segmental phonology does extend successfully to R-representation, itdoes have one peculiar quirk, not mentioned by Raimy. Although it is easy to determinethe successor to a particular phoneme, or more generally the “right environment” of thephoneme, it is much more difficult to determine the “left environment.” The predecessorrelation is not coded directly in the representation. Determining the predecessorrequires a search starting at the # mark. This asymmetry in the structure is questionable.One might introduce a more complex structure, a double linked list, which codes boththe successor and predecessor relations directly, but Raimy (p.c.) rejects this.

References

Anderson, Stephen. 1992. A-Morphous Morphology. Cambridge: Cambridge Univer-sity Presss.Applegate, Richard. 1972. Ineseno Chumash Grammar. Doctoral dissertation, Univer-sity of California, Berkeley.Aronoff, Mark. 1988. Head Operations and Strata in Reduplication: A Linear Treat-ment. In Yearbook of Morphology 1, eds. Booij, Geert and Jaap van Marle, 1–16.Dordrecht: Kluwer.Austin, Peter. 1981. A Grammar of Diyari, South Australia. Cambridge, CambridgeUniversity Press.Bagemihl, Bruce. 1991. Syllable structure in Bella Coola. Linguistic Inquiry 22.4:589–646.Banksira, Degif Petros. 2000. Sound Mutations: The morphophonology of Chaha.Philadelphia/Amsterdam, John Benjamins.Barker, M. A. R. 1964. Klamath Grammar. Berkeley and Los Angleles: University ofCalifornia Press.Blevins, Juliette. 1996. Mokilese reduplication. Linguistic Inquiry 27.3: 523–530.Boersma, Paul, and Bruce Hayes. 2001. Empirical Tests of the Gradual LearningAlgorithm. Linguistic Inquiry 32.1: 45–86.Booker, Karen M. 1980. Comparative Muskogean: Aspects of Proto-Muskogean VerbMorphology. Ph. D. Disseration University of Kansas, Lawrence.Bromberger, Sylvain and Morris Halle. 1988. Conceptual issues in morphology. Ms.,MIT.Broselow, Ellen, and John McCarthy. 1983 A theory of internal reduplication. TheLinguistic Review 3: 25–98.Calabrese, Andrea. 1995. Linguistic Inquiry 26.3: 373–463.Chomsky, Noam, and Morris Halle. 1968. The Sound Patterns of English. New York:Harper and Row.Chung, Chin Wan. 1999. Reduplication in Korean. Doctoral dissertation. Universityof Indiana.

226 References

Crowley, Terry. 1998. An Erromangan (Sye) Grammar. Oceanic Linguistics SpecialPublication No. 27. Honolulu: University of Hawai‘i Press.Davis, Stuart. 2002. The Interaction of Nasal Substitution and Reduplication inPonapean. Presentation, Graz Reduplication Conference.Dell, Francois and Mohamed Elmedlaoui. 1988. Syllabic consonants in Berber: Somenew evidence. Journal of African Languages and Linguistics, 10: 1–17.Demers, R., F. Escalante and E. Jelinek. 1999. Prominence in Yaqui words. Interna-tional Journal of American Linguistics, 65:1, 40-55.Diffloth, Gerard, 1976. Expressives in Semai. In (eds.) Jenner, Thomson, and Starosta,Austroasiatic Studies, Vol. II. Honolulu: Univeristy Press of Hawaii.Dixon, Robert M. W. 1977. A Grammar of Yidiny. Cambridge, Cambridge UniversityPress.Eisner, Jason. 1997. Efficient generation in primitive Optimality Theory. Proceedingsof the 35th Annual Meeting of the ACL (jointly with the 8th Conference of the EACL),Madrid.Emeneau, M. B. 1955. Kolami: A Dravidian Language. University of CaliforniaPulications in Linguistics, 12. Berkeley, University of California Press.Fitzgerald, Coleen. 2000. Vowel hiatus and faithfulness in Tohono O’odham redupli-cation. Linguistic Inquiry 31.4: 713–722.Frampton, John. 2001. SPE Extensions: conditions on representations and defectdriven rules. To appear in M. Halle and B. Vaux (eds), Phonological Perspectives:Rules and constraints in contemporary phonological theory.Frampton, John. 2002. What kind of a thing is a language faculty? GLOT Interna-tional 6.1.Frampton, John and Sam Gutmann. 1999. Cyclic Computation, a ComputationallyEfficient Minimalist Syntax. Syntax 2.1:1–27.Gafos, Adamantios. 1999. On the proper characterization of ‘nonconcatenative’languages. Ms, Johns Hopkins.Gafos, Adamantios. 2001. Deriving Greenberg’s Asymmetry in Arabic. Ms. New YorkUniversity.Goldsmith, John. 1976. An overview of autosegmental phonology. Linguistic Analysis2: 23–68.Goldsmith, John. 1990. Autosegmental and Metrical Phonology. Oxford: Blackwell.Hale, Ken and Abanel Lacayo Blanco. 1989. Diccionario elemental del Ulwa (Sumumeridional). Cambridge, MA: Center for Cognitive Science. MIT.Halle, Morris. 1990. An approach to morphology. Proceedings of NELS 20, 150–84.Amherst: GLSA, University of Massachusets.Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry26.1: 1–46.Halle, Morris. 1997. On stress and accent in Indo-European. Language 73.2:275–313.Halle, Morris and Alec Marantz. 1993. Distributed morphology and the pieces ofinflection. In K. Hale and S. J. Keyser (eds), The View from Building 20: Essays inLinguistics in Honor of Sylvain Bromberger (pp. 111–176). Cambridge, MA: MITPress.Halle, Morris, and William Idsardi. 1995. In (ed) Goldsmith, John, The Handbook ofPhonological Theory. Oxford: Blackwell.Halle, Morris and Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, MA:MIT Press.

References 227

Harrison, Sheldon P. 1973. Reduplication in Micronesian Languages. Oceanic Lin-guistics 12: 407–454.Harrison, Sheldon P. 1976. Mokilese Reference Grammar. Honolulu: University ofHawaii Press.Haas, Mary R. 1977. Nasals and Nasalization in Creek. Proceeding of the Third AnnualMeeting of the Berekeley Linguistics Society, 194–203. Berkeley: Berkeley LinguisticsSociety, University of California, Berkeley.Haugen, Jason. 2003. Allomorphy in Yaqui Reduplication. In Studies in Uto-Aztecan,eds. Luis Barragan and Haugen, Jason, 75–103. MIT Working Papers on Endangeredand Less Familiar Languages 5.Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62: 321–351.Hayes, Bruce and May Abad. 1989. Reduplication and syllabification in Ilokano.Lingua 77: 331–374.Healy, P. M. 1960. An Agta Grammar. Manila: SIL.Healy, Alan, Ambrose Isoroembo, and Martin Chittleborough. 1969. Papers in NewGuinea Linguistics 9: 33–64.Hudson, Grover. 1995. In (ed) Goldsmith, John, The Handbook of PhonologicalTheory. Oxford: Blackwell.Hyman, Larry and Ngessimo Mutaka. 1990. Syllables and morpheme integrity inKinande reduplication. Phonology 7: 73–119.Hyman, Larry, Sharon Inkelas and Galen Sibanda. 1999. Morphosyntactic correspon-dence in Bantu reduplication. ms. University of California, Berkeley.Inkelas, Sharon and Cheryl Zoll. 2004. Reduplication as morphological doubling.Manuscript, UC Berkeley and MIT.Jacobsen, William. 1964. A grammar of the Washo language. Doctoral dissertation,University of California at Berkeley.Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Oxford: Blackwell.Kenstowicz, Michael and Degif Petros Banksira, 1999. Reduplicative identity in Chaha.Linguistic Inquiry 30.4: 573–585.Kim, Eunsook. 2002. Presentation, NELS 2002.Kiparsky, Paul. 2003. Paradigms and Opacity. Unpublished book manuscript. StanfordUniversity.Krause, S. R. 1980. Topics in Chukchee Phonology and Morphology. Doctoraldissertation, University of Illinois, Urbana-Champaign, Illinois.Levin, Juliette. 1985. A metrical theory of syllabicity. Doctoral Dissertation, MIT.Lieber, Rochelle. 1992. Deconstructing Morphology. Chicago: University of ChicagoPress.Lowenstamm, Jean. 1996. CV as the only syllable type. In J. Durand and B. Laks, eds.,Current Trends in Phonology: Models and Methods, University of Salford Publications.Martin, Jack B., and Margaret McKane Mauldin. 2000. A Dictionary of Creek/Muskogee, with notes on the Florida and Oklahoma Seminole dialects of Creek. Lin-coln/London: University of Nebraska Press.Marantz, Alec. 1982. Re reduplication. Linguistic Inquiry 13.3: 435–482.McCarthy, John. 1984. Ponapean reduplication. Ms., University of Texas.McCarthy, John. 1986. OCP effects: Gemination and antigemination. LinguisticInquiry 17.2: 207–263.McCarthy, John and Alan Prince. 1986. Prosodic morphology. Ms., University ofMassachusetts and Brandeis. Reprinted in J. Goldsmith (ed), 1999. PhononologicalTheory: The Essential Readings. Oxford: Blackwell.

228 References

McCarthy, John and Alan Prince. 1990. Foot and word in prosodic morphology: TheArabic broken plural. Natural Language and Linguistic Theory, 8: 209–283.McCarthy, John and Alan Prince. 1995. Prosodic morphology. In J. Goldsmith (ed),The Handbook of Phonological Theory, (pp. 318–366). Oxford, Oxford UniversityPress.McCarthy, John and Alan Prince. 1998. Faithfulness and reduplicative identity. In R.Kager, H. van der Hulst, and W. Zonneverld, eds., The Prosody-Morphology Interface.Cambridge University Press, Cambridge, U. K.McCarthy, John and Alan Prince. 2001.Faithfulness and reduplicative identity. Prosodic Morphology: Constraint Interactionand Satisfaction. Unpublished book manuscript, University of Massachusetts andRutgers University.Merlan, F. 1982. Mangarayi. Lingua Descriptive Series 4: 1–242.Moravcik, Edith. 1978. Reduplicative constructions. In (ed) J. Greenberg, Universalsof human language, vol. 3: Word Structure. Stanford: Stanford University Press, 297-334.Mutaka, Ngessimo. 2000. Interaction of imbrication and reduplication in Kinande.Journal of West African Languages XVIII.1 (2000/2001): 33–47.Nash, David. 1979. Yidiny stress: A metrical account. CUNY Forum 7/8: 112–30.Nash, David. 1980. Topics in Warlpiri Grammar. Doctoral Dissertation, MIT.Newman, Paul. 2000. The Hausa Language. New Haven: Yale University Press.Odden, David and Mary Odden. 1985. Ordered reduplication in Kıhehe. LinguisticInquiry 16.3: 497–503.Onn, Farid. 1976. Aspects of Malay phonology and morphology: A generativeapproach. Doctoral dissertation. University of Illinois, Urbana.Payne, David. 1981. The phonology and morphology of Axininca Campa. TheSummer Institute of Linguistics and the University of Texas at Arlington.Pensalfini, Rob and Gavan Breen. 1999. Arrente: A language with no syllable onsets.Linguistic Inquiry 30.1: 1–25.Prince, Alan and Paul Smolensky. 1993. Optimality theory: Constraint interaction ingenerative grammar. Ms., Rutgers University and University of Colorado, Boulder.Raimy, Eric. 2000 The morphology and morphophonology of reduplication. Berlin:Mouton de Gruyter.Raimy, Eric. 2000. Deconstructing prosodic morphology. Talk given at ESCOL.Rehg, Kenneth, and Damien Sohl. 1981. Ponapean Reference Grammar. Honolulu:University of Hawaii Press.Riggle, Jason. 2004. Nonlocal Reduplication. Proceedings of NELS 34.Rose, Sharon. 2001. Triple Take: Tigre and the case of internal duplication. To appearin R. Hayward, J. Ouhalla and D. Perret, Studies in Afroasiatic grammar. Amsterdamand Philadelphia: John Benjamins.Sagey, Elizabeth. 1988. On the ill-formedness of crossing association lines. LinguisticInquiry 19: 109–118.Sapir, Edward. 1922. The Takelma Language of Southwestern Oregon. In Boaz, Franz,ed., Handbook of American Indian Laanguages, Part II.Saxton, Dean, Lucille Saxton, and Susie Enos. 1983. English-Papago/Pima Dictionary,2nd edition. Tuscon: University of Arizona Press.Schein, Barry and Donca Steriade. 1986. On Geminates, Linguistic Inquiry 17.4: 691–744.

References 229

Shaw, Pat. 1992. Templatic Evidence of the Syllable Nucleus. Paper presented atNELS 23, University of Ottawa.Sommerstein, Alan. 1974. On phonotactically motivated rules. Journal of Linguistics10: 71–94.Spring, Cari. 1990. Implications of Axininca Campa for prosodic morphology andreduplication. Doctoral Dissertation, University of Arizona.Steriade, Donca. 1988. Reduplication and syllable transfer in Sanskrit and elsewhere.Phonology 5: 73–155.Stevens, A. 1968. Madurese phonology and morphology. New Haven: AmericanOriental Society.Stonham, 1990. Current issues in morphological theory. Doctoral dissertation, StanfordUniversity.Urbanczyk, Suzanne. 1996. Patterns of reduplication in Lushootseed. Ph.D. disserta-tion. University of Massachusetts.Urbanczyk, Suzanne. 2000. The bases of double reduplication. WCCFL 19 Proceed-ings. ed. Billerey and Lillehaugen.Wiesemann, U. 1972. Die phonologische und grammatische Struktur der KaingangSprache. The Hague: Mouton.Wilbur, Ronnie Bring. 1973. The Phonology of Reduplication. Bloomington: IndianaUniversity Linguistics Club.Wilkinson, Karina. 1988. Prosodic structure and Lardil phonology. Linguistic Inquiry19: 25–34.Winter, Werner. 1970. Reduplication in Washo: A restatement. International Journalof American Linguistics 36.3: 190–198.Yu, Alan. 2001. Washo internal duplication revisited. ms. University of California atBerkeley.Zepeda, Ofelia. 1983. A Papago grammar. Tuscon: University of Arizona Press.Zoll, Sharon. 2002. Vowel reduction and reduplication in Klamath. Linguistic Inquiry33.3: 520–527.

Glossary of Abbreviationsand Rules

Rules pageJuncture insertion rules:×-rule ∅ → ] /Left × 77×∗-rule ∅ → [ /Left × 77C∗ν-rule ∅ → /Left ν (narrow interpretation) 77C∗V-rule ∅ → /Left ν (broad interpretation) 77DC Default Closure 38

Transcription rules:Trscr Transcription 7Suboperations of Transcription

Init 54Reduce 54CopyShift 54Combine 62CopyDelete 63

Prosodic adjustment rules:PrAdj Prosodic Adjustment 88FCVL First Conjunct Vowel Lengthening 65FCVE First Conjunct Vowel Epenthesis 64

NCCR NCC Repair 29

Glossary of Abbreviations and Rules 231

T&R Trscr followed by NCCR (composite) 39Readj readjustment 33

AbbreviationsDDR defect driven rule 94DDRS defect driven rule schema 94DR Distributed Reduplication 2ISDC Initial Syllable Duplicant Condition (Yaqui) 111NCC No Crossing Constraint 7PM Prosodic Morphology 5RC Retraction Condition 203t-juncture transcription juncture 2WFSS well-formed syllable structure 115

Subject Index

The subject index is preliminary and largely confined to the first six chapters of themonograph. It needs a final editing, but is included because it may be useful even in itspresent state.

×∗-rule 77, 78, 89, 108, 141affix allomorphy 130allomorphy 33, 35, 98, 145, 150applicativized passive 128atemplatic reduplication 5augmentation to minimal word 131augmentation to polysyllabicity 144autosegmental representation 200autosegmental tier 14Bantu verbal morphology 127base-reduplicant conditions 7bimoraic duplicant 172bimoraic foot 80bimoraic minimal word 103bimoraicity desideratum 109, 113bimoraicity 4, 108, 124, 144bimorphemic verb forms 136bisyllabic duplicant 91bisyllabic foot 5, 59, 78bisyllabic minimal word condition 131bisyllabicity 4, 124broad interpretation 74C∗ν-reduplication 75, 77, 170C∗ν-rule 110, 112, 113C∗ν juncture insertion 172C-final desideratum 109

C-finality 108, 111, 112, 119, 178Classical Arabic Metathesis 189Classical Arabic 189complementarity 104complex duplicant 70consecutive syllable reduplication 117constraint on possible grammars 15constraints on feature spreading 16constraints on repair rules 120cophonology 183correspondence relations 214Correspondence Theory 7, 188, 197, 212counter-cyclicity 182crossed association lines 12crossed structure 8, 11, 16, 129crossing violation 149, 188crossing violations 13, 85C∗V-reduplication 45, 77C∗V-rule 111, 112C∗V juncture insertion 38C∗V root reduplication 45cycle 31, 43, 99cyclic binary footing 151cyclic morphology 123cyclic rule block 33cyclicity 175

Subject Index 233

Default Closure 38, 141, 165default leftmost rule application 38defect driven rule scheme 205defect driven rule 94derivational constraint 8, 13desideratum driven rule schema 94Diphthong Preference Hierarchy 223direction of transcription 11, 41, 42, 73dissimilation 171Distributed Morphology 30Distributed Reduplication 2domain designation 121domain of a juncture insertion rule 57,

59domain of a reduplicative affix 77, 121domain of an affix 4domain selection 5double reduplication 71, 98, 99, 134,

175double stem reduplication 138duplicant augmentation 91duplicant syllabification 90duplication juncture 53edge truncation 173embedded duplication juncture 164embedded metathesis 71embedded t-juncture 57, 93embedded t-junctures 73embedded word boundary 58English past tense morphology 33epenthetic vocalism 189exponent of a lexical item 31, 57falling diphthong 169feature spreading 8final syllable prefixation 115final syllable suffixation 116first conjunct bisyllabicity 123first conjunct syllable well-formedness

147First Conjunct Vowel Epenthesis 64,

124, 172, 177, 178First Conjunct Vowel Lengthening 65,

93, 101, 109, 124first conjunct 64, 95first-conjunct deletion 137fission 14full grade 166

geminate inalterability 8, 16, 17, 41,129, 131, 187

geminate interpretability 190heavy syllable reduplication 92, 104heavy syllable suffixation 114hiatus avoidance mechanism 149hiatus avoidance 158Identity Condition 6Identity Effects 3, 7, 10imbrication 130immediate repair 15infixation 9, 65inherently reduplicated root 188initial consonant reduplication 177initial onsetless syllable 82Initial Syllable Duplicant Condition 111initial timing slot reduplication 77interface condition 15, 200internal reduplication 145, 182inventory of juncture insertion rules 73juncture insertion rule 37juncture insertion 5, 36late lexical insertion 30late syllabification 189leading edge truncate 65leading edge truncation 55, 146left-edge truncate 95lexical insertion operation 31lexical insertion 49lexical marking 169lexical pointers 213lexically specified irregularity 154linear order 31linearization algorithm 212locus of affixation 77long-distance geminate 7, 10, 186lookahead 15m-structures 30medial truncation 137, 173medial truncation 62melody copying 145metathesis 9, 65, 204MIC 133monosyllabic root 60monosyllabic roots 98, 134monosyllabic stem 66, 71, 101, 150monosyllabic stems 145moraic consonants 193

234 Subject Index

Moravcik’s Generalization 4, 112, 117morpheme association 40morpheme boundaries 117, 120morpheme boundary effect 90morpheme boundary 90, 143morpheme contiguity 41Morpheme Integrity Constraint 133morpheme metathesis 127morpheme realization rule 131morpheme 30morphemic association 89morphological doubling 183morphological hierarchy 141morphological pointers 213morphologically conditioned rule 5, 40morphologically conditioned 34morphophonology 30motor instructions 200, 201multiple association 187multiple linking 16multiply associated phoneme 83narrow interpretation 74NCC repair 84, 128, 150, 171, 174, 179NCC 12, 200negative prosodic circumscription 67nested duplicant transcription 98nested duplicants 60, 165, 192non-reduplicating roots 135nonconcatenative morphology 9nonlinear timing tier 7, 210nonmoraic coda consonants 111nontrivial duplicant 75nuclear skeleton 73nuclear structure 168null exponent 31null lexical item 31occurence of a phoneme 17occurence 41occurrence 149, 186onset incorporation 88, 140onset-coda asymmetry 117onsetless syllable 91onsetless syllables 58opacity 100opaque processes 198optimal linearization 214output condition on phonology 200output condition 14, 15

output interface 14overapplication 41, 128, 130, 181, 185,

190overapply 17overweight roots 135partial morpheme reduplication 133passivized applicative 128permutation 65persistence of foot structure 151phoneme fission 7, 40, 84phoneme tiers, unordered 202phonology-phonetics interface 15, 200phonotactically driven rule 94phonotactics 158polymorphemic stem 117positive prosodic circumscription 77possible condition on repair 120possible juncture insertion rule 75post-transcription modification 72post-transcription structure 95postcyclic rule block 33postcyclic rules 31pre-transcription structure 95pretranscription modification 72pretranscription representation 146pretranscription source 93pretranscription structure 55primacy of weight desiderata 112prosodic adjustment rule 91, 108prosodic adjustment schema 120prosodic adjustment strategies 173prosodic adjustment to bimoraicity 96prosodic adjustment 5, 69Prosodic Adjustment 135prosodic adjustment 88prosodic constituency 114prosodic desiderata 88, 119, 135prosodic desideratum 94, 141Prosodic Morphology 4, 58, 67, 77, 120,

185, 188prosodic target 88prosodic weight desideratum 143pseudo-roots 136R-representations 210readjustment rule 136readjustment rules 33recoverability of deletion 86reduplicant truncation 55, 137

Subject Index 235

reduplicant 64reduplicant-remnant boundary 84, 106reduplicated monosyllable 151reduplicative prefix 58, 67reduplicative suffix 58remnant 64repair 13, 84, 90Resyllabification Condition 168Retractability Condition 203retraction 201right-edge truncate 95rising nucleus 169root vowel reduction in Klamath 39root-template association 13rule design 177Sanskrit, intensive reduplication 165,

172Sanskrit, low vowel syncope 166Sanskrit, perfective reduplication 165,

170second conjunct 64secondary prosodic desiderata 5, 108secondary prosodic desideratum 111Semitic morphology 13Shortcut Repair 84, 106, 165simple duplicant 69Sino-Korean words in Korean 25spirantization 17split phoneme tiers 14stem ablaut 130, 148stem augmentation 142, 160stem syllable structure 90stem 37string copy 53subword 9, 59surface prefix 92syllable copy reduplication 3syllable epenthesis 131, 143syllable nucleus 73syllable well-formedness conditions

119, 143t-juncture 2, 54, 120

target prosodic desideratum 5target prosodic shape 124tasty-shmasty reduplication 58, 145temporal simultaneity 200theory of word formation 30Tier Conflation 14timing of NCC repair 131total reduplication 6, 42, 77total root reduplication 48, 125total stem reduplication 9, 138trailing edge truncate 65trailing edge truncation 63transcription factor 1transcription juncture 2, 54transcription of genes 1transcription 7, 53Truncated Trailing Edge Epenthesis 64truncation juncture 54truncation 9underapplication 17, 41, 187underlying bare timing slots 148unfooted material 81unpaired duplication juncture 75variation 72vocabulary choice 30vowel hiatus 85, 102vowel syncope 147well-formed duplicant syllable structure

90, 141well-formed syllable structure desidera-

tum 146well-formed syllable structure 115, 132whole morpheme effect 133Wilbur Effect 27Wilbur’s Identity Condition 3Wilbur’s Identity Constraint 188Yaqui geminating reduplication 69Yaqui light syllable reduplication 69,

110, 112zero grade 166

Language Index

Agta 9, 57, 92, 97, 100, 101Arrernte 58, 65, 150, 153Asheninca Campa 9, 91, 122, 138, 215Bella Coola 81Biblical Hebrew 17Chaha 9, 57, 122, 185, 206, 216Choctaw 201Choktaw 66Chukchee 55, 118Chumash 181Classical Arabic 206Creek 71Diyari 78Endegen 193English 33, 58, 66Erromangan 46Ezha 193German 70Hausa 86, 165Ilocano 9, 35, 92, 97, 100Imdlawn Tashlhiyt Berber 83Kaingang 116, 118Kıhehe 88, 140Kinande 9, 62, 90, 120, 122, 133Klamath 39Kolami 84, 165Korean 25, 117, 118

Lardil 108, 111Levantine Arabic 160, 163Lushootseed 9, 175Madurese 55, 115, 116, 118Malay 6, 23, 122, 198, 211Manam 80Mangarayi 75, 147, 150Mokilese 9, 64, 71, 92, 93, 101, 118,

119, 124, 134, 172, 178Ndebele 9, 64, 122, 123, 133, 134, 135Nuu-chah-nulth 49, 108, 109, 111, 113Orokaiva 82Ponapean 9, 87, 92, 97, 103, 111, 112,

120, 165Sanskrit 9, 62, 74, 113, 122, 165Semai 162Tagalog 44Takelma 61Temiar 122, 160, 164Tigre 191, 194Tohono O’odham 74, 113, 122, 154Ulwa 79Washo 9, 145Western Aranda 82Yaqui 68, 90, 108, 110, 112, 120Yidiny 4, 78Yoruba 56, 57, 202